ZETA FUNCTIONS OF

REDUCTIVE GROUPS
AND THEIR ZEROS

10723_9789813231528_TP.indd 1 23/1/18 9:21 AM

 b2530   International Strategic Relations and China’s National Security: World at the Crossroads

This page intentionally left blank

b2530_FM.indd 6 01-Sep-16 11:03:06 AM

 ZETA FUNCTIONS OF
REDUCTIVE GROUPS
AND THEIR ZEROS

Lin Weng
Kyushu University, Japan

World Scientific
NEW JERSEY • LONDON • SINGAPORE • BEIJING • SHANGHAI • HONG KONG • TAIPEI • CHENNAI • TOKYO

10723_9789813231528_TP.indd 2 23/1/18 9:21 AM

 Published by
World Scientific Publishing Co. Pte. Ltd.
5 Toh Tuck Link, Singapore 596224
USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601
UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE

Library of Congress Cataloging-in-Publication Data

Names: Weng, Lin, 1964– author.
Title: Zeta functions of reductive groups and their zeros / by Lin Weng
(Kyushu University, Japan).
Description: New Jersey : World Scientific, 2018. | Includes bibliographical
references and index.
Identifiers: LCCN 2017053916 | ISBN 9789813231528 (hardcover : alk. paper)
Subjects: LCSH: Functions, Zeta. | Linear algebraic groups.
Classification: LCC QA351 .W46 2018 | DDC 515/.56--dc23
LC record available at https://lccn.loc.gov/2017053916

British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library.

Copyright © 2018 by World Scientific Publishing Co. Pte. Ltd.

All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means,
electronic or mechanical, including photocopying, recording or any information storage and retrieval
system now known or to be invented, without written permission from the publisher.

For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance
Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy
is not required from the publisher.

For any available supplementary material, please visit

http://www.worldscientific.com/worldscibooks/10.1142/10723#t=suppl

Printed in Singapore

LaiFun - 10723 - Zeta functions of reductive groups.indd 1 19-12-17 9:39:38 AM

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page v

Dedicated to my wife and

our daughter and son

v
 b2530   International Strategic Relations and China’s National Security: World at the Crossroads

This page intentionally left blank

b2530_FM.indd 6 01-Sep-16 11:03:06 AM

January 9, 2018 12:12 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page vii

Introduction

As a naive generalization of the Dedekind zeta function of a number field F,

for a positive integer n, there is a natural integration
Z
Θ(Λ) vol(Λ) s dµ(Λ) <(s) > 1. (0.1)
Mtot
F,n

Here, MtotF,n denotes the moduli space of (OF -)lattices Λ of rank n (with OF the
integer ring of F), Θ(Λ) and vol(Λ) denote the theta series and the co-volume of
Λ, respectively. If it were convergent when n ≥ 2, (0.1) would be viewed as a rank
n non-abelian zeta function of F, since this integration recovers the Dedekind zeta
function of F if n = 1, and the rank n lattices are associated to the group GLn , for
which only GL1 is commutative. Obviously, using the Mellin transform, (0.1) can
be rewritten as
Z ∞Z
dT
Θ(Λ) vol(Λ) s dµ(Λ)
0 Mtot
F,n [T ]
T
Z Z ∞ !
1 dT
= Θ(T 2n · Λ) T s dµ(Λ) (0.2)
tot
MF,n [1] 0 T
Z
= E(Λ,
b s) dµ(Λ),
Mtot
F,n [1]

where, for T > 0, Mtot F,n [T ] denotes the moduli space of rank n lattices of co-
volumes T > 0, and E(Λ, b s) denotes the (complete) Epstein zeta functions of Λ.
Since Epstein zeta functions are special kinds of Eisenstein series and hence are
well-known to be of slow growth, and Mtot F,n [T ] are not compact, e.g. in the case
F = Q, Mtot F,n [1] is isomorphic to SL n (Z)\SL n (R)/SOn , all the integrations above
(over moduli spaces) diverge.
The first task of this book is to remedy the above constructions to obtain con-
vergent integrations. Motivated by Mumford’s stability in algebraic geometry, it
is only natural to truncate the moduli spaces Mtot F,n [T ] using the stability condition
so as to obtain the compact moduli spaces MF,n [T ] and hence also the moduli
space MF,n of semi-stable lattices of rank n.

vii
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page viii

viii Introduction

ζF,n (s) of a number field F

Definition 0.1. The rank n non-abelian zeta function b
is defined by
Z
ζF,n (s) :=
b Θ(Λ) vol(Λ) s dµ(Λ) <(s) > 1. (0.3)
MF,n

Theorem 0.1. Non-abelian zeta functions satisfy all basic zeta properties.
Namely,

(0) Up to a constant factor depending on n and the local and global units of F,
ζF,1 (s) = b
b ζF (s).
(1) The integration (0.3) defines a homomorphic function in s and admits a unique
meromorphic continuation b ζF,n (s) to the whole complex s-plane.
ζF,n (s) satisfies the standard functional equation
(2) Zeta function b
ζF,n (1 − s) = b
b ζF,n (s).
(3) Zeta function bζF,n (s) has two singularities only, i.e. two simple poles at s =
0, 1, and the residues admit the following geometric interpretations
ζF,n (s) = vol MF,n [1] .


Res s=1b

One of the central themes of this book is to expose algebraic, analytic and
geometric structures of these zeta functions. In particular, a weak Riemann hy-
pothesis for the rank n non-abelian zeta function will be established, ensuring that
all but finitely many zeros of b ζQ,n (s) lie on the central line <(s) = 1/2 when n ≥ 2.
To explain this, as in (0.2) above, the Mellin transform transforms (0.3) into
Z
bζF,n (s) = E(Λ,
b s) dµ(Λ). (0.4)
MF,n [1]

Since the Epstein zeta function E(Λ, s) coincides with the Eisenstein series
E SLn /Pn−1,1 (1, g; s), induced from the constant function one on the Levi subgroup
of the maximal subgroup Pn−1,1 of SLn corresponding to the ordered partition
n = (n − 1) + 1, the rank n non-abelian zeta function may be viewed as a special
geometric Eisenstein period, defined as the integration of Eisestein series over
some compact moduli spaces.
Naturally associated to zeta functions, Eisenstein series play a central role in
number theory as well. From their theories, ranging from the classical Siegel to the
modern Langlands, and the trace formula techniques, of the primitive Selberg and
the powerful Arthur, for an L2 automorphic function ϕ on a Levi subgroup M of a
parabolic subgroup P of a reductive group G, with respect to a sufficiently regular
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page ix

Introduction ix

parameter T , there is an induced Eisenstein series E G/P (ϕ, g; λ) on G(F)\G(A)1 /K

and hence an analytic Eisenstein period
Z  
ΛT E G/P (ϕ, g; λ) dg. (0.5)
G(F)\G(A)1 /K
Here A denotes the adelic ring of F, K denotes a maximal compact subgroup of
G(A), and ΛT denotes the Arthur’s analytic truncation. In general, it is difficult
to calculate an analytic Eisenstein period. However, when ϕ is cuspidal, there is
a way to evaluate, thanks to the advanced Rankin-Selberg formula obtained by
Jacquet-Lapid-Rogowski with a use of regularized integrations over cones. For
example, when P = B is minimal and ϕ is the constant function 1 on the maximal
torus, based on the structures of the constant terms of E G/B (1, g; λ) exposed by
Siegel in lower rank and Langlands in general, it is proved that (0.5), for suffi-
ciently regular T , coincides with the period in the following:

Definition 0.2. The T -period of a split reductive group G over a number field F
is defined by
X ehwλ−ρ,T i Y ζF (hλ, αi)
ωG;T .
b
(λ) := (0.6)
F
ρ, αi
Q
α>0, wα<0 ζF (hλ, αi + 1)
w∈W α∈∆ hwλ − b
Here W, ∆, ρ, > and < 0 denote the Weyl group, the set of simple roots, the Weyl
vector and the positive and negative roots in the induced root system, respectively.

In addition, by Langlands’ theory on Eisenstein systems, being induced from

a special L2 -automorphic form, the Eisenstein series E SLn /Pn−1,1 (1, g; s) admits in
principle a realization as multiple residue of some Eisenstein series induced from
cusp forms over Levi subgroups of some higher co-rank parabolic subgroups.
Practically, in this book, as an SLn -analogue of a result of Diehl, the following ex-
plicit realization of the single variable E SLn /Pn−1,1 (1, g; s) is obtained as a multiple
residue of a several variables E SLn /P1,...,1 (1, g; λ), where P1,....1 denotes the minimal
parabolic subgroup of SLn associated to the partition n = 1 + · · · + 1.

Lemma 0.1. For SLn , denote by λi n−1


i=1 the fundamental dominant weights.
n−1
X
Then, with λ = si λi + ρ ∈ Cn−1 and s = sn−1 ,
i=1
SLn /Pn−1,1
(1, g; s) = Res sn−2 =1,...,s2 =1, s1 =1 E SLn /P1,...,1 (1, g; λ) .
 
E (0.7)

As a by-product, up to a certain normalization factor Norm(s) to effectively

collect the zeta factors in the denominators of the terms on the right hand side of
(0.8) below, there is the following:
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page x

x Introduction

ζFSLn (s) over a number field F is defined by

Definition 0.3. The SLn -zeta function b
bζFSLn (s)  
:= Res sn−2 =1,...,s2 =1, s1 =1 ωSL
F
n ;0
(λ)
Norm(s)
   (0.8)
X   1 Y ζF (hλ, αi) 
=  .
b
Res s =1,...,s =1, s =1  Q
w∈W
 n−2 2 1 
α∈∆ hwλ − ρ, αi α>0, wα<0 ζF (hλ, αi + 1)
b

Up to this point, despite all the main constructions have emerged, fine struc-
tures for non-abelian zeta functions have not been exposed because of the lack of
any connection between geometric Eisenstein periods and analytic Eisenstein pe-
riods. Our next main theme is to show that these two types of Eisenstein periods
coincide, by disclosing an intrinsic relation between Arthur’s analytic truncation
over SLn (F)\SLn (A)1 /K and the geometric truncation of stability on Mtot F,n . This
is a number theoretic analogue of a result of Lafforgue and leads to the following:

Theorem 0.2. Under the identification SLn (F)\SLn (A)1 /K ' Mtot
F,n [1],

(1) (Analytic Aspect of Stability and Parabolic Reduction) For g ∈ SLn (A)
and a normalized convex polygon p on [0, n],
1(pg ≤ p) = (−1)|P|−1 1(pδg
X X
P >P p). (0.9)
P δ∈P(F)\G(F)

Here p denotes the canonical polygon of the lattice Λ(g) associated to g.

g

(2) (Equivalence of Analytic and Geometric Eisenstein Periods) For the char-
acteristic function 1MF,n [1] of MF,n [1] ⊂ Mtot
F,n [1],

Λ0 1 =1MF,n [1] , (0.10a)

Z Z
E(ϕ, Λ; λ) dµ(Λ) = Λ0 E(ϕ, g; λ) dµ(g). (0.10b)
MF,n [1] SLn (F)\SLn (A)1 /K

All these then lead to the following:

Theorem 0.3 (Special Uniformity of Zeta Functions). For n ≥ 2,

ζF,n (s) = b
b ζFSLn (ns − n). (0.11)

Therefore, the rank n non-abelian zeta function b ζF,n (s) can be calculated
through a multiple residue of the SLn -period ωSL
F
n
(λ) in (0.8), and hence becomes a
linear combination of (products of) complete Dedekind zeta functions with ratio-
nal function coefficients. This result, apparently surprising, confirms once again
a hidden general principle that non-abelian invariants are built up from abelian
invariants via non-abelian symmetric structures.
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page xi

Introduction xi

To go further to write down b ζFSLn (s) and hence b

ζF,n (s) explicitly, we need to
understand for which special Weyl elements, the corresponding terms in (0.8)
admit non-trivial successive residues.

Definition 0.4. An element w ∈ W is called (n − 1)-special if

∆r{αn−1 } = α ∈ ∆r{αn−1 } : wα ∈ ∆ ∨ wα < 0 ,

(0.12)
where αn−1 ∈ ∆ denotes the simple root of SLn corresponding to Pn−1,1 .

There are totally 2n−3 (n + 2) special elements, as exposed jointly with Kim.
Moreover, when working on the special uniformity of zeta functions for function
fields, jointly with Zagier, we are able to characterize all these (n − 1)-special
elements, or better, the special permutations, after identifying the Weyl group W
of SLn with the symmetric group Sn on n symbols. The up-shot is the following

Theorem 0.4. The SLn -zeta function b ζFSLn (s) and hence the rank n non-abelian
zeta function bζF,n (s) of F are given explicitly by
n
X
bζF,n (s) = b ζFSLn (ns − n) = F,n (ns − n) · ζF (ns − n + a).
R(a) b (0.13)
a=1

Here the R(a)

F,n (s)’s are the rational functions in s defined by
n
X X vF,k1 . . .b
b vF,k p 1
F,n (s) =
R(a) (−1) p+r−1 ·
a=1 k1 ,...,k p >0
(k1 + k2 ) . . . (k p−1 + k p ) s + a + k p
k1 +···+k p =n−a
(0.14)
X 1 vF,l1 . . .b vF,lr
,
b
× ·
l1 ,...,lr >0
−s − a + 1 + l1 (l1 + l2 ) . . . (lr−1 + lr )
l1 +···+lr =a−1
m
Y  
vF,m =
where b ζF (k) and b
b ζF (1) := Res s=1 bζF (s) .
k=1

Our first application of this theorem is a closed formula for the volume of the
moduli space MF,n [1], since, by Theorems 0.1(3) and 0.4,
 n 
X (a)
vol MF,n [1] = Res s=1  RF,n (ns − n) · ζF (ns − n + a) .


b (0.15)
a=1

Corollary 0.1 (Semi-Stable Volume). The volume of the moduli space MQ,n [1]
of semi-stable OF -lattices of rank n and volume one is given by
n

X X vF,n1 · · ·b vF,nk
vol MF,n [1] = (−1)k−1 .
b
(0.16)
k=1 n ,...,n >0
(n1 + n2 ) · · · (nk−1 + nk )
1 k
n1 +...+nk =n
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page xii

xii Introduction

Our second application of Theorem 0.4 is the following:

Theorem 0.5 (Weak Riemann Hypothesis). Let n ≥ 2. All but finitely many
1
ζQ,n (s) lie on the central line <(s) = .
zeros of the zeta function b
2
When n = 2, 3, 4 and 5, Suzuki-Lagaruas, Suzuki and Ki establish the full
Riemann hypothesis for b ζQSLn (s), respectively. Our first proof of this theorem was
obtained using the techniques developed in these works and a joint work on the
functional equation of bζFSLn (s) with Kim. In this book, as claimed above, we give
a simplified proof using Theorem 0.4. The key is to use a meromorphic decom-
position of the rank n zeta function induced by the refined symmetric structures
hidden in (0.13). Namely,
≤ n+1
 
≥ n+1 ≤ n+1
≥ n+1
 ζQ 2 (s) 
b
ζQ 2 (s) + b
ζQ,n(s) = b ζQ 2 (s) = b
ζQ 2 (s) ·  1 + (0.17)
b  
≥ n+1
ζ 2 (s)
 
b
Q

where
≥ n+1 X0
ζQ 2 (s) :=
b
F,n (ns − n) · ζF (ns − n + a),
R(a) b
a≥ n+1
2
n+1 X0 (0.18)
≤
ζQ 2 (s) :=
b
F,n (ns − n) · ζF (ns − n + a)
R(a) b
a≤ n+1
2
P
X0 
 a≥ n+1

 2
n even X0
with n+1
:=  and a similar . Surely,
a≤ n+1
a≥ 2
 
 a≥ n+1 − 1 a= n+1

 P P
n odd 2
2 2 2
(0.17) is compatible with the functional equation since
n+1 n+1
≥ ≤
ζQ 2 (1 − s) = b
b ζQ 2 (s). (0.19)

Hence, to locate the zeros of rank n zeta function b ζQ,n (s), it suffices to expose
≥ n+1 ≤ n+1 . ≥ n+1
the structures of the factor ζ
b 2
Q (s) and the quotient ζ 2 (s) b
b
Qζ 2 (s). Being the
Q
combinations of Riemann zeta functions with coefficients of rational functions,
≥ n+1 ≤ n+1
ζQ 2 (s) and hence also b
b ζQ 2 (s) admit Hadamard products. Say,
 n 
≥ n+1 Y  ≥ n+1
ξQ (s) :=  (ns − n + k) · b
2
ζQ 2 (s) = κn ecn s Pn (s) Hn (s) (0.20)
k=1

where

(1) κn ∈ R∗ and cn is constant depending only on n,

(2) Pn (s) ∈ R[s] is a polynomial with real coefficients,
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page xiii

Introduction xiii

(3) Hn (s) is uniformly convergent on every compact subset of C in the form

∞  !
Y s  s 
Hn (s) = 1− 1+ , (0.21)
m=1
ρm ρm
where =(ρm ) , 0 for all m.
n+1
≥
In fact, it is not too difficult to see that cn is real and hence bζQ 2 (s) can be approx-
imated by the polynomials
m !
 cn s m Y  s  s
wm (s) := P(s) 1 + 1− 1+ . (0.22)
m k=1
ρk ρk
Hence, following an idea of Ki, with an elementary result of de Bruijn on the
zeros of polynomials wm (s) ± wm (1 − s), the weak Riemann Hypothesis above
≥ n+1
ζ 2 (s) lie in the region of
holds, provided that all but finitely many zeros of b Q
cn < <(s) < 1/2 for a suitable constant cn . In addition, it suffices to treat the zero
≥ n+1
free region of bζQ 2 (s) on the right, since its zero free region on the left can be
deduced using the same argument with the help of the functional equation of the
Riemann zeta function.
The idea then becomes extremely simple – there should be a term in the sum-
≥ n+1
mation of bζ 2 (s) which gives a dominant wave of zeros. Since the zero free
Q
regions for the Riemann zeta function and the estimations for both the Riemann
zeta function and the Gamma function are very much classical, it suffices to locate
where are the dominant rational factor. Or equivalently, by multiplying the least
common multiple of their denominators, we must detect the sign of the coeffi-
cient for the highest order term of the resulting polynomial. This then leads to the
following:

Definition 0.5. The discriminant ∆F,n of b

ζF,n (s) is defined by
n−1
X X vF,n1 · · ·b vF,nk
∆F,n := (−1)k−1 .
b
(0.23)
k=1 n1 ,...,nk >0
(n1 + n2 ) · · · (nk−1 + nk )
n1 +...+nk =n−1

n+1
≥
ζQ 2 (s) in the
Theorem 0.6. Let n ≥ 2. There are at most finitely many zeros of b
region of <(s) ≥ 1/2 and hence bζQ,n (s) satisfies the weak Riemann hypothesis,
provided that ∆Q,n > 0.

By Corollary 0.1, ∆Q,n = vol MQ,n−1 [1] when n ≥ 2. Therefore. ∆Q,n is

always positive. This then completes the proof of Theorem 0.5 on a weak Riemann
ζQ,n (s) when n ≥ 2.
Hypothesis for b
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page xiv

xiv Introduction

With the weak Riemann hypothesis established, for each n ≥ 2, there is a

zero sequence of the zeta function b ζQ,n (s), lying on the central line <(s) = 1/2,
ready to be studied and used. It is natural to ask what are their distributions. Let
n1 √ o
+ −1 γ be the set of zeros of b ζQ,n (s). Arranging the γ’s in an increasing
2
order
0 ≤ γn,1 ≤ γn,2 ≤ · · · ≤ γn,3 ≤ · · · , (0.24)
and, as usual, denote by
Nn (T ) := # k : 0 < γn,k < T

(0.25)
the number of zeros of bζQ,n (s) whose imaginary parts lie between 0 and T .

ζQ,n (s), when n ≥ 3,1 we have

Theorem 0.7. For the zeros of b
n  n  
(1) Nn (T ) = T log T − 1 + O(1).
2π 2π
2π k
!!
log log k
(2) γn,k = 1+O .
n log k log k
2π 1
!
log log k
(3) γn,k+1 − γn,k = +O .
n log k log2 k
Apparently, they are similar to the corresponding statements for the Riemann
zeros (under the Riemann hypothesis). However, this is not true.

ζQ,n (s) is defined by

Definition 0.6. The pair correlation sequence for zeros of b
 n
  n 
δn,k := γn,k+1 − γn,k · log γn,k . (0.26)
2π 2π
ζQ,n (s),
Theorem 0.8. Let n ≥ 3. For the pair correlation sequence of the zeros of b
!
log log k
δn,k = 1 + O . (0.27)
k log k
In other words, the pair correlations of our zeros only yield the Dirac distri-
butions. This is very different from the distributions of the Riemann zeros, which
conjecturally coincide with that for GUE, the Gaussian Unitary Ensemble. It took
quite some time for us to understand and accept this. The clue is yet another re-
fined level of structures for the zeros of non-abelian zeta functions, first exposed
in our studies for non-abelian zeta functions of function fields.

Definition 0.7. The secondary big Delta pair correlation sequence for the zeros
ζQ,n (s) are defined by
of b
 n 
∆n,k := δn,k − 1 · log γn,k .


(0.28)
2π
1 Here and in the sequel, when n = 2, stronger results hold. For details, please see Part 6.
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page xv

Introduction xv

Motivated by this consideration of secondary structures, Suzuki proves a re-

fined result for the correlations distributions of rank 2 zeta zeros. This in turn
motivates the following result of ours for all rank n (n ≥ 2) zeta zeros. To explain
this, we introduce the well-known M-function
Z X ∞

1 λz (n)λz̄ (n)   
M(z, σ) :=


 2σ
 · exp − i<(z̄w) dw. (0.29)
2π C n=1
n
∞  ∞ !k
X i k Λk (n) k X Λk (n) ζ0
Here λz (n) := − z with Λk (n) defined by := (s) .
k=0
2 k! n=1
ns ζ
The M-function M(z, σ) is real valued, decays rapidly as |z| → ∞. Since
λz (m)λz (n) = λz (mn) for (m, n) = 1, M(z, σ), being a Fourier transform, admits a
natural convolution Euler product.

Theorem 0.9. Let ϕ(x) be a function in C 1 (R)!such that ϕ0 (x)  1 for |x| < 1,
d ζ0  n 
ϕ0 (x)  x−2 for |x| ≥ 1, and ϕ < + int is bounded on R. Then for n ≥ 2,
dt ζ 2
Z ∞ 
1 X 1 n
ϕ ∆n,k = ϕ(u) du.


lim m u, (0.30)
T →∞ Nn (T ) 2π −∞ 2
0<γk ≤T
Z ∞
Here m(u, σ) := M(u + iv, σ) dv.
−∞

With all these said, reemerged is the role played by the stability condition for
non-abelian zeta functions b ζF,n (s). Note that, instead of T = 0 above, correspond-
ing to the stability condition, every positive T ≥ 0 can be used to introduce a
ζF,n
T -version of b T
(s) satisfying bζF,n
0
(s) = b
ζF,n (s). Even many tautological zeta prop-
erties can be similarly established for ζF,n bT
ζQ,3
(s), numerical calculations with b T
(s)
indicate that, only when T = 0, the Riemann hypothesis holds. So, stability does
play a key role for non-abelian zeta functions to satisfy the Riemann hypothesis.

We will not explore the deep reasons behind. Instead, in this book, we de-
velop a new theory for zeta functions of general split reductive groups G, or
better, of split reductive groups G and their maximal parabolic subgroups P, over
number fields. The searching process for these zeta functions b ζFG/P (s) is faithfully
recorded in Part 4, with a theoretic preparation in Part 3, in which genuine mul-
tiple L-functions for reductive groups are introduced using Langlands’ theory of
Eisenstein series and analytic and geometric truncations. The up-shot is an ana-
logue of Definitions 0.2 and 0.3. The functional equation of bζFG/P (s) is established
essentially following Komori in Part 4 as well. In Parts 5 and 6, we prove a weak
Riemann Hypothesis for the zeta functions b ζFG/P (s). It consists of two aspects.
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page xvi

xvi Introduction

The first by Ki-Komori-Suzuki on algebraic and analytic structures of our zeta

functions in Part 5. Here, for a Chevalley group G over Q, they deduce a weak
Riemann hypothesis for b ζFG/P (s) to the positivity of its associated discriminant,
based on a refined decomposition for b ζFG/P (s) in terms of Lie theoretic structures.
All these are very technical. Then, the second more recent one by the author on
the geometric structures of our zeta functions developed in Part 6. Here, we first
introduce the notions of arithmetic G-torsors and their stability, then establish an
analogue of Theorem 0.2 for reductive groups, and finally, show that the discrim-
inant above coincides the volume of the moduli space of the arithmetic principle
G-torsors of slopes zero which admit P as their canonical parabolic subgroups.
This part is rather challenging and finally completes a proof of the following:

Theorem 0.10 (Wean Riemann Hypothesis). Let G be a Chevalley group over

ζFG/P (s) lie on the central line.
Q. All but finitely many zeros of b

This book ends with five appendices on our joint works with K. Sugahara on
new arithmetic cohomologies for arithmetic varieties, ranging from the construc-
tions of arithmetic adelic cohomology groups in higher dimensions, to a theory of
ind-pro topologies in arithmetic dimension two, and the reciprocity laws for arith-
metic surfaces. They may be viewed as natural continuations of the cohomology
theory in arithmetic dimension one of Part 1.

Acknowledgement. I would like to thank Hida for his consistent support. The
first suggestion to the author on writing a book about the zetas for reductive groups
came from him. I also would like to thank Lafforgue for his help. Not only the
quantitive relation between stability and parabolic reduction is motivated by his
work, but the title of the book was proposed by him. The zeta project, as a part of
the Geometric Arithmetic program, was initiated in 1999. Working alone mainly,
the author has benefit from the visits to IHES, MPIM, and UCLA, where many
components of this project were done. I would like to thank them for providing
excellent research environments. Without the works of Suzuki, Komori and all our
co-authors, we cannot present a complete story on the weak Riemann hypothesis
of our zeta functions. I would like to thank them for their contributions. Special
thanks are due to Sugahara for agreeing to publish our joint works as the appen-
dices to this book. At last, but not the least, I would like to thank my family for
their understanding, trust and support given to me in all these years.
The zeta project has been partially supported by JSPS.
December 21, 2017 14:18 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page xvii

Contents

Introduction vii

Non-Abelian Zeta Functions 1

1. Semi-Stable Lattice 3
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Projective Modules . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2.1 Invertible Modules . . . . . . . . . . . . . . . . . . . . 7
1.2.2 General Projective Modules . . . . . . . . . . . . . . . 8
1.3 Stability of Lattices . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3.1 Lattices . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3.2 Covolumes of Lattices . . . . . . . . . . . . . . . . . . 10
1.3.3 Stability of Lattices . . . . . . . . . . . . . . . . . . . 11
1.3.4 Canonical Filtration . . . . . . . . . . . . . . . . . . . 13
1.4 Volume of Lattice: Special Linear Group . . . . . . . . . . . . . 17
1.4.1 Metrics of Lattices . . . . . . . . . . . . . . . . . . . . 17
1.4.2 Special Metrics of Lattices . . . . . . . . . . . . . . . . 18
1.5 Automorphisms of Lattices . . . . . . . . . . . . . . . . . . . . 20
1.5.1 General Automorphisms . . . . . . . . . . . . . . . . . 20
1.5.2 Special Automorphisms . . . . . . . . . . . . . . . . . 21
1.5.3 Unit Automorphisms . . . . . . . . . . . . . . . . . . . 23
1.6 Compact Moduli Spaces of Semi-Stable Lattices . . . . . . . . . 25

xvii
December 21, 2017 14:18 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page xviii

xviii Contents

2. Geometry of Numbers 27
2.1 Global Cohomology . . . . . . . . . . . . . . . . . . . . . . . . 27
2.1.1 Adelic Ring . . . . . . . . . . . . . . . . . . . . . . . 27
2.1.2 Adelic Cohomology Groups . . . . . . . . . . . . . . . 28
2.1.3 Topological Duality: Local and Global Pairings . . . . 29
2.2 Duality and Riemann-Roch Theorem in Arithmetic . . . . . . . 30
2.2.1 Nine-Diagram . . . . . . . . . . . . . . . . . . . . . . 30
2.2.2 Arithmetic Riemann-Roch Theorem . . . . . . . . . . . 32
2.3 Arithmetic Vanishing Theorem . . . . . . . . . . . . . . . . . . 36
2.3.1 Elementary Properties of 0-th Cohomology . . . . . . . 36
2.3.2 Ampleness and Vanishing Theorem . . . . . . . . . . . 39

3. Non-Abelian Zeta Functions 43

3.1 Moduli Spaces of Semi-Stable Lattices . . . . . . . . . . . . . . 43
3.1.1 Stability . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.1.2 Effective Vanishing Theorem . . . . . . . . . . . . . . 44
3.2 Non-Abelian Zeta Function . . . . . . . . . . . . . . . . . . . . 46
3.2.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . 46
3.2.2 Zeta Properties . . . . . . . . . . . . . . . . . . . . . . 46
3.2.3 Relation with Epstein Zeta Function . . . . . . . . . . 48
3.2.4 Riemann Hypothesis . . . . . . . . . . . . . . . . . . . 51

Rank Two Zeta Functions 53

4. Distances to Cusps and Fundamental Domains 55

4.1 Upper Half Space Model . . . . . . . . . . . . . . . . . . . . . 55
4.1.1 Upper Half Plane . . . . . . . . . . . . . . . . . . . . . 56
4.1.2 Upper Half Space . . . . . . . . . . . . . . . . . . . . 57
4.1.3 Rank Two OK -Lattices: Upper Half Space Model . . . . 59
4.2 Cusps and Ideal Classes . . . . . . . . . . . . . . . . . . . . . . 60
4.2.1 Generators of Fractional Ideals . . . . . . . . . . . . . 61
4.2.2 Special Transformations . . . . . . . . . . . . . . . . . 62
4.2.3 Cusps and Ideal Classes for Total Real Fields . . . . . . 63
January 18, 2018 9:46 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page xix

Contents xix

4.2.4 Cusps and Ideal Classes . . . . . . . . . . . . . . . . . 65

4.3 Stabilizers of Cusps and Their Fundamental Domains . . . . . . 68
4.3.1 Upper Half Plane . . . . . . . . . . . . . . . . . . . . . 68
4.3.2 Upper Half Space . . . . . . . . . . . . . . . . . . . . 70
4.3.3 Fundamental Domains for Stabilizer Groups (I) . . . . 71
4.3.4 Fundamental Domains for Stabilizer Groups (II) . . . . 78
4.4 Distance to Cusp and Fundamental Domain . . . . . . . . . . . 81
4.4.1 Upper Half Plane . . . . . . . . . . . . . . . . . . . . . 81
4.4.2 Upper Half Space . . . . . . . . . . . . . . . . . . . . 83
4.4.3 Fundamental Domain (I) . . . . . . . . . . . . . . . . . 85
4.4.4 Fundamental Domain (II) . . . . . . . . . . . . . . . . 93

5. Rank Two Zeta Functions 103

5.1 Distance to Cusp and Stability . . . . . . . . . . . . . . . . . . 103
5.1.1 Parameter Space of Semi-Stable Rank Two Lattices . . 103
5.1.2 Rank One Sub-Lattices . . . . . . . . . . . . . . . . . 105
5.1.3 Stability and Distances to Cusps . . . . . . . . . . . . . 106
5.1.4 Example: Truncated Fundamental Domain . . . . . . . 108
5.1.5 Moduli Space of Rank Two Semi-Stable Lattices . . . . 109
5.2 Rank Two Non-Abelian Zeta Function . . . . . . . . . . . . . . 111
5.2.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . 111
5.2.2 Relations with Epstein Zeta Functions . . . . . . . . . 113
5.3 Epstein Zeta Function and Its Fourier Expansion . . . . . . . . . 114
5.3.1 Automorphic Function in One Variable . . . . . . . . . 114
5.3.2 Upper Half Space . . . . . . . . . . . . . . . . . . . . 118
5.3.3 Eisenstein Series for Rank Two Lattice . . . . . . . . . 125
5.4 Explicit Formula of Rank Two Zeta Function . . . . . . . . . . 133
5.4.1 Classical Rankin-Selberg Method . . . . . . . . . . . . 133
5.4.2 Generalization of Rankin-Selberg Method . . . . . . . 140
5.4.3 Rank Two Non-Abelian Zeta Function . . . . . . . . . 143
5.5 Zeros of Rank Two Non-Abelian Zeta Functions . . . . . . . . . 153
5.5.1 Product Formula for Entire Function of Order 1 . . . . 153
5.5.2 Zeros of Rank Two Non-Abelian Zeta Function of Q . . 154
5.5.3 A Simple Generalization . . . . . . . . . . . . . . . . . 156
January 18, 2018 9:46 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page xx

xx Contents

Eisenstein Periods and Multiple L-Functions 159

6. Multiple L-Functions 161

6.1 Reduction Groups . . . . . . . . . . . . . . . . . . . . . . . . . 161
6.1.1 Parabolic Subgroups . . . . . . . . . . . . . . . . . . . 161
6.1.2 Roots, Coroots, Weights and Coweights . . . . . . . . . 163
6.1.3 Example with Special Linear Group . . . . . . . . . . . 165
6.1.4 Logarithmic Map . . . . . . . . . . . . . . . . . . . . 169
6.1.5 Compatible Haar Measures . . . . . . . . . . . . . . . 170
6.1.6 Classical Reductive Theory . . . . . . . . . . . . . . . 171
6.2 Multiple L-Functions . . . . . . . . . . . . . . . . . . . . . . . 172
6.2.1 Growth Condition . . . . . . . . . . . . . . . . . . . . 172
6.2.2 Automorphic Forms . . . . . . . . . . . . . . . . . . . 173
6.2.3 Eisenstein Series and Their Constant Terms . . . . . . . 176
6.2.4 Stability Truncation on Adelic Space . . . . . . . . . . 178
6.2.5 Multiple L-Functions . . . . . . . . . . . . . . . . . . 181
6.2.6 Meromorphic Extension and Functional Equations . . . 183
6.2.7 Holomorphicity and Singularities . . . . . . . . . . . . 183

7. Periods of Reductive Groups 185

7.1 Arthur’s Analytic Truncation . . . . . . . . . . . . . . . . . . . 185
7.1.1 Positive Cone and Positive Chamber . . . . . . . . . . 185
7.1.2 Preliminary Estimations . . . . . . . . . . . . . . . . . 186
7.1.3 Langlands’ Combinatorial Lemma . . . . . . . . . . . 187
7.1.4 Langlands-Arthur’s Partition: Reduction Theory . . . . 188
7.1.5 Arthur’s Analytic Truncation . . . . . . . . . . . . . . 189
7.1.6 Basic Properties . . . . . . . . . . . . . . . . . . . . . 190
7.2 Analytic Arthur Periods and Geometric Eisenstein Periods . . . 191
7.2.1 ΛT 1 as Characteristic Function of Compact Set . . . . 191
7.2.2 Geometric Eisenstein Period . . . . . . . . . . . . . . . 194
7.2.3 Regularized Integration over Cone . . . . . . . . . . . 195
7.2.4 Regularized Period of Automorphic Form . . . . . . . . 197
7.2.5 Regularized Periods . . . . . . . . . . . . . . . . . . . 200
7.3 Periods of Reductive Groups . . . . . . . . . . . . . . . . . . . 201
January 18, 2018 9:46 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page xxi

Contents xxi

7.3.1 Eisenstein Periods for Cusp Forms . . . . . . . . . . . 201

7.3.2 Gindikin-Karpelevich Formula . . . . . . . . . . . . . 202
7.3.3 Periods of Reductive Groups . . . . . . . . . . . . . . 202
7.3.4 Volumes of Truncated Fundamental Domains . . . . . . 203

Zeta Functions for Reductive Groups 207

8. Zeta Functions for Reductive Groups 209

8.1 Zeta Function for SLn : Genuine but Different . . . . . . . . . . 209
8.1.1 Non-Abelian Zeta Function and Eisenstein Period . . . 209
8.1.2 Epstein Zeta Function and Siegel-Eisenstein Series . . . 210
8.1.3 Langlands’ Eisenstein and Siegel’s Eisenstein . . . . . 213
8.1.4 Zeta Function for SLn . . . . . . . . . . . . . . . . . . 214
8.2 From SLn to Sp2n : Role of Periods . . . . . . . . . . . . . . . . 217
8.2.1 Sp2n -Periods . . . . . . . . . . . . . . . . . . . . . . . 217
8.2.2 Siegel Eisenstein Series . . . . . . . . . . . . . . . . . 218
8.2.3 Siegel Eisenstein Series and Langlands Eisenstein Series 219
8.2.4 Siegel-Maaβ-Eisenstein Period and Zeta Function of Sp2n 219
8.3 Zeta Functions for G2 . . . . . . . . . . . . . . . . . . . . . . . 220
8.3.1 Looking for Zeta Functions of G2 . . . . . . . . . . . . 220
8.3.2 RH for Zeta Functions of G2 I: Preparations . . . . . . 228
8.3.3 Riemann Hypothesis for G2 . . . . . . . . . . . . . . . 236
8.4 Zeta Functions for (G, P) . . . . . . . . . . . . . . . . . . . . . 242
8.4.1 Singular Hyperplanes Located . . . . . . . . . . . . . . 242
8.4.2 Main Definition . . . . . . . . . . . . . . . . . . . . . 242
8.4.3 Basic Properties . . . . . . . . . . . . . . . . . . . . . 245

9. Zeta Function of (SLn , Pn−1,1 ) 247

9.1 Special Weyl Elements . . . . . . . . . . . . . . . . . . . . . . 247
9.1.1 General Setting . . . . . . . . . . . . . . . . . . . . . . 247
9.1.2 Special Weyl Elements . . . . . . . . . . . . . . . . . . 249
9.1.3 Special Permutations . . . . . . . . . . . . . . . . . . . 252
9.1.4 Working Site . . . . . . . . . . . . . . . . . . . . . . . 253
9.1.5 Special Permutations . . . . . . . . . . . . . . . . . . . 255
December 21, 2017 14:18 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page xxii

xxii Contents

9.2 Explicit Formula for Zeta Function of (SLn , Pn−1,1 ) . . . . . . . 258

9.2.1 Rough Formula for Period of (SLn , Pn−1,1 ) . . . . . . . 258
9.2.2 Applications of Special Permutations . . . . . . . . . . 260
9.2.3 Explicit Formula for Zeta Functions of (SLn , Pn−1,1 ) . . 263

10. Functional Equation 265

10.1 Preparations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
10.1.1 Statement . . . . . . . . . . . . . . . . . . . . . . . . 265
10.1.2 Lie Theoretic Relations (I) . . . . . . . . . . . . . . . . 267
10.2 Normalization and Micro Functional Equation . . . . . . . . . . 269
10.2.1 Overdone Normalization and Micro Functional Equation 269
10.2.2 Normalization . . . . . . . . . . . . . . . . . . . . . . 273
10.3 Functional Equation . . . . . . . . . . . . . . . . . . . . . . . . 274
10.3.1 Proof of Functional Equation . . . . . . . . . . . . . . 274
10.3.2 Constant cG,P . . . . . . . . . . . . . . . . . . . . . . . 276
10.4 T -Version for SL3 . . . . . . . . . . . . . . . . . . . . . . . . . 277

Algebraic, Analytic Structures and Rieman Hypothesis 279

SLn /Pn−1,1
ζQ
11. Conditional Weak RH for b (s) 281
11.1 Refined Symmetric Structure . . . . . . . . . . . . . . . . . . . 281
11.1.1 Zeta Functions for (SLn , Pn−1,1 )/Q . . . . . . . . . . . 281
11.1.2 Refined Symmetric Structure . . . . . . . . . . . . . . 283
11.2 Zero-Free Region of Upper Half Function . . . . . . . . . . . . 284
11.2.1 Dominant Term . . . . . . . . . . . . . . . . . . . . . 284
11.2.2 Zero Free Region on Right Half Plane . . . . . . . . . 286
11.2.3 Zero Free Region on Left Half Plane . . . . . . . . . . 289
n+1
≥
ξQ
11.3 Refined Hadamard Product for b 2
(s) . . . . . . . . . . . . . 291
≥ n+1
ξQ 2 (s) . . . . . . .
11.3.1 Distributions of Zeros for b . . . . . 291
≥ n+1
11.3.2 Refined Hadamard Product for b ξQ 2 (s) . . . . . . . . . 293
SL /P
11.4 Conditional Weak Riemann Hypothesis for b ζQ n n−1,1(s) . . . . . 295
11.4.1 Elementary Lemma . . . . . . . . . . . . . . . . . . . 295
SL /P
11.4.2 Conditional Weak Riemann Hypothesis for bζQ n n−1,1(s) 298
December 21, 2017 14:18 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page xxiii

Contents xxiii

12. Algebraic and Analytic Structures and Weak Riemann Hypothesis 299
12.1 Criterion for Weak Riemann Hypothesis . . . . . . . . . . . . . 299
12.1.1 Discriminant of (G, P) . . . . . . . . . . . . . . . . . . 299
12.1.2 Criterion of Weak Riemann Hypothesis . . . . . . . . . 301
12.2 Refined Symmetries of Zeta Functions . . . . . . . . . . . . . . 301
12.2.1 Positively Oriented . . . . . . . . . . . . . . . . . . . . 301
12.2.2 Refined Symmetries . . . . . . . . . . . . . . . . . . . 302
12.3 Dominate Term . . . . . . . . . . . . . . . . . . . . . . . . . . 304
12.3.1 Entire Function Oriented . . . . . . . . . . . . . . . . . 304
12.3.2 Terms with Maximal Discrepancy . . . . . . . . . . . . 306
12.3.3 Relation between W p and W p . . . . . . . . . . . . . . 309
12.3.4 Leading Polynomials . . . . . . . . . . . . . . . . . . 312
12.4 Zero Free Regions . . . . . . . . . . . . . . . . . . . . . . . . . 313
12.4.1 Lie Theoretic Structures Involved . . . . . . . . . . . . 313
12.4.2 Estimations on Right Half Plane . . . . . . . . . . . . . 318
12.4.3 Normalization Factor . . . . . . . . . . . . . . . . . . 319
12.4.4 Zero Free Region on Right Half Plane . . . . . . . . . 320
12.4.5 Zero Free Region on Left Half Plane . . . . . . . . . . 322
12.5 Hadamard Product . . . . . . . . . . . . . . . . . . . . . . . . 325
12.5.1 Distributions for Zeros of A p (s) . . . . . . . . . . . . . 325
12.5.2 Hadamard Product . . . . . . . . . . . . . . . . . . . . 328
12.6 Proof of Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 332
12.7 Discriminant and Residue of Period . . . . . . . . . . . . . . . 333
12.7.1 Residue of Period . . . . . . . . . . . . . . . . . . . . 333
12.7.2 Period of Pseudo Root System . . . . . . . . . . . . . . 335
12.7.3 Residue and Leading Coefficient . . . . . . . . . . . . 337
12.A Decomposition of Σ p (1) . . . . . . . . . . . . . . . . . . . . . . 339
12.B Geometric Interpretation of c p . . . . . . . . . . . . . . . . . . 348

13. Weak Riemann Hypothesis for Zeta Functions of Exceptional Groups 349
13.1 Weyl Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
13.1.1 Root Systems . . . . . . . . . . . . . . . . . . . . . . 349
13.1.2 Weyl Group for D5 . . . . . . . . . . . . . . . . . . . . 350
December 21, 2017 14:18 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page xxiv

xxiv Contents

13.1.3 Weyl Group for E6 . . . . . . . . . . . . . . . . . . . . 351

13.1.4 Weyl Group for E7 . . . . . . . . . . . . . . . . . . . . 352
13.1.5 Weyl Group for E8 . . . . . . . . . . . . . . . . . . . . 353
13.2 Weak Riemann Hypothesis for Zeta Functions of En . . . . . . . 356
13.2.1 Conditions for Weak Riemann Hypothesis . . . . . . . 356
13.2.2 Special and Very Special Weyl Elements . . . . . . . . 358
13.2.3 Constants hλP , α∨w i . . . . . . . . . . . . . . . . . . . . 360
13.2.4 Discriminants ∆QEn ,P . . . . . . . . . . . . . . . . . . . 360
13.2.5 Weak Riemann Hypothesis for Zeta Functions of
Exceptional Groups . . . . . . . . . . . . . . . . . . . 366

Geometric Structures and Riemann Hypothesis 367

14. Analytic and Geometric Truncations: Special Linear Groups 369

14.1 Geometric Truncation and Parabolic Reduction . . . . . . . . . 369
14.1.1 Canonical Polygon and Compactness . . . . . . . . . . 369
14.1.2 Parabolic Reduction and Geometric Truncation . . . . . 370
14.2 Analytic and Geometric Truncations . . . . . . . . . . . . . . . 375
14.2.1 Micro Bridge . . . . . . . . . . . . . . . . . . . . . . . 375
14.2.2 Analytic and Geometric Truncations . . . . . . . . . . 380

15. Special Uniformity of Zeta Functions and Weak Riemann Hypothesis 381
15.1 Special Uniformity of Zeta Functions . . . . . . . . . . . . . . . 381
15.1.1 Basic Properties of Truncations . . . . . . . . . . . . . 381
15.1.2 Analytic and Geometric Periods . . . . . . . . . . . . . 383
15.2 Weak Riemann Hypothesis for Non-Abelian Zeta Functions . . . 384
15.2.1 Parabolic Reduction, Stability and Volumes . . . . . . . 384
15.2.2 Existence of Stable OF -Lattices . . . . . . . . . . . . . 386
15.2.3 Weak Riemann Hypothesis . . . . . . . . . . . . . . . 386

16. Analytic and Geometric Truncations: Reductive Groups 387

16.1 Lie Theoretic Preparation . . . . . . . . . . . . . . . . . . . . . 387
16.1.1 Maximal Split Sub or Quotient Torus . . . . . . . . . . 387
December 21, 2017 14:18 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page xxv

Contents xxv

16.1.2 Generalized Root System . . . . . . . . . . . . . . . . 388

16.1.3 Partial Order Induced by Positive Weyl Cone . . . . . . 390
16.1.4 Classical Example . . . . . . . . . . . . . . . . . . . . 394
16.1.5 Relative ‘Root System’ . . . . . . . . . . . . . . . . . 396
16.1.6 Relative Positive Chamber and Cone . . . . . . . . . . 397
16.1.7 Partial Order in Relative Theory . . . . . . . . . . . . . 399
16.2 Arithmetic Principal Torsors and Their Stability . . . . . . . . . 401
16.2.1 Compatible Metrics . . . . . . . . . . . . . . . . . . . 401
16.2.2 Principal Lattices . . . . . . . . . . . . . . . . . . . . 405
16.2.3 Parabolic Reduction . . . . . . . . . . . . . . . . . . . 407
16.2.4 Torsors in Arithmetic . . . . . . . . . . . . . . . . . . 410
16.2.5 Canonical Parabolic Subgroup Scheme . . . . . . . . . 413
16.3 Analytic and Geometric Truncations . . . . . . . . . . . . . . . 415
16.3.1 Stability and Parabolic Reduction . . . . . . . . . . . . 416
16.3.2 Equivalence of Analytic and Geometric Truncations . . 422
16.3.3 Properties of Analytic and Geometric Truncations . . . 422

17. Weak Riemann Hypothesis for Zeta Functions of Reductive Groups 427
17.1 Volumes of Semi-Stable Moduli Spaces . . . . . . . . . . . . . 427
17.1.1 Parabolic Reduction, Stability and the Volumes . . . . . 427
17.2 Riemann Hypothesis for Zeta Functions of Reductive Groups . . 431
17.2.1 Geometric Interpretation of Discriminant . . . . . . . . 431
17.2.2 Riemann Hypothesis for Weng Zeta Functions . . . . . 432

18. Distributions of Zeros for Zeta Functions of Reductive Groups 433

18.1 Distributions of Zeta Zeros . . . . . . . . . . . . . . . . . . . . 433
18.1.1 Distributions in Classical Style . . . . . . . . . . . . . 433
18.1.2 New Secondary Distributions . . . . . . . . . . . . . . 434
18.2 Proof of Theorems . . . . . . . . . . . . . . . . . . . . . . . . 437
18.2.1 Rank Two Zeta Zeros . . . . . . . . . . . . . . . . . . 437
18.2.2 Proof of Theorems 18.3 and 18.4 . . . . . . . . . . . . 438
18.2.3 Proof of Theorems 18.1 and 18.2 . . . . . . . . . . . . 444
December 21, 2017 14:18 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page xxvi

xxvi Contents

Five Essays on Arithmetic Cohomology

(Joint with K. Sugahara) 445

Appendix A Arithmetic Adelic Complexes 447

A.1 Parshin-Beilinson’s Theory . . . . . . . . . . . . . . . . . . . . 447
A.1.1 Local Fields for Reduced Flags . . . . . . . . . . . . . 447
A.1.2 Adelic Cohomology Theory . . . . . . . . . . . . . . . 449
A.1.3 Example . . . . . . . . . . . . . . . . . . . . . . . . . 451
A.2 Arithmetic Cohomology Groups . . . . . . . . . . . . . . . . . 452
A.2.1 Adelic Rings for Arithmetic Surfaces . . . . . . . . . . 452
A.2.2 Adelic Spaces at Infinity . . . . . . . . . . . . . . . . . 454
A.2.3 Arithmetic Adelic Complexes . . . . . . . . . . . . . . 455
A.2.4 Cohomology Theory for Arithmetic Curves . . . . . . . 458

Appendix B Cohomology for Arithmetic Surfaces 459

B.1 Local Residue Pairings . . . . . . . . . . . . . . . . . . . . . . 459
B.1.1 Residue Maps for Local Fields . . . . . . . . . . . . . 459
B.1.2 Local Residue Maps . . . . . . . . . . . . . . . . . . . 461
B.2 Global Residue Pairing . . . . . . . . . . . . . . . . . . . . . . 461
B.2.1 Global Residue Pairing . . . . . . . . . . . . . . . . . 462
B.2.2 Non-Degeneracy . . . . . . . . . . . . . . . . . . . . . 462
B.3 Adelic Subspaces . . . . . . . . . . . . . . . . . . . . . . . . . 463
B.3.1 Level Two Subspaces . . . . . . . . . . . . . . . . . . 463
B.3.2 Perpendicular Subspaces . . . . . . . . . . . . . . . . . 465
B.4 Arithmetic Cohomology Groups for Arithmetic Surfaces . . . . 471
B.4.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . 471
B.4.2 Inductive Long Exact Sequences . . . . . . . . . . . . 472
B.4.3 Duality of Cohomology Groups . . . . . . . . . . . . . 474

Appendix C Ind-Pro Topologies in Arithmetic Dimension Two 477

C.1 Ind-Pro Topology on Adelic Spaces . . . . . . . . . . . . . . . 477
C.1.1 Ind-Pro Topology Spaces and Their Duals . . . . . . . 477
C.1.2 Adelic Spaces and Their Ind-Pro Topologies . . . . . . 481
C.1.3 Adelic Spaces are Complete . . . . . . . . . . . . . . . 482
December 21, 2017 14:18 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page xxvii

Contents xxvii

C.1.4 Adelic Spaces are Compactly Oriented . . . . . . . . . 485

C.1.5 Dual Adelic Spaces . . . . . . . . . . . . . . . . . . . 485
C.2 Adelic Spaces and Their Duals . . . . . . . . . . . . . . . . . . 486
C.2.1 Continuity of Scalar Products . . . . . . . . . . . . . . 486
C.2.2 Residue Maps are Continuous . . . . . . . . . . . . . . 488
C.2.3 Adelic Spaces are Self-Dual . . . . . . . . . . . . . . . 488

Appendix D Closeness of Adelic Subspaces and Finiteness of

Har1 for Arithmetic Surfaces 491
D.1 Arithmetic Cohomology Groups: A Review . . . . . . . . . . . 491
D.2 Closeness of Adelic Sub-Quotient Spaces . . . . . . . . . . . . 493
D.3 Topological Duality of Arithmetic Cohomology Groups . . . . . 498

Appendix E Central Extensions and Reciprocity Laws for

Arithmetic Surfaces 501
E.1 Algebraic Aspect . . . . . . . . . . . . . . . . . . . . . . . . . 504
E.1.1 Reciprocity Law around Point/Along Vertical Curve . . 504
E.1.2 K2 -Central Extensions . . . . . . . . . . . . . . . . . . 505
E.1.3 Sketch of Proof . . . . . . . . . . . . . . . . . . . . . 507
E.2 Arithmetic Aspect . . . . . . . . . . . . . . . . . . . . . . . . . 508
E.2.1 Arithmetic Adelic Complex . . . . . . . . . . . . . . . 508
E.2.2 Numerations in Terms of Arakelov Intersection . . . . . 509
E.3 Numerations for Arithmetic Adelic Cohomologies . . . . . . . . 513
E.4 Arithmetic Central Extension . . . . . . . . . . . . . . . . . . . 515
E.5 Splitness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 516
E.6 Reciprocity Law along Horizontal Curve . . . . . . . . . . . . . 517

Bibliography 519

Index 525
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 1

PART 1

Non-Abelian Zeta Functions

Non-abelian zeta functions for number fields are constructed as the integrations
over the moduli spaces of semi-stable lattices together with the integrants involv-
ing a newly developed arithmetic cohomology theory. Like their classical ana-
logues, these new yet genuine zeta functions are well-defined, satisfy the standard
zeta properties, including a weak Riemann Hypothesis. In this part, we first re-
call the basic theory for the lattices over the ring of integers of number fields,
introduce the stability conditions and construct the moduli spaces for these lat-
tices. This makes the integral domains easily accessible. Then, we develop an
arithmetic cohomology theory for arithmetic curves by introducing two types of
cohomology groups. We show that not only the arithmetic counts for these groups
satisfy a duality and an arithmetic Riemann-Roch theorem, but its refined arith-
metic vanishing theorem under the stability condition implies naturally that the
integrations above are well-defined. In addition, all the zeta properties can be
established tautologically in this language. For example, the functional equation
comes from the numerical duality and the arithmetic Riemann-Roch theorem, and
the geometric interpretation of the residues at one of the zeta functions as the
volumes of the moduli spaces of semi-stable lattices with fixed volumes comes
from the construction and the vanishing theorem. To end this part, we express all
the non-abelian zeta functions as the integrations of the (complete) Epstein zeta
functions over some compact moduli spaces of semi-stable lattices.

1
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 3

Chapter 1

Semi-Stable Lattice

1.1 Motivation

By an algebraic number field, we mean a finite field extension of the field of

rational numbers. Let K be an algebraic number field. We denote by OK , resp.
CLK , resp. ∆K , the ring of integers in K, resp. the class group of K, resp. the
absolute value of discriminant of K. It is well known that OK is a Dedekind ring,
i.e. all the fractional ideals of K admit unique factorizations in terms of prime
ideals of OK , and CLK is a finite abelian group. Denote by S , resp. S ∞ , resp. S fin ,
the set of all (inequivalent) normalized places of K, resp. the set of Archimedean
places of K in S , resp. the set of non-Archimedean places of K in S . Easily,
S = S fin t S ∞ . Let r1 and r2 be the number of real places and complex places of
K, respectively.
For each place v ∈ S , we denote by Kv the v-completion of K. Then Kv is a
local field of dimension one. When v ∈ S fin is a finite place, Kv is a finite extension
of the field Q p of p-adic numbers, for a certain prime number p. For this reason,
we call Kv a p-adic number field. On the other hand, if σ ∈ S ∞ , Kσ equals either
R or C. Accordingly, we call σ real or complex.
For a p-adic number field Kv , we denote by Ov , resp. mv , resp. kv its valuation
ring, resp. its maximal ideal, resp. its residue field. Let ν : Kv  Z be the
associated normalized valuation and N : Kv → R the associated norm map. Then
n o n o
Ov := a ∈ Kv : νv (a) ≥ 0 , mv := a ∈ Kv : νv (a) ≥ 1 and kv := Ov mv .


For example, when K = Q, S fin naturally corresponds to p : prime number in Z ,


Q p is the field of p-adic numbers, O p = Z p is the ring of p-adic integers, mv = pZ p
and k p = F p is the finite field with p elements. Quite often, we call Ov the ring
of integers of Kv , since it can also be characterized as the subring consisting of all
integral elements of Kv over Z p for a unique p. From the definition, N(p) = p−1

3
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 4

4 Non-Abelian Zeta Functions

and N(a) = p−ν(a)Nv for a ∈ Kv , where Nv := [kv : F p ]. More generally, let a be a

fractional ideal of K with the prime ideal factorization a = ki=1 pm
Q i
i , where pi are
Y k
prime ideals of OK and mi ∈ Z. We define the norm of a by N(a) := mi |OK /ai |.
i=1
By definition, the idele group IK of K is a restricted product of Kv∗ with respect
to O∗v . That is to say, IK := 0v Kv∗ ⊂ v∈S Kv∗ , consisting of elements (av )v∈S
Q Q
satisfying the condition that av ∈ O∗v , for all but finite many v ∈ S fin . With respect
to the component-wise multiplication, IK becomes an abelian group. This makes
IK a topological group induced from the open sets UA := v∈A Kv∗ × v<A O∗K ,
Q Q
where A runs over all finite sets of places containing S ∞ .
Denote by N : IK → R≥0 , (av )v 7→ v∈S N(av ) the norm map. Here, for
Q
σ ∈ S ∞ , we set N(aσ ) = kaσ kσ = |aσ |Nσ with Nσ := [Kσ : R]. In parallel,
we define the (additive) degree map on ideles deg : IK → R by the composition
N log
IK → R∗+ → R.
For later use, we introduce a few canonical subgroups of IK by
I0K := a = (av ) ∈ IK : deg(a) = 0 ,


K ∗ := a = (α) ∈ I0K : av = α ∈ K ∗ ∀v ∈ S ,


U K := a = (av ) ∈ I0K : |av |v = 1 ∀v ∈ S ,


(1.1)
IK,fin := a = (av ) ∈ IK : av = 1 ∀v ∈ S ∞ ,


IK,∞ := a = (av ) ∈ IK : av = 1 ∀v ∈ S fin .



Set U K,fin := U K ∩ IK,fin . In addition, let Pic(OK ) be the (Arakelov) Picard group of
K, defined as the abelian group of isometric classes invertible OK -lattices and let
Pic0 (OK ) be the subgroups consisting of degree zero classes. Then, with respect
to the natural locally compact (ind-pro) topology on IK , we have

Proposition 1.1. ( [66]) Let K be a number field.

(1) K ∗ is a discrete subgroup of IK and the quotient group I0K /K ∗ is compact.

(2) IK = IK,fin × IK,∞ ' I0K × R∗+ .
(3) The natural inclusion U K ,→ IK , resp. U K,fin ,→ IK,fin is compact, resp. both
open and compact. Moreover, they induce an isomorphism
I : K ∗ \IK,fin U K,fin '

CLK
a = [(av )] 7→ v∈S fin mνvv (av )
Q

such that N(a) = N(I(a))−1 . In particular, the class group CLK is finite.
(4) Pic(OK ) of K, resp. Pic0 (K), is isomorphic to U K,fin \IK /K ∗ , resp.
U K,fin \I0K /K ∗ . In particular, the degree and the norm map factor through
Pic(OK ) with Pic0 (K) taking values one and zero, respectively.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 5

Semi-Stable Lattice 5

We may write an idele a ∈ IK as a = (afin .a∞ ) with afin ∈ IK,fin and a∞ ∈ IK,∞ .
In particular, if dµ(a) denotes the normalized Haar measure on IK , induced by
vol(Ov ) = 1 for all v ∈ S fin , and the canonical Euclidean, resp. Hermitian, metric
on R, resp. on C, (see e.g. [118]), we obtain a measure dµ(a) on IK and hence on
Pic(OK ) via dµ(a) = dµ(afin ) · dµ(a∞ ).

Definition 1.1. Let K be a number field. The (complete) Dedekind zeta function
of K is defined by
s
 s
 s r1  r2 X
ζK (s) := ∆K2 2π− 2 Γ
b (2π)−s Γ(s) N(I)−s <(s) > 1. (1.2)
2 0,I E O K

Z ∞OK -ideals of K, and Γ(s) is the standard

Here I runs over all non-zero integral
dx
Gamma function defined by Γ(s) := x s e−x .
0 x
Our first purpose of this book is to construct a non-abelian analogue for the
Dedekind zeta function of F. As a starting point, we give an abelian interpretation
of these zeta functions following Iwasawa [45] and Tate [113].
For an idele (av ) ∈ IK , we let L be the associated element in Pic(OK ). We
define the 0-th cohomology group of L−1 by
H 0 (K, L−1 ) := α ∈ K ∗ : αav ∈ Ov ∀v ∈ S fin = I(L−1 ),

(1.3)
and define arithmetic 0-th cohomology for the idele class of L−1 by
 
 X  X X 
h0 (K, L−1 ) := log  exp − π |gv α|2 − 2π |gv α|2  .

(1.4)
α∈H 0 (K,L−1 )r{0} v:R v:C

In addition, denote by wK the number of units of K.

Proposition 1.2. Let K be a number field. Then, up to a constant factor

1
depending only on units of K,
wK vol(U K,fin )
s
Z  0 
ζK (s) = ∆K
b 2
eh (K,L) − 1 N(L)−s dµ(L) <(s) > 1. (1.5)
Pic(OK )

Proof. Introduce a characteristic function e(afin ) and a weight function e(a∞ ) by

  
1, if I(afin ) ⊆ OK ,
e(afin ) :=  and e(a∞ ) := exp −π
  X 2 X
|av |  .
 2

 av − 2π
0, if I(afin ) * OK , v:R v:C

They induce a counting function e(a) on IK defined by

e (afin .a∞ ) := e(afin ) · e(a∞ ) ∀afin ∈ IK,fin , ∀a∞ ∈ IK,∞ .


(1.6)
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 6

6 Non-Abelian Zeta Functions

In terms of these functions, for s ∈ C with <(s) > 1,

s
 s
 s r1  r2 X
ζK (s) = ∆K2 2π− 2 Γ
b (2π)−s Γ(s) N(I)−s
2 0,IEOK
Z  
s X Y  X 2 X Y ∗
=∆ −s s
 2
2
K N(I) |tv |v exp −π tv − 2π |tv |  d tv
0,I⊂OK tv ∈Kv ,v∈S ∞ v v:R v:C v
s
Z
N(a∞ ) s e(a∞ )dµ(a∞ )
X
= ∆K 2
N(I)−s ·
0,I⊂OK IK,∞
Z Z !
s 1
= ∆K ·
2
N(afin ) s e(afin )dµ(afin ) · N(a∞ ) s e(a∞ )dµ(a∞ )
vol(U K,fin ) IK,fin IK,∞
s
∆K2 Z
= N(afin )N(a∞ ) s e(afin ) e(a∞ ) dµ(afin )dµ(a∞ ) .




vol(U K,fin ) IK,fin ×IK,∞
−s 
Z
ζK (s) = N(a) s e(a)dµ(a). Moreover,K ∗ ,→ IK

This implies that vol(U K,fin ) ∆K2 b
I
e(a∞ ), if I(afin ) E OK ,

is discrete by Proposition 1.1(1). Thus, by e(afin a∞ ) := 


0,
 otherwise,
and the product formula,
 −s
 Z Z 
vol(U K,fin )∆K2 · bζK (s) = N(αa) s e(αa)dα dµ([a])
IK /K ∗ K∗
Z X  Z X 
= e(αa) N([a])s dµ([a]) = N([a]) s dµ([a]) e(αafin )e(αa∞ )
IK /K ∗ α∈K ∗ IK /K ∗ α∈K ∗
Z  
e(αa∞ )
X
= N([a]) s dµ([a])
IK /K ∗ α∈K ∗ ,I(αafin )EOK
Z  
e(αa∞ ) .
X
= N([a]) s dµ([a])
IK /K ∗ α∈K ∗ ,αav ⊂Ov ,∀v∈S fin
Here, the notation I E OK means that I ⊆ OK is an OK -ideal. Therefore,
 
−s
ζK (s)
vol(U K,fin ) · ∆K 2 · b
Z
1 X  X X 
= exp − π |gv α|2 − 2π |gv α|2 N(L) s dµ(L)
Pic(OK ) wK α∈H 0 (K,L−1 )r{0} v:R v:C
Z Z
1  h0 (K,L−1 )  1  0 
= e − 1 N(L) s dµ(L) = eh (K,L) − 1 N(L)−s dµ(L)
Pic(OK ) wK wK Pic(OK )
Z h0 (K,L)
e − 1 deg(L) −s
 
= e dµ(L).
Pic(OK ) #Aut(L)
Here, the last step, we have used the fact that wK = #Aut(L) with Aut(L) the
automorphism group of the OK -lattice L. 
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 7

Semi-Stable Lattice 7

1.2 Projective Modules

1.2.1 Invertible Modules

Let K be an algebraic number field with OK its ring of integers.

Definition 1.2. Let M be a finitely generated OK -module.

(1) The rank of the OK -module M, denoted by rankOK (M), or simply rank(M), is
defined to be the dimension of the K-linear space MK := M ⊗OK K.
(2) The OK -module M is called projective if there exists an OK -module N such
that M ⊕ N is a free OK -module.

An OK -module is called invertible if its rank is one.

Proposition 1.3. Invertible projective OK -modules are classified as follows.

(1) A fractional ideal of K is a rank one and projective OK -module.

(2) A rank one projective OK -submodule in K is a fractional ideal.
(3) Two fractional ideals of K are rationally equivalent if and only if they are
isomorphic as OK -modules.

Proof. (1) Let a be a fractional ideal of K. By definition, there exists an element

α ∈ K ∗ such that α a ⊆ OK . Since α a ⊗OK K ⊆ OK ⊗OK K = K, we conclude that
αa and hence a are rank one OK -modules.
To prove a is projective, for any surjective morphism π : M → a of OK -
modules, it suffices to construct a section of π. For this, denote by a−1 the inverse
of a as a fractional ideal of K. There exist xi ∈ a and yi ∈ a−1 (i = 1, . . . , n) such
that 1 = ni=1 xi yi . Since π is surjective, for all i = 1, . . . , n, there exists zi ∈ M such
P
that π(zi ) = xi . Accordingly, we obtain a morphism s : a → M, x 7→ ni=1 (xyi )zi .
P
Moreover, for any x ∈ a, π(s(x)) = π( i=1 (xyi )zi ) = i=1 (xyi )π(zi ) = x ni=1 xi yi =
Pn Pn P
x. Hence, s is a section of π.
(2) Let I ⊂ K be a non-zero projective OK -module. From the definition, there
exists an OK -module N such that I ⊕ N ' O⊕m K for some positive integer m. Since
I is finitely generated, without loss of generality, we may and hence will assume
that I ⊆ OK . Denote by ι : I ,→ O⊕m K the induced inclusion, π : OK → I the
⊕m

induced projection, and, for each j = 1, . . . m, the j-th projection p j : O⊕m K → OK

and π j = p j ◦ ι. Fix a ∈ I r {0} and let J := h a−1 π1 (a), . . . , a−1 πm (a) iOK be the
OK -module generated by the a−1 π j (a)’s. Then J is a fractional ideal of K, since
a J = h π1 (a), . . . , πm (a) iOK ⊆ OK , We claim that I · J = OK . For this, we first
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 8

8 Non-Abelian Zeta Functions

prove that 1 ∈ I J. For j = 1, . . . m, let e j be j-th standard basis vector of the

K-linear space K m . Since π j (a) ∈ OK and p j is a morphism of OK -modules, we
have 1 = a−1 π(ι(a)) = a−1 π mj=1 π j (a)e j = mj=1 a−1 π j (a) π(e j ) ∈ I J. This
P
P

implies that OK ⊆ I J. On the other hand, for x ∈ I and j = 1, · · · , m, we have
x (a−1 π j (a)) = a−1 π j (xa) = a−1 (aπ j (x)) = π j (x) ∈ OK , since I ⊆ OK , a ∈ I and the
π j ’s are OK -morphisms. This implies that I · J ⊆ OK . Therefore, I · J = OK and
hence I is a fractional ideal of K.
(3) Let a, b be two fractional ideals of K. Without loss of generality, we as-
sume that a, b ⊆ OK . If a and b are rationally equivalent, there exists α ∈ K ∗ such
that αa = b. Clearly α : a → b defined by x 7→ α x is an OK -isomorphism. Con-
versely, if a ' b is an OK -isomorphism, OK ' a−1 b. That is to say, as OK -modules,
a−1 b is generated by a single element, say, α ∈ K ∗ . So α a = b, as required. 

1.2.2 General Projective Modules

There is also a classification for all projective OK -modules.

L r projective OK -module. There exists a fractional

Proposition 1.4. Let P be a rank
ideal a such that P ' O⊕(r−1)
K a.

Proof. We prove this by an induction on the rank r of P. If r = 1, this is proved

in Proposition 1.3. Assume that the statement holds for all projective OK -modules
of rank ≤ r − 1. If P is a projective OK -module of rank r, denote by ai : i =

1, . . . , r ⊂ P be a K-basis of the K-vector space PK . Then the map detr−1,1 : P →
OK a1 ∧ ∧a2 ∧ . . . ∧ ar defined by x 7→ a1 ∧ ∧ . . . ∧ ar−1 ∧ x is a surjective


morphism of OK -module. Since detr−1,1 (P) is a rank one OK -module, there exists
a submodule P1 of rank r − 1 such that P ' P1 ⊕ detr−1,1 (P). Therefore, by the
induction hypothesisLr and Proposition 1.3, there exist fractional ideals a1 , . . . , ar of
K such that P ' i=1 ai .

Lemma 1.1. Let a and b be two fractional ideals of K. As OK -modules,

a ⊕ b ' OK ⊕ a b.

Proof. Without loss of generality, we may assume that both a and b are integral.
From the short exact sequences of OK -modules
0 → a ∩ b → a ⊕ b → a + b → 0, (1.7)
we have a ⊕ b ' (a + b) ⊕ (a ∩ b). If a and b are co-prime in OK , then
a + b ' OK , and a ⊕ b ' OK ⊕ a b. We are done. For general a and b,
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 9

Semi-Stable Lattice 9

from the unique factorization property for integral ideals of the Dedekind do-
main OK , there exist α, β ∈ K such that α a and β b are co-prime. Therefore,
a ⊕ b ' α a ⊕ β b ' OK ⊕ (α a) (β b) ' OK ⊕ a b. 

Applying this lemma repeatedly,

P ' ⊕ri=1 ai ' OK ⊕ a1 a2 ⊕ ⊕ri=3 ai ) ' . . . ' O⊕r−1
K ⊕ (a1 a2 . . . ar ).
In particular, det(P) ' a1 a2 . . . ar . 

For r ∈ Z>0 and a ⊂ K a fractional ideal, let Pr:a := O⊕(r−1) K ⊕ a. By Proposi-

tion 1.3, Pr:a is a projective OK -module of rank r. Conversely, by Proposition 1.4,
every rank r projective OK -module is isomorphic to Pr:a for a certain fractional
ideal a of K. In addition, by Proposition 1.3(3), Pr:a ' Pr:b if and only if a is
rationally equivalent to b. Fix the integral
n OK -idealso a1 , . . . , ah satisfying
CLK = [a1 ], . . . , [ah ] . (1.8)
Then what we just said proves the following

Corollary 1.1. Every projective OK -module of rank r is, up to isomorphism, in

the form Pr:ai = O⊕(r−1)
K ⊕ ai for one and only one i (1 ≤ i ≤ h).

1.3 Stability of Lattices

1.3.1 Lattices

Definition 1.3. An OK -lattice Λ of rank r is defined to be a pair (P, ρ) consisting

of

(1) a projective OK -module P = P(Λ) of rank r, and

(2) a collection ρ = {ρσ }σ∈S∞ of metrics ρσ on Vσ := P⊗OK Kσ for all σ ∈ S ∞ such
that when σ ∈ S ∞ is real, resp. complex, ρσ is Euclidean, resp. Hermitian.

For later use, we denote by rankOK (Λ) the rank of the OK -module P and denote
by rankZ (Λ) the rank of the Z-module P. Since an OK -ideal is of rank n as a Z-
module, where n = r1 + 2r2 = [K : Q] is the extension degree of K over Q,
rankZ (Λ) = rankOK (Λ) · rankZ (OK ) = rank(Λ) · n. (1.9)
Accordingly, V(P) := P ⊗Z R = σ∈S∞ V(P)σ is an n-dimensional real space.
Q
Moreover, ρ induces a natural metric on V(P), which we also denote by ρ by an
abuse of notations.

Definition 1.4. An isometry ϕ : (P1 , ρ1 )  (P2 , ρ2 ) of OK -lattices is defined to

be an isomorphism ϕ : P1 ' P2 of OK -modules such that ρ1 coincides with the
φ-pullback ϕ∗ ρ2 of ρ2 .
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 10

10 Non-Abelian Zeta Functions

1.3.2 Covolumes of Lattices

There is a procedure called restriction of scalars, which naturally associates a Z-

lattice to an OK -lattice. To explain this, let Λ = (P, ρ) be an OK -lattice of rank r.
On V(P) = P ⊗Z R = σ∈S∞ Vσ , we introduce a new inner product:
Q

X X
hu, wi∞ := huσ , wσ iσ + <(huσ , wσ iσ ) ∀u, w ∈ V(P). (1.10)
σ:R σ:C
We denote by ρZ the induced metric on V(P).

Definition 1.5. Let Λ = (P, ρ) be an OK -lattice.

(1) The restriction of scalars from K to Q of Λ, denoted by ResK/Q Λ, is defined

to be the Z-lattice (P, ρZ ).
(2) The Lebesgue (co)volume of Λ, denoted by volLeb (Λ), is defined to be the
covolume of the full rank Z-lattice ResK/Q Λ of the metrized space (V(P), ρZ ).

By definition, if {ai } is a Z-basis of ResK/Q Λ and {e j } is an orthonormal basis

of V(P) with respect to the inner product h·, ·i∞ , then volLeb (Λ) = | det hai , e j i|. In
particular, for a Z-module Λ, if rankZ Λ = 1, volLeb (Λ) is the length of its genera-
tor, and, if rankZ Λ = 2, volLeb (Λ) is the area of its fundamental parallelepiped.

Example 1.1. Let a be a fractional ideal of K. Induced from the Minkowski

embedding K ,→ σ∈S∞ Kσ , we obtain a natural inclusion a ,→ K ,→ σ∈S∞ Kσ .
Q Q
Denote by ρLeb the metric on σ∈S∞ Kσ induced by the inner product defined in
Q
(1.10), and define a := (a, ρLeb ). It is well known that, see e.g. Lemma 2 in § V.2
of [66], for the invertible OK -lattice a,
 p 
volLeb a = 2−r2 · N(a) · ∆K .

To eliminate the factor 2−r2 , we introduce

Definition 1.6. Let Λ be an OK -lattice. The canonical volume of Λ is defined

to be the modification of the Lebesgue (co)volume of Λ by the factor 2r2 rankOK (Λ) .
That is to say, if we denote by volcan (Λ), or simply vol(Λ), the canonical volume
of Λ, then
volcan (Λ) = 2r2 rankOK (Λ) volLeb (Λ).

Example 1.2. By Example 1.1,

p
vol a = N(a) · ∆K .


(1.11)

√
In particular, when a = OK , we have vol OK = ∆K .
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 11

Semi-Stable Lattice 11

The canonical volume is theoretically ‘correct’. To explain this, recall that

the Arakelov-Euler characteristic χ(OK , Λ) of an OK -lattice Λ is defined to be the
opposite of the log of the canonical volume, that is,
χ(OK , Λ) := − log vol(Λ) .


(1.12)
Proposition 1.5 (Arakelov Riemann-Roch Theorem). Let Λ be a OK -lattice of
rank r. Denote by degar (Λ) the Arakelov degree of Λ. Then
r
χ(OK , Λ) = degar (Λ) − log ∆K . (1.13)
2
We will give a refinement of this Arakelov-Riemann-Roch theorem in Chapter
2, based on the Fourier analysis over one dimensional adelic spaces in [113].

1.3.3 Stability of Lattices

Let Λ = (P, ρ) be an OK -lattice of rank r. For a projective sub OK -module P1 ⊆ P

of rank r1 , there is a natural embedding V(P1 ) ,→ V(P). We denote by ρ1 := ρ|V(P1 )
the collection of metrics obtained from the restrictions of the metrics ρσ on V(P)σ
to its subspace V(P1 )σ . The resulting OK -lattice Λ1 := (P1 , ρ1 ) of rank r1 is called
the OK -lattice induced from P1 . For simplicity, we write Λ1 =: Λ ∩ P1 and use
Λ1 ≤ Λ to indicate the above relation between OK -lattices Λ1 and Λ.
On the other hand, if P/P1 is projective, for σ ∈ S ∞ , there exists a natu-
ral orthogonal projection πσ : V(P)σ → V(P1 )⊥σ , where V(P1 )⊥σ denotes the or-
thogonal complement of V(P1 )σ in V(P)σ . This induces a natural isomorphism
(P/P1 ) ⊗OK Kσ ' V(P1 )⊥σ , and hence a natural OK -lattice structure on P/P1 , which
we denote by Λ Λ ∩ P1 .


Definition 1.7. Let Λ = (P, ρ) be an OK -lattice and let P1 be a projective sub

OK -module of P.

(1) Λ ∩ P1 is called a sub OK -lattice of Λ if P/P1 is a projective OK -module.

(2) If Λ∩P1 is a sub OK -lattice of Λ, the OK -lattice Λ Λ∩P1 is called the quotient

of Λ by Λ ∩ P1 .
(3) An OK -lattice Λ1 is called a sub OK -lattice of Λ if there exists an isometry
Λ1  Λ ∩ P1 for some projective sub OK -module P1 of P.

We are now ready to introduce the following main

Definition 1.8. An OK -lattice Λ is called semi-stable (resp. stable), if, for all sub
OK -lattices Λ1 of Λ,
vol(Λ1 )rankOK (Λ) ≥ (resp. >) vol(Λ)rankOK (Λ1 ) . (1.14)
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 12

12 Non-Abelian Zeta Functions

Remark 1.1. (1) This definition of stability for lattices is motivated by Mum-
ford stability for vector bundles over integral, regular, projective curves defined
over algebraic closed fields. Indeed, by the Arakelov-Riemann-Roch theorem, i.e.
Proposition 1.5, the inequality (1.14) is equivalent to
degar (Λ1 ) degar (Λ)
≤ or equivalently µ(Λ1 ) ≤ µ(Λ) ∀Λ1 ≤ Λ (1.15)
rankOK(Λ1 ) rankOK(Λ)
where, by definition, for an OK -lattice Λ, its µ-invariant is defined by
degar (Λ)
µ(Λ) := . (1.16)
rankOK(Λ)
In this form, (11.2) is the same as the µ-semi-stability condition of Mumford for
vector bundles over curves.
(2) The main reason for Mumford to introduce the stability condition is to con-
struct a projective moduli spaces for geometric objects. In the case of vector bun-
dles over a proper regular integral curve over algebraic closed field, the space of
isomorphic classes of vector bundles of rank r and degree d is not even Hausdorff.
However, if we consider only stable bundles, their moduli space becomes quasi-
projective. In addition, by adding the so-called S-classes of semi-stable bundles,
we obtain a projective moduli space.
(3) Stability condition for lattices was introduced when we constructed the non-
abelian zeta function for number fields in 1999. See, however, [104, 105].

Similar to the stability of vector bundles, we have the following elementary:

Lemma 1.2. Let Λ be an OK -lattice. The following statements are equivalent.

(1) Λ is semi-stable.
(2) For all proper sub OK -lattices Λ1 of Λ,
volLeb (Λ1 )rankOK (Λ) ≥ volLeb (Λ)rankOK (Λ1 ) .
(3) For all quotient OK -lattices Λ2 of Λ,
volLeb (Λ)rankOK (Λ2 ) ≥ volLeb (Λ2 )rankOK (Λ) .

Consequently, when verifying stability for an OK -lattice, either canonical or

Lebesgue volumes can be used.

Proof. The equivalence between (2) and (3) is a direct consequence of

Definition 1.7. Indeed, by Definition 1.7, there is a canonical isometry Λ 
Λ1 ⊕ Λ2 , when Λ1 , resp. Λ2 , is a sub (resp. a quotient) OK -lattice of Λ and
Λ2 (resp. Λ1 ) is the associated quotient (resp. sub-lattice).
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 13

Semi-Stable Lattice 13

To prove the equivalence between (1) and (3), it suffices to prove

Claim 1.1. Let Λ = (P, ρ) be an OK -lattice. The following two are equivalent.

(1) Λ is semi-stable.
(4) For all projective sub OK -modules P1 of P,
vol(Λ ∩ P1 )rankOK (Λ) ≥ vol(Λ)rankOK (Λ∩P1 ) .

Proof. (4) implies (1) is clear. The opposite direction is slightly complicated.
The point is that while Λ ∩ P1 is an OK -lattice, it may not be a sub OK -lattice
of Λ. However, it is well known that, for P1 ⊆ P, there exists a projective sub
OK -module P1env , the envelop of P1 in P, such that P1 ⊆ P1env is a finite index sub
OK -module and the quotient P/P1env is projective. In addition, by Definition 1.5,
we certainly have vol(Λ ∩ P1 ) = vol(Λ ∩ P1env ) [P1env : P1 ]. Therefore
env
vol(Λ ∩ P1env )rankOK (Λ) ≥ vol(Λ)rankOK (Λ∩P1 ) ,
implies (4). This proves the claim and hence the lemma. 

1.3.4 Canonical Filtration

Let Λ = (P, ρ) be an OK -module of rank r. For a fixed constant t ∈ R∗+ , let tρ be

the metric on P ⊗Q R obtained by multiplying the metric ρ by t. Denote by tΛ the
OK -lattice by (P, tρ). Obviously, vol(tΛ) = trn · vol(Λ), and, if P1 is a non-trivial
OK -module of P, vol((tΛ) ∩ P1 ) = tr1 n · vol(Λ ∩ P1 ). Moreover, an easy calculation
implies that
vol(Λ ∩ P1 )rankOK (Λ) ≥ vol(Λ)rankOK (Λ∩P1 )
(1.17)
⇐⇒ vol(tΛ ∩ P1 )rankOK (tΛ) ≥ vol(tΛ)rankOK (tΛ∩P1 ) .
Therefore, when studying the (semi-)stability for OK -lattices, instead of working
over all OK -lattices, it suffices to study on OK -lattices of a fixed volume, say, one.
For each projective sub OK -module P1 of P, assign to the sub OK -lattice
Λ1 = Λ ∩ P1 a point Q(Λ1 ) := (rankOK Λ1 , − log vol(Λ1 )) in the standard xy-
plane. For example, when Λ1 = 0, Q(0) = (0, 0), and, when Λ1 = Λ,
Q(Λ) = (r, − log vol(Λ)). Consequently, for a non-trivial sub OK -lattice Λ1 , we
obtain two line segments, namely,

(a) one, denoted by L(0, Λ1 ), of slope µ(Λ1 )− 21 log ∆K , connecting Q(0) to Q(Λ1 ),
(b) the other, denoted by L(Λ1 , Λ), of slope µ(Λ/Λ1 )− 21 log ∆K , connecting Q(Λ1 )
to Q(Λ).
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 14

14 Non-Abelian Zeta Functions

This, together with the line segment L(0, Λ), forms a triangle ∆(0, Λ1 , Λ) in the
xy-plane. In this language, the stability condition is naturally translated.

Corollary 1.2. For the OK -lattice, the following conditions are equivalent.

(1) Λ = (P, ρ) is semi-stable.

(2) For all projective sub OK -module P1 of P, the slope of L(0, Λ ∩ P1 ) is less
than or equal to the slope of L(0, Λ).
(3) For all projective sub OK -module P1 of P, the slope of L(0, Λ) is less than or
equal to the slope of L(Λ ∩ P1 , Λ).
(4) For all projective sub OK -module P1 of P, the triangle ∆(0, Λ1 , Λ) is below
the line L(0, Λ).

Therefore, for a lattice Λ of volume one, Λ is semi-stable if and only

∆(0, Λ1 , Λ) is a triangle under the x-axis for all sub OK -lattices Λ1 of Λ. To treat
more general cases, we first note that, among OK -lattices Λ1 of a fixed rank, the
smaller the volume of Λ1 is, the larger the slope of L(0, Λ1 ) is.

Lemma 1.3. Let Λ be a fixed OK -lattice. For any constant c ∈ R∗+ , there exist
only finitely many sub OK -lattices Λ1 of Λ such that vol(Λ1 ) ≤ c. In particular,
the subset vol(Λ1 ) : Λ1 ⊆ Λ of R≥0 is discrete and bounded from below.


Proof. By the restriction of scalars, we may reduce the proof to Z-lattices. Fur-
thermore, as we did in the proof of Lemma 1.1, passing to the enveloping sub
Z-modules, it suffices to treat the cases Λ1 ≤ Λ. For this, we use an induction on
r. Let r1 be the rank of Λ1 = Λ ∩ P1 . If r1 = 1, the Λ1 ’s are lattices vectors in Λ
and hence form a discrete subset in the metrized space ΛR := Λ ⊗Z R. Obviously,
this discrete subset admits only finitely many intersections with the compact ball
of radius c, centered at the origin. Hence, when r1 = 1 or r = 1, we are done. As-
sume now the same holds for all Z-lattices of rank ≤ r − 1. To treat general r1 , we
use the map Λ 7→ ∧r1 Λ, which sends Λ1 to ∧r1 Λ1 . By the induction hypothesis,
we may assume that ∧r1 Λ1 is not degenerate. That is, det Λ1 := L is a Z-lattice of
rank one. By the case r1 = 1 proved above, there are only finitely many L’s such
that vol(L) ≤ c. This then completes the proof, since vol(det Λ1 ) = vol(Λ1 ) and,
for a fixed L, there are only finitely many Λ1 such that det Λ1 = L. 

Hence, within Λ, there exists a unique sub OK -lattice Λ1 such that, not only

(a) the slope of L(0, Λ1 ) attains the maximum at a sub OK -lattices Λ1 of Λ, but
(b) the rank of Λ1 is also maximum at a sub OK -lattices Λ01 satisfying that the
slope of L(0, Λ01 ) is equal to the slope of L(0, Λ1 ).
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 15

Semi-Stable Lattice 15

All these, together with Corollary 1.2, imply, in addition,

(c) Λ1 itself is a semi-stable OF -lattice and µ(Λ1 ) > µ(Λ/Λ1 ).

We call Λ1 the (first) canonical sub OK -lattice of Λ and denote by it by Λ1 .

Similarly, by applying the same argument to the OK -lattice Λ/Λ1 , we obtain, up
to isometry, a unique sub OK -lattice Λ2 of Λ such that

(d) Λ2 Λ1 is the (first) canonical sub OK -bundle of Λ/Λ1 . In particular, Λ2 Λ1 is

 
a semi-stable OF -lattice and µ Λ2 Λ1 > µ(Λ/Λ2 ).


We call Λ2 the second canonical sub OK -lattice of Λ. Repeating this process, we

then obtain the existence part of the following

Proposition 1.6 (Canonical Filtration). Let Λ be an OK -lattice. There exists a

unique filtration of sub OK -lattices of Λ
Λ∗ : 0 = Λ0 ⊂ Λ1 ⊂ Λ2 ⊂ · · · ⊂ Λk = Λ (1.18)
such that the following conditions are satisfied. For all i = 1, . . . , k − 1,

(1) The i-th successive graded OK -lattice gri (Λ) := Λi+1 Λi is semi-stable.



(2) Not only the slope of L 0, gri (Λ) attains the maximum among all sub OK -
lattices of Λ Λi , but the rank of Λi+1 attains the maximum among all sub

OK -lattices Λi+1 satisfying the same conditions.
(3) µ gri (Λ) > µ gri+1 (Λ) . Or equivalently,

vol(Λi+1 /Λi ) rank(Λi /Λi−1 ) > vol(Λi /Λi−1 ) rank(Λi+1 /Λi ) .

Proof. It suffices to prove uniqueness. For this, we make a preparation.

Lemma 1.4. Let Λ1 and Λ2 be two sub OK -lattices of Λ. Then

vol(Λ1 ∩ Λ2 ) vol(Λ1 + Λ2 ) ≤ vol(Λ1 ) vol(Λ2 ). (1.19)

Proof. By the restriction of scalars, we may simply work over Z. Let P1 and P2
be the projective Z-modules associated to Λ1 and Λ2 , respectively. Since Λ1 and
Λ2 are sub OK -lattices of Λ, Λ1 ∩Λ2 = Λ1 ∩ P2 = Λ2 ∩ P1 and Λ1 = (Λ1 +Λ2 )∩ P1
are sub OK -lattices of Λ2 and Λ1 + Λ2 , respectively. Here, as usual, Λ1 + Λ2 is a
sub OK -lattice of Λ generated by Λ1 and Λ2 . Consequently, it suffices to prove
vol(Λ1 + Λ2 /Λ1 ) ≤ vol(Λ2 /Λ1 ∩ Λ2 ). (1.20)
Obviously, there exists a full filtration P1 ∩ P2 = P12,0 ⊂ P12,1 ⊂ · · · ⊂
P12,k = P2 , such that, as Z-modules, ((P12,i + P1 )/P1 ) ((P12,i−1 + P1 )/P1 ) '

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 16

16 Non-Abelian Zeta Functions

(P12,i /P1 ∩ P2 ) (P12,i−1 /P1 ∩ P2 ) are of relative rank one. In terms of lattices,


since (P12,i + (P12,i−1 + P1 )) (P12,i−1 + P1 ) = (P12,i + P1 ) (P12,i−1 + P1 ) and
 
P12,i P12,i−1 = P12,i (P12,i ∩ (P12,i−1 + P1 ), it suffices to verify that
 

P12,i + P12,i−1 + P1 P12,i−1 + P1 ' P12,i P12,i ∩ P12,i−1 + P1 .

 

Here ∗ denotes the lattice induced by ∗. These may be viewed as special cases of
(1.20) if we set Λ1 = P12,i−1 + P1 and Λ2 = P12,i . Hence it suffices to prove (1.20)
in the case of relative rank one. For this, replacing Λ1 and Λ2 by Λ1 Λ1 ∩ Λ2 and


Λ2 Λ1 ∩Λ2 , respectively, we may further assume that Λ1 ∩Λ2 = 0. Accordingly,


let s be a generator of Λ2 and π1 be the orthogonal projection to Λ1 ⊗Z Q. Then
vol(Λ2 ) = |s| ≥ |π1 (s)| = vol Λ2 Λ1 ∩ Λ2 , as desired.





Back to the proof of the proposition. Let Λ1 and Λ01 be two canonical sub OK -
lattices of Λ. By Lemma 1.4, Λ1 + Λ01 is also canonical. Thus, by the maximality
of the rank, Λ1 +Λ01 = Λ1 = Λ01 . This implies that the first canonical sub OK -lattice
of Λ, and hence all levels canonical sub OK -lattices of Λ, are unique. 

Definition 1.9. Let Λ be an OK -lattice.

(1) The filtration (1.18) is called the canonical filtration of Λ.

(2) Let Λ∗ : 0 = Λ0 ⊂ Λ1 ⊂ Λ2 ⊂ · · · ⊂ Λn = Λ be a filtration of sub OK -lattices
of Λ. Its associated polygon pΛ∗ is defined to be the polygon characterized by
the conditions that, for all i = 1, . . . , n,
(a) pΛ∗ is affine in the interval [rankOK Λi−1 , rankOK Λi ],
(b) pΛ∗ (0) = 0 and pΛ∗ rankOK Λi = − log vol Λi .




(3) pΛ∗ is called the canonical polygon of Λ.

Let p and q be two polygons defined over [a, b]. We write that p ≥ q if
p(x) ≥ q(x) for all x ∈ [a, b]. Then a translation of Proposition 1.6 and its proof in
the language of polygons gives the following

Corollary 1.3. Let Λ be an OK -lattice. The canonical polygon of Λ is the maximal

one among all polygons associated to filtrations of Λ. In addition, it is convex.

Similar to vector bundles over curves, there is a notation of Jordan-Hölder

Filtration for OK -lattices.

Proposition 1.7 (Jordan-Hölder Filtration). Let Λ be a semi-stable O-lattice.

Then there exists a filtration of proper sub OK -lattices,
0 = Λl+1 ⊂ Λl ⊂ · · · ⊂ Λ0 = Λ
such that
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 17

Semi-Stable Lattice 17

(a) For all j = 0, . . . , l, Λ j /Λ j+1 is stable. And

(b) For all j = 1, . . . , l, vol(Λ j /Λ j+1 ) rank(Λ /Λ ) = vol(Λ j−1 /Λ j ) rank(Λ /Λ ) .

j−1 j
j j+1

Ll
j=0 Λ /Λ
j j+1
Moreover, up to isometry, the graded lattice Gr(Λ) := is uniquely
determined by Λ.

Even the Jordan-Hölder filtrations play a key role in introducing S-classes for
vector bundles, in this book, it will not be used. We leave its proof to the reader.

1.4 Volume of Lattice: Special Linear Group

In this section, we construct a natural parameter space for OK -lattices of rank r.

1.4.1 Metrics of Lattices

Let Λ = (P, ρ) be an OK -lattice of rank r, with P a projective OK -module of rank

r, and ρ a metric structure on V(P) = P ⊗Z R = σ∈S∞ Vσ . By Proposition 1.1, we
Q
⊕(r−1)
identify P with one of the Pi := Pr:ai := OK ⊕ ai for a certain 1 ≤ i ≤ h.
Induced from the Minkowski embedding κR : K ,→ KR := Rr1 × Cr2 of K, we
obtain a natural inclusion P = O⊕(r−1)
K ⊕ ai ,→ K r , and hence an embedding
κP : P ,→ Rr r1 × Cr r2 .



(1.21)
Consequently, ρ may simply be understood as metrics on the space R × Cr r2 ,
r
r1

or equivalently, a collection of Euclidean metrics on Rr , r1 of them, and of Her-
mitian metrics on Cr , r2 of them.

Lemma 1.5.

(1) For g ∈ GL(r, R), resp. g ∈ GL(r, C), there is an associated Euclidean metric
ρ(g) on Rr , resp. Hermitian metric ρ(g) on Cr , defined by
hx, yiρ(g) := hg x, g yi ∀x, y ∈ Rr , resp. ∀x, y ∈ Cr .
Here on the right hand side, h·, ·i denotes the canonical pairing.
(2) Each Euclidean metric on Rr , resp. Hermitian metric on Cr , is determined by
h·, ·iρ(g) for some g ∈ GL(r, R), resp. g ∈ GL(r, C).
(3) Let g, g0 ∈ GL(r, R), resp. ∈ GL(r, C). The following two are equivalent.
(a) On Rr , resp. on Cr , h·, ·iρ(g) ≡ h·, ·iρ(g0 ) . n o
(b) There is an orthogonal matrix O ∈ O(n) = O ∈ GL(r, R) OOt = Ir , resp. a
n o
unitary matrix U ∈ U(n) = U ∈ GL(r, C) UU t = Ir , such that g0 = g · O,
resp. g0 = g · U.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 18

18 Non-Abelian Zeta Functions

That is to say, the metric structures on Rr , resp. on Cr , are parametrized by the

quotient space GL(r, R)/O(n), resp. GL(r, C)/U(n). Therefore, all metric struc-
tures on Rr r1 × Cr r2 are parametrized by the space

GL(r, R) O(n) r1 × GL(r, C) U(n) r2 .




This then completes a proof of the following:

Proposition 1.8. The metrics on all OK -lattices of rank r are parametrized by

G
GL(r, R) O(n) r1 × GL(r, C) U(n) r2 .



a∈{a1 ,...,ah }

1.4.2 Special Metrics of Lattices

Let Λ = (P, (ρσ )) be an OK -lattice of rank r. Since, for t > 0 vol(tΛ) = trn vol(Λ),
where n := [K : Q], we conclude that

(1) The parametrizing space of metric classes over OK -lattices of rank r natu-
rally decomposes into a disjoint union of the subspaces indexed by T ∈ R∗+ ,
parametrizing the metric classes over OK -lattices of rank r and volume T .
(2) For fixed T , T 0 ∈ R∗+ , the above scaling induces a natural bijection between
the sub collection associated to T and the sub collection associated to T 0 .

For this reason, in the sequel, we concentrate on OK -lattices with a fixed volume.
Fix a OK -ideal a = ai , let Λ0 = (P0 , ρ0 ) be the OK -lattice of rank r such that
(i) P0 = O⊕(r−1)X ⊕ a, and (ii) ρ0 is the metric on V(P0 ) corresponding to the image
of (Ir , . . . , Ir ) in  GL(r, R) O(n) r1 × GL(r, C) U(n) r2 . Then, by Example 1.1,



r−1 1 r
vol(Λ0 ) = ∆K2 · N(a) · ∆K2 = N(a) · ∆K2 . As usual, set then

SO(r) := O ∈ O(n) det O = 1 and SU(r) := U ∈ U(n) det U = 1

 

be the special orthogonal group and the special unitary group, respectively. Then,
by Proposition 1.8, what we have just said proves the following:

Lemma 1.6. Let a be a fixed OK -ideal, and let ρ be a metric on V(O⊕(r−1) ⊕ a)

K
corresponding to an element of SL(r, R) SO(r) r1 × SL(r, C) SU(r) r2 . Then



r
vol O⊕(r−1) ⊕ a, ρ = N(a) ∆K2 .


K
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 19

Semi-Stable Lattice 19

There is a natural procedure to reduce parameter spaces from the general linear
group to the special linear group. To explain this, we give a decomposition of
GL(r, C)/U(n) with respect to a fixed branch of r-th root of unity on C defined by
' '
GL(r, C) −→ SL(r, C) × C∗ −→ SL(r, C) × S1 × R∗+
   
 1  1 det g (1.22)
, | det g|
 
g 7→  pr g, det g 7→  p g,
det g r
det g | det g|

and
!
' 1
U(n) −→ SU(r) × S , 1
U 7→ √r U, det U , (1.23)
det U
where S1 := z ∈ C : |z| = 1 denotes the unit circle in C∗ . As a direct consequence,


GL(r, C) U(n)  SL(r, C) SU(r) × R∗+ .

 

(1.24)
.
The quotient space SL(r, C) SU(r) × R∗+ induced from the spacial linear group


SLr will be used to reparametrize the metric structures on Cr .
pr
Unfortunately, this does not work for GL(r, R)/O(n), simply because p det g
is not well-defined in the reals. For examples, when det g is negative, det g does
not exists in R. Hence (1.22) cannot be applied directly. To remedy this, we
introduce the subgroup

GL+ (r, R) := g ∈ GL(r, R) : det g > 0 and O+ (r) := O ∈ O(n) : det g > 0 .
n o n o

Obviously,

(a) O+ (r) = SO(r) and GL(r, R) O(n)  GL+ (r, R) SO(r).

 
(b) There is an identification
 
+ '  1
GL (r, R) −→ SL(r, R) × R+ , g, det g .
∗

g 7→  pr

det g

This yields a natural identification

GL(r, R) O(n)  SL(r, R) SO(r) × R∗+ ,

 

(1.25)

and hence proves the following

Proposition 1.9. The metric structures of OK -lattices of rank r are parametrized

by
G 
SL(r, R) SO(r) r1 × SL(r, C) SU(r) r2 × (R∗+ )r1 +r2 .



(1.26)
a∈{a1 ,...,ah }
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 20

20 Non-Abelian Zeta Functions

1.5 Automorphisms of Lattices

In the previous section, a parameter space for the metrics of OK -lattices of rank r is
constructed, using classifications of projective OK -modules and metrics on Rr and
Cr . For arithmetic applications, in this section, we construct a moduli space for
the OK -lattices of rank r, by examining the automorphism groups of the associated
projective OK -modules.

1.5.1 General Automorphisms

We use the same notation as before. In addition, we denote by U K be the

unit group of K, and, for an OK -ideal a and P = Pr,a = O⊕(r−1)
K ⊕ a, we set
GL(P) := AutOX (O⊕(r−1)
K ⊕ a) to be the automorphism group of O ⊕(r−1)
K ⊕ a (as
an OK -module).

By Proposition 1.10, the metrics of OK -lattices Λ = (P, ρ) of rank r are

parametrized by the space SL(r, R) SO(r) r1 × SL(r, C) SU(r) r2 × (R∗+ )r1 +r2 .




Therefore, to construct the moduli space for the isometry classes of OK -lattices
Λ = (P, ρ) of rank r, it suffices to study the quotient space

SL(r, R) SO(r) r1 × SL(r, C) SU(r) r2 × (R∗+ )r1 +r2 .

/ 
AutOX (O⊕(r−1)




K ⊕ a)

For this purpose, we first study elements of AutOX (O⊕(r−1) K ⊕ a), viewed as a sub-
group of GL(r, K).
 
Let A ∈ AutOK O⊕(r−1) ⊕ a . Then (i) det A ∈ U K , (ii) A O⊕(r−1) ⊕a ⊆ O⊕(r−1)


K K K ⊕
−1 ⊕(r−1)
⊕(r−1)
a and (iii)
 A O K ⊕ a ⊆ O K ⊕ a. As an element in GL(r, K), we write
A = ai j with ai j ∈ K. In this way, by examining how the entires ai j ’s of A act on
lattice vectors x1 , . . . , xn t of O⊕(r−1)


K ⊕ a, we conclude that

arr , ai j ∈ OK
 


 


 
 −1 
 ⊕(r−1)   ⊕(r−1)    air ∈ a, ar j ∈ a  
⊕ a = GL OK .
 
AutOK OK ⊕ a := (ai j ) ∈ GL(r, K) :
 
i, j = 1, · · · , r − 1

 



 



 

 det(a ) ∈ U
ij K


This implies the following:

Lemma 1.7. The automorphic group of the projective OK -module O⊕(r−1)

K ⊕ a is
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 21

Semi-Stable Lattice 21

determined by
   

  a  

.

 

.

 
 
 

.
  \   
O
 ⊕(r−1)
⊕a = .
 
AutOK OK A ∈ GL(r, K) K  : det A ∈ U K 
 
 
 






 a 





a . . . a−1 OK

  −1  

 

To facilitate ensuing discussion, we define Aut+OK O⊕(r−1)

 
K ⊕ a to be the sub-
 ⊕(r−1) 
group of AutOK OK ⊕ a consisting of elements whose local determinants
at all real places are positive. Since GL(r, R) O(n) ' GL+ (r, R) O+ (r) and
 
O+ (r) = SO(r), weobtain a natural  / identification:
r

AutOK OK ⊕(r−1)
⊕a GL(r, R)/O(n) 1 × GL(r, C)/U(n) r2

y'



Aut+OK OK GL+ (r, R)/O+ (r) r1 × GL(r, C)/U(n) r2 .

 ⊕(r−1)  /


⊕a
Therefore, to construct the moduli space of isometric classes of OK -lattices Λ =
(P, ρ) of rank r, it suffices
/ to studythe quotient space
+
⊕ a) SL(r, R) SO(r) × SL(r, C) SU(r) r2 × (R∗+ )r1 +r2 .
⊕(r−1)
r1 
 
AutOX (OK

1.5.2 Special Automorphisms

From the discussions in § 1.4.2, there are the following natural identifications:
' '
GL(r, R)/O(n) −→ GL+ (r, R)/SO(r) −→ SL(r, R)/SO(r) × R∗+


1
[A+ ] A+ , det A+ ,
 
[A] 7→ 7→ √r
det A+
' 1
'
GL(r, C)/U(n) −→ SL(r, C) × C SU(r) × S −→ SL(r, C)/SU(r) × R∗+



 1 
[A] 7→ [A] 7→ √r A , | det A| .
det A
To go further, we construct a corresponding decomposition for the group
AutOK O⊕(r−1)
K ⊕ a . Define then
U K+ := ε ∈ U K : εσ > 0, ∀σ ∈ S ∞ , real .
n o

This is still not enough, since we cannot take the r-th root of units in K. To over- 
come this, observe that, if ε ∈ U K is a unit, diag(ε, . . . , ε) ∈ AutOK O⊕(r−1) K ⊕a ,
and its determinant εr belongs to
U Kr := εr : ε ∈ U K .


Accordingly, we introduce a subgroup U Kr,+ of U K+ by

U Kr,+ := U Kr ∩ U K+ . (1.27)

Lemma 1.8. The quotient groups U K /U K+ and U K+ U Kr,+ are all finite.

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 22

22 Non-Abelian Zeta Functions

Proof. By Dirichlet’s unit theorem, the image log kU K k of U K (under the natural
logarithm-norm map) is a Z-lattice of rank r1 + r2 − 1 in Rr1 +r2 . Moreover, kU Kr k =
rkU K k, consisting of elements which can be written as the r-powers of elements
in kU K k. Therefore, rankZ kU Kr k = rankZ kU K k. Consequently, up to a finite torsion
subgroup consisting of the roots of unity in K, U Kr is a finite index subgroup of
U K . But U Kr,+ ⊂ U Kr ⊂ U K . This also shows that U Kr,+ is a finite index subgroup of
U K since U K2r ⊆ U Kr,+ . 

With these preparations, we are now ready

 to introduce
 a compatible decom-
position of the automorphism group AutOK O⊕(r−1)
K ⊕ a . We define first
   
SL O⊕(r−1)
K ⊕ a := SL(r, K) ∩ GL O⊕(r−1)
K ⊕a ,

to be the special automorphism

 ⊕(r−1)  group of the OK -lattice O⊕(r−1)
K ⊕ a.
 Then, the 
⊕ a · U K · diag(1, . . . , 1) is a subgroup of GL O⊕(r−1)


product SL OK K ⊕a ,
 
n
and the intersection is SL O⊕(r−1)
K ⊕ a ∩ U K ·diag(1, . . . , 1) = ε·diag(1, . . . , 1) :
εr = 1 . Accordingly, let ε1 , . . . , εµ+ (r,K) ∈ U K+ be a set of complete representatives
o

for the quotient group U K+ U Kr,+ , where µ+ (r, K) denotes


the cardinal number of
U K+ U Kr,+ . Obviously, Ai := diag(1, . . . , 1, εi ) ∈ GL+ O⊕(r−1)
  
K ⊕ a .

Lemma 1.9. The set {A1 , . . . , Aµ+ (r,K) } is a complete

 ⊕(r−1) set of representatives of the
coset Aut+OK O⊕(r−1) r,+
. . . , 1) .


K ⊕ a with respect to SL O K ⊕ a · U K · diag(1,

Proof. For A ∈ GL+ O⊕(r−1) ⊕ a , det A ∈ U K+ . Write det A = εi · εr for a certain

 
K
εr ∈ U Kr,+ and a unique i among 1, . . . , µ+ (r, K)). Obviously, the element defined by
+ ⊕(r−1)

A · A−1i diag(ε −1
, . . . , ε −1
) belongs to GL O ⊕ a and its determinant equals
 K⊕(r−1) 
(εi · ε ) · εi ε = 1. Therefore, A ∈ Ai SL OK
r −1 −r
⊕ a · U K · diag(1, 1) .
r,+

Conversely, if Ai and A j are in the same coset of Aut+OK O⊕(r−1)

 
K ⊕ a with re-
   ⊕(r−1) 
spect to SL O⊕(r−1) ⊕ a · U Kr,+ · diag(1, 1) , then Ai A−1 ⊕ a · U Kr,+ ·


K j ∈ SL OK
diag(1, . . . , 1) . This implies that εi ε−1 j ∈ U K , and hence i = j.
r,+




Accordingly, we obtain a natural decomposition of the automorphism group

Aut+OK O⊕(r−1) ⊕ a n U K+ .
   
K ⊕ a ' SL O⊕(r−1)
K (1.28)

Moreover, for U K+ , we have a decomposition by the group U Kr,+ as follows

µ(r,K)
G
U K+ diag(1, . . . , εi ) × U Kr,+ · diag(1, . . . , 1) .


' (1.29)
i=1
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 23

Semi-Stable Lattice 23

Now, we define kU Kr,+ k := { kεk ∈ R∗+ : ε ∈ U Kr,+ } to be the norm group of U Kr,+ .
Note that this eliminates possible ambiguity caused by the roots of unity contained
in K. As a direct consequence from the above discussion, we have (proved) the
following:

Proposition 1.10. The isometry classes of OK -lattices of rank r are parametrized

by
G 
⊕a SL(r, R) SO(r) r1× SL(r, C) SU(r) r2 × kU Kr,+ k (R∗+ )r1 +r2 .


  
SL O⊕(r−1)

 
K
a∈{a1 ,...,ah }

For simplicity, we denote this space by Mtot

K,r .

1.5.3 Unit Automorphisms

From the previous subsection, to understand moduli space of OK -lattices of rank

r, it suffices to study the following two spaces

(1) The quotient space kU Kr,+ k R∗+ r1 +r2 , and



 
(2) The moduli space SL O⊕(r−1) SL(r, R)/SO(r) r1 × SL(r, C)/SU(r) r2 .




K ⊕ a

In this subsection, we use Dirichlet’s unit theorem, to offer a natural decomposi-

tion for the first (quotient) space kU Kr,+ k R∗+ r1 +r2 .


To start with, we introduce the logarithm map

log k k : (R∗+ )r1 +r2 →  Rr1 +r2 

(x s )σ∈S∞ 7→ log kxσ kσ .
σ∈S ∞

This map transforms the multiplicative space (R∗+ )r1 +r2 into the additive space
Rr1 +r2 isometrically. By Dirichlet’s unit theorem, the image log kU K k of the unit
group U K of K is a full rank Z-lattice of the metrized hyper-plane
 X 
H := (xσ )σ∈S∞ ∈ Rr1 +r2 : [Kσ : R] xσ = 0 .
σ∈S ∞

Consequently, by Lemma 1.9, the image log kU Kr,+ k of the subgroup U Kr,+ is a full
rank sub Z-lattice in H as well. We denote by Dr,+ K a fundamental parallelepiped
of the lattice log kU Kr,+ k in H.
Similarly, let S be the norm-one hypersurface in (R∗+ )r1 +r2 defined by
 Y 
S := (xσ )σ∈S∞ ∈ (R∗+ )r1 +r2 : xσ[Kσ :R] = 1 . (1.30)
s∈S ∞
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 24

24 Non-Abelian Zeta Functions

There is a natural decomposition (R∗+ )r1 +r2 = R∗+ × S. Let Fr,+

K be the pre-image
∗ r1 +r2
of Dr,+
K in (R + ) with respect to log k k. From the definitions, the funda-
r,+
mental
G domain F K cuts the norm-one hyper-surface S into the disjoint union
S= +
η FK . This then proves the following
r r,+
η∈U F

Theorem 1.1. Let Fr,+ r,+

K be the fundamental domain of U K . Then the isometry
classes of OK -lattices of rank r are parametrized by
G   ⊕(r−1) 


Fr,+ SL(r, R) SO(r) r1 × SL(r, C) SU(r) r2 × R∗+ .


K × SL OK ⊕a
a∈{a1 ,...,ah }

This decomposition is very important. Indeed, for a fixed T ∈ R∗+ , the isometry
classes of OK -lattices of rank r and volume T form a natural subspace of that for
all lattices. Denote by Mtot
K,r [T ] the moduli space of isometry classes of OK -lattices
of rank r and volume T .This subspace can be read from the above decomposition
easily, since the final factor R∗+ represents this volume parameter. That is to say,
we have the following:

Theorem 1.2. For T ∈ R∗+ , Mtot

K,r [T ] is isomorphic to the space
G   ⊕(r−1) 


Fr,+ SL(r, R) SO(r) r1 × SL(r, C) SU(r) r2 × {T }.


K × SL OK ⊕a
a∈{a1 ,...,ah }

The volume of Fr,+

K can be calculated. Indeed, being induced from the standard
1 +r2
rY
d(xi ) dxi
Haar measure := on Rr1 +r2 = S×R∗+ , we denote by d∗ (x) the natural
(xi ) i=1
x i
d(xi ) dt
Haar measure on S such that = d∗ (x) × .
(xi ) t

Lemma 1.10. The volume of the fundamental domain Fr,+

K of the unit sub-group
U Kr,+ in S with respect to d∗ (x) is given by
+ +
K ) = µ (r, K) · RK .
vol(Fr,+

Here R+K is the narrow regulator of K (to be defined at the end of the proof below).

Proof. Let F+K be the fundamental domain of the unit sub-group U K+ in S. The log-
d(x)
arithm log k k : (R∗+ )r1 +r2 → Rr1 +r2 = σ∈S∞ R transforms
Q
into the Lebesgue
(x)
measure d(x) on Rr1 +r2 . Hence,

vol(F+K ) = volRr1 +r2 log kFrK × Ik ,

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 25

Semi-Stable Lattice 25

where I := [0, 1] denotes the unit interval. By Dirichlet’s unit theorem, there exist
ε1 , . . . , εr1 +r2 −1 such that U K+ , modulo roots of unity, is generated by them. Thus,
the full rank sub lattice log |F+K × I| is generated by (r1 + r2 − 1) vectors
t
e j := log |σ1 (εi )|σ1 , . . . , log |σr1 (ε j )|σr1,log |σr1 +1 (ε j )|2σr1 +1 , . . . log |σr1 +r2 (ε j )|2σr1 +r2


where 1 ≤ j ≤ r1 + r2 − 1, together with e := (eσ1 , · · · , eσr1 +r2 )t n, a norm one


vector in Rr1 +r2 , where eσi = 1 or 2 depending whether σi is real or complex.
For simplicity, write e j =: e1, j , . . . , ei, j , . . . , er1 +r2 , j for each j. Then the volume


F+K × I and hence the volume vol(F+K ) are given by the absolute value of
 1 
 e1,1 · · · er1 +r2 −1,1 n eσ1  
 · · · · · · ··· · · · 

1  .

det  e1,i · · · er1 +r2 −1,i n eσi 

 
 · · · · · · ··· · · · 
e1,r1 +r2 · · · er1 +r2 −1,r1 +r2 n1 eσr1 +r2 −1
 

Since the ε j ’s are units, adding the first r1 +r2 −1 lines to the last one, all entries of
r1 +r2
1 X
the new line becomes 0 except the last one, which is ep = 1. By definition,
n i=1 i
the absolute value of the determinant of this latest matrix is called the narrow
regulator of K and denoted by R+K . Therefore, we have vol(F+K ) = R+K . This
completes the proof since |U K+ /U Kr,+ | = µ+ (r, K). 

1.6 Compact Moduli Spaces of Semi-Stable Lattices

Recall that for an OK -lattice Λ with canonical filtration

0 = Λ0 ⊂ Λ1 ⊂ · · · ⊂ Λk = Λ,
the canonical polygon pΛ : [0, r] → R of Λ is a convex polygon characterized by
the following conditions:

(1) pΛ (0) = 0.
 
(2) pΛ is affine over the closed interval rankOK (Λi ), rankOK (Λi+1 ) , and
(3) pΛ (rankOK Λi ) = degar (Λi ) − rankOK (Λi ) · µ(Λ).

Obviously, the canonical polygon is well-defined on the space Mtot K,r and its sub-
∗
spaces Mtot
K,r [T ] for a fixed T ∈ R+ .
Recall also that, for a positive real number t, the scaling assignment Λ 7→ tΛ
induces a natural isomorphism between the moduli spaces
Mtot tot r[K:Q]
K,r [T ] ' MK,r [t T] ∀T ∈ R∗+ .
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 26

26 Non-Abelian Zeta Functions

For this reason, when necessary, we fix the volume, say, to be one, for the lattices
to be considered.
On the other hand, by Theorem 1.2, the moduli spaces of OK -lattices of rank
r are not compact, because of the existence of parabolic cusps. To overcome the
problem, we have introduced the stability condition.

Theorem
n 1.3.h Letip : [0, r] →o R be a fixed convex polygon. Then,
h r/2the
i subspace
[Λ] ∈ MtotK,r ∆ r/2
K : p Λ ≤ p is a compact subspace of M tot
K,r ∆ K . Inh partic-
i
ular, the moduli spaces of semi-stable OK -lattices of rank r and volume ∆r/2 K is
compact.

Proof. It suffices to treat lattices in Mtot

K,r [1] with a fixed underlying projective
module P, say, P = Pr.a . Set then MK,r.P [1] the moduli spaces parametrizing
(isometry classes of ) OK -lattices corresponding to points of Mtot [1]. What we
n o K,r
want to prove is that the space [Λ] ∈ MK,r.P [1] : pΛ ≤ p is compact. By
Theorem 1.2,
 ⊕(r−1) 
MK,r.P [1] ' Fr,+ SL(r, R) SO(r) r1 × SL(r, C) SU(r) r2 .





K × SL OK ⊕a
Hence, there are two parts which should be dealt with, namely, the fundamental
domain Fn,+K and the rest. By the Dirichlet unit theorem, or better, Lemma 1.10,
Fn,+
K is compact. So it suffices to treat the remaining factor. For this, we use
the classical Minkowski’s reduction theory. Indeed, since the volumes of lattices
involved are fixed, semi-stability condition implies that the first Minkowski suc-
cessive minimums of these lattices admit a natural lower bound away from 0 (de-
pending only
n on r). Hence by the standard
o reduction theory, see e.g. [12, 13], the
subspace [Λ] ∈ MK,r.P [1] : pΛ ≤ p is compact.
To prove the final statement, it suffices to take p to be identically zero. 

For later use, for T > 0, denote by MK,n [T ] the moduli space of OK -lattices
of rank r and volume T .
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 27

Chapter 2

Geometry of Numbers

An arithmetic cohomology theory for number fields is developed here. We first

introduce two types of topological cohomology groups for vector adeles, and es-
tablish a topological duality between them. Then we assign some natural weights
for the elements of these groups and hence are able to measure the associated to-
talities. Finally, we prove that these totalities satisfy an arithmetical duality, an
arithmetic Riemann-Roch theorem, and an arithmetic vanishing theorem.

2.1 Global Cohomology

2.1.1 Adelic Ring

Let K be an algebraic number field. Denote by OK its ring of integers. Let S fin ,
resp. S ∞ , be the set of inequivalent normalized non-Archimedean places, resp.
Archimedean places, of K, and set S = S fin S ∞ . For each v ∈ S , write Kv the
F
v-completion of K, and, for each v ∈ S fin , denote by Ov the ring of integers in Kv .
Finally, let r1 , resp. r2 , be the number of real, resp. complex, places of K in S ∞ .
The adelic ring A = AK :=
Q0
v∈S Kv of K is defined to be the restricted
Q
product of the Kv ’s with respect to the Ov ’s, i.e. the collection of (xv )v∈S ∈ v∈S Kv
satisfying xv ∈ Ov for almost all v ∈ S fin . Set Afin := 0v∈S fin Kv and call it
Q
the ring of finite adeles of K and set A∞ := σ∈S ∞ Kv = Rr1 × Cr2 . There is a
Q
natural topology on A, and hence on Afin , generated by the basis of open subsets
Q Q
v∈I Uv × v<S Ov , where I ⊂ S are finite and contains S ∞ and Uv ⊂ Kv are open.

Proposition 2.1. (see e.g. [66])

(1) A and Afin are locally compact separated topological rings.

(2) Kv ,→ Afin (v ∈ S fin ), Kv ,→ A (v ∈ S ) and Afin ,→ A are all closed.

27
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 28

28 Non-Abelian Zeta Functions

(3) For an open compact subgroup Kfin in Afin ,

K + Kfin = Afin and K + Kfin + A∞ = A.
(4) K is dense in Afin and discrete in A and A/K is compact. Moreover, for
K := v∈S fin Ov , K\A/K ' A∞ /OK .
Q

2.1.2 Adelic Cohomology Groups

As usual, denote by GLn (A) the adelic general linear group, consisting of ele-
Q
ments (gv )v∈S of v∈S GLn (Kv ) satisfying gv ∈ GLn (Ov ) for almost all v ∈ S fin .
For convenience, we also write elements (gv )v∈S of GLn (A) as (gp .gσ )p∈S fin ,σ∈S ∞ or
simply (gp .gσ ) if there is no confusion.

Definition 2.1. Let g = (gp .gσ ) be an element of GLn (A),

(1) The auxiliary subspace An (g) of An is defined by

∃a ∈ K n , s.t. xv = a, ∀v ∈ S ∞
( )
An (g) := x = (xv ) ∈ An .
gp · xp ∈ Onp , gp · a ∈ Onp , ∀p ∈ S fin
To introduce a topology on it, let ρ(g) be a new metric on An∞ defined by
altering the standard one at all v ∈ S ∞ with the positive definite matrices
gtσ · gσ , resp. ḡtτ · gτ , for real v = σ, resp. complex v = τ. This, together with
the topology on Afin then induces a topology on An and hence on An (g).
(2) The 0-th and the 1-st arithmetical cohomology groups of g on K is defined by
H 0 (K, g) := An (g) ∩ K n H 1 (K, g) := An An (g) + K n .


and

With respect to the induced topology on An , H 0 (K, g) and H 1 (K, g) becomes

topological groups. Since over a finite dimensional R-vector space, the topologies
induced from different metrics are equivalent, all the topologies on An associated
to g ∈ GLn (A) are equivalent. Since A is locally compact, GLn (A), An and its
sub-quotient spaces H i (K, g) are locally compact as well.

Proposition 2.2. For g ∈ GLn (A),

(1) H 0 (K, g) is discrete, and

(2) H 1 (K, g) is compact.
0
(K, g) := x ∈ K n gp (x) ∈ Onp ∀p ∈ S fin . Then H 0 (K, g) '

Proof. (1) Set Hfin
0
Hfin (K, g), and, via the Minkowski embedding,
i  n
H 0 (K, g) ' Hfin
0
(K, g) ,→ K n ,→ Rr1 × Cr2 = An∞ .
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 29

Geometry of Numbers 29

This implies that H 0 (K, g) is discrete by Proposition 2.1(4).

(2) Equipped An∞ ' σ:R Rn × τ:C Cn with the twisted metric structure ρ(g)
Q Q
obtained from the positive definite matrix (gtσ · gσ .gtτ · gτ ), H 0 (K, g) ' Hfin
0
(K, g)
becomes a full rank lattice Λ(g) within. Then it suffices to prove that,
 
Y n Y n  .
1
H (K, g) '  R × C , ρ(g∞ ) Λ(g), (2.1)
σ:R τ:C

or equivalently, the following

Lemma 2.1. There is a natural isomorphism

 
An /(An (g) + K n ) ' An∞ /i Hfin
0
(K, g) .

Proof. Denote by φ : An∞ → An /(An (g) + K n ) the morphism induced from the
embedding A∞ ,→ A. It is sufficient to show that

(i) φ is surjective,
 and 
(ii) Ker φ = i Hfin0
(K, g) .

(ii) trivially comes from the definition of H 0 (K, g). To prove (i), we show the
equivalent condition that Anfin ⊆ An (g) + K n . But this is a direct consequence
of Proposiiton 2.1(4). Indeed, for any g = (gp .gσ ) ∈ An , there exists a finite set
S 0 ⊆ S fin such that gq ∈ GLn (Oq ) for q < S 0 . But, by the strong approximation
theorem, there exists a ∈ K n such that gp (ap ) − a ∈ Onp , for all p ∈ S 0 , and that
a ∈ Onq for all q < S 0 . This proves the lemma and hence also the proposition. 

2.1.3 Topological Duality: Local and Global Pairings

There is a canonical duality between H 0 and H 1 . To see this, we use the pairing
h·, ·i : A × A → R/Z ' S1 ⊂ C∗
(2.2)
(x, y) 7→ v∈S hxv , yv iv .
P

n
Y √
Here hxv , yv iv := e2π −1·χv (xv,i ·yv,i )
with χv := λv ◦ TrFv /Qv , and
i=1

Qv  Qv /Zv ,→ Q/Z ,→ R/Z, v = p ∈ S fin

λv := 

(2.3)
R  R/Z,
 v ∈ F∞ .
bn ' An (as
By Tate [113], this global pairing induces a natural isomorphisms A
n ⊥ n
locally compact groups) and (K ) ' K (as discrete subgroups). On the other
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 30

30 Non-Abelian Zeta Functions

hand, for the local pairings h·, ·ip of finite places p, by a direct local calculation
from the definition (see e.g. [113]), we have a⊥p ' a−1 p · ∂p , where ∂p denotes
the local differential module at p (dual to Op ). Put this globally, denote by κF
−ord (∂ )

the canonical idele πp p p .1 , where πp denotes a uniformizer of Op , Then the
above proof shows that the dual lattice Λ(g)∨ of Λ(g) is given by Λ(g)∨ ' Λ(κF ·
g−1 ). In addition, since H 0 is insensitive when local or global units are multiplied,
the cohomology group H 0 (K, κF · g−1 ) is well-defined since it does not depend on
the πp ’s used. Thus by (2.1), there is an isomorphism
Y Y .
H 1 (K, g) ' Rn × Cn , ρ(g∞ ) Λ(g).
σ:R τ:C

Consequently, if we denote by X
b the Pontryagin dual of a locally compact space
X, what we have said proves the following:

Proposition 2.3 (Topological Duality). The global pairing h·, ·i : A × A → C∗

induces a natural isomorphism between locally compact groups

(K, g) ' H 0 (K, κF ⊗ g−1 ).

H 1[

2.2 Duality and Riemann-Roch Theorem in Arithmetic

2.2.1 Nine-Diagram

Let p ∈ S fin . The 0-th cohomology groups of g ∈ GLn (Kp ) is defined by

H 0 (Kp , g) := {x ∈ Kpn : g · x ∈ Op }.

And two elements g1 , g2 of GLn (Kp ) are called equivalent, denoted by g1 ∼ g2 ,

if there is an isomorphism H 0 (Kp , g1 ) = H 0 (Kp , g2 ). Clearly, g1 ∼ g2 if and only
if there exists g ∈ GLn (Op ) such that g2 = g · g1 . Accordingly, over the quotient

space GLn (Kp ) GLn (Op ), introduce a partial order ≤ by

[g1 ] ≤ [g2 ] if H 0 (Kp , g1 ) ⊆ H 0 (Kp , g2 ).

Consequently, we obtain an equivalence relation ∼ on GLn (Afin ) (and more

generally on GLn (Afin ) × {g∞ } for a fixed g∞ ∈ GLn (A∞ )), characterized by
g1 ∼ g2 if and only if Hfin 0
(F, g1 ) = Hfin0
(F, g2 ), that is, g2 = g · g1 for
Q 
a certain g ∈ GLn O
p∈S fin p . Similarly, we obtain a partial order ≤ on
 Q 
GLn (Afin ) GLn p∈S fin Op characterized by

0 0
[g1 ] ≤ [g2 ] if and only if Hfin (F, g1 ) ⊆ Hfin (F, g2 ).
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 31

Geometry of Numbers 31

For g ∈ GLn (A), there is a 9-diagram with exact rows and columns given by

0 0 0
↓ ↓ ↓
0→ H 0 (K, g) → An (g) → An (g)/An (g) ∩ K n → 0
↓ ↓ ↓
0→ K n
→ An
→ A /K n
n
→0
↓ ↓ ↓
0 → K n /An (g) ∩ K n → An /An (g) → H 1 (K, g) →0
↓ ↓ ↓
0 0 0.

Consequently, for any two elements g = (gp .gσ ) and g0 = (g0p .g0v ) satisfy the
condition that gv = g0v ∀v ∈ S ∞ and [(gp )] ≤ [(g0p )], since the middle row remains
unchanged in the corresponding 9 diagrams, there exists a long exact sequence

H 0 (K, g) ,→ H 0 (K, g0 ) → An (g0 ) An (g) → H 1 (K, g)  H 1 (K, g0 ).



With g∞ fixed, the quotient An (g0 ) An (g) measures only differences at finite

places. Namely degfin (g0 ) − degfin (g0 ) = deg(g0 ) − deg(g0 ), where
X
degfin (g) := degfin (det g) = ordp (detgp ) · Np log p.
p

This then suggests that, to obtain an arithmetic Riemann-Roch theorem, the arith-
metic counts hi (i=0,1) of H i should satisfy the following axioms.

Axiom 1 For g, g0 ∈ GLn (A) with a common g∞ ,

h0 (K, g0 ) − h1 (K, g0 ) − deg(g0 ) = h0 (K, g) − h1 (K, g) − deg(g).

Axiom 2 On K\GLn (A)/GLn (K), the function h0 (K, g) − h1 (K, g) − deg(g) is

a constant, denoted by ∆(K, n).

In particular, by examining the diagonal elements of GLn (A), which are also
elements of n-copies of GL1 (A), it is natural to introduce

Axiom 3 There exists a constant ∆(K) depending only on K such that

∆(K, n) = n · ∆(K).

Axiom 4 For all elements g ∈ GLn (A),

h1 (K, g) = h0 (K, κK ⊗ g−1 ).

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 32

32 Non-Abelian Zeta Functions

2.2.2 Arithmetic Riemann-Roch Theorem

By the Arakelov-Riemann-Roch formula, i.e. Proposition 1.5, for g ∈ GLn (A),

n  Y Y . 
deg(g) − log ∆K = χ(OK , g) := − log vol Rn × Cn , ρ(g∞ ) Λ(g) .
2 σ:R τ:C

Axioms (1∼ 4) for h0 and h1 is equivalent to

Axiom 0 For all elements g ∈ GLn (A),

 
  Y n Y n .
h (K, g) − h (K, g) = − log vol C , ρ(g∞ ) Λ(g) .
0 1

R ×
σ:R τ:C
Axiom 4 For all elements g ∈ GLn (A),
h1 (K, g) = h0 (K, κK ⊗ g−1 ).

To introduce the hi ’s to satisfy these axioms, we first assign some natural

weights to the elements of H 0 (K, g) using the function
eA ((xv )) := 1Onv (xv ) ×
Y Y 2
e−π·Nσ ·|xσ |can,σ .
v∈S fin σ∈S ∞

Here 1Onv denotes the characteristic function of Onv in Kvn , and | · |can,σ denotes the
norm induced by the canonical metric for each of the Archimedean place σ of K.

Definition 2.2. Let g ∈ GLn (A).

(1) The arithmetical count of the OK -lattice H 0 (K, g) is defined by

Z
#Ar (H 0 (K, g)) := eA (g · x) dµ(x)
x∈H 0 (K,g)

1Onv (gv · xv )
X Y Y 2
X Y 2
= e−πNσ |gσ ·xσ |can,σ = e−πNσ |gσ ·xσ |can,σ .
x∈H 0 (K,g) v∈S fin σ∈S ∞ x∈H 0 (K,g) σ∈S ∞

(2) The 0-th numerical cohomology h0 (K, g) of g is defined by

eA (g · x)
   X 
h0 (K, g) := log #Ar (H 0 (K, g)) = log
x∈H 0 (K,g)
 
 X  X 
= log  exp −π Nσ |gσ · xσ |2can,σ  .

x∈H 0 (K,g) σ∈S ∞

Remark 2.1. This definition is motivated by h0 (X, L) for a line bundle L on an

integral, regular, proper curve X over a finite field Fq . Indeed, the weight function
then is taken to be the constant function one, and hence
X 0
#H 0 (X, L) = 1 = qh (X,L) .
x∈H 0 (X,L)
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 33

Geometry of Numbers 33

To find h1 (K, g), we introduce a new axiom.

Axiom 5. Let G be a locally compact group, either discrete or compact, the

arithmetic counts #Ar for G and its Pointrjagin dual G
b coincide. That is,

#Ar (G) = #Ar (G).

b

Recall that H 1 (K, g) is compact with a projection An  An /(An (g) + K n ) =

1
H (K, g). Consequently, for the discrete H 1[ (K, g), there exists a natural injection
(K, g) ,→ A
H 1[ cn . Fix a test function f on An , and denote by b
√ P f its Fourier transform
with respect to the character e−2π −1 v∈S χv (∗) on A
cn = An . That is
√ P
Z
f (x) :=
b f (y)e−2π −1 v∈S χv (xy) dy.
An

Here χv denotes an additive character of Kv . Then b f | H[1 (K,g) makes sense.

 
1
In addition, for the pairing H 1 (K, g), H (K, g) , with respect to the character
[
√ P
e−2π −1 v∈S χv (∗) again, define the Fourier transform b f | H[
f of b 1 (K,g) by
c

X √ P
−1 v∈S χv (ξv ηv ) b
f (η) := e−2π f (ξ), ∀η ∈ H 1 (K, g).
c
b
ξ∈H[
1 (K,g)

Assume further that

f |H 0 (K,g) ∈ L2 H 0 (K, g) , 2
(K, g) ,
H 1[ f ∈ L2 H 1 (K, g) .




1 (K,g) ∈ L
f H[
b c
b

Here, as usual, for a measurable space X, L2 (X) denotes the associated space
of L2 -functions. Accordingly, we introduce the arithmetic counts for the groups
H 0 (K, g) and H 1 (K, g) with respect to the test function f by setting
Z
2 2
X
0
f (x) dx = f (x) ,


#Ar, f H (K, g) :=
H 0 (K,g) x∈H 0 (K,g)
Z
2 2
X
#Ar, f H 1[ f (y) dy = f (y) ,


(K, g) := b b
1 (K,g)
H[ 1 (K,g)
y∈H[
Z
2
H 1 (K, g) :=


#Ar, f f (z) dz.
c
b
H 1 (K,g)

By the Plancherel formula (see e.g [26]),

(K, g) = #Ar, f H 1 (K, g) .

#Ar, f H 1[

That is, #Ar verifies Axiom 4.

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 34

34 Non-Abelian Zeta Functions

Therefore, to define h1 counting H 1 , it suffices to choose a good test function.

For our limited purpose, we take the canonical Gaussian distributions on An
1
eA2 := 1Onp ×
Y Y 1
e− 2 πNv |∗|can,v .
p∈S fin v∈S ∞

[
1
[
That is to say, we count H 1 (K, g) by the numeration function e(∗) := eA2 (g∗) ,
2

1
since 1Onp = 1O2 n .
p

Definition 2.3. Let g ∈ GLn (A).

(1) The arithmetic count for H 1 (K, g) if defined by

Z
#Ar H 1 (K, g) := e(x)dx.
H 1 (K,g)

(2) The 1-st numerical cohomology h1 (K, g) of g is defined by

Z
e(x) dµ(x) .
  
h1 (K, g) := log #Ar (H 1 (K, g)) = log
h1 (K,g)

By a direct calculation (see, e.g. [113]),

ecA = 1(∂−1p )n ×
Y Y 2
e−πNv |∗|can,v .
p∈S fin v∈S ∞

This, together with arithmetical duality H 1 (K, g) ' H 0 (K,[κF · g−1 ), implies that
ecA (g−1 a) = #Ar H 0 (K, κF · g−1 ),
X
(K, g) =
#Ar H 1[
1 (K,g)
a∈H[

since, in the definition of #Ar H 0 (K, g), a shift by g is used. In particular,

#Ar (H 1 (K, g)) = #Ar H 1[
(K, g).
This verifies Axiom 5. In addition,
1Onp (gp (a)) × eA (g(a)),
X Y Y 2
X
#Ar H 0 (K, g) = e−πNv |gv (a)|can,v =
a∈K n p∈S fin v∈S ∞ a∈K n

eA (κF · g−1 (a)) = eA (g−1 (a)).

X X
#Ar H 0 (K, κK · g−1 ) = b
a∈K n a∈K n

To verify Axiom 0, we use the Tate Riemann-Roch theorem ([113]), obtained from
the Poisson summation formula for adelic spaces K n ,→ An . Namely,
1
eA (g(a)) = eA (g−1 (a)).
X X
· b
a∈K n kdetgk a∈K n
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 35

Geometry of Numbers 35

In terms of our language,

1
#Ar H 0 (K, g) = · #Ar h0 (K, κF · g−1 ).
kdetgk
Or equivalently, by using the arithmetic duality,
h0 (K, g) − h1 (K, g) = χ(OK , g).
All these then prove the following fundamental:

Theorem 2.1. Let g ∈ GLn (A).

(1) (Numerical Duality)

h1 (K, g) = h0 (K, κF · g−1 ).
(2) (Arithmetic Riemann-Roch Theorem)
n
h0 (K, g) − h1 (K, g) = deg(g) − · log ∆K .
2

The above discussion applies to lattices as well.

Definition 2.4. Let Λ be an OK -lattice of rank n.

(1) The 0-th cohomolog group H 0 (K, Λ) is defined to be Λ. .

(2) The 1-st cohomolog groups H 1 (K, Λ) is defined to be Λ ⊗Q R Λ.

In particular, H 0 (K, Λ) is discrete and H 1 (K, Λ) is compact. Let ωK be the

canonical lattice of K associated to the inverse of the global differential module
∂K := (∂v ) of OK equipped with the standard metric. By Proposition 2.3, we have

Proposition 2.4 (Arithmetic Duality). As locally compact groups,

(K, Λ) ' H 0 (K, ωF ⊗ Λ∨ ).
H 1[

Similarly, we define
X
Nσ |x|2ρσ
#Ar H 0 (K, Λ) := .
P
e−π σ∈S ∞

x∈H 0 (K,Λ)

This coincides with k0 (F, Λ) in [32] obtained using arithmetic effectivity. Set
h0 (K, Λ) := log k0 (F, Λ). (2.4)
1
To define h , we use the counting function e(∗) on H (K, Λ) obtained using the
1
P 2
Fourier transformZfrom the function e−π σ∈S ∞ Nσ |∗|ρσ . That is
#Ar H 1 (K, Λ) := e(x)dµ(x) and h1 (K, Λ) := log #Ar H 1 (K, Λ).
x∈H 1 (K,Λ)
As a direct consequence of the Plancherel theorem, the Poisson summation for-
mula, and the Arakelov Riemann-Roch theorem, we have the following:
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 36

36 Non-Abelian Zeta Functions

Theorem 2.2. Let Λ be an OK -lattice of rank n.

(1) (Arithmetic Duality)

h1 (K, Λ) = h0 (K, ωK ⊗ Λ∨ ).
(2) (Arithmetic Riemann-Roch Theorem)
n
h0 (K, Λ) − h1 (K, Λ) = degar (Λ) − · log ∆K .
2
We end this discussion with the following:

Definition 2.5. The Gamma invariant γK of a number field K is defined by

γK := h0 (K, (OK , ρcan )).

For example, when K = Q,

X 2
X 2
γQ = e−πn = 1 + 2 e−πn ' 1.086.
n∈Z n≥1

2.3 Arithmetic Vanishing Theorem

2.3.1 Elementary Properties of 0-th Cohomology

We recall some elementary results concerning h0 for lattices following [32, 35].

Lemma 2.2. Let Λ be a (Z-)lattice in a (finite dimensional) metrized

X space V.
2
0
Then the function #HAr (Q, ·) : V/Λ → R defined by v + Λ 7→ e−π|v+x| at-
x∈Λ
tains a unique maximum in Λ.

Proof. Let f : V → C be a rapidly decreasing function, and for a fixed y0 ∈ V,

set g(x) = f (x + y0 ). Then, the Fourier transform of g is given by
Z Z
g(x) =
b f (y + y0 )e−2πihx,yi dy = f (y)e−2πihx,y−y0 i dy
V V
Z
= f (y)e2πihx,y0 i e−2πihx,yi dy = bf (x) · e2πihx,y0 i .
V
Moreover, by the Poisson summation formula,
X X 1 X
f (x) = g(x) = ∨)
e2πihv,yi b
f (y). (2.5)
x∈v+Λ x∈Λ
vol(Λ y∈Λ∨

Specify now f (x) = e−πhx,xi , then bf = f . Easily, the sum on the right hand side of
(2.5) is maximal if e2πihv,yi = 1 for all y. This implies that v ∈ Λ. 
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 37

Geometry of Numbers 37

Corollary 2.1. Let Λ1 be a sub lattice of Λ. Then

0
(Q, Λ) ≤ #HAr
0
(Q, Λ1 ) · #HAr
0
(Q, Λ Λ1 ),

#HAr
and the equality holds if and only Λ = Λ1 ⊕ Λ Λ1 .


Proof. Let Λ be a full rank lattice in an Euclidean space V. Let π be an orthogonal

projection on a subspace V 2 of V such that π(Λ) is discrete. Denote the associated
π
lattice of V 2 by Λ2 , let Λ1 be the kernel of Λ → Λ2 and let V 1 be the subspace
of V associated to Λ1 . Then Λ1 is a sub lattice of Λ and we have a short exact
sequence of lattices
0 → Λ1 → Λ → Λ2 → 0.
Clearly, all sub lattices Λ1 of Λ can be realized in this way, and Λ2 is simply
Λ Λ1 . Consequently, for x ∈ Λ1 , y ∈ Λ2 , we have e−πhx,xi e−πhy,yi = e−πhx+y,x+yi .

Hence #HAr 0
(Q, Λ1 ⊕ Λ2 ) = #HAr0
(Q, Λ1 ) #HAr
0
(Q, Λ2 ). In addition, for y ∈ Λ2 , if
we choose r(y) ∈ π−1 (y) ⊂ Λ,
X X
0
#HAr (Q, Λ1 ⊕Λ2 ) = 0
#HAr (Q, Λ1 +y) & #HAr 0
(Q, Λ) = 0
#HAr (Q, r(y)+Λ1 ).
y∈Λ2 y∈Λ2

On the other hand, for x ∈ V 1 , we have hx + y, x + yi = hx, xihy, yi. Hence

0
#HAr (Q, Λ1 + y) = e−hy,yi #HAr
0
(Q, Λ1 ),
0
#HAr (Q, Λ1 + r(y)) = e−hy,yi #HAr
0
(Q, Λ1 + r(y) − y).
Consequently, by Lemma 2.2, #HAr 0
(Q, Λ1 +y) ≥ #HAr0
(Q, Λ1 +r(y)) and the equal-
ity holds if and only r(y) ∈ Λ1 + y. This implies that
X
#HAr0
(Q, Λ1 ) #HAr
0
(Q, Λ2 ) = #HAr
0
(Q, Λ1 ⊕ Λ2 ) = 0
#HAr (Q, Λ1 + y)
y∈Λ2
X
≥ 0
#HAr (Q, Λ1 + r(y)) = #HAr
0
(Q, Λ),
y∈Λ2

and the equality holds if and only if Λ = Λ1 ⊕ Λ2 . 

Definition 2.6. Let Λ be a full rank Z-lattice in an Euclidean space V = (V, h·, ·i).

(1) The first Minkowski successive minimum λ1 (Λ) of Λ is defined to be the min-
imal length among the lengths of all non-zero vectors in Λ. Namely,
λ1 = λ1 (Λ) := min kxk : x ∈ Λ, x , 0 .


(2) For each t ∈ R∗+ , we set

NΛ (t) := # x ∈ Λ : hx, xi ≤ λ1 (Λ)2 t .

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 38

38 Non-Abelian Zeta Functions

For simplicity, we write N(t) instead of NΛ (t), if there is no confusion.

Lemma 2.3. Let Λ be a lattice.

 √ n
(1) NΛ (t) ≤ 2 t + 1 .
Z ∞
2
(2) #H 0 (Q, Λ) = πλ21 N(t) e−πλ1 t dt.
0 Z
∞  √ n 2
(3) #H 0 (Q, Λ) ≤ 1 + πλ21 2 t + 1 e−πλ1 t dt.
1

Proof. It suffices to prove the first two since (3) then comes automatically.
(1) For a positive real number t and an element x0 in V, set

D(t) := x ∈ Λ : hx, xi ≤ λ21 t Bx0 (t) := x ∈ V : |x − x0 | < t .

 
and

From the definition, for any two distinct point x, y ∈ D(t),

kx − yk ≥ λ1 and Bx (λ1 /2) ∩ By (λ1 /2) = ∅.

 √  
Consequently, x∈D(t) Bx (λ1 /2) ⊂ B0 t + 1/2 λ1 . Therefore, by comparing
S
the volumes,
√ !n 
( t + 1/2)λ1 √ n
N(t) = #D(t) ≤ = 2 t+1 .
λ1 /2

(2) From the definition,

X XZ ∞ Z ∞
#H (Q, Λ) =
0 −πhx,xi
= πe
−πt
dt = π x ∈ Λ : hx, xi ≤ t e−πt dt.

e
x∈Λ x∈Λ hx,xi 0

Hence, a change from t to λ21 t implies (2). 

Corollary 2.2.
√
(1) Let Λ be a Z-lattice of rank n such that λ1 (Λ) ≥ n. Then
3n π 2
#H 0 (Q, Λ) ≤ 1 + e−πλ1 (Λ) .
π − log 3
(2) Let K be a number field of degree n over Q and let Λ be a rank one OK -lattice
satisfying degar (Λ) ≤ 0. Then
3n π − 2 degar (Λ)
h0 (K, Λ) ≤ 1 + e−πn e n .
π − log 3
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 39

Geometry of Numbers 39

√
Proof. (1) Write λ for λ(Λ). If t ≥ 1, 2 t + 1 ≤ 3t . Hence, by Lemma 2.3(3),
Z ∞
2
#H 0 (Q, Λ) − 1 ≤ πλ21 3nt e−πλ1 t dt
1
∞ 3n πλ21 3n π
Z
e(−πλ1 +n log 3)t dt =
2 2 2
= πλ21 e−πλ1 ≤ e−πλ1 .
1 πλ21 − n log 3 π − log 3

(2) We apply (1) to Λ. Then, it suffices to show that

2
kxk2 ≥ n e− n degar (Λ) ∀x ∈ Λ, x , 0. (2.6)

Recall that, by definition,

X
degar (Λ) = log # Λ OK x − Nσ log kxkσ /Nσ .

σ∈S ∞

kxkσ
Here Nσ appearing in comes from the difference between canonical metric
Nσ
and Lebesgue metric. Hence, by the geometric-arithmetic mean inequality,
!2  1
X X kxk σ
 Y kxkσ !2Nσ  n
kxk2 := kxk2σ ≥ Nσ ≥ n  
σ∈S ∞ σ∈S ∞
Nσ σ∈S ∞
Nσ (2.7)
2
  2
= n #(Λ OK x)e− degar (Λ) n ≥ n e− n degar (Λ) .


This gives (2.6), and hence (2) since h0 (K, Λ) ≤ #H 0 Q, ResK/Q (Λ) − 1.




2.3.2 Ampleness and Vanishing Theorem

Similar to the prevalent cohomology theory in algebraic geometry, there is a type

of vanishing theorems for the cohomology groups just introduced. We treat here
only invertible OK -lattices. For high ranks, please refer to §3.1.2 below.

Definition 2.7. Let a = (ap .aσ ) ∈ IK = GL1 (AK ).

(1) a is called positive if

X X
deg(a) := ordp (ap ) · Np log p − Nv log |av |v > 0.
p v

(2) a is called ample if for each g = (gp .gσ ) ∈ GLn (A) and for a sufficiently

positive m, the unit ball B0 (1) centered at 0 of the metrized space Rr1 ×

Cr2 n , ρ(am 0 m


∞ · g∞ ) contains a basis of the lattice H (K, a · g) whose elements
are positive at finite places.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 40

40 Non-Abelian Zeta Functions

Theorem 2.3 (Vanishing Theorem). Let a ∈ IK . The conditions below are

equivalent.

(1) a is ample.
(2) a is positive.
(3) lim h1 (K, am · g) = 0 for all g ∈ GLn (A).
m→∞

Proof. (1) ⇒ (2). This is obvious.

(2) ⇒ (1). This is essentially Theorem 2.2 proved in [138]. A technical point
here is that, for a fixed n and for all possible m, a and g, the H 0 (K, am · g) ⊆
K n ’s are full rank lattices in the metrized spaces based on (the same vector space)
Rr1 × Cr2 n .

Indeed, if a is positive, by the Arakelov Riemann-Roch theorem, for suffi-

ciently positive m, H 0 (K, am ) contains a strictly effective section l, i.e. a section
` satisfying klk < 1. Thus, from the technical point above, H 0 (K, am · g) contains
a full rank sub-lattice generated by strictly effective sections `e1 , `e2 , . . . , `eN ,
where N := n[K : Q] and {e1 , e2 , . . . , eN } is a fixed basis of H 0 (K, g). (Here,
we view l, resp. ei , as an element of the field K, resp. of the K-vector space K n .)
This, together with Lemma 1.7 of [138], then shows that, for sufficiently positive
m, H 0 (K, am · g) is generated by strictly effective sections.
(3) ⇒ (2). By the duality, h0 (K, a−m · g) → 0 when m → ∞ for g ∈ GLn (A).
Recall that, by definition,
 
 X 
2
h (K, a ) = log 1 +  .
P
0 −m −π N kxk
 e v v v

x∈h0 (K,a−m )r{0}
P 2
Consequently, for any non-zero x(m) ∈ h0 (K, a−m ), e−π v Nv kx(m)kv → 0, or equiv-
alently, v kx(m)kv → ∞ as m → ∞. Applying this to x(m) := em ∈ h0 (K, a−m ),
P
we conclude that kekv > 1 for some v and all non-zero sections e ∈ H 0 (K, a−1 ).
Therefore, deg(a) ≥ 0. Indeed, if deg(a) < 0 or the same deg(a−1 ) > 0, by apply-
ing the equivalence of (1) and (2) to a−1 , (replacing a with al for sufficiently large
l if necessary), we conclude that H 0 (K, a−1 ) consists of a Z-basis {e1 , e2 , . . . , eN }
such that kei kv < 1, a contradiction.
In addition, we claim that deg(a) , 0. Otherwise, choose g0 ∈ GLn (A) such
n
that χ(F, g0 ) = deg(g0 ) − log ∆K > 0. Then, by the Arakelov Riemann-Roch,
2
 n 
h0 (K, a−m · g0 ) − h1 (K, a−m · g0 ) = deg(g0 ) − log ∆K = χ(F, g0 ).
2
Consequently, limm→∞ h (K, a · g) ≥ χ(F, g0 ) > 0, a contradiction as well.
0 −m

Therefore, deg(a) > 0. This proves (2).

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 41

Geometry of Numbers 41

(2) ⇒ (3). By the numerical duality, it suffices to show that

lim h0 (K, a−m · g) = 0 for any fixed g ∈ GLn (A).
m→∞

Denote by λ1 (−m) the first

 Minkowski n successive minimum of the associated lat-
0 −m r1 r2
tice H (K, a · g) ⊆ R × C . Then, by Proposition 4.4 of [35], namely,
Corollary 2.2(1), it suffices to show that limm→∞ λ1 (−m) = +∞, since the Z-rank
of the OK -lattice H 0 (K, a−m · g) = H 0 (K, a−1 )⊗m ⊗ H 0 (K, g) remains n · [K : Q].
Suppose, otherwise, that there exists an increasing sequence {mk } of natural
numbers such that λ1 (−mk ) remains bounded. We use the stability condition to
deduce a contradiction. Denote by
{0} ⊂ Λ1 ⊂ Λ2 ⊂ · · · ⊂ Λ s = H 0 (K, g)
the Harder-Narasimhan filtration of the OK -lattice H 0 (K, g). Then the Harder-
Narasimhan filtration of the lattice H 0 (K, a−m · g) is given by
{0} ⊆ H 0 (K, a−m ) · Λ1 ⊂ · · · ⊆ H 0 (K, a−m ) · Λ s = H 0 (K, a−m · g).
In particular, H 0 (K, a−m )·Λ1 is, by definition, a maximal semi-stable OK -sublattice
of H 0 (K, a−m · g) with the biggest slope and highest rank, and
   
lim deg H 0 (K, a−m ) ⊗ Λ1 = lim − m · rank(Λ1 ) · deg(a) + deg(Λ1 ) → −∞.
m→∞ m→∞

Now, fix a minimal vector e(−mk ) ∈ H 0 (K, a−mk · g) such that ke(−mk )k =
λ1 (−mk ), and denote by Λ1 (−mk ) the rank one OK -sublattices of H 0 (K, a−mk · g)
generated by e(−mk ). Then, by the biggest slope and highest rank property of
H 0 (K, a−m ) · Λ1 in the canonical filtration of h0 (K, a−mk · g), we conclude that
  deg H 0 (K, a−mk ) · Λ1

deg Λ1 (−mk ) ≤
.
rank H 0 (K, a−mk ) · Λ1
Consequently,
lim deg Λ1 (−mk ) = −∞.


k→∞

Hence, by applying Proposition 7.1 of [35], or the same, estimation (2.6), to the
rank one Λ1 (−mk ), we see that


λ1 (−mk ) = ke(−mk )k ≥ [K : Q] · e− [K:Q ] ·deg Λ1 (−mk ) .
2

Hence, λ1 (−mk ) are unbounded, a contradiction. 

We will apply this cohomology theory to the construct of non-abelian zeta

functions in the next chapter. In addition, for its partial generalization to higher
dimensions, please refer to the Appendix of this book.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 43

Chapter 3

Non-Abelian Zeta Functions

3.1 Moduli Spaces of Semi-Stable Lattices

3.1.1 Stability

Let K be a number field with OK the integer ring and ∆K the absolute value of the
discriminant. Denote by MK,n (d) the moduli space of semi-stable OK -lattices of
rank n and degree d. By the Arakelov-Riemann-Roch theorem,
n
− log vol(Λ) = degar (Λ) − log ∆K ,
2
h n/2 i
MK,n (d) is simply the moduli space MK,n ∆K e introduced in the first chapter,
−d

and by Theorem 1.2, MK,n (d) is parametrized by

G
Fn,+ ⊕(n−1)
⊕ a) SL(n, R) SO(n) r1 × SL(n, C) SU(n) r2 ss .
 




K × SL(OK
a∈{a1 ,...,ah }

Here a1 , . . . , ah denote by integral OK -ideals such that the class group CLK of K is
given by CLK = [a1 ], [a2 ], . . . , [ah ] , Fr,+

K is the fundamental domain of the unit
group U Kr,+ := U Kr ∩ U K+ , and the lower index ss stands for the collections of the
modular points corresponding to the semi-stable OK -lattice of rank r (and volume
one). In addition, by Theorem 1.3, MK,n (d) is compact.
More generally, By Theorem 1.1, the moduli space Mtot K,n of isometric classes
of OK -lattices of rank n admits a natural modular representation
G
Fn,+ ⊕(n−1)
⊕ a) SL(n, R) SO(n) r1 × SL(n, C) SU(n) r2 ss × R∗+ .
 




K × SL(OK
a∈{a1 ,...,ah }

Consequence, there are natural induced measures dµ on Mtot

K,n and MK,n (d).

Example 3.1. If n = 1, the stability condition is automatic, and the moduli spaces
d
Mtot
K,n and MK,n (d) coincide with the Arakelov-Picard groups Pic(K) and Pic (K)
(see e.g [66]). It is well known that Picd (K) is compact.

43
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 44

44 Non-Abelian Zeta Functions

3.1.2 Effective Vanishing Theorem

Since h0 (K, Λ) is a smooth

n functiono on the moduli space MK,n (d), there is a natural
analytic stratification MK,n (d)≤T of MK,n (d) defined by
T ∈R>0
n o
MK,n (d)≤T := Λ ∈ MK,n (d) : h0 (K, Λ) ≤ T . (3.1)

Similarly, there are subspaces MK,n (d)<T , MK,n (d)≥T , MK,n (d)>T and MK,n (d)=T .
By the compactness of MK (n, d), we have the following:

Proposition 3.1. For (n, d) ∈ Z>0 × Z, there exist MK (n, d), mK (n, d) ∈ R such
that MK,n (d)<m = MK,n (d)>M = ∅ unless mK (n, d) ≤ n and M ≤ MK (n, d).

Denote by mK (n, d) and MK (n, d) the minimal and maximal values of h0 on

MK,n (d), respectively, and call ∆K (n, d) := MK (n, d) − mK (n, d) the complexity of
MK (n, d).
Recall that, if L is a rank one OK -lattice, det(Λ∨ ⊗ L) = det(Λ)∨ ⊗ (L)⊗rankΛ .
This induces a natural isomorphism MK,n (d) ' MK n, n log ∆K − d , Λ 7→ Λ∨ ⊗


ωK , where ωK denotes the canonical lattice of K. Hence, by arithmetic duality
and Riemann-Roch, we have the following:

Lemma 3.1. (Duality) For the extremal values of h0 ,

n 
MK n, n log ∆K − d =MK (n, d) + log ∆K − d ,


 n2 
mk n, n log ∆K − d =mK (n, d) + log ∆K − d .


2
Proof. Since the arguments are similar, we only treat MK . The morphism
Λ 7→ Λ∨ ⊗ ωK transforms h0 (K, Λ) to h0 (K, Λ∨ ⊗ ωK ) = h1 (K, Λ) by
the duality. On the other hand, by the Arakelov Riemann-Roch theorem,
n
h0 (K, Λ∨ ⊗ ωK ) = h0 (K, Λ) + log ∆K − d . Therefore,
2
MK n, n log ∆K − d = max h0 (K, Λ0 ) : Λ0 ∈ MK n, n log ∆K − d



= max h0 (K, Λ∨ ⊗ ωK ) : Λ ∈ MK n, d


n n 
o
= max h0 (K, Λ) + log ∆K − d : Λ ∈ MK n, d
2
n 
=MK (n, d) + log ∆K − d ,
2
as desired. 

For the limit behaviors of MK (n.d) and mK (n.d) when d → ∞, we have

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 45

Non-Abelian Zeta Functions 45

Theorem 3.1 (Effective Vanishing Theorem). Let Λ be a semi-stable OK -lattice

of rank n.
n log n
(1) If degar (Λ) ≤ −[K : Q] · ,
2
2deg(Λ)
3n·[K:Q]  −π·[K:Q] e− n·[K:Q]
h (K, Λ) ≤
0
· e .
1 − logπ 3
n log n
(2) If degar (Λ) ≥ [K : Q] · + n log ∆K ,
2
2 2deg(Λ)
3n·[K:Q]  −π·[K:Q]·∆− [K:Q ] e n·[K:Q ]
1
h (Λ) ≤ · e K .
1 − logπ 3
If n = 1, (1) is simply Corollary 2.2(2), originally proved in [35] (see also
[32]).

Proof. By Theorem 2.2(1), it suffice to establish (1). For any non-zero vector x
of Λ, denote by L(x) = OK · x, the OK -lattice generated by x in Λ. Then, by
2
estimation (2.6), kxk2 ≥ [K : Q] · e− [K:Q] ·degar (L(x)) . On the other hand, since Λ is
semi-stable,
degar (Λ) log n
degar (L(x)) ≤ ≤ −[K : Q] · .
n 2
Consequently, kxk ≥ λ1 ≥ [K : Q] · n = rankZ Λ. Since the OK -lattice Λ, viewed
2

as a Z-lattice, satisfies the condition in Corollary 2.2(1),

3n·[K:Q] −π·[K:Q]·e− 2deg(Λ)
#Ar H 0 (K, Λ) ≤ 1 + ,
n·[K:Q ]
·e
1 − logπ 3
2deg(Λ)
if we use the inequality λ21 ≥ [K : Q] · e− n·[K:Q ] , deduced from
− 2deg(Λ)
kxk2 ≥ [K : Q] · e . Hence, to complete the proof, it suffices to note that
n·[K:Q ]

h (K, Λ) = log #Ar H (K, Λ) ≤ #Ar H 0 (K, Λ) − 1.

0 0


Corollary 3.1 (Uniform Boundness). For ε > 0, these exists an effectively

computable constant dK (n.ε), depending only on K, n and ε, such that, for all
d ≥ dK (n.ε),
n n
d − log ∆K ≤ mK (n.d) ≤ MK (n.d) < ε + d − log ∆K .


2 2
In particular, limd→∞ ∆K (n.d) = 0.

Proof. The first and the second statements are obvious. Recall that, from the
definition and the Arakelov Riemann-Roch theorem, MK (n.d) − d − n2 log ∆K =


max h1 (K, Λ) : Λ ∈ MK,n (d) . Hence the third inequality is a direct consequence

of Theorem 3.1(2). Finally, to prove limd→∞ ∆K (n.d) = 0, it suffices to notice that,
for a fixed ε, the constant dK (n, ε) chosen for the three inequalities is effective. 
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 46

46 Non-Abelian Zeta Functions

3.2 Non-Abelian Zeta Function

In this section, as a direct application of arithmetic cohomology theory developed,

we show that the non-abelian zeta functions are well-defined and satisfy the stan-
dard zeta properties.

3.2.1 Definition

Let K be an algebraic number field. Denote by OK its ring of integers and denote
by ∆K the absolute value of its discriminant. For n ∈ Z>0 , denote by MK,n :=
F
d∈R MK,n (d) the moduli space of semi-stable OK -lattices of rank n. Denote by
dµ the natural measure on MK,n .

Definition 3.1. The rank n non-abelian zeta function bζK,n (s) of K is defined by
 n s Z  0   deg(Λ)
ζK,n (s) := ∆K2 ·
b eh (K,Λ) − 1 · e−s dµ(Λ) <(s) > 1. (3.2)
Λ∈Mtot
K,n

3.2.2 Zeta Properties

Like classical zeta functions, such as the Dedekind zeta functions, these rank n
ζK,n (s) satisfy the standard zeta properties.
non-abelian zeta functions b

Theorem 3.2. Let K be a number field.

(0) Up to constants depending only on local and units of K and n,

ζK,1 (s) = b
b ζK (s).
(1) (Meromorphic Continuation) b ζK,n (s) is a well-defined holomorphic function
in s when <(s) > 1, and admits a meromorphic continuation, denoted also by
ζK,n (s), to the whole complex s- plane.
b
(2) (Functional Equation) b ζK,n (s) satisfies the standard functional equation
ζK,n (1 − s) = b
b ζK,n (s).
(3) (Singularities & Residues) As a meromorphic function over the complex s-
ζK,n (s) admits two singularities, i.e. two simple poles at s = 0, 1, and
plane, b
ζK,n (s) = vol MK,n [1] .


Res s=1b

Proof. There are two proofs presented in this book. Namely, one proof given here
uses the arithmetic cohomology theory, and the other proof, to be given in the next
part, uses the theory of Eisenstein series.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 47

Non-Abelian Zeta Functions 47

The assertion (0) is proved in Proposition 1.2. To prove the rest, denote by
MK,n [≥ T ], resp. MK,n [≤ T ], the moduli spaces of rank n semi-stable OK -lattices
of volumes ≥ T , resp. ≤ T . Obviously,
[ G
MK,n = MK,n [≤ T ] MK,n [≥ T ] = MK,n [T ].
T >0
Moreover, if we denote by dµT the natural induced volume form on MK,n [T ], then
dT
dµ = · dµT . Accordingly,
T Z  0 
ζK,n (s) =
b eh (K,Λ) − 1 · vol(Λ) s · dµ(Λ)
Λ∈Mtot
K,n

= I(s) + A(s) − α(s)

where
Z
0
 
I(s) := eh (K,Λ) − 1 · vol(Λ) s · dµ(Λ)
Λ∈MK,n [≥1]
Z
0
(K,Λ)
A(s) := eh · vol(Λ) s · dµ(Λ)
Λ∈MK,n [≤1]
Z
α(s) := vol(Λ) s · dµ(Λ).
Λ∈MK,n [≤1]
We next treat the integrations I(s), A(s) and α(s) separately.
I(s) By the effective vanishing theorem, namely Theorem 3.1, I(s) is holo-
morphic over the whole complex  0 s-plane,  since it is the integration of the T -
exponentially decay function eh (K,Λ) − 1 · vol(Λ) s over the space MK,n [≥ 1] =
F
T ≥1 MK,n [T ].
A(s) Note that if Λ is semi-stable, so is κF ⊗ Λ∨ . Consequently, Λ 7→ κF ⊗ Λ∨
defines a natural involution on MK,n , and interchanges MK,n [≥ 1] and MK,n [≥ 1]
by the duality and the Arakelov Riemann-Roch theorem. Thus
Z
eh (K,Λ) · e−s·χ(OK ,Λ) · dµ(Λ)
0
A(s) :=
Λ∈MK,n [≤1]
Z
0 ∨ ∨
= eh (K,κF ⊗Λ ) · e−s·χ(F,κF ⊗Λ ) dµ(Λ)
Λ∈MK,n [≥1]
Z Z
eh (K,Λ) e s·χ(OK ,Λ) dµ(Λ) = eh (K,Λ) e(s−1)·χ(OK ,Λ) dµ(Λ)
1 0
=
Λ∈MK,n [≥1] Λ∈MK,n [≥1]
Z
0
= eh (K,Λ) · vol(Λ)1−s dµ(Λ) = I(1 − s) + β(s)
Λ∈MK,n [≥1]
Z
where β(s) := vol(Λ)1−s dµ(Λ). Since I(1 − s) is a holomorphic func-
Λ∈MK,n [≥1]
tion on s, it suffices to understand β(s).
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 48

48 Non-Abelian Zeta Functions

α(s) and β(s) From the definitions,

Z Z 1 Z
dT
α(s) = vol(Λ) s · dµ(Λ) = vol(Λ) s · dT µ(Λ)
Λ∈MK,n [≤1] 0 T Λ∈MK,n [T ]
Z 1 Z Z 1 Z
dT s dT
= T · dT µ(Λ) =
s
T · dT µ(Λ).
0 T Λ∈MK,n [T ] 0 T Λ∈MK,n [T ]

Note that there is a natural isomorphism between MK,n [1] → MK,n [T ], say by
log T
sending Λ 7→ Λ ⊗ Λ0 for a fixed rank one lattice Λ0 of degree . Hence
n
Z 1  1
dT   
α(s) = Ts · vol MK,n [1] = vol MK,n [1] · .
0 T s
Similarly,
Z   1
β(s) = vol(Λ)1−s · dµ(Λ) = −vol MK,n [1] · .
Λ∈MK,n [≥1] 1−s
From all these, we have
ζK,r (s) = I(s) + I(1 − s) − α(s) + β(s)
b
!
  1 1
= I(s) + I(1 − s) + vol MK,n [1] · − .
s−1 s
This completes the proof since I(s) is a holomorphic function in s. 

3.2.3 Relation with Epstein Zeta Function

Let a1 = OK , a2 , n. . . , ah be integral
o OK -ideals such that the ideal class group CLK
of K is given by [a1 ], . . . , [ah ] . From the classification of projective OK -modules
in Corollary 1.1, for a rank n projective OK -module P, there exists one and only
one a ∈ a1 , . . . , ah such that P is isomorphic to Pn.a = O⊕(n−1)

K ⊕ a.
Let MK,n be the total moduli space of OK -lattices of rank n, and let Mtot
tot
K,a,n be
the moduli space of OK -lattices of rank n with underlying projective OK -module
Pn.a . From above,
h
G
K,n =
Mtot K,ai ,n .
Mtot
i=1

From Theorem 1.1, up to the factor Fn,+

the moduli space Mtot
K , K,a ,n corresponds to
 ⊕(n−1)  
r1  i
r2 
the space SL(OK ⊕ ai ) SL(n, R) SO(n) × SL(n, C) SU(n) × R.
Set, as before, MK,a,n to be the subspace of MtotK,a,n consisting of (the isometric
T tot
classes of) semi-stable OK -lattices and set MK,a,n [T ] := MK,a,n MK,n [T ]. Then,
by Theorems 1.2 and 1.3, we have:
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 49

Non-Abelian Zeta Functions 49

(1) There is a natural decomposition MK,a,n = T ∈R>0 MK,a,n [T ], and

F
(2) For a fixed T ∈ R∗+ , the moduli space MK,a,n [T ] is compact.

Definition 3.2. Let a be an integral OK -ideal and n ∈ Z>0 . We define a partial

ζK,n.a (s) by
non-abelian zeta function b
Z
0
ζK,n.a (s) := eh (K,Λ) − 1 · (e−s )− log vol(Λ) dµ(Λ), <(s) > 1.
b

MK,a,n

Similar to non-abelian zeta functions, (with almost the same proof we con-
clude that) these partial zeta functions satisfy the standard zeta properties with
a modified functional equation. Indeed, for integral OK -ideals a, b, if [a] = [b]
as ideal classes, then b ζK,n.a (s) = b
ζK,n.b (s). Hence, we may generalize the above
ζK,n.a (s) for an integral OK -ideal a to the case when a is a fractional
definition of b
OK -ideal. Under this modification, it is then rather easy to check that
ζK,n.a (1 − s) = b
b ζK,n.dnK a−1 (s),

since dual lattices of the OK -lattices with underlying projective module O⊕(n−1)
K ⊕a
have d(n−1)
K ⊕ d K a−1
as their underlying projective modules. Here d K denotes the
differential ideal of K. All this then gives a proof of the following:

Lemma 3.2. Let K be a number field and let a be OK -ideal of K. In addition, let
a1 , . . . , ah be integral OK -ideals such that the ideal class group CLK of K is given
by CLK = [a1 ], . . . , [ah ] . Then,


Xh
(0) b ζK,n (s) = bζK,n.ai (s) and b
ζK,a,n (s) = b
ζK,b,n , where b ∈ [a] is an OK -ideal.
i=1
(1) (Meromorphic Continuation) ζK,a,n (s) is a well-defined holomorphic function
b
in s when <(s) > 1, and admits a meromorphic continuation, denoted also by
ζK,a,n (s), to the whole complex s-plane.
b
(2) (Functional Equation) b ζK,a,n (1 − s) = b
ζK,dnK a−1 ,n (s).
(3) (Singularities & Residues) As a meromorphic function over the complex s-
plane, b ζK,a,n (s) admits only two singularities, namely, two simple poles at
s = 0, 1 with the residues ±vol MK,a,n [1] .

To go further, we examine how the integrand

 0 
eh (K,Λ) − 1 · (e−s )− log vol(Λ) dµ(Λ)
behaves on the space
 /
r1
r2 
(n−1)
Fn,+
K × SL(O K ⊕ a) SL(n, R)/SO(n) × SL(n, C)/SU(n) × R∗+ .
ss
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 50

50 Non-Abelian Zeta Functions

By definition,
 
X  X X
h0 (K,Λ)
−1= kxσ kρσ  .
2 2

e exp −π kzσ kρσ − 2π
z∈Λr{0} σ:R σ:C

Thus, in terms of the composition of embeddings

n
z ∈ Λ = O(n−1)
K ⊕ a ,→ K (n) ,→ Rr1 × Cr2 ' (Rn )r1 × (Cn )r2 ,

a point z in Λ maps to the corresponding point (zσ ) in (Rn )r1 ×(Cn )r2 , whose norms
are given by kzσ kρσ = kgσ zσ k. Here, we have assumed that the metric ρσ is defined
by gσ · gtσ for a certain gσ ∈ GL(n, R) when σ is real, and by gσ · ḡtσ for a certain
gσ ∈ GL(n, C) when σ is complex.
Recall that kgσzσ k is O(n), resp. U(n) invariant for real, resp. complex, σ.
0
Hence vol(Λ) and eh (K,Λ) − 1 · (e−s )− log vol(Λ) are well-defined over the space
 r1  r2
GL(n, R)/O(n) × GL(n, C)/U(n) .

Moreover, there is a scalar operation Λ 7→ tΛ for t > 0, which keeps the un-
derlying projective module unchanged, but rescale the metrics by a constant t.
Accordingly,
 
X X X
h0 (K,tΛ)
  
e −1= exp  − π 2
kxσ kρ − 2π kxσ kρ · t  ,
σ
2 2
σ
x∈Λr{0} σ:R σ:C

the volume vol(tΛ), resp. the volume form dµ(Λ), is changed to tn[K:Q] vol(Λ),
dt
resp. · dµ1 (Λ1 ). Here dµ1 (Λ1 ), or simply dµ(Λ1 ) if there is no confusion,
t
denotes the corresponding volume form on the moduli space of semi-stable OK -
 
lattices of rank n corresponding to the points in MK,a,n 1 . This later space is
simply isomorphic to
 /  r1  r2 
(n−1)
Fn,+
K × SL(O K ⊕ a) SL(n, R)/SO(n) × SL(n, C)/SU(n) × {1} . (3.3)
ss
h n i
In the sequel, we identify the moduli space MK,a,n N(a) · ∆K with this realization
2

in terms of SL as in (3.3) above. Moreover, since OK -units have norm 1, units do

not affect neither h0 nor the volume of the lattice. This implies that, for <(s) > 1,
n s
  Z dt
ζK,n.a (s) = RK N(a) · ∆K
b n,+ 2
ts
∗
R+ t
Z  
X  X X  n
exp  − π 2 2 [K:Q]

×  kxσ kρσ− 2π kxσ kρσ t 2  dµ1 (Λ).
Λ∈MK,a,n 1 x∈Λr{0} σ:R σ:C
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 51

Non-Abelian Zeta Functions 51

Z ∞
dtB 1 s
s
Therefore, by applying Mellin transform e−At t s= · A− B · Γ (when-
0 t B B
ever the two sides make sense), we obtain that, for <(s) > 1,
ζK,n.a (s)
b Z
 n s
 = Rn,+
K   dµ1 (Λ)
N(a)∆K2 Λ∈MK,a,n 1

X Y  − ns2  ns   Y  − ns2 !
× πkxσ kρσ Γ · 2πkxσ kρσ Γ(ns)
x∈Λr{0} σ:R
2 σ:C
 
 ns  ns  r 
 r
Z  X 1 
1 −ns 2
= Rn,+K π− 2 Γ 2π Γ(ns) 
ns
 dµ(Λ).
2 Λ∈MK,a,n 1 x∈Λr{0} kxkΛ
   

Definition 3.3. Let Λ be an OK -lattice of rank n. We define a completed Epstein

bΛ (s) associated to Λ by
zeta function E
 ns ns r1 
−ns
bΛ (s) := π− 2 Γ( )
 r 2  n s X 1 
E 2π Γ(ns) · N(a) · ∆K2 (<(s) > 1).
2 x∈Λr{0}
kxkns
Λ

Then, what we have just said proves the following:

Proposition 3.2 (Non-Abelian Zeta = Integration of Epstein Zeta). Let K be a

number field and let a be OK -ideals of K. Then
Z
ζK,a,n (s) = RK
b n,+ bΛ (s) dµ,
 E <(s) > 1.
MK,a,n 1

Here Rn,+
K denotes the (co)volume of the unit lattice U Kn,+ := U Kn ∩ U K+ .

3.2.4 Riemann Hypothesis

Being a natural zeta function, it is natural to formulate the following:

Conjecture 3.1 (Riemann Hypothesis). Let K be a number field and n ∈ Z>0

be a positive integer. Then, all the zeros of the rank n non-abelian zeta function
1
ζKin (s) of K lie on the central line <(s) = .
b
2
When n = 1, this is a well-known conjecture made by Riemann, as one of the
central problems in mathematics. In this book, we will prove the following:

Theorem 3.3 (Weak Riemann Hypothesis). For n ≥ 2, all but finitely many
1
ζK,n (s) lie on the central line <(s) = .
zeros of b
2
For details, please refer to Parts 4, 5 and 6.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 53

PART 2

Rank Two Zeta Functions

Matrices of rank two lattices are parametrized by arithmetic quotients of copies

of upper half planes and upper half-spaces. If the base field is totally real, this
quotient was studied by Siegel, who constructed the associated fundamental do-
main using a function on modular points and the cusps in question. Based on
this idea, it is not too difficult to treat the general cases, opened for some while,
since it is classically known that the cusp classes are finite. In addition, once
Siegel’s functions are suitably normalized, they may be viewed as a sort of natural
distances between the cusps and the modular points, and are proved to be very
natural and extremely useful. For example, we show that, with the help of a work
of T. Hayashi, a lattice of volume one is semi-stable if and only if the distances of
its associated modular point to all cusps are bigger than or equal to one. This then
can be used to write down the rank two zeta functions for number fields explicitly
via an advanced Rankin-Selbarg and Zagier method, based on the associated SL2 -
trace formula. With arbitrary number fields, this later study is certainly new and in
fact quite complicated. We end this part with the work of Suzuki-Lagarias on the
Riemann hypothesis for our rank two zeta functions, based on some elementary
analytic number theoretic arguments on our explicit formulas. As such, the ap-
proach of this part is rather classical, and many fine pieces of traditional algebraic
and analytic number theory are beautifully used.

53
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 55

Chapter 4

Distances to Cusps and

Fundamental Domains

In this chapter, for a number field K, with r1 real embeddings and r2 complex
embeddings, and a fractional ideal a of K, we first define a fractional action of
the modular group SL(a ⊕ OK ) on the hyperbolic space H r1 × Hr2 , where H,
resp. H, denotes the complex upper half plane, resp. the hyperbolic 3 space.
Then we establish a natural correspondence between the associated cusps and
ideal classes, and construct a fundamental domain for the stabilizer subgroup of
a cusp in SL(a ⊕ OK ). Finally, we introduce a notion of distances to cusps, and
construct a fundamental domain for the action SL(a ⊕ OK ) on H r1 × Hr2 .
To make the constructions and structures here more transparent, most of the
sections are divided into three layers. To be more precise, the first reviews the
classical theory for the action of Fuchsian groups of first kind on the upper half
plane, the second recalls Siegel’s famous theory on the actions of the modular
group SL(a ⊕ OK ) associated to totally real field K on H r1 , and the final develops
a new theory for the action of the modular group SL(a⊕OK ) associated to a general
number field K on the hyperbolic space H r1 × Hr2 . We claims neither credits nor
the materials for the theories and constructions appeared in the first two layers,
which mainly follows Kubota ([62]) and Siegel ([101]), respectively.

4.1 Upper Half Space Model

In this section, we first recall some basic facts on the upper half plane and the
hyperbolic 3-space and their associated modular actions. Then we construct a
hyperbolic parameter space for all isometric classes of rank two OK -lattices.

55
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 56

56 Rank Two Zeta Functions

4.1.1 Upper Half Plane

Let H ⊂ C be the upper half plane defined by

H := {z = x + iy ∈ C : x ∈ R, y ∈ R∗+ }. (4.1)
There is a natural hyperbolic metric on H defined by the line element
dx2 + dy2
ds2 := . (4.2)
y2
Obviously, its volume form, resp. its hyperbolic Laplace-Beltrami operator, is
given by
dx ∧ dy ∂2 ∂2

dµ := , resp. ∆ := y2 + 2 . (4.3)
y2 ∂x 2 ∂y
It is well-known that, see e.g. [62], the associated geodesics, sometimes called
hyperbolic lines, are half circles or half lines in H, which are orthogonal to the
boundary real line R of H, in Euclidean geometry.
There is a natural fractional action of the special linear group SL(2, R) on H
given by:
az + b
!
ab
(M, z) 7→ M (z) := , ∀M = ∈ SL(2, R), ∀z ∈ H. (4.4)
cz + d cd
For a point x = x + iy ∈ H (with x, y ∈ R), if we write M(z) = x∗ + iy∗ with
x∗ , y∗ ∈ R, then, by an easy calculation, we have,
(ax + b)(cx + d) + acy2 y
x∗ = , y∗ = > 0. (4.5)
(cx + d)2 + c2 y2 (cx + d)2 + c2 y2
In particular, y∗ depends only on z and the second column of M.
As said, H admits the real line R as its boundary. To compatify it, we add to it
the"associated
# real projective line P1 (R). As usual, we" write
# the elements of P1 (R)
x 1
as , where x, y ∈ R, (x, y) , (0, 0), and set ∞ := . Furthermore, we extend
y 0
the action of SL(2, R) on H in (4.4) naturally to an action on P 1 (R) by
SL(2, R) × P1 (R) −→ P1 (R)
ax + by .
! " #! " #
ab x (4.6)
, →
cd y cx + dy
ax + by
" # " #
x x
Sometimes, we write as . In particular, simply becomes
y y cx + dy
ax + by a yx + b
= x .
cx + dy cy + d
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 57

Distances to Cusps and Fundamental Domains 57

" #
0
Back to H itself. If M is a stabilizer of i = ∈ H in SL(2, R), then
1
ai + b
!
ab
= i. This implies that a = d, b = −c, or the same, ∈ SO(2). Hence,
ci + d cd
the stabilizer group of i in SL(2, R) is equal to SO(2). Moreover, for (x, y) ∈ H,
! √ !
1 x  y 0  1x
if M := ·   ∈ SL(2, R), then M(i) = (i y) = x + iy. Hence, the
01 0 √1y  01
action of SL(2, R) on H is transitive. Therefore, there is a natural identification
between the quotient SL(2, R)/SO(2) and H:
'
SL(2, R)/SO(2) −→ H
(4.7)
[g] 7→ g · (i).

This identification may be further interpreted as follows. For M ∈ SL(2, R),

the coset M · SO(2) naturally corresponds to a positive definite matrix Y = M · M t .

This induces an embedding of the quotient space SL(2, R) SO(2) into the space
P+ of positive definite matrices of size 2. By Iwasawa decomposition, there is a
! √  √ x 
1 x  y 0 y √
unique (x, y) ∈ R2 (y > 0) such that = y
  
  

·   is a representative
01 0 √1y   0 √1y 
.
for the coset M · SO(2). Hence, we can parametrize the image of SL(2, R) SO(2)
 x2 +y2 
x
+
in P by the matrix Y = M · M =  yx 1y  . Therefore, identification (4.7) can
t  
y y
be written as
! √  !
1 x  y 0  1x
M 7→ z M := ·  (i) = (i y) = x + iy ∈ H. (4.8)
0 √1y 

01 01

Moreover, a direct computation shows this explicit identification is SL(2, R)-

equivalent.

4.1.2 Upper Half Space

Similar to the upper half plane model in plane hyperbolic geometry, the follow-
ing upper half space H in Euclidean 3-space R3 is a convenient model for 3-
dimensional hyperbolic space

H := C × R>0 = (z, r) : z = x + iy ∈ C, r ∈ R>0


(4.9)
= (x, y, r) : x, y ∈ R, r ∈ R>0 .


We use both (z, r) and (x, y, r) as coordinate systems of H below.

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 58

58 Rank Two Zeta Functions

To simplify our computations, we view H as a subset of Hamilton’s quater-

nions as usual. We write 1, i, j, k for the standard R-basis of the quaternions with
i2 = j2 = k2 = −1, i j = jk = ki = 1, ji = k j = ik = −1. (4.10)
Then, all points P in H can be written as and hence characterized by
P = (z, r) = (x, y, r) = z + r j where z = x + iy, j = (0, 0, 1). (4.11)
There is a natural hyperbolic metric on H defined by the line element
dx2 + dy2 + dr2
ds2 := . (4.12)
r2
Obviously, its volume form, resp. its hyperbolic Laplace-Beltrami operator, is
given by
2 ∂ ∂2 ∂2
∂
2
dx ∧ dy ∧ dr
dµ := , resp. ∆ := r + + − r . (4.13)
r 3 ∂x 2 ∂y 2 ∂r 2 ∂r
Geodesics with respect to this hyperbolic metric, sometimes called hyperbolic
lines, are half circles or half lines in H which are orthogonal to the boundary com-
plex plane C, in Euclidean geometry. The hyperbolic planes (also called geodesic
hyperplanes) is defined to be the isometrically embedded copies of 2-dimensional
hyperbolic space. They are Euclidean hemispheres or half-planes in Euclidean
geometry, which are perpendicular to the boundary C of H.
There is a natural fractional action of SL(2, C) on H given by
!
ab
(M, P) 7→ M P := (aP + b)(cP + d)−1 , ∀M = ∈ SL(2, C), (4.14)
cd
where the inverse on the right is taken in the skew field of quaternions. Indeed, if
we write P = z + r j ∈ H and set M(z + r j) = z∗ + r∗ j with z∗ ∈ C, r∗ ∈ R, then by
a direct computation, we have
(az + b)(c̄z̄ + d̄) + ac̄r2 r r
z∗ := , r∗ := = , (4.15)
|cz + d| + |c| r
2 2 2 |cz + d| + |c| r
2 2 2 kcP + dk2
where kcP+dk denotes the Euclidean norm of the vector cP+d ∈ R4 , or better, the
square root of the norm of cP + d in the quaternions. Therefore, r∗ > 0 depends
only on P and the second column of M and M(z + r j) ∈ H.
1
The hyperbolic
" # space H admits a natural boundary P (C). As " # usual, we write
x 1
its elements as where x, y ∈ C, (x, y) , (0, 0), and let ∞ := . With a similar
y 0
computation as in § 4.1.1, we may extend the above action of SL(2, C) on H to an
action of SL(2, C) to the boundary P 1 (C) naturally as follows
SL(2, C) × P 1 (C) −→ P 1 (C)
ax + by . (4.16)
! " #! " #
ab x
, →
cd y cx + dy
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 59

Distances to Cusps and Fundamental Domains 59

Similarly to the upper half plane H, the upper half 3-space H also admits a
quotient space interpretation. Indeed, for the action of SL(2, C) on H above, the
stabilizer group of j = (0, 0, 1) ∈ H in SL(2, C) is equal to SU(2), and this action
of SL(2, C) on H is transitive. Therefore, there is a natural identification between
the quotient SL(2, C)/SU(2) and H:
'
SL(2, C)/SU(2) −→ H
(4.17)
[M] 7→ M · ( j).

Equivalently, for M ∈ SL(2, C), the coset M · SU(2) is naturally related to a

positive definite matrix Y = M · M t . This induces an embedding of the quotient
space SL(2, C) SU(2) into the space P+C of positive definite hermitian matrices of

size 2. Thus, by Iwasawa decomposition, there is a unique z ∈ C, r > 0 such
!√  √ z 
1 z  r 0   r √r 
that  =   (z ∈ C, r > 0) gives a representative of the coset
0 1  0 √1r   0 √1r 


M · SU(2). Hence, we can parametrize the image of SL(2, C) SU(2) in P+C by the
.
 |z|2 +r2 
z
t
matrix Y = M · M =  rz̄ 1r  . Therefore, the above identification can be more
 
r r
explicitly written as
!  √  !
1 z  r 0  1z
M 7→ P M :=  ( j) = (r j) = z + r j ∈ H. (4.18)
0 √1r 

01 01

Moreover, a direct computation shows this explicit identification is SL(2, C)-

equivalent.

4.1.3 Rank Two O K -Lattices: Upper Half Space Model

Let K be an algebraic number field. We denote by OK be the ring of integers

of K, denote by CLK be its class group, and denote by U K be its unit group.
Let a1 , . . . , ah be OK -ideals such that CLK = [a1 ], [a2 ], . . . , [ah ] . Let S ∞ be

the set of Archimedean places of K, and set U K+ to be a subgroup of U K defined
by UK+ = ε ∈ U K : εσ > 0, ∀σ real, σ ∈ S ∞ . Moreover, as in Part 1, let

Y
S = (xσ )σ∈S∞ ∈ (R∗+ )r1 +r2 : xσ[Kσ :R] = 1 , and denote F+K the fundamental
s∈S ∞
domain of U K+ in S. Define now the group SL(OK ⊕ a) to be the subgroup of
GL(OK ⊕ a) consisting of elements in of determinant one. Namely
!
OK a
SL(OK ⊕ a) = SL(2, K) ∩ −1 . (4.19)
a OK
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 60

60 Rank Two Zeta Functions

In other words,

ad − bc = 1,

! 


ab 
a, d ∈ OK , b ∈ a,

∈ SL(OK ⊕ a) if and only if (4.20)

cd



c ∈ a−1 .



By Theorem 1.1, all isometry classes of rank two OK -lattices are parametrized
by
G
F+K × SL(OK ⊕ a) SL(2, R)/SO(2) r1 × SL(2, C)/SU(2) r2 × R∗+ .
 


a∈{a1 ,...,ah }

Consequently, by (4.7) and (4.17), there are natural identifications

H ' SL(2, R)/SO(2) and H ' SL(2, C)/SU(2). (4.21)

This then proves the following:

Theorem 4.1. All isometry classes of rank two OK -lattices are parametrized by
G
F+K × SL(OK ⊕ a) H r1 × Hr2 × R∗+ .
 

(4.22)
a∈{a1 ,...,ah }

Put this in a more concrete term, for rank two OK -lattices Λ = (OK ⊕ a, ρ),
when the volumes of Λ’s are fixed, the associated metrics ρ = (ρσ ) are repre-
sented by the matrices g = (gσ )σ∈S∞ with gσ ∈ SL(2, Kσ ). Moreover, by (4.7) and
(4.17), the rank two Ok -lattice Λ associated to a fixed g corresponds uniquely to
the equivalence class of the point g(ImJ) = (gσ τσ )σ∈S∞ ∈ H r1 × Hr2 , where

iσ := (0, 1)
 σ real
ImJ := ( i, . . . i, j, . . . , j ) τσ = 

and (4.23)
|{z} | {z }  jσ := (0, 0, 1) σ complex.

r1 r2

For this reason, in the rest of this chapter, we study only the quotient space
SL(OK ⊕ a) H r1 × Hr2 .


4.2 Cusps and Ideal Classes

In this section, we prove a structural result on the correspondence between cusp

classes and ideal classes for the action of a modular group SL(a ⊕ OK ) on the
associated hyperbolic space H r1 × Hr2 .
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 61

Distances to Cusps and Fundamental Domains 61

4.2.1 Generators of Fractional Ideals

We start with the following well-known remainder theorem. As above, we write

K for an algebraic number field and denote by OK its ring of integers.

Theorem 4.2. (Sun Tsu’s Theorem) Let p1 , . . . , p s be mutually distinct prime

ideals of OK , and let e1 , . . . , e s be positive integers. Then, the natural product-
 e

quotient morphism of rings π : OK → sj=1 OK p j j induces a canonical isomor-
Q
phism
. Ys e j Ys  e

OK pj ' OK p j j .
j=1 j=1

In terms of congruence, this means that, for given x1 , . . . , x s ∈ OK , there exists an

element x ∈ OK satisfies the conditions that

(1) x ≡ x1 mod αe11 , . . . , x ≡ x s mod pess , and

e
(2) its mod sj=1 p j j -class is uniquely determined.
Q

A proof of this classical result can be easily found in literatures, and hence
will be omitted.
As a direct consequence, there is the following elegant result on minimal gen-
erators of a fractional ideals of an algebraic number field, even many standard
textbooks on algebraic number theory omit it.

Lemma 4.1. Let a, b be two fractional ideals of K. There exist two elements
α, β ∈ K such that

OK · α + a · β = b. (4.24)

In particular, every fractional ideal of K is generated by two elements.

Proof. We first reduce the proof to the case b is an integral OK -ideal. Indeed,
by definition, there exist b ∈ K and an integral OK -ideal b0 such that b = b · b0 .
Hence, if there exist two elements α0 , β0 ∈ K such that OK · α0 + a · β0 = b0 , then
b = b · b0 = OK · (bα0 ) + a · (bβ0 ). Therefore, α = bα0 and β = bβ0 would do the job
for the original a, b.
Similarly, we can reduce the proof to the case a is an integral OK -ideal. Indeed,
by definition, there exists an a ∈ K ∗ such that a · a =: a0 is integral. Thus, if there
exist two elements α0 , β0 ∈ K such that OK ·α0 +a0 ·β0 = b, then b = OK ·α0 +a0 ·β0 =
OK · α0 + a · (a−1 β0 ). This implies that α = α0 and β = a−1 β0 would do the job for
the original a, b.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 62

62 Rank Two Zeta Functions

Hence, from now on, we may and will assume that both a, b are integral OK -
ideal. Choose then β ∈ a−1 b\{0} so that a · β ⊆ a · a−1 b = b. By the unique factor-
ization theorem of integral OK -ideals into product of prime ideals for Dedekind
domain OK (see e.g. [66]), there exist distinct prime ideals pi ’s and integers ni ’s
and mi ’s satisfying 0 ≤ ni ≤ mi (1 ≤ i ≤ l) such that
Yl Yl
b= pni ⊇ a · β = pmi .
i=1 i i=1 i

Choose, accordingly, bi ∈ pni i pni i +1 for all i = 1, . . . , l. By Sun Tsu’s Remainder

.

Theorem just cited, there exists an element α ∈ OK such that α ≡ bi mod pni i +1
(1 ≤ i ≤ l). Since, in terms of local orders at each pi , νpi (α) = νpi (b), we conclude
that α ∈ b. Consequently, OK · α + a · β ⊆ b.
On the other hand, if p is a prime ideal of OK which does not lie in the collec-
tion p1 , . . . , pl , then 0 = νp (β) = νp (b). This means that, for all primes p of OK ,

νp (OK · α + a · β) = inf νp (OK · α), νp (a · β) = νp (b). Therefore, OK · α + a · β = b.


As for the last statement, it suffices to take a = OK . 

4.2.2 Special Transformations

We next introduce a special kind of transformations, which interplay a point in the

affine line A1 (K) and the infinity in P 1 (K) canonically.

Lemma 4.2. Let a, b be two fractional ideal of K, let α, β be two elements of K

such that OK α + aβ = b , {0}. There exist two elements α∗ , β∗ of K satisfying the
following conditions.

α α∗
!
(1) ∈ SL(2, K), and
β β∗
(2) OK β∗ + a−1 α∗ = b−1 .

α α∗ α
! " #! " #
1
1
In particular, when acting on ∞ ∈ P (K), we have = .
β β∗ 0 β

Proof. By definition, 1 ∈ OK = b · b−1 = (OK α + aβ) · b−1 = b−1 · α + (ab−1 ) · β.

Hence, there are elements β∗ ∈ b−1 , α∗ ∈ ab−1 such that αβ∗ − βα∗ = 1. This
proves (1).
For (2), it suffices to show that (OK β∗ +a−1 α∗ )·(OK α+aβ) = OK . One inclusion
is clear. Indeed, by our construction, 1 = αβ∗ − βα∗ ∈ (OK β∗ + a−1 α∗ ) · (OK α + aβ).
Hence (OK β∗ + a−1 α∗ ) · (OK α + aβ) ⊇ OK . For the opposite inclusion, we first note
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 63

Distances to Cusps and Fundamental Domains 63

that
(OK β∗ + a−1 α∗ ) · (OK α + aβ) = OK · β∗ α + a−1 · α∗ α + aa−1 · α∗ β + a · ββ∗ .

Moreover, since b = OK α + aβ, we have α ∈ b, β ∈ a−1 b. This, together with

β∗ ∈ b−1 , α∗ ∈ ab−1 , then implies that
(OK β∗ + a−1 α∗ ) · (OK α + aβ)
⊆ OK · b−1 · b + a−1 · (ab−1 ) · b + aa−1 · (ab−1 ) · (a−1 b) + a · (a−1 b)b−1

= OK .
This then proves (2). 

4.2.3 Cusps and Ideal Classes for Total Real Fields

Before treating general number fields, we here give an example treated by Siegel
in [101]. Let K be a number field of degree n over Q. K is called to-
tally real if all Archimedean places of K are real. In this case, we denote by
K = K, K (2) , . . . , K (n) the collection of all conjugates of K. Thus, if we set
 (1)
Hb := H t P 1 (R), there is a natural embedding
P1 (K) ,→ P1 (R) n ⊂ H b
n , a 7→ (a(1) , · · · , a(n) ) =: a.


(4.25)
" #
a
In the sequel, for a point µ ∈ P 1 (K), we choose a, b ∈ OK such that µ = . In
b
a 1
this case, for simplicity, we also write µ = . For example, ∞ = .
b 0
αβ
!
Let G = SL(2, K), Z = {±I} and M = ∈ G. For z = (z1 , . . . , zn ) ∈ H b
n ,
γδ
α( j) z j + β( j)
we set z M := (z∗1 , . . . , z∗n ), where z∗j := ( j) for all j = 1, . . . n. This induces
γ z j + δ( j)
a natural action of G, and hence an action of G/Z, on H b
n defined by
G× H b
n −→ H b
n
(4.26)
(M, z) 7→ z M .
For simplicity, if z j = x j + iy j and z∗j = x∗j + iy∗j with x j , y j , x∗j , y∗j ∈ R, we set
x := (x1 , . . . , xn ), y := (y1 , . . . , yn ), x M := (x1∗ , . . . , xn∗ ), y M := (y∗1 , . . . , y∗n ).
Hence, if z = x + iy ∈ H b
n , then z M = x M + iy M ∈ H b
n . In addition, we
αz + β
" #
1
symbolically denote z M by . For example, in the case ∞ = , we have
γz + δ 0
α(1) α(n)
!
∞M = , . . . , . Accordingly, by the calculations in §4.1.1, we have
γ(1) γ(n)
z M M2 = (z M ) M ∀z ∈ H b
n , ∀M1 , M2 ∈ G. (4.27)
1 2 1
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 64

64 Rank Two Zeta Functions

αβ
!
Let now Γ := SL(2, OK ) be the subgroup of G consisting of elements
γδ
where α, β, γ, δ ∈ OK . The quotient group ΓK := Γ/Z is traditionally called narrow
Hilbert modular group.
By definition, two elements!µ, ν in P1 (K) are called Γ-equivalent, denoted by
αβ αµ + β
µ ∼ ν, if there exists M = ∈ Γ such that ν = µ M = . By (4.27), this
γδ γµ + δ
∼ is an equivalence relation. Denote, as usual, by [µ] the equivalence class of
µ ∈ P 1 (K).
a
For µ = ∈ P1 (K) with a, β ∈ OK , we denote by a := OK a + OK b its
b
associated integral ideal.

Lemma 4.3. The assignment ϕ : [µ] 7→ [a] is well-defined.

αµ + β αβ
!
Proof. Indeed, if ν = µ M = ∼ µ for ∈ SL(2, OK ), then the ideal
γµ + δ γδ
αβ
!
OK (αp + β) + OK (γp + δ) corresponding to ν coincides with a, since is
γδ
uni-modular. Thus, all elements in the equivalence class [µ] associated to µ in
P1 (K) correspond to the same ideal class [a]. On the other hand, if a1 = (c)a is an
integral ideal in the ideal class of [a], then c ∈ K and a1 = OK (ca) + OK (cb)
ca
with ca, cb ∈ OK . Hence, we may take µ = ∈ P1 (K). Conversely, if
cb
a1 is an integral ideal associated to µ ∈ P 1 (K), there exists a1 , b1 ∈ OK such
a a1
that a1 = OK a1 + OK b1 and µ = = . This implies ab1 = a1 b and hence
b b1
OK a OK b1 = OK a1 OK b. Therefore, up to factor of rational ideals, we may and
hence will assume that OK a and OK b as also OK a1 and OK b1 are mutually co-
OK a OK b1 OK a1 OK b
prime. Since = , this implies that, up to factor of rational
a a1 a1 a
O K a O K a1 OK b OK b1
ideals, = and similarly = . This means that a1 = (c)a for
a a1 a a1
a certain c ∈ K. Consequently, for this c ∈ K, we have a1 = ac and b1 = bc. 

Proposition 4.1. Let h be the class number of K. Then there exist h elements
µ1 , . . . , µh ∈ P 1 (K) such that
n o
P1 (K) ∼ = [µ1 ], . . . , [µh ] .

(4.28)

Proof. Let a1 , . . . , ah be OK -ideals such that CLK = [a1 ], . . . , [ah ] . By


Lemma 4.1, for each i, there exist two elements ai , bi ∈ OK such that ai =
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 65

Distances to Cusps and Fundamental Domains 65

ai
OK ai + OK bi . In this way, we obtain an element µi := ∈ P 1 (K) such that
bi
ϕ([µi ]) = [ai ] for all 1 = 1, . . . , h. This proves the surjectivity of ϕ.
a a1
Conversely, for a fixed i, assume that the ϕ-images of both µ = , µ1 = ∈
b b1
P (K) with a, b, a1 , b1 ∈ OK are the same ideal class [ai ]. By Lemma 4.3, we may
1

simply assume that ai = OK a + OK b = OK a1 + OK b1 . Thus, by Lemma 4.1 again,

there exist elements c, d, c1 , d1 in a−1 such that ad − bc = 1 and a1 d1 − b1 c1 = 1.
! i !
ac a1 c1
Consequently, if we set A := and A1 := , then A, A1 ∈ G = SL(2, R).
bd b1 d1
αβ
! ! !
a c d −c
Moreover, A1 A−1 = 1 1 · = , with αδ − βγ = 1. Therefore,
b1 d1 −b a γδ
proceeding in the same way as in the proof of Theorem 4.3 below, we have that
all four elements α = a1 d − c1 b, β = −a1 c + c1 a, γ = b1 d − d1 b, δ = −b1 c + d1 a
belong to OK . Hence, A1 A−1 ∈ Γ and thus
a1 αa + βb α b + β αµ + β
a
µ1 = = = a = = µA1 A−1 ∼ µ.
b1 γa + δb γb + δ γµ + δ
This proves the infectivity of ϕ. 

4.2.4 Cusps and Ideal Classes

We are ready to deal with general number fields. Let K be an algebraic number
field with OK the ring of integers. Denote by r1 , resp. r2 , the number of real, resp.
complex, places of K. Let a be an integral OK -ideal. Induced from the Minkowski
embedding K ,→ Rr1 × Cr2 , we obtain two embeddings

K 2 ,→ Rr1 × Cr2 2 ' R2 r1 × C2 r2 and P 1 (K) ,→ P 1 (R)r1 × P 1 (C)r2 . (4.29)

" #
1
Denote the image of the infinity of P1 (K) by := ∞ 7→ (∞(r1 ) , ∞(r2 ) ).
0
Let
!
OK a
 
SL(OK ⊕ a) := A ∈ −1 : det A = 1 . (4.30)
a OK
We may view SL(2, K), and hence also SL(OK ⊕ a), as a subgroup of the prod-
uct SL(2, R)r1 × SL(2, C)r2 . Obviously, SL(OK ⊕ a) is a discrete subgroup of
SL(2, R)r1 × SL(2, C)r2 . Induced from the canonical fractional transformation ac-
tion of SL(2, R), resp. SL(2, C), on the upper half plane H, resp. on the upper half
space H in §4.1, via (4.29), we obtain a natural action of SL(OK ⊕ a) on the space
H r1 × Hr2 . In the sequel, we study the quotient space SL(OK ⊕ a) H r1 × Hr2 .


November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 66

66 Rank Two Zeta Functions

The product space H r1 × Hr2 admits a natural boundary Rr1 × Cr2 , or better,
1
P (R)r1 × P 1 (C)r2 . By extending the above fractional action to this boundary,
the natural action of SL(OK ⊕ a) on P 1 (K) then induces a SL(OK ⊕ a)-equivalent
embedding P 1 (K) ,→ P 1 (R)r1 × P 1 (C)r2 . Accordingly, we introduce the following

Definition 4.1. We call SL(OK ⊕ a)-orbits in P1 (K) the cusps of the quotient space
SL(OK ⊕ a) H r1 × Hr2 , or the cusps of the arithmetic group SL(OK ⊕ a). Quite


often, we also call representatives of these orbits as cusps of SL(OK ⊕ a).

The cusps of SL(OK ⊕ a) play an important role in the sequel. For later use,
we here give an elegant classification of cusps, rooted back to Maaβ.

Theorem 4.3. (Correspondence between Cusps and Ideal Classes) There is a

natural one-to-one correspondence between the ideal class group CLK of K and
the collection CΓ of the cusps for Γ = SL(OK ⊕ a) acting on H r1 × Hr2 , defined by
α
" # h i
CΓ → CLK , 7→ OK α + a β . (4.31)
β

α
" #
Proof. To start with, let P = be a point in P1 (K) so that α, β ∈ K, and define
β
π(P) to be the ideal class [OK ·α+a·β] associated to the fractional ideal OK ·α+a·β.

Claim 4.1. With the same notation as above, we have

(1) π : P1 (K) → CLK is well-defined.

(2) π factors through the orbit space SL(OK ⊕ a)\P1 (K).

α α
" # " #
Proof. (1) Indeed, if P = 1 = 2 , we have, as ideal classes,
β1 β2
h i h β2 i h α1 i h i
OK · α1 + a · β1 = (OK · α1 + a · β1 ) = OK · β2 · + a · β2 = OK · α2 + a · β2 .
β1 β1
!
ab
(2) For γ = ∈ SL(OK ⊕ a), we have that a, d ∈ OK , b ∈ a, c ∈ a−1 . Note
cd
aα + bβ
" #

h i
that γ(P) = . Hence π γ(P) = OK · (aα + bβ) + a · (cα + dβ) . Now, by
cα + dβ
definition,
OK · (aα + bβ) + a · (cα + dβ) = (aα) · OK + (bβ) · OK + (cα) · a + (dβ) · a
⊆ α · OK + β · a + α · (a−1 · a) + β · a = α · OK + β · a.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 67

Distances to Cusps and Fundamental Domains 67

This implies also that OK · (aα + bβ) + a · (cα + dβ) ⊇ α · OK + β · a, since the
determinant of γ is one. Therefore, we have

h i h i
π γ · P = OK · (aα + bβ) + a · (cα + dβ) = OK · α + a · β .
This verifies (2). 

Back to the proof of Theorem 4.3. Let h = hK denote the class number of K.
Choose now integral OK -ideals a1 , . . . , ah such that CLK = [a1 ], . . . , [ah ] . We

want to show that the elements of P1 (K) are divided into h equivalence classes by
the action of SL(OK ⊕ a). From Claim 4.1, there is a well-defined map
α
" #
Π : SL(OK ⊕ a) P1 (K) → CLK , 7→ [OK · α + a · β].

(4.32)
β
It suffices to show that Π is a bijection.

Surjectivity of Π can be easily verified. Indeed, by Lemma 4.1, for all ideals
" # αi , βi ∈ OK such that ai = OK · αi + a · βi . This
ai (1 ≤ i ≤ h), there exist elements
α
implies that the Π-image of i is exactly ai .
βi
To prove that Π is an injection, let ζ1 , ζ2 be two elements of P1 (K) such " # that
α
Π(ζ1 ) = Π(ζ2 ) = [b] for a fixed OK -ideal b ∈ a1 , . . . , ah . Write ζ1 := 1 and

β1
α2
" #
ζ2 := . We have ( f1 )b = OK α1 + aβ1 and ( f2 )b = OK α2 + aβ2 for suitable
β2
f1 , f2 ∈ K. We want to show that there exists an element γ ∈ SL(OK ⊕ a) such that
γζ2 = ζ1 .
For this, we first assume that ( f1 )b = ( f2 )b = b. By definition, this is equivalent
to the conditions that
OK α1 + aβ1 = b and OK α2 + aβ2 = b. (4.33)
α1 α∗1 α α ∗
! !
By Lemma 4.68, there exist M1 := and M2 := 2 ∗2 ∈ SL(2, K) such
β1 β∗1 β2 β2
that
OK β∗1 + a−1 α∗1 = b−1 and OK β∗2 + a−1 α∗2 = b−1 . (4.34)
Clearly, M1 (∞) = ζ1 and M2 (∞) = ζ2 . Hence, if we set γ := M1 · M2−1 , then
α2 α
" #! " #
γ ∈ SL(2, K), and γ = 1 . In other words, γ(ζ2 ) = ζ1 . Thus, to complete
β2 β1
our proof in this special situation, it suffices to verify the following:

Claim 4.2. We have γ = M1 · M2−1 ∈ SL(2, OK ⊕ a).

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 68

68 Rank Two Zeta Functions

α1 β∗2 − α∗1 β2 −α1 α∗2 + α∗1 α2

!
Proof. By a direct calculation, M1 M2−1 = . More-
β1 β∗2 − β∗1 β2 −β1 α∗2 + β∗1 α2
over, from (4.33) and (4.34),
α1 , α2 ∈ b, β1 , β2 ∈ a−1 b and α∗1 , α∗2 ∈ ab−1 , β∗1 , β∗2 ∈ b−1 . (4.35)
Consequently, we have

(a) α1 β∗2 − α∗1 β2 ∈ bb−1 + (ab−1 )(a−1 b) = OK .

(b) −α1 α∗2 + α∗1 α2 ∈ b(ab−1 ) + (ab−1 )b = a.
(c) β1 β∗2 − β∗1 β2 ∈ (a−1 b)b−1 + b−1 (a−1 b) = a−1 .
(d) −β1 α∗2 + β∗1 α2 ∈ (a−1 b)(ab−1 ) + b−1 b = OK .

Hence γ ∈ SL(OK ⊕ a). 

Back to the proof of Theorem 4.3. If, more generally, ( f1 )b = OK α1 + aβ1

α1 β1
and ( f2 )b = OK α2 + aβ2 for suitable f1 , f2 ∈ K. Then b = OK +a and
f1 f1
α2 β2
b = OK + a . Thus, by the discussion just made, there exists γ ∈ SL(OK ⊕ a)
f1 " f2 #! "
α2 / f2 α /f α2 α
# " #! " #
such that γ = 1 1 . Therefore, γ = 1 . 
β2 / f2 β1 / f1 β2 β1

To end this subsection, we mention that, for the bijection

α
" # h i
Π : SL(2, OK ⊕ a) P (K) ' CLK , 7→ OK α + aβ := b ,
 1
(4.36)
β
its inverse can be easily written down. Indeed, for b, by Lemma 4.1, there exist
αb , βb "∈ K# such that OK αb + a βb = b. Hence, Π−1 ([b]) "is #simply the class of the
α α
point b in SL(2, OK ⊕ a) P 1 (K). Moreover, for η = ∈ P 1 (K), there exists

βb β
α α∗
!
Mη := ∈ SL(2, K) such that Mη (∞) = η.
β β∗

4.3 Stabilizers of Cusps and Their Fundamental Domains

4.3.1 Upper Half Plane

By definition, a subgroup Γ ⊂ SL(2, R) is called a discontinuous group if for every

z ∈ H and every sequence (T n )n≥1 of distinct elements of Γ, the sequence (T n z)n≥1
has no accumulation point in H. And a subgroup Γ ⊂ SL(2, R) is called a discrete
group if the corresponding space in SL(2, R) ⊂ R4 is discrete with respect to the
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 69

Distances to Cusps and Fundamental Domains 69

induced topology. It is well known that a subgroup Γ ⊂ SL(2, R) is discontinuous

if and only if it is discrete. These groups are called Fuchsian groups.
Let Γ ⊂ SL(2, R) be a Fuchsian group. For a point P ∈ H b := H ∪ P1 (R), as
usual, we define the stabilizer group ΓP of P in Γ by
ΓP := γ ∈ Γ : γ P = P .

(4.37)
For example, if ζ ∈ P 1 (R), Γζ can be described by the stabilizer group Γ∞ of ∞,
together with a transformation Aζ : ∞ 7→ ζ. Indeed, if Aζ is an element in SL(2, R)
satisfies A(∞) = ζ, then
Γζ = Aζ · Γ∞ · A−1
ζ . (4.38)

a+b·0
! " # " #
ab 1
Moreover, from definition, γ = ∈ Γ∞ if and only if = . That
cd c+d·0 0
!
a b
means that γ = . Thus, if we let
0 a−1
( ! ) ( ! )
a b 1b
B(R) := : 0 , a ∈ R, b ∈ R and N(R) := : b ∈ R (4.39)
0 a−1 01
be the Borel subgroup B(R) of SL(2, R) and the associated unipotent radical N(R),
respectively, then Γ∞ = Γ ∩ B(R). Note that N(R) is isomorphic to the additive
group R+ , and B(R) is isomorphic to the semi-direct product R∗ by R+ . Hence the
group Γ∞ has a relatively simple structure.

Definition 4.2. A point ζ ∈ P1 (R) is called a cusp of Γ if its stabilizer group Γζ

contains a free abelian group of rank one. That is to say, for a certain d > 0,
( !m ) ( !)
1d 1 dZ
Γζ = Aζ Γ∞ Aζ = Aζ ±
−1
: m ∈ Z Aζ = Aζ ±
−1
ζ .
A−1
01 0 1
Here Aζ is an element of SL(2, R) satisfying Aζ (∞) = ζ. Moreover, Γ is said to be
reduced at the cusp ζ if d = 1.

Let dZ be the rank one Z-lattice associated to the cusp ζ. A standard funda-
mental domain for the action of Γζ on H is set to be
( ) ( )
d d d d
Aζ · z = x+iy ∈ H : − < x ≤ = Aζ · z = x+iy ∈ C : y > 0, − < x ≤ .
2 2 2 2
In the sequel, to simplify our discussions, we assume that Γ is reduced at all of its
cusps, unless otherwise is stated. This implies, in particular, that, for a cusp ζ of
Γ, Γζ /Z(Γ) ' Z, where Z(Γ) := Γ ∩ {±I} denotes the center of Γ.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 70

70 Rank Two Zeta Functions

Let now CΓ be the collection of all cusps of Γ and set

H ∗ := HΓ∗ := H t CΓ . (4.40)
It is a classical fact that the hyperbolic metric on H descends down to Γ\H ∗ . By
definition, Γ is called a Fuchsian group of the first kind if Γ\H ∗ is compact. This
is equivalent to the conditions that Γ is Fuchsian and that the hyperbolic volume of
Γ\H ∗ is finite. In the sequel, for simplicity, we further assume that all subgroups
Γ of SL(2, R) appeared are Fuchsian groups of first kind.
We end this discussion with the following remark on cusps and parabolic ele-
ments. By definition, ! an element γ ∈ SL(2, R) is called parabolic if it conjugates
11
in SL(2, R) to , or equivalently, if Tr(γ) = 2. Easily, an element γ , I in
01
SL(2, R) is parabolic if and only if it has one and only one fixed point on the
boundary P1 (R) of H. Easily, z ∈ P 1 (R) is a cusp of Γ if and only if it is a fixed
point of a parabolic element of Γ.

4.3.2 Upper Half Space

By definition, a subgroup Γ ⊂ SL(2, C) is called a discontinuous group if for every

P ∈ H and for every sequence (T n )n≥1 of distinct elements of Γ, the sequence
(T n P)n≥1 has no accumulation point in H; and a subgroup Γ ⊂ SL(2, C) is called a
discrete subgroup if the corresponding subspace in SL(2, C) ⊂ C4 is discrete with
respect to the induced topology. It is well known that a subgroup Γ ⊂ SL(2, C) is
discontinuous if and only if it is discrete.
b := H ∪ P 1 (C), we define its stabilizer group ΓP in Γ by
For a point P ∈ H
ΓP := γ ∈ Γ : γP = P .

(4.41)
Similarly as above, if Aζ is an element in SL(2, C) satisfying A(∞) = ζ, then
Γζ = Aζ · Γ∞ · A−1
ζ . (4.42)
! !
ab a b
Moreover, γ = ∈ Γ∞ if and only if γ = . Hence, if we set
cd 0 a−1
( ! ) ( ! )
a b 1b
B(C) := : 0 , a ∈ C, b ∈ C and N(C) := : b ∈ C (4.43)
0 a−1 01
to be the Borel subgroup B(C) of SL(2, C) and its unipotent radical N(C), respec-
tively, then Γ∞ = Γ ∩ B(C). This implies that Γ∞ has a very simple structure.

Definition 4.3. An element ζ ∈ P 1 (C) is called a cusp of a discrete subgroup Γ of

SL(2, C) if Γζ contains a free abelian group of rank two.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 71

Distances to Cusps and Fundamental Domains 71

If Λ is the rank two Z-lattice naturally associated to the cusp ζ, there is then a
standard fundamental domain for the action of Γζ on H given by:
n o n o
Aζ · (x+iy)+r j ∈ H : (x, y) ∈ Λ = Aζ · x+iy+r j : x, y, r ∈ R, r > 0, (x, y) ∈ Λ .

We write CΓ for the set of cusps of Γ. Clearly, the group Γ leaves CΓ invariant,
and breaks CΓ into Γ-equivalent classes. It is known that if Γ ⊂ SL(2, C) is a dis-
crete group of finite co-volume, then Γ has only finitely many Γ-classes of cusps.
In this latter case, we denote these finite cusps by η1 , . . . , ηh , and, choose, once and
hence for all, elements A1 , . . . , Ah ∈ SL(2, C) such that η1 = A1 ·∞, . . . , ηh = Ah ·∞.
Similarly, as in the case of upper half plane, cusps are closely related with
the concept of parabolic elements. By definition, an element γ ∈ SL(2, C) r {±I}
is called parabolic if | Tr(γ) | = 2. It is not difficult to check that an !element
11
γ ∈ SL(2, C) is parabolic if and only if it is conjugate in SL(2, C) to . Hence
01
ζ ∈ P 1 (C) is a cusp of Γ if and only if ζ is the fixed point of a certain parabolic
element in Γ (on the boundary P 1 (C) of H).

4.3.3 Fundamental Domains for Stabilizer Groups (I)

In this subsection, to motivate our later discussions for general number fields,
we recall the corresponding constructions for totally real fields, mainly following
Siegel ([101]).

4.3.3.1 Hilbert Modular Group

Let K be a totally real field of degree n over Q with OK its ring of integers. With
respect to the Minkowski embedding K ,→ K ⊗Q R = Rn , for a ∈ K, denote by
(a(i) ) the image of a in Rn .
Let Γ = SL(2, OK ) be the Hilbert modular group and Hn := H n be the product
of n copies of the upper half-plane. Then Γ may be viewed as a group of analytic
automorphisms of Hn via the assignment
αz + β α( j) z j + β( j) αβ
! !
z = (z1 , . . . , zn ) 7→ z M := := ( j) ∀M = ∈ Γ. (4.44)
γz + δ γ z j + δ( j) γδ
In the sequel, we shall identify M ∈ Γ with the modular substitution z 7→ z M if
there is no confusion.
As usual, for a point z of Hn , let Γz := M ∈ Γ : z M = z be the isotopy group

of z in Γ. More generally, for two subsets D and D0 of Hn , we define ΓD,D0 to
be the collection of elements M ∈ Γ satisfying the condition that D M ∩ D0 , ∅.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 72

72 Rank Two Zeta Functions

Here, for an element M of Γ, we denote by D M := z M : z ∈ D the M-image of


D. In other words, an element M of Γ belongs to ΓD,D0 if and only if there exists
w = (w1 , . . . , wn ) ∈ D such that w M = (w01 , . . . , w0n ) ∈ D0 .

Lemma 4.4. Let D, D0 be two compact subsets of Hn . We have

M ∈ Γ : D M ∩ D0 , ∅ < ∞.


In particular, Γz is finite for all z ∈ Hn .

αβ
!
Proof. Let M = ∈ ΓD,D0 . Assume that w = (w1 , . . . , wn ) ∈ D and write
γδ
w M = (w01 , . . . , w0n ) ∈ D0 . For each j = 1, . . . , n, write w j = u j + iv j and
w0j = u0j + iv0j with u j , u0j ∈ R, v j , , v0j ∈ R>0 . By a direct calculation, we have
vj
v0j = ( j) . Since D and D0 are compact, there exists a constant c > 0,
|γ w j + δ( j) |2
depending only on D and D0 but not on the j’s, such that |γ( j) w j + δ( j) | < c for all
j = 1, 2, . . . , n. This implies that, for all M ∈ ΓD,D0 , the associated γ’s and δ’s are
with bounded norms and hence should all belong to a certain finite subset of OK .
δ −γ
! !
0 1
To go further, let A := . Obviously, A · M · A = M =
t −t
.
−1 0 −β α
Accordingly, introduce the modular transformation ι : z 7→ zA on Hn . Note that
D M ∩ D0 , ∅ if and only if ι(D) M−t ∩ ι(D0 ) , ∅. Even ι(D) and/or ι(D0 ) may not be
compact, since D and D0 are compact, by a direct calculation as above but tracing
back to D and D0 , we conclude that, for all M ∈ ΓD,D0 , the associated α’s and β’s
are with bounded norms and hence should all belong to a certain finite subset in
OK . Therefore, ΓD,D0 itself is finite.
In particular, if we choose D = D0 = {z} ⊂ Hn , from above, we conclude that
Γz = Γ{z},{z} is finite. 

Lemma 4.5. Let z ∈ Hn . There exists an open neighborhood U of z such that

there are only finitely many elements M of Γ satisfying the following conditions:

(1) U M intersects U, and

(2) If U M ∩ U , ∅, then M ∈ Γz .

Proof. Let z ∈ Hn and let U 0 be an open neighborhood of z such that the closure
U 0 of U 0 in Hn is compact. By applying Lemma 4.4 to the case D = D0 = U 0 ,
we conclude that there exist at most finitely many M ∈ Γ, denoted by M1 , . . . , Mr ,
such that D M ∩ D , ∅. By reordering if necessary, we may and hence will assume
that M1 , . . . , M s are the only elements which do not belong in Γz . Accordingly,
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 73

Distances to Cusps and Fundamental Domains 73

there exist an open neighborhood U 00 of z such that U M

00
j
∩ U 00 = ∅ for j = 1, . . . , s.
Therefore, we may take U to be the open neighborhood U := U 0 ∩ U 00 of z, since,
by the discussion above, U M ∩ U = ∅ if M , M s+1 , . . . , Mr . 

In particular, if z is not a fixed point of any M in Γ r {±I}, then there exists an

open neighborhood U of z in Hn such that U M ∩ U = ∅ for M ∈ Γ r {±I}. All
these then prove the following:

Proposition 4.2. The Hilbert modular group SL(2, OK ) acts properly and discon-
tinuously on Hn .

4.3.3.2 Stabilizer Groups of Cusps

The action of Γ on Hn above admits an extension to an action to P 1 (K) as fol-

α β αµ + β α( j) µ( j) + β( j)
! !
lows: For µ ∈ P 1 (K), M = ∈ Γ, (M, µ) 7→ µ M := := ( j) ( j) .
γδ γµ + δ γ µ + δ( j)
Denote by Γµ the stabilizer group Γµ of µ in Γ, and by CΓ = [µ1 ], . . . , [µh ] the

collection of cusps classes of Γ.
By definition, for each µ ∈ P 1 (K), there exists a unique i such that µ = (µi ) Mi
αβ
!
for a certain M = Mi ∈ Γ. By an abuse of notation, write Mi as M = . Then,
γδ
αµi + β
Γµ = Mi Γµi Mi−1 and = µ. Since OK is a Dedekind domain, we may and
γµi + δ
ρi ρ
will choose OK elements ρ, σ, ρi , σi such that µi = , µ = . By Theorem 4.3
σi σ
and its proof, since both M and M −1 belong to Γ, as fractional ideals of K, αρi +
βσi , γρi + δσi = ρi , σi = ( f ) ρ, σ) for a certain f ∈ K ∗ . Denote this ideal by ai .



αµi + β
Since = µ, as elements of K, (αρi + βσi ) · ( f σ) = (γρi + δσi ) · ( f ρ). Thus,
γµi + δ
in terms of fractional ideals of K, we have
(αρi + βσi ) ( f σ) (γρi + δσi ) ( f ρ)
· = · .
ai ai ai ai
( f ρ) ( f σ)
By changing f is necessary, we may assume that and are prime to each
ai ai
other. This then implies that
(αρi + βσi ) ( f ρ) (γρi + δσi ) ( f σ)
= , and = .
ai ai ai ai
Therefore, there exists a certain unit ε ∈ O∗K such that αρi + βσi = ε( f ρ) and
γρi + δσi = ε( f σ). Consequently, if we choose a special transformation Ai :=
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 74

74 Rank Two Zeta Functions

ρi ρ∗i ρi bζi∗ ε 0
! ! !
as in Lemma 4.2, then Ai · ∞ = λi , and MAi = , where both
σi σ∗i σi η∗i 0 ε
elements b ζi∗ = (αρ∗i + βσ∗i )ε and η∗i = (γρ∗i + δσ∗i )ε lie in a−1 i . By definition,
ρi σi − σi ρ∗i = 1 = ρi η∗i − σib
∗
ζi∗ . We obtain ρi (η∗i − σ∗i ) = σi (ξ∗ − ρ∗ ). This implies
that
(ρi ) (η∗i − σ∗i ) (σi ) (ξ∗ − ρ∗ )
· = · . (4.45)
ai a−1
i
ai a−1
i
(ρi ) (σi ) (ρi )
Again, since and are prime to each other, we conclude that divides
ai ai ai
(ξ∗ − ρ∗ )
. Hence, there exists an OK -ideal b such that (ξ∗ − ρ∗ ) = a−2
i b (ρi ), or
a−1
i
i such that ζi = ρi + ρi ζ. Consequently, by (4.45), we
the same, there exists ζ ∈ a−2 b∗ ∗

have ηi = σi + σi ζ. Therefore,
∗ ∗

ε ζε−1
!
MAi = Ai .
0 ε−1
This then completes our proof of the following:

α
" #
Proposition 4.3. (Siegel) Let K be a totally real field. Let λ = ∈ P 1 (K) such
β
α α∗
!
that α, β ∈ OK , and that, as OK ideals, (α) is prime to (β). Denote by A = ∈
β β∗
SL(2, K) a special transformation in Lemma 4.2. Then, the stabilizer group Γµ of
λ in Γ = SL(2, OK ) is given by;
ε ζ
( ! )
Γλ = A : ζ ∈ ai , ε ∈ U K A−1 .
−2
0 ε−1

Note that a−2

i is an OK -lattice, and hence a rank two Z-lattice. This justifies the
notion that elements of P 1 (K) are cusps of SL(2, OK ).

4.3.3.3 Fundamental Domain of Cusp Stabilizers

Our aim here is to construct a fundamental domain Dλ for Γµ in Hn for a fixed

α
" #
λ ∈ CΓ , following §2, Chapter III of [101]. We write λ = ∈ P 1 (K) with
β
α α∗
!
α, β ∈ OK , and let A = ∈ SL(2, K) be a fixed matrix of Lemma 4.2, satis-
β β∗
fying the conditions that OK α + OK β = a and α∗ , β∗ ∈ a−1 . Moreover, for a given
z = (z1 , . . . , zn ) ∈ H n , we write N(z) := ni=1 zi for its norm and Tr(z) := ni=1 zi
Q P
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 75

Distances to Cusps and Fundamental Domains 75

for its trace. If λ ∈ K, N(λ) and Tr(λ) coincide with the usual norm and trace. In
addition, we denote zA−1 = (z∗1 , . . . , z∗n ), and write z j = x j + iy j and z∗j = x∗j + iy∗j
with x j , x∗j ∈ R, y j , y∗j ∈ R>0 . For later use, set y∗ := yA−1 := (y∗1 , . . . , y∗n ).

We begin with a system of ‘local coordinates’ for points z in Hn with respect

to λ in terms of the following 2n-tuple:
 
 1
, Y1 , . . . , Yn−1 , X1 , . . . , Xn 

 p
∗
N(y )
where Y1 , . . . , Yn−1 , X1 , . . . , Xn are uniquely determined by the following linear
equations
 
1  y∗k 
Y1 log ε1 + . . . + Yn−1 log εn−1 = log  p
(k) (k)   , 1 ≤ k ≤ n − 1,
2 N(y∗ ) (4.46)
X1 α(l)
1 + . . . + Xn αn =xl ,
(l) ∗
1 ≤ l ≤ n.
We want to know how this system changes under modular transformations in Γµ .
By Proposition 4.3, the group Γµ consists of modular substitutions z 7→ zAT A−1
ε ζε−1
!
where T = with ε ∈ U K and ζ ∈ a−2 . Obviously, z 7→ zAT A−1 is equivalent
0 ε−1
to the transformation zA−1 7→ zT A−1 = ε2 zA−1 + ζ, which is a composition of the
dilation zA−1 7→ ε2 zA−1 and a translations zA−1 7→ zA−1 + ζ. Hence, if we ε1 , . . . , εn−1
be the generators of the unit group U K of K, and α1 , . . . , αn be a Z-basis of the
OK -lattice a−2 , there exist rational integers k1 , . . . , kn−1 , m1 , . . . , mn such that ε =
±εk11 · · · εkn−1
n−1
and ζ = m1 α1 + . . . + mn αn . Consequently, for M = AT A−1 , with
respect to the modular transformation z 7→ z M , a composition of a dilation and a
translation, the above local coordinate system changes as follows (as to be easily
verified):

1
(a) The first coordinate p of z is preserved. Indeed, by definition, we have
N(y∗ )
N(yA−1 M ) = N(yT A−1 ) = N(ε2 )N(yA−1 ) = N(yA−1 ) = N(y∗ ).
(b) The next (n − 1) coordinates Y j ’s are shifted by k j ’s, i.e.
 
Y1 , Y2 , . . . , Yn−1 7→ Y1 + k1 , Y2 + k2 , . . . , Yn−1 + kn−1 .

(c) For the last n coordinates X j ’s, we have

(i) If ε2 = 1, they are shifted by a translation. That is,
 
X1 , . . . , Xn 7→ X1 + m1 , . . . , Xn + mn .
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 76

76 Rank Two Zeta Functions

(ii) If ε2 , 1, they are not merely shifted by a translation, but changed by an

affine transformation as follows:
 
X1 , . . . , Xn 7→ X1∗ + m1 , . . . , Xn∗ + mn .

Here X1∗ , X2∗ , . . . , Xn∗ = X1 , X2 , . . . , Xn · UVU −1 , where U denotes

the n-rowed square matrix (α(i j) ) and V denotes the diagonal matrix
 
(ε(1) )2 , . . . , (ε(n) )2 .

Following Siegel, a point z ∈ Hn is called reduced with respect to Γµ , if

1 1
− ≤Yi < , i = 1, 2, . . . , n − 1,
2 2 (4.47)
1 1
− ≤X j < , j = 1, 2, . . . , n.
2 2
Claim 4.3. For any z ∈ Hn , there exists an M ∈ Γµ such that the equivalent point
z M is reduced with respect to Γµ .

ε 0
!
Proof. Indeed, for ε = ±εk11 · · · εkn−1
n−1
and Mε = A A−1 , the changes on
0 ε−1

Y1 , Y2 , . . . , Yn−1 under the substitution z 7→ z Mε are given by:
 
Y1 , Y2 , . . . , Yn−1 7→ Y1 + k1 , Y2 + k2 , . . . , Yn−1 + kn−1 .
Hence, by choosing k1 , . . . , kn−1 properly, we may assume that the coordinates
Y1 , . . . , Yn−1 of z Mε satisfy (4.47). Moreover, since, for ζ = m1 α1 + . . . + mn αn and
1 ζ −1
! 
Mζ = A A , the changes on X1 , X2 , . . . , Xn under the substitution z 7→ z Mζ
01
are given by:
 
X1 , X2 , . . . , Xn 7→ X1 + m1 , X2 + m2 , . . . , Xn + mn .
Hence, we can choose suitable m1 , . . . , mn such that coordinates X1 , . . . , Xn of
z Mζ Mε satisfy (4.47). Therefore, z Mζ Mε is reduced with respect to Γµ . 

Let now z = x + iy, w = u + iv ∈ Hn be reduced and equivalent with respect

to Γµ . Then we have zA−1 = ε2 wA−1 + ζ for A as above, ε ∈ U K and ζ ∈ a−2 . Write
the local coordinates of z, resp. of w, relative to λ as:
1
p , Y1 , . . . , Yn−1 , X1 , . . . , Xn ,
N(yA−1 )
resp.
1
√ , Y1∗ , . . . , Yn−1
∗
, X1∗ , . . . , Xn∗ .
N(vA−1 )
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 77

Distances to Cusps and Fundamental Domains 77

Then, from the fact that

1 1
Yi∗ ≡ Yi (mod 1), and − ≤ Yi , Yi∗ < , i = 1, 2, . . . , n − 1,
2 2
we have first Yi∗ = Yi , i = 1, 2, . . . , n − 1 and hence ε2 = 1. Moreover, since
1 1
X ∗j ≡ X j (mod 1), and − ≤ X j , X ∗j < , i = 1, 2, . . . , n
2 2
we conclude that X ∗j = X j , j = 1, 2, . . . , n. This implies that ζ = 0 and zA−1 = wA−1 .
Therefore, z = w.
Motivated by all this, denote by Dλ the set of z ∈ Hn whose local coordinates
Y1 , . . . , Yn−1 , X1 , . . . , Xn satisfy (4.47). Then, what we have just verified implies:

(1) For z ∈ Hn , there exists a point of Dλ which is equivalent to z with respect to

the Γµ -action. And
(2) There are no distinct points of Dλ which are Γµ -equivalent.

That is, we have proved the following:

Proposition 4.4. (Siegel) Dλ is a fundamental domain for the Γµ -action in Hn .

For example, when n =n 1, a fundamental domain o of Γµ in H can be described

by the vertical strip D := − 21 ≤ x∗ < 12 , y∗ > 0 with reference to the canonical
coordinate zA−1 = x∗ + iy∗ . In particular, referring to the coordinate z, the vertical
strips x∗ = ± 21 , y∗ > 0 are mapped into semi-circles passing through λ and or-
thogonal to the real axis.

For later use, we give the following:

Corollary 4.1. All points z = x + iy ∈ Dλ satisfying c1 ≤ N(yA−1 ) ≤ c2 lie in

a compact set in Hn , depending only on c1 , c2 and on the choices of generators
ε1 , . . . , εn−1 of U K and generators α1 , . . . , αn of a−2 in K.
 
 y∗i 
Proof. From (4.47), there exist constants c3 , c4 > 0 such that log  p   ≤ c3
N(y∗ )
for all i = 1, 2, . . . , n − 1, and x∗j ≤ c4 for all j = 1, 2, . . . , n. Hence, there exist
y∗
constants c5 , c6 > 0 such that c5 ≤ p i ≤ c6 for all j = 1, 2, . . . , n − 1. By our
N(y∗ )
assumption, c1 ≤ N(y∗ ) ≤ c2 . Hence, there exist constants c7 , c8 > 0 such that
y∗
c7 ≤ p i ≤ c8 , i = 1, 2, . . . , n.
N(y∗ )
January 18, 2018 9:46 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 78

78 Rank Two Zeta Functions

Consequently, there exist constants c9 , c10 > 0 such that

c9 ≤ y∗i ≤ c10 , |x∗j | ≤ c4 i, j = 1, 2, . . . , n
where c9 , c10 , c4 depend only on c1 , c2 and the choices of ε1 , . . . , εn−1 and
α1 , . . . , αn in K. Therefore, zA−1 and therefore z lies in a compact set in Hn , de-
pending on c1 , c2 and K. 

4.3.4 Fundamental Domains for Stabilizer Groups (II)

Let K be an algebraic number field with r1 , resp. r2 , real, resp. complex, embed-
dings of K. Denote by OK the ring of integers of K. Let a be an OK -ideal of K.
There is a natural action of Γ := SL(OK ⊕ a) on H r1 × Hr2 where, as above, H,
resp. H, is the complex upper half plane, resp. the upper half space. Our main aim
in this subsection is to construct a canonical fundamental domain for the quotient
space SL(OK ⊕ a) H r1 × Hr2 , which parametrize rank two OK -lattices of volume


one.

4.3.4.1 Stabilizer Groups

α
" #
Fix a point η = ∈ P 1 (K). Here, we assume, as we may, that α, β ∈ OK . Denote
β
by Γη its associated stabilizer group. Namely,
Γη := γ ∈ SL(OK ⊕ a) : γη = η .

(4.48)
Our main purpose here is to construct a fundamental domain for this action. For
this, we first determine the structure of Γη .

We first transform η to ∞. By Lemma 4.2 on special transformations, there

α α∗
!
exists A = ∈ SL(2, K) satisfying
β β∗

OK β∗ + a−1 α∗ = b−1 , where b := OK α + aβ. (4.49)

α
" #! " #
1
Obviously, A(∞) = A = and Γη = A · Γ∞ · A−1 . Moreover, as a general-
0 β
ization of Proposition 4.3, we have the following:

Theorem 4.4. The stabilizer group Γη is characterized by

ε z
( ! )
A−1 Γη A = : ε ∈ U K , z ∈ ab−2
. (4.50)
0 ε−1
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 79

Distances to Cusps and Fundamental Domains 79

Since ab−2 is a OK -lattice, we call η a cusp of Γ and ab−2 the lattice associated to
the cusp η.

Proof. Since all elements in A−1 Γη A fix ∞, so they are given by upper triangle
! !
−1 a b u z
matrices. Accordingly, we write an element A A of A Γη A as
−1
for
cd 0 u−1
suitable u, z ∈ K. This implies that
α α∗ u z a b α α∗ aα + bβ ∗
! ! ! ! ! !
uα ∗
= and hence = .
β β∗ 0 u−1 c d β β∗ uβ ∗ cα + dβ ∗
In particular, as OK -modules, (uα) = (aα + bβ) and (uβ) = (cα + dβ). This implies
that, as fractional OK -ideals,
(u)(α, β) = (uα, uβ) = (aα + bβ, cα + dβ). (4.51)
Recall that, from (4.30), a, d ∈ OK , b ∈ a and c ∈ a . From d(aα + bβ) − b(cα +
−1

dβ) = (da − bc)α = α, we have, as fractional OK -ideals,

(aα+bβ, cα+dβ) = (α, bβ, dβ) ⊇ (α, bβ, adβ) = (α, bβ, (1−bc)β) = (α, β). (4.52)
This together with (4.51), implies that
(u)(α, β) = (aα + bβ, cα + dβ) = (α, β). (4.53)
Therefore, u ∈ O∗K .
To complete our proof, it suffices to verify the following:

Claim 4.4. We have z ∈ ab−2 .

Proof. Since
β∗ −α∗ a b α α∗
! ! ! !
u z
· · = , (4.54)
−β α cd β β∗ 0 u−1
by a direct calculation, we have that
z = (a − d)α∗ β∗ − c(α∗ )2 + b(β∗ )2 . (4.55)
For entries involved, as mentioned above, a, d ∈ OK , b ∈ a, c ∈ a . Moreover, −1

by (4.49), α ∈ b, β ∈ a−1 b and β∗ ∈ b−1 , α∗ ∈ ab−1 . Therefore,

z ∈ OK · (ab−1 ) · b−1 + a−1 · (ab−1 )2 + a · (b−1 )2 = ab−2 ,

as required. 

ε z ε
! " #! " # " #
1 1
Finally, if z ∈ ab−2 , and ε ∈ U K , we have = = . Conse-
0 ε−1 0 0 0
ε z
!
quently, ∈ A−1 Γη A. 
0 ε−1
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 80

80 Rank Two Zeta Functions

4.3.4.2 Fundamental Domain of Stabilizer Group

Now we are ready to construct a natural fundamental domain for the action of
Γη ⊂ SL(OK ⊕ a) on H r1 × Hr2 .

α α∗
!
Let A = ∈ SL(2, K) be as in the previous subsection. Obviously,
β β∗

α
" # !
u 0
 
A(∞) = η = and Γη = Γ0η × A A−1
: u ∈ U K . (4.56)
β 0 u−1
! !

1 z −1

u 0 uz
Here Γη := A
0
A : z ∈ ab . Since
−2
(z) = −1 = u2 z, the isotropy
01 0 u−1 u
group of η, or the same, the stabilizer group of η, in A−1 SL(OK ⊕ a)A is generated
by:

(i) Translations τ 7→ τ + z where z ∈ ab−2 , and

(ii) Dilations τ 7→ uτ where u runs through the unit group U K2 = ε2 : ε ∈ U K .


Next we analysis how these two types affect our construction of a fundamental
domain for Γη . First, to deal with translations, we introduce the following:

Definition 4.4.

(0) If z = x + iy ∈ H, resp. P = z + r j ∈ H, with x ∈ R, y ∈ R>0 , resp.

z ∈ C, r ∈ R>0 , we set <(z) := x and =(z) = y, resp. Z(P) := z and J(P) = r.
(1) We define a map ImJ by
1 +r2
ImJ : H r1 × Hr2 → Rr>0 , 
(z1 , . . . , zr1 ; P1 , . . . , Pr2 ) 7→ =(z1 ), . . . , =(zr1 ); J(P1 ), . . . , J(Pr2 ) ,
(2) We define a map ReZ by
ReZ : H r1 × Hr2 → Rr1 × Cr2 , 
z1 , . . . , zr1 ; P1 , . . . , Pr2 7→ <(z1 ), . . . , <(zr1 ); Z(P1 ), . . . , Z(Pr2 ) .

Obviously, the canonical ImJ map induces a natural quotient map

 1 +r2
A−1 Γη A H r1 × Hr2 → U K2 Rr>0 ,



 1 +r2
which exhibits A−1 Γη A H r1 × Hr2 as a torus bundle over U K2 Rr>0



with fiber
a real n = r1 + 2r2 dimensional torus Rr1 × Cr2 ab−2 .



Having factored out all translations, we next construct a fundamental domain

1 +r2
for the action of U K2 on Rr>0 . We start with the action of the unit group U K2
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 81

Distances to Cusps and Fundamental Domains 81

1 +r2 1 +r2
n o
on the norm-one hyper-surface S := y ∈ Rr>0 : N(y) = 1 in Rr>0 . By taking
logarithms, S is transformed bijectively into a trace-zero hyperplane H, which is
isomorphic to Rr1 +r2 −1
log Xr1 +r2
S ←→ H := (a1 , . . . , ar1 +r2 ) ∈ Rr1 +r2 : ai = 0 ' Rr1 +r2 −1 ,
n o
i=1
log y1 , . . . , log yr1 +r2 .


y 7→
Accordingly, the actions of unite in U K2 on S are transformed into the actions on
Rr1 +r2 −1 by translations: ai 7→ ai + log ε(i) . By Dirichlet’s Unit Theorem, the
logarithm of U K2 is a full rank lattice in Rr1 +r2 −1 . Consequently, the exponential
map transforms a fundamental parallelepiped of this lattice back to a fundamental
domain, denoted by SUK2 , for the action of the unit group U K2 on the hyper-surface
1 +r2
S. Hence, the cone based on SUK2 , that is the subset R>0 · SUK2 ⊂ Rr>0 , is a
2 r1 +r2
fundamental domain for the action of U K on R>0 .
Therefore, if we denote by T a fundamental domain for the action of the
translations by elements of ab−2 on Rr1 × Cr2 , then what we have just said proves
the following:

Theorem 4.5. Let η ∈ P 1 (K) be a cusp of SL(OK ⊕ a). Then

n o
E := τ ∈ H r1 × Hr2 : ReZ (τ) ∈ T , ImJ (τ) ∈ R>0 · SUK2

is a fundamental domain for the action of A−1 Γη A on H r1 × Hr2 .

For later use, we also set Fη := A−1

η · E.

Surely, as in Siegel’s discussion, we may introduce Y1 , . . . , Yn−1 , X1 , . . . , Xn

together with a ‘reduced norm’ of τ ∈ H r1 × Hr2 to precisely written done this
fundamental domain in a simple form. We leave the details to our later discussions
on a generalization of the classical Rankin-Selberg method.

4.4 Distance to Cusp and Fundamental Domain

4.4.1 Upper Half Plane

Back to the case of a single upper half plane. Let Γ be a Fuchsian group of first
kind. By definition, a connected domain F of H is called a fundamental domain
for the action of Γ on H if the following conditions are satisfied:

(a) There is a disjoint decomposition H = γ∈Γ γF.

F
(b) F = U where U is an open set consisting of all the interior points of F.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 82

82 Rank Two Zeta Functions

(c) For all γ ∈ Γ Z(Γ), γU ∩ U = ∅.



To construct a fundamental domain of Γ, we use the standard hyperbolic dis-

tance d on H. Fix a point z0 ∈ H, which is not an elliptic point of Γ, i.e. z0 is not
a point fixed by any element α ∈ Γ satisfying Tr(α)2 < 4 det(α). And, for every
γ ∈ Γ Z(Γ), we set


Fγ := z ∈ H : d(z, z0 ) ≤ d(z, γz0 )


(4.57)
and denote by Uγ , resp. Cγ , the interior, resp. the boundary, of Fγ .
Introduce a subset F of H by
\ n o
F := Fγ := z ∈ H : d(z, z0 ) ≤ d(z, γz0 ) ∀γ ∈ γ and Lγ := F ∩ γF.
γ∈Γ

Uγ , and for γ ∈ Γ Z(Γ). Then we have the following:

T 
Let U := γ∈Γ\Z(Γ)

Proposition 4.5. (See e.g. [62]) The set F is a fundamental domain for the action
of Γ on H. Moreover, we have

(1) If C is a geodesic joining two points of F. Then C is contained in F.

(2) For all γ ∈ Γ, there is a natural inclusion Lγ ∈ Cγ .
(3) For any compact subset M ⊂ H, the set { γ ∈ Γ : M ∩ γF , ∅ } is finite.

The boundary of F can be described in more details. Indeed, if Lγ , ∅, then

Lγ either consists of a single point or is a (hyperbolic) geodesic. In the later case,
Lγ is called a side of F. It is well known that (see, e.g. [62]):

(i) The boundary of F consists of sides of F.

(ii) Two distinct sides of F are either disjoint or intersected at exactly one point.

To go further, let L and L0 be two sides of F. If L and L0 intersects at p, we

call p a vertex of F in H. Moreover, L and L0 are called linked, denoted as L ∼ L0 ,
if either L = L0 or there exist finitely many distinct sides L1 , . . . , Ln of F such that
L = L1 , L0 = Ln and Lν ∩ Lν+1 , ∅, for 1 ≤ ν ≤ n − 1. Easily, the connected
component of the boundary of F containing a side L of F is a union of all sides
L0 which are linked to L. In the case when a side L of F has no end, we call the
intersection points of the extension of L with P 1 (R) := R ∪ {∞} the vertices of F
on P 1 (R). Here extension is taken by viewing L, a geodesic, as a part of a circle
or a line orthogonal to the real axis. It is also well known that (see, e.g. [62]):

(iii) For a vertex p of F on P 1 (R), if p is an end of two sides and is fixed by a

non-scalar element γ of Γ, then p is a cusp of Γ.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 83

Distances to Cusps and Fundamental Domains 83

(iv) For Γ, a Fuchsian group of the first kind, every vertex of F on P1 (R) is a cusp
of Γ.
(v) Every cusp of Γ is equivalent to a vertex of F on P 1 (R).

To end this discussion, we point out that Γ\H admits a natural Riemann sur-
face structure. To explain this, let CΓ ⊆ P 1 (R) be the set of all cusps of Γ and
H ∗ := HΓ∗ := H ∪ CΓ . And, for a fixed l > 0, let Ul := z ∈ H : =(z) > l

and set Ul∗ := Ul ∪ {∞}. Then, there exists a natural topology structure on H ∗
characterized by

( H ) If z ∈ H, a fundamental neighborhood system at z in H ∗ is defined to be that

of z in H;
(H ∗ ) If x ∈ CnΓ , a fundamental
o neighborhood system at z in H ∗ is defined to be the
family σUl ∗
, where σ ∈ SL(2, R) is an element satisfying σ(∞) = x.
l>0

α α
! " #
ab
Naturally, H is Hausdorff. Indeed, if σ =
∗
∈ Γ and x = = ,
cd β β
( )
=(z)
we have σ(Ul ) = z ∈ H : > l . In particular, σ(Ul ) is contained in a
|cz + d|2
circle of radius (2l2 )−1 which is tangent to the x-axis. As usual, if x ∈ CΓ , we call
σ(Ul ) a neighborhood of x in H as well. For example, if ∞ is a cusp of Γ, then a
fundamental neighborhood system of Γ\H ∗ at ∞ is given by F ∩ Ul∗ l>0 . In fact,

it is easy to check that F ∩ Ul∗ = z ∈ H ∗ : =(z) > l hz 7→ z + mi is a rectangle
 
of width m, starting from y = l and towards infinity.
Tautologically, the action of Γ on H extends naturally to H ∗ . And, with re-
spect to the induced quotient topology Γ H ∗ is Hausdorff as well. Moreover, we

have the following:

Proposition 4.6. The coordinates above induce a compact Riemann surface

structure on Γ H ∗ .


For details, please refer to [100].

4.4.2 Upper Half Space

Let Γ ⊂ SL(2, C) be a discrete group which acts on H discontinuously. By defini-

tion, a closed subset F ∈ H is called a fundamental domain for the action of Γ on
H if

(1) F meets each Γ-orbit at least once,

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 84

84 Rank Two Zeta Functions

(2) The interior F 0 meets each Γ-orbit at most once,

(3) the boundary of F has Lebesgue measure zero.

To construct a fundamental domain for the action of Γ on H, let

Hλ := z + r j : z ∈ C, r > |λ| .

(4.58)
1λ
!
belongs to Γ for a certain λ ∈ C, then, Hλ ∩ M · Hλ = ∅ for


Note that, if
01
!
ab
every M = ∈ Γ with c , 0. Consequently, we have (see e.g. [25])
cd

(i) Two points of Hλ are Γ-equivalent if and only! if they are Γ∞ -equivalent.
1 λi −1
(ii) If A1 , A2 ∈ SL(2, C) satisfying that Ai A ∈ Γ for a certain λi ∈ C r {0},
0 1 i
and that A1 (∞), A2 (∞) ∈ P 1 (C) are Γ-equivalent, then A1 ·Hλ1 ∩ A2 ·Hλ2 = ∅



for all γ ∈ Γ.
(iii) For ζ ∈ P 1 (C), if Γζ contains a parabolic element, then ζ is a cusp of Γ.
Namely, Γµ is associated with a rank two Z-lattice.

Note that, if A ∈ SL(2, C) and λ ∈ C\{0}, the subset A · Hλ is either an open

upper half-space or an open ball in Hλ touching P 1 (C). For this reason, we call
A(Hλ ) a horoball. We have (see e.g. [25])

(iv) All horoballs induce a natural topology on H t P 1 (C).

Indeed, over H, the topology induces by these hornbills is equivalent to the stan-
dard topology. To extend this topology to H∗ , it suffices to add the horoballs
touching P 1 (C) at the λ as systems of open neighborhoods for λ ∈ P 1 (C). Tauto-
logically, we have

(v) SL(2, C) acts continuously on H ∪ P 1 (C).

(vi) Γ H is not compact if Γ contains parabolic elements.

(vii) Γ admits only finitely many cusps if Γ H is with finite volume.


Assume from now on that Γ is of finite co-volume. Let η1 := A1 (∞), . . . , ηh :=

Ah (∞) ∈ P 1 (C) be a complete set of representatives for Γ-cusps. Here
A1 , · · · , Ah ∈ SL(2, C). Since A−1 i Γηi Ai contains a rank two Z-lattice of C, we
may and hence will fix a parallelepiped fundamental domain Pi for the action
of the stabilizer group A−1 i Γηi Ai on P (C) r {∞} = C. Moreover, for Y > 0,
1

let Fi (Yi ) := Ai Fei (Y) where Fei (Y) := z + r j : z ∈ Pi , r ≥ Y . Then, for


Y1 , . . . , Yh ∈ R∗+ large enough, Fi (Yi )’s are contained in the horospheres Ai (Hi )’s.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 85

Distances to Cusps and Fundamental Domains 85

Proposition 4.7. (See e.g. [25])

(1) There exists a compact set F0 ⊂ H such that

F := F0 ∪ F1 (Y1 ) t · · · t Fh (Yh )
is a fundamental domain for Γ.
(2) The compact F0 and the ends Fi (Yi )’s can be chosen to satisfy
(i) The intersections F0 ∩ Fi (Yi ) are all contained in the boundary of F0 , and
hence have Lebesgue measure 0, and
(ii) The intersections Fi (Yi ) ∩ F j (Y j ) is empty if i , j.

As this result will not be used below, we omit its proof.

4.4.3 Fundamental Domain (I)

Let K be a totally real field of degree n with OK its ring of integers. Let Γ :=
SL(2, OK ) be the Hilbert modular group. Denote by Hn := H n be the product of
n copies of the upper half plane H. With respect to the natural action of Γ on Hn ,
the quotient space Γ\Hn is not compact. To compatify it, we add the cusps of Γ.
To motivate our later discussion for general number fields, in this subsection, we
follow Siegel ( [101]) closely to first introduce a Γ-invariant function measuring
the difference between a cusp of Γ and a modular point in Γ\Hn , and then to use
this function to construct a fundamental domain for the action of Γ on Hn .

4.4.3.1 Primitive Distance To Cusp

To facilitate our ensuing discussion, we first introduce some notations. For a point
z = (zk ) = (xk + iyk ) ∈ Hn with xk ∈ R, yk ∈ R>0 , we denote x = xz := ! (xk ) ∈ R ,
n

ab
y = yz := (yk ) ∈ Rn>0 , and let N(y) := nk=1 yk . Moreover, if M =
Q
∈ SL(2, K),
cd
a(k) z + b(k)
!
we write z M for M(z) = (k) . Here, for an element α ∈ K, α(k) denotes its
c z + d(k)
image of the k-th real embedding K → R. Accordingly, we write y-part of M(z) !
yk
as y M(z) , or simply y M . An easy calculation implies that y M = (k) . In
! |c zk + d(k) |2
yk y
the sequel, we denote the n-tuple (k) simply by .
|c zk + d(k) |2 |cz + d|2
ρ ρ
" #
Let µ = = ∈ P 1 (K) be a cusp of Γ. As before, we assume that ρ, σ ∈ OK
σ σ
ρ ξ
!
and set b := OK ρ + OK σ. Accordingly, fix a special transformation A := ∈
ση
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 86

86 Rank Two Zeta Functions

SL(2, K) as in Lemma 4.2. This implies, in particular, b−1 = OK ξ+OK η. Following

Siegel, we introduce the following:
ρ
Definition 4.5. ( [101]) Let z ∈ Hn and µ = ∈ P 1 (K) be a cusp of Γ. We define
σ
a function ∆(z, µ) of (z, µ) by
!1 !1
1 | − σz + ρ|2 2 | − σz + ρ|2 2
∆(z, µ) := 1
= N = N .
N(yA−1 ) 2 y y
n
| − σz + ρ|2 (−σ(k) xk + ρ(k) )2 + (σ(k) yk )2
! Y !
Here, N := .
y k=1
y

1
For example, if µ = ∞, then ∆(z, ∞) = p . Hence, the larger is the N(y),
N(y)
the smaller is the ∆(z, ∞).

For later use, we next give some basic properties for the function ∆(z, µ).

Lemma 4.6. ([101])

(1) ∆(z, µ) is Γ-invariant. Namely, ∆(z, µ) satisfies

∆(z M , µ M ) = ∆(z, µ) ∀M ∈ Γ. (4.59)

(2) ∆(z, µ) is well-defined. That is, it does not depend on the elements ρ, σ and
the matrix A used.

Proof. These are very easy to verify. Indeed, for (1), by definition,
 − 1  − 1
1
∆(z M , µ M ) = N y(zM )A−1 M−1 2 = N yz(A−1 M−1 M) 2 = N yA−1 − 2 = ∆(z, µ).
ρ1 ρ ζ
!
For (2), let µ = and A1 = 1 1 ∈ SL(2, K) with ρ1 , σ1 ∈ OK and A1 satisfy-
σ1 σ1 η1
ing the conditions described before the definition. Then, A−1 A1 ∈ Γ!∞ . Therefore,
ε ζ
by Theorem 4.4, or better, Proposition 4.3, we have A−1 A1 = for a certain
0 ε−1
unit ε in K. This implies that N(yA−1 1
) = N(yA−1 )N(ε−2 ) = N(yA−1 ). 

Lemma 4.7. ([101]) For z = x + iy ∈ Hn , there exists a cusp µ0 of Γ such that,

for all cusps µ of Γ,

∆(z, µ0 ) ≤ ∆(z, µ).

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 87

Distances to Cusps and Fundamental Domains 87

Proof. By definition, it suffices to show that there exists a pair of OK -integers

(ρ0 , σ0 ) such that, for all pairs of OK -integers (ρ, σ),

 −(σ0 x + ρ0 )2 + σ20 y2  −(σx + ρ)2 + σ2 y2

  !
N   ≤ N . (4.60)
y y

As above, write OK ρ+OK σ = a. Assume first that a attaches the minimal norm
among OK -ideals in its associated ideal class [a]. By Dirichlet’s unit theorem, for
every α = (α1 , . . . , αn ) ∈ (K ∗ )n , there exists a unit ε in K such that, for a certain
constant c1 > 0 depending only on K,
pn
αi · ε(i) ≤ c1 |N(α)| (i = 1, . . . , n). (4.61)

−(σx + ρ)2 + σ2 y2
!
Since N keeps unchanged when (ρ, σ) is changed to an as-
y
sociated pair (ερ, εσ) for an ε ∈ U K , there exists a constant c2 > 0, depending
only on c1 , such that, up to associations, for all (ρ, σ)’s,
2
(−σ(i) xi + ρ(i) )2 + σ(i) y2i
≤ c2 , i = 1, 2, . . . , n. (4.62)
yi

This implies that (−σ(i) xi + ρ(i) ) and σ(i) and, hence, ρ(i) and σ(i) are all bounded.
Therefore, up to associations, there exist only finitely many pairs of (ρ, σ) sat-
isfying (4.62). Among these finite pairs, obviously, there ! exists at least a pair,
−(σx + ρ)2 + σ2 y2
say (ρ1 , σ1 ), at which the values of N attaches its minimum.
y
Obviously, (ρ1 , σ1 ) satisfies the inequality (4.60).
Next, to treat general case, let a ∈ K be an element satisfying a =
(a) OK ρ1 + OK σ1 . In the inequality (4.60), if we put ρ = ρ0 := aρ1 and


σ = σ0 := aσ1 , by the minimum property of (ρ1 , σ1 ), we have N(a) ≥ 1.
On the other hand, since a is of minimum norm among OK -ideals of its class,
N(a) ≤ N(a−1 a). Hence N(a) ≤ 1 and N(a) = 1. This then essentially com-
ρ0
pletes the proof. Indeed, by definition, a = OK ρ0 + OK σ0 , and, for λ0 := , we
σ0
1 1

 −(σ0 x + ρ0 ) + σ0 y   −(σ1 x + ρ1 ) + σ1 y 

 2 2 2 2  2 2 2 2
have ∆(z, λ0 ) = N   = N  . There-
y y
fore, by the inequality (4.60) for (ρ1 , σ1 ), we have ∆(z, λ0 ) ≤ ∆(z, µ) for all cusps
µ of Γ. 

Lemma 4.8. ([101]) There exists a constant d > 0, depending only on K, such
that, if ∆(z, µ1 ) < d and ∆(z, µ2 ) < d for all z = x + iy ∈ Hn , then µ1 = µ2 .
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 88

88 Rank Two Zeta Functions

ρ1 ρ2
Proof. Write µ1 = and µ2 = and assume that ρ1 , σ1 , ρ2 , σ2 ∈ OK . Then
σ1 σ2
the conditions ∆(z, µ1 ) < d and ∆(z, µ2 ) < d are equivalent to that

1
1
N − (σ1 x + ρ1 )2 y−1 + σ21 y 2 < d and N − (σ2 x + ρ2 )2 y−1 + σ22 y 2 < d.
As in the proof of Lemma 4.7, by multiplying ρ1 and σ1 with a suitable unit ε ∈
U K , if necessary, we may assume that there exists a constant c1 > 0 such that
(i) 2
1 xi − ρ1 ) yi + σ1 yi < c1 d
(σ(i) (i) 2 −1 2/n
for all i = 1, 2, . . . , n. This implies that
1
(i) − 2 √ 1 √ 1/n
− σ(i)
1 xi + ρ1 yi < c1 d1/n and σ(i)
1 yi <
2
c1 d .
Similarly, there exists a constant c2 > 0 such that
1
(i) − 2 √ 1 √
− σ(i)
2 xi + ρ2 yi < c2 d1/n and σ(i)
2 yi <
2
c2 d1/n .
Therefore,
N(ρ1 σ2 − ρ2 σ1 ) < 2c1 c2 d2/n n ,

1 1 1 1
(i)
− 2 (i)
− 2
1 σ2 −ρ2 σ1 = −σ1 xi +ρ1 yi ·σ2 yi − −σ2 xi +ρ2 yi ·σ1 yi . Hence,
since ρ(i) (i) (i) (i) (i) (i) 2 (i) (i) 2

if we set d = 2c1 c2 −n/2 , N(ρ1 σ2 − ρ2 σ1 ) < 1. This implies that ρ1 σ2 − ρ2 σ1 = 0

since ρ1 σ2 − ρ2 σ1 ∈ OK . Therefore, µ1 = µ2 . 

Lemma 4.9. ([101]) There exists a constant c > 0, depending only on K, such
that, for each z = x + iy ∈ Hn , there exists a cusp µ such that ∆(z, µ) < c.

Proof. Let α1 , . . . , αn be a Z-basis of OK and let c1 , . . . , cn , d1 , . . . , dn be fixed 2n

constants. We first analysis the following system of 2n linear inequalities in 2n
variables a1 , . . . , an , b1 , . . . , bn :
 − 1 (1) − 12
y1 2 α1 a1 + . . . + α(1) n an − x1 y1 α1 b1 + . . . + αn bn
(1) (1)






 ≤ c1






 ·························································
−1 −1

 yn 2 α1 a1 + . . . + αn an − xn yn 2 α1 b1 + . . . + αn bn ≤ cn


 (n) (n)
(n) (n)


1 (4.63)
y12 α(1)
1 b1 + . . . + αn bn
(1)


≤ d1







···························





 y 12 α(n) b + . . . + α(n) b
≤ d .



n 1 1 n n n

Obviously, the determinant of the coefficient matrix of this system is equal to

det(α(i j) )2 = ∆K , the absolute value of the discriminant of K. Therefore, by
Minkowski’s theorem on geometry of numbers, see e.g. [101], the system (4.63)
admits a non-trivial rational integral solution, provided that
p
c1 , . . . , cn , d1 , . . . , dn ≥ ∆K .
2n
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 89

Distances to Cusps and Fundamental Domains 89

p
This is certainly satisfied when ci = d j = ∆K (i, j = 1, 2, . . . , n). In
2n

this latter case, we denote the corresponding non-trivial solution by

a1 , . . . , an , b1 , . . . , bn ∈ Z as well, with an abuse of notation. Accordingly, set
ρ := a1 α1 + . . . + an αn and σ := b1 α1 + . . . + bn αn . Then what we have just said
implies that
(i) 2
pn
(−σ(i) xi + ρ(i) )2 y−1
i +σ yi ≤ 2 ∆ K ∀i = 1, . . . , n.
 
In particular, N −(σx + ρ) y + σ y ≤ 2 ∆K .
2 −1 2 n

Let now a be an OK -ideal such that its norm is minimal among all OK -ideals
of in the (ideal) class [OK ρ + OK σ]. Then, there exist an element a of K such that
aρ ρ
a OK ρ + OK σ = a and N(a) ≤ 1. Therefore, for the cusp µ := = ,


aσ σ
 1
1
∆(z, µ) = N(a) · N −(σx + ρ) y + σ y ≤ 2 ∆K 2 .
2 −1 2 2 n

is valid for every z ∈ Hn . 

4.4.3.2 Fundamental Domain in Case of Totally Real Field

We next use the properties established for ∆(z, µ) above to construct a fundamental
domain for the action of Γ = SL(2, OK ) in Hn following Siegel ([101]). To begin
with, for a cusp µ ∈ P 1 (K) and r > 0, let
Uµ,r := z ∈ Hn : ∆(z, µ) < r

(4.64)
be an influence sphere of µ.

Lemma 4.10. ([101]) We have

(1) Let Dλ is the fundamental domain of Γµ as in §4.3.3.3. Then the subset Uµ,r is
Γµ -invariant. In particular, Dµ ∩ Uµ,r is a fundamental domain of Γµ in Uµ,r .
(2) Let d be the constant in Lemma 4.8. Then the Uµ,d ’s are mutually
n o disjoint.
(3) Let c be the constant in Lemma 4.9. Then the collection Uµ,c is an
µ∈P1 (K)
open covering of Hn .

Proof. The assertion (2), resp. (3), is a direct consequence of Lemma 4.8, resp.
Lemma 4.9. As for (1), note that, if M ∈ Γµ , then µ M = µ. Hence, by
Lemma 4.6(1), we have
∆(z M , µ) = ∆(z M , µ M ) = ∆(z, µ) < r ∀z ∈ Uµ,r
This implies that z M ∈ Uµ,r . Therefore, Uµ,r is Γµ -invariant. 

Definition 4.6. Let z be a point of Hn and µ ∈ P 1 (K) be a cusp of Γ.

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 90

90 Rank Two Zeta Functions

(1) The ∆-invariant ∆(z) of z ∈ Hn is defined by ∆(z) := inf λ∈P 1 (K) ∆(z, µ).
(2) If ∆(z, µ) = ∆(z), z is defined to be semi-reduced with respect to the cusp µ.
(3) The subset Fµ to be the collection of all elements in Hn which are semi-
reduced with respect to µ.

By Lemma 4.7, ∆(z) = ∆(z, µ) holds for at least one cusp µ of Γ. So the assertion
(2) makes sense.

Lemma 4.11. ([101]) We have

(1) If ∆(z) = ∆(z, µ), then ∆(z) is Γµ -invariant. That is, for z ∈ Hn and M ∈ Γµ ,
∆(z M ) = ∆(z). In particular, Fµ is Γµ -invariant.
(2) For the constant c in Lemma 4.9, Fµ ⊂ Uµ,c .
Proof. Since ∆(z, µ) is Γ-invariant,
∆(z M ) = inf ∆(z M , µ) = inf ∆(z, µ M−1 ) = inf ∆(z, µ) = ∆(z).
µ µ µ

This proves (1). (2) comes directly from the constructions of Fµ and Uµ,c . 

Let now µ1 (= (∞, . . . , ∞)), µ2 , . . . , µh be the h inequivalent base cusps of Hn .

For i = 1, . . . , h, denote by Dµi the fundamental domain of Γµi constructed in
§4.3.3.3, and denote by Dµi its topological closure in Hn . Moreover, for our own
convenience, set
\
Di := Fµi Dµi .
Then, in terms of local coordinates X1 , . . . , Xn , Y1 , . . . , Yn−1 introduced in §4.3.3.3,
Di can be described as the collection of z ∈ Hn satisfying the conditions that

− 2 ≤ Xk , Yl ≤ 2 , ∀k = 1, . . . , n, l = 1, . . . , n − 1

 1 1
(4.65)
∆(z, µ) ≥ ∆(z, µi ),

 ∀µ ∈ P1 (K).

Hence, all Di ’s are closed in Hn . Set now F := hi=1 Di .

S

Definition 4.7.

(1) A point z ∈ Hn is called reduced (with respect to Γ) if z is semi-reduced with

respect to some one of the h cusps µ1 , . . . , µh , say µi , and z ∈ Dµi .
(2) (i) A point z ∈ Di is called an inner point of Di if, in the coordinates above,

−1/2 < Xk , Yl < 1/2,

 ∀k = 1, 2, . . . , n, ∀1 ≤ l ≤ n − 1,
(4.66)
∆(z, µ) > ∆(z, µi ),

 ∀µ ∈ P 1 (K) r {µi }.

Denote by D0i the set of inner points of Di .

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 91

Distances to Cusps and Fundamental Domains 91

(ii) If at least one equality holds in the inequalities (4.65), z is called a bound-
ary point of Di . We denote by ∂Di the set of boundary points of Di .
(iii) A point z ∈ F is called an inner point of F , if z is an inner point of some
Di . Similarly, we may define a boundary point of F and hence obtain the
boundary ∂F of F (consisting of all boundary points of F ).

Lemma 4.12. ([101]) We have

(1) The set Di coincides with the topological closure D0i of D0i .
(2) The set of inner points of F is open in Hn .

Proof. (1) Let Ai be a special transformation associated to the cusp µi charac-

terized in Lemma 4.2. Then µi = ∞Ai . Accordingly, for z ∈ Di , let z∗ := zA−1 =
ρ
i

x∗ +iy∗ with x∗ = <(z∗ ), y∗ = =(z∗ ) and ν := µA−1 = , ∞. By definition, σ , 0,

i σ
and
∆(z, µ) ∆(z∗ , ν)  1
= = N (−σx ∗
+ ρ)2
+ (σy∗ 2
2
) .
∆(z, µi ) ∆(z∗ , ∞)
Note that, if y∗ is replaced by t y∗ for t ∈ R>0 , by (4.46), the associated coordinates
X1 , . . . , Xn , Y1 , . . . , Yn−1 of z remain unchanged. Since N (−σx∗ + ρ)2 + (σty∗ )2 is


a strictly increasing function of t, if z ∈ Di , for z(t) = (x∗ + ity∗ )Ai with t > 1, the
inequalities ∆(z(t) , µ) > ∆(z(t) , µi ) are satisfied for all µ , µi . On the other hand, a
small change in the coordinates X1 , . . . , Xn , Y1 , . . . , Yn−1 preserves the inequalities
(4.66). Consequently, if z ∈ Di , every neighborhood of z in Hn must intersect
with D0i . Therefore, D0i = Di .
(2) It is sufficient to show that all the D0i ’s are open in Hn . Recall that

1 ρ
∆(z, µ) = N (−σx + ρ)2 y−1 + σ2 y 2 ∀z = x + iy ∈ Hn , ∀mu = ∈ P 1 (K).
σ
(i)
2 −1 (i) 2
Since the − σ xi + ρ yi + σ yi ’s are positive-definite quadratic forms
(i)

in (σ(i) , ρ(i) )’s, there exists a constant c = c(z) which depends continuously on
1
z, but not on µ, such that ∆(z, µ) ≥ c N(σ2 + ρ2 ) 2 . Consequently, if z0 ∈ D0i ,
there is a sufficiently small neighborhood U of z0 such that, for all z ∈ U,
c0 1
∆(z, µ) ≥ N(σ2 + ρ2 ) 2 , where c0 = c(z0 ). Moreover, by striking U if neces-
2
sary, we may and hence will assume that ∆(z, µi ) ≤ 2∆(z0 , µi ) for all z ∈ U. This
implies that
c0 1 ρ
∆(z, µ) − ∆(z, µi ) ≥ N(σ2 + ρ2 ) 2 − 2∆(z0 , µi ) ∀z ∈ U, ∀µ = ∈ P 1 (K).
2 σ
Hence, following the same arguments in the proof of Lemma 4.7, there
are only finitely many non-associated pairs of integers (ρ, σ) such that
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 92

92 Rank Two Zeta Functions

c0 1
N(σ2 + ρ2 ) 2 ≤ 2∆(z0 , µi ). Hence ∆(z, µ) > ∆(z, µi ) for all µ , µi , except for
2
finitely many points λ1 , . . . , λk ∈ P 1 (K). Therefore, by the continuity of ∆(z, λ j )’s
in z, there exists a neighborhood V of z0 such that, for j = 1, . . . , k, we have
∆(z, λ j ) > ∆(z, µi ), for all z ∈ U and λ j , µi . Consequently, if µ , µi , we have
∆(z, µ) > ∆(z, µi ) for all z ∈ U ∩ V. Furthermore, it is easy to see that, if necessary,
by shrinking V, the inequalities (4.47) hold for all z ∈ V, in addition. This implies
that the open neighborhood U ∩ V of z0 is contained entirely in D0i . 

Consequently, for z ∈ Di , the curve defined by z(t) = (x∗ + ity∗ )Ai , t ≥ 1 lies
entirely in Di . Hence, for sufficiently large t, the z(t) ’s are all contained in D0i .
This can be used to show that Di and D0i are connected. For details, please refer
to [101].
Now we are ready to stated the following main result of Chapter III in
[101].Recall that for F ⊆ Hn and M ∈ Γ, FM is defined to be the collection of
al the z M ’s for all z ∈ F .

Theorem 4.6. ([101]) F is a fundamental domain for Γ in Hn . More precisely,

we have

(1) Denote by FM := z M : z ∈ F . Then { F M } M∈Γ is a covering of Hn .
(2) Distinct inner points of F are not Γ-equivalent.
(3) F intersects only finitely many of its neighborhoods F M1 , . . . , F Ms at points
on the boundary.

Proof. (1) By the comments after Definition 4.6, for every z ∈ Hn , there exists a
cusp µ such that ∆(z) = ∆(z, µ). Since there exist a certain 1 ≤ i ≤ h and an M ∈ Γ
satisfying µ = (µi ) M , we have ∆(z M−1 ) = ∆(z) = ∆(z, µ) = ∆(z M−1 , µi ). Hence
z M−1 ∈ Fµi . Furthermore, by definition, there is an Mi ∈ Γµi such that (z M−1 ) Mi is
reduced with respect to Γµi . Therefore, z Mi M−1 ∈ F .
(2, 3) Let z1 , z2 ∈ F . Assume that z1 ∈ Di , z2 ∈ D j for some 1 ≤ i, j ≤ h, and
that z1 = (z2 ) M for a certain M in Γ r {±I2 }. Since z1 ∈ Fµi , by the Γ-invariance,
we have ∆(z1 , µi ) ≤ ∆ z1 , (µi ) M−1 = ∆(z2 , µ j ). Similarly, we have ∆(z2 , µ j ) ≤


∆ z2 , (µi ) M = ∆(z1 , µi ). Therefore, ∆(z1 , µi ) = ∆(z2 , µi ) = ∆ z1 , (µ j ) M . This



implies that, with the constants c and d as in Lemmas 4.8 and 4.9, either one
of the following two statements holds: (i) ∆(z1 , µi ) = ∆(z2 , µ j ) < d, or (ii)
d ≤ ∆(z1 , µi ), ∆(z2 , µ j ) ≤ c.
We first treat the case ∆(z1 , µi ) = ∆(z2 , µ j ) < d. By Lemma 4.8, µi = (µ j ) M ,
since ∆(z1 , µi ) < d and ∆(z1 , (µ j ) M ) < d. By definition, for different i and j, µi
and µ j are not Γ-equivalent. Hence, i = j and M ∈ Γµi . Furthermore, since both
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 93

Distances to Cusps and Fundamental Domains 93

z1 and z2 are contained in Dµi and z1 = (z2 ) M for a certain M ∈ Γµi , both z1 and z2
are contained in ∂F . In addition, there are only finitely many such M’s in the set
i=1 Γµi .
Sh

Next we deal with the case d ≤ ∆(z1 , µi ), ∆(z2 , µ j ) ≤ c. By Lemmas 4.8 and
4.9, µi and µ j are not Γ-equivalent. To go further, for each 1 ≤ k ≤ h, denote by
Bk the set of z ∈ Hn satisfying d ≤ ∆(z1 , µk ) ≤ c and z ∈ Dµk . By Lemma 4.1, Bi
and hence B = hi=1 Bi are compact. Hence, both z2 and z1 = (z2 ) M belong to a
S
compact set. Consequently, by Lemma 4.4, there are only finitely many such M in
Γ, depending only on B, and hence only on K. Since ∆(z1 , µi ) = ∆ z1 , (µ j ) M and


(µ j ) M , µi , z1 and similarly z2 are contained in the boundary ∂F . This implies that
there are points z, arbitrarily close to z1 and z2 , such that ∆(z1 , µi ) , ∆ z, (µ j ) M .

Therefore, for each of the two cases, there are no two distinct inner points
of F which are Γ-equivalent. Moreover, F intersects only finitely many of its
neighborhoods, namely the F M ’s in both cases, at points on the boundary. 

As a direct consequence, the fundamental domain F just constructed con-

sists of exactly h connected ‘pieces’ corresponding to the Γ-inequivalent cusps
µ1 , . . . , µh , and is bounded by a finite number of manifolds, defined by

∆(z, µi ) = ∆ z, (µ j ) M , ∀i, j = 1, . . . , h, ∀M from the second case,







 (i)


 Xk = ± 21 , ∀k = 1, . . . , n,

Y (i) = ± 1 ,

 ∀l = 1, . . . , n − 1,
l 2

where X1(i) , . . . , Xn(i) , Y1(i) , . . . , Yn−1

(i)
are local coordinates relative to the cusp µi .

Remark 4.1. The manifolds defined by the ∆ equation above are seen to be gen-
ρi
eralizations of the isometric circles for a Fuchsian group. Indeed, for µi = and
σi
ρ
(µ j ) M = , the condition becomes N |σi z − ρi | = N |σz − ρ| . Hence, when n = 1



σ
and µi = ∞, the condition becomes |σz − ρ| = 1. This is simply the ‘isometric
ηz − ξ
circle’ corresponding to the transformation z 7→ on H.
−σz + ρ

4.4.4 Fundamental Domain (II)

Let K be a number field of degree n over Q. Denote by OK its ring of integers and
by ∆K the absolute value of its discriminant. Denote by r1 , resp. r2 , the number of
real embeddings, resp. complex embeddings, of K. Then n = r1 + 2r2 . Moreover,
for an element a ∈ K, denote (a(i) ) its image under the Minkowski embedding
K ,→ Rr1 × Cr2 .
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 94

94 Rank Two Zeta Functions

Fix an OK -ideal a, and let Γ := SL(OK ⊕ a) be the associated modular group.

There is a natural proper and discontinuous action of Γ on H r1 ×Hr2 . By §4.1.3, up
to a factor of F+K dealing with units, the quotient space SL(OK ⊕ a) H r1 × Hr2 can


be viewed as the moduli space of rank two OK -lattices of volume one. Here we
introduce a normalized distance from a modular point in H r1 × Hr2 to a cusp η ∈
P 1 (K) of Γ, and construct a fundamental domain for the action of Γ on H r1 × Hr2 .

4.4.4.1 Distance to Cusps

Recall that, for a point τ = (z1 , . . . , zr1 ; P1 , · · · , Pr2 ) ∈ H r1 × Hr2 , we have let
ImJ(τ) := =(z1 ), . . . , =(zr1 ), J(P1 ), . . . , J(Pr2 ) ∈ Rr1 +r2 , where =(z) = y for z =


x + iy ∈ H with x ∈ R, y ∈ R>0 and J(P) = v for P = z + v j ∈ H with z ∈ C, v ∈
R>0 . Set now
r1 r2

Y Y
N(τ) := N ImJ(τ) = J(P j )2 = y1 · . . . · yr1 · v1 · . . . · vr2 2 . (4.67)



=(zi ) ·
i=1 j=1

By a direct computation, (see e.g. §§4.7 and 4.17,) we have

!
N(ImJ(τ)) ab
N ImJ(γ · τ) = = ∈ Γ.


∀γ (4.68)
kN(cτ + d)k2 cd
It is easy to see that the norm N ImJ(γ · τ) depends only on τ and the second row


of γ.

To introduce a distance to from a modular point τ ∈ H r1 × Hr2 to a cusp

η ∈ P 1 (K) of Γ, recall that, by Theorem 4.3, there is a canonical correspondence
between # Γ-equivalence classes and the ideal classes of K. To be more precise,
" the
α
if η = ∈ P1 (K) is a cusp of Γ, the corresponding ideal class of K is the one
β
associated to the fractional ideal b := OK · α + a · β. From now on, assume that
α, β are all contained
! in OK . By Lemma 4.2, there exists a special transformation
α α∗
A = Aη = ∈ SL(2, K) such that OK β∗ + a−1 α∗ = b−1 and A(∞) = η.
β β∗
Moreover, by Theorem 4.4, the corresponding stabilizer group Γη of η in Γ is given
by
!
u z
 
A−1 Γη A = γ = ∈ Γ : u ∈ U K , z ∈ ab−2
. (4.69)
0 u−1

Definition 4.8. Let τ = z1 , . . . , zr1 : P1 , . . . , Pr2 be a point in H r1 × Hr2 , and

α
" #
η= be a cusp of Γ in P1 (K). We define the reciprocal distance µ(η, τ) between
β
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 95

Distances to Cusps and Fundamental Domains 95

η and τ by
=(z1 ) · · · =(zr1 ) · J(P1 )2 · · · J(Pr2 )2
µ(η, τ) := N a−1 · (OK α + aβ)2 · Qr1


i=1 |(−β zi + α )| j=1 k(−β P j + α )k
(i) (i) 2
Qr2 ( j) ( j) 2

1 N(ImJ(τ))
= · .
N(ab ) kN(−βτ + α)k2
−2

As it stands, this definition is a generalization, and, more importantly, a nor-

malization of the function of Siegel for totally real fields introduced in the previous
subsection.
Our distance plays a key role in the sequel as well. To start with, we summarize
its basic properties in the following tautological lemmata.

Lemma 4.13. Let τ ∈ H r1 × Hr2 and η ∈ P 1 (K). For the reciprocal distance
µ(η, τ) between η and τ, we have

(1) It is well defined. Namely, it does not depend on the choices of α and β.
(2) It is SL(OK ⊕ a)-invariant. That is to say,
µ(γη, γτ) = µ(η, τ), ∀γ ∈ SL(OK ⊕ a).
α α
" # " 0#
Proof. (1) If η = = in P1 (K), there exists λ ∈ K ∗ such that
β β0
α
" 0#
α0 = λ · α and β0 = λ · β. Therefore, µ(η, τ) in terms of 0 is given by
β
1 N(ImJ(τ))
· 0 τ + α0 )k2
where b = OK α + aβ = (λ) · b. Hence, µ(η, τ),
0 0 0
N(ab0 −2 ) kN(−β
α
" 0#
in terms of 0 , is equal to
β
N(λ)2 N(ImJ(τ)) 1 N(ImJ(τ))
· = · .
N(ab ) N(λ) · kN(−βτ + α)k
−2 2 2 N(ab ) kN(−βτ + α)k2
−2

α
" #
This is nothing but µ(η, τ) in terms of .
β
! (2) for a fixed pair α, β of the cusp η. Assume
(2) By (1), it suffices to verify
α α∗
then α, β ∈ OK . Let Aη = be one of the special transformations introduced
β β∗
β∗ −α∗
!
just before Definition 4.8. Since A−1 η = , the second row of A−1
η is to-
−β α
tally determined by α and β, even A may not be unique. With this, by a direct
calculation, we have
1
µ(η, τ) = · N ImJ A−1 η (τ) .



(4.70)
N(ab−2 )
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 96

96 Rank Two Zeta Functions

To complete the proof, it suffices to verify the following:

Claim 4.5. The denominator and the numerator of the right hand of (4.70) are
Γ-invariant. In particular, µ(η, τ) is Γ-invariant.
!
ab
Proof. Let γ = ∈ Γ. By definition, a, d ∈ OK and b ∈ a and c ∈ a−1 . Since
cd
aα + bβ α
" # " #
γ(η) = =: 1 with α1 = aα + bβ and β1 = cα + dβ. Then
cα + dβ β1
b1 := OK α1 +aβ1 = OK aα+OK bβ+acα+adβ ⊆ OK α+aβ+OK α+aβ = OK α+aβ = b.
Since γ, being uni-modular, belongs to Γ, a similar argument as above implies that
b ⊆ b1 and hence b1 = b. Therefore, N(ab−2 1 ) = N(ab ). Hence, the denominator
−2

of the right hand of (4.70) is Γ-invariant. 

To show that the numerator is also Γ-invariant, we first verify the following

Claim 4.6. Let η ∈ P 1 (K) and let γ ∈ Γ be one of the special permutations associ-
ated to the cusp η in Lemma 4.2. Then γAη is a special permutation associated to
the cusp γ(η). Namely
Aγη = γAη . (4.71)

α α α∗
" # ! !
a b
Proof. Let η = with α, β ∈ OK , γ = ∈ Γ and Aη = ∈ SL(2, K).
β c d β β∗
Then
aα + bβ aα∗ + bβ∗ aα + bβ
! !
γAη = ∈ SL(2, K) and γ(η) = ∈ P 1 (K).
cα + dβ cα∗ + dβ∗ cα + dβ
From the proof of the previous claim, OK (aα + bβ)a(cα + dβ) = OK α + aβ = b.
Moreover,
OK (cα∗ + dβ∗ ) + a−1 (aα∗ + bβ∗ ) ⊆ b−1 + b−1 + b−1 + b−1 = b−1
since OK β∗ + a−1 α∗ = b−1 and a, d ∈ OK , b ∈ a, c ∈ a−1 . Hence, similar to the
proof of the previous lemma, this implies that OK (cα∗ +dβ∗ )+a−1 (aα∗ +bβ∗ ) = b−1 ,
since γ ∈ Γ is unimodular. 

γη (γτ) = Aη (τ). Hence,

Back to the proof of the lemma, we have A−1 −1

−1
γη (γτ) = γAη
A−1 η γ (γτ) = Aη γ γτ = Aη (τ).
(γτ) = A−1 −1 −1 −1
−1

Therefore, the numerator of the right hand of (4.70) is Γ-invariant as well. 

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 97

Distances to Cusps and Fundamental Domains 97

Lemma 4.14. There exists a positive constant C, depending only on K and a,

such that, if µ(η, τ) > C and µ(η0 , τ) > C for τ ∈ H r1 × Hr2 and η, η0 ∈ P 1 (K),
then η = η0 .

Proof. By the Cusp-Ideal Class correspondence and the Γ-invariance property

" 0 # of
α α
" #
distance to cusp just proved, we may write our cusps η = , η0 = 0 to
β β
satisfy the conditions that α, β, α0 , β0 ∈ OK , and that both b := OK α + aβ and
b0 := OK α0 + aβ0 have their norms less than a constant C depending only on K.
For our later use, set
1 1
∆(η, τ) := · . (4.72)
N(ab ) µ(η, τ)
−2

It suffices to verify the following:

Claim 4.7. There exists a positive constant c, depending only on K, such that, if
∆(η, τ) < c and ∆(η0 , τ) < c for τ ∈ H r1 × Hr2 and η, η0 ∈ P 1 (K), then η = η0 .

Proof. Similar to the proof of Lemma 4.7, for every (r1 + r2 )-tuple (t1 , . . . , tr1 +r2 )
of non-zero real numbers, by Dirichlet’s unit theorem, there exists a unit ε ∈ K
such that
1
ti ε(i) ≤ c · N(t) r1 +r2 ∀i = 1, . . . , r1 + r2
Qr1 Qr1 +r2 2
where N(t) := i=1 ti · j=r1 +1 t j and c is a constant depending only on K. Hence,
for a suitable constant T > 0 to be determined later, by multiplying α and β with
a suitable uint in K if necessary, we may and hence will assume that
 
2
max =(zi )−1 − β(i) zi + α(i) , J(P j )−2 − β( j) P j + α( j)
1 2 1 2
−r −r
≤ c · ∆(η, τ) 1 +r2 · C r1 +r2 ≤ c · T 1 +r2 · C r1 +r2 .
This implies
n o
max − β(i) <(zi )+α(i) · =(zi )−1/2 , − β( j) Z(P j )+α( j) · J(P j )−1
− 2(r 1+r 1
≤ c1/2 · T 1 2) · C r1 +r2 , (4.73)
− 2(r 1+r ) 1
n o
max β(i) · =(zi )1/2 , β( j) · J(P j ) ≤ c1/2 · T 1 2 ·C r1 +r2
.
In parallel, for α0 and β0 , we obtain similar inequalities. On the other hand, for
real places
α(i) (β0 )(i) − β(i) (α0 )(i) = − β(i) <(zi ) + α(i) =(zi )−1/2 · (β0 )(i) =(zi )1/2

− − (β0 )(i) <(zi ) + (α)(i) =(zi )−1/2 · β(i) =(zi )1/2 ,

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 98

98 Rank Two Zeta Functions

and, for complex places,

α( j) (β0 )( j) − β( j) (α0 )( j) = − β( j) Z(P j ) + α( j) J(P j ) · (β0 )( j) J(P j )

− − (β0 )( j) Z(P j ) + (α0 )( j) J(P j ) · β( j) J(P j ).

Consequently, N αβ0 − βα0 ≤ (2c)r1 +r2 · T −1 · C 2 . Therefore, by (4.73), N αβ0 −

βα0 < 1 if we choose the constant T to satisfy the condition T > (2c)r1 +r2 · C 2 .


Since the norm of the algebraic integer αβ0 − βα0 is strictly less than 1, we have
αβ0 − βα0 = 0, or the same, η = η0 , as desired. This then verifies the claim and
hence also proves the lemma. 

Lemma 4.15. There exists a positive real number T := T (K), depending only on
K, such that, for τ ∈ H r1 × Hr2 , there exists a cusp η such that µ(η, τ) > T .

Proof. By Theorem 4.3, there are finitely many inequivalent cusps. Hence, it
suffices to show that, there exist α, β in OK , both of which are not zero at the same
time, satisfying the inequality
2
N(−βτ + α) · N ImJ(τ) −1 ≤ T −1 .

For this, we examine the following stronger inequalities in terms of Minkowski

coordinates, for some constants of c’s and d’s, depending only on K:

− β(i) <(zi ) + α(i) · =(zi )−1/2 ≤ ci , β(i) · =(zi )1/2 ≤ di , 1 ≤ i ≤ r1

(4.74)
− β( j) Z(P j ) + α( j) · J(P j )−1 ≤ c j , β( j) · J(P j ) ≤ d j , 1 ≤ j ≤ r2 .

Let ω1 , . . . , ωr1 +2r2 be a Z-basis of OK . Then, by changing final r2 inequalities

to the 2r2 inequalities involving only real numbers with the help of complex
conjugations, we arrive at a system of r1 + 2r2 linear inequalities for the lat-
tice points of O  K . Clearly, the coefficient matrix of this final system is simply
given by ω(k) i . Consequently, by the well-known Minkowski theorem on ge-
ometry of numbers, see e.g Theorem 3 in § V.3 of [66], there exists a solution
Xn Xn
α= ai ωi , β = bi ωi with ai , bi ∈ Z for this system of linear inequalities,
i=1 i=1
Y 2
Y 
provided that the number ci · d2j is no less than ω(k)
i = ∆K . Hence, if
1
r +2r
we let ci = d j = ∆K1 2 , the system of inequalities (4.74) certainly have solu-
tions in α, β ∈ OK . Therefore, it suffices to take T as the induced value, namely
2n ∆K . 
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 99

Distances to Cusps and Fundamental Domains 99

Definition 4.9.

(1) For a modular point τ ∈ H r1 × Hr2 and a Γ-cusp η ∈ P 1 (K), we define the
distance d(η, τ) of the modular point τ from the cusp η by
1
d(η, τ) := .
µ(η, τ)1/2
α
" #
(2) For a Γ-cusp η = ∈ P 1 (K), we define the influence sphere Sη of η by
β
n o
Sη := τ ∈ H r1 × Hr2 : µ(η, τ) ≥ µ(η0 , τ), ∀η0 ∈ P1 (K) .

−1/2
For example, if η = ∞, the distance d(η, ∞) is simply N(τ) · N(a) , since,
N(a(OK · 1 + a · 0)2 )N(τ)
by definition, µ(∞, τ) = = N(a) · N(τ).
|N(−0τ + 1)|2

Lemma 4.16. Let η ∈ P 1 (K) be a fixed Γ-cusp. Let τ ∈ H r1 × Hr2 be a modular

point and γ ∈ Γ. If both τ and γτ belong to S0η , then γτ = τ. In particular, the
boundary of Sη consists of pieces of ‘generalized isometric circles’ defined by
equalities µ(η, τ) = µ(η0 , τ) for η0 , η.

Proof. We have, by definition of Sη and the Γ-invariance of the distance to cusps,

for τ, γτ ∈ Soη , we have that

d(γ−1 η, τ) ≥ d(η, τ)
k k
d(η, γτ) ≤ d(γη, γτ)
and that the inequalities are strict if γη , η. 

4.4.4.2 Fundamental Domain

Now we are ready to construct a fundamental domain for Γ = SL(OK ⊕ a) in

H r1 × Hr2 . From Lemma 4.16, the action of Γ in the interior Soη of Sη reduces to
the action of the stabilizer group Γη of η. Hence, for each cusp η of Γ, there exists
a natural embedding

iη : Γη Sη ,→ SL(OK ⊕ a) H r1 × Hr2 .
 

(4.75)

Moreover, from the discussions above, particularly, by Lemma 4.15, in terms of

the maps iη ’s, we obtain a decomposition for the orbit space SL(OK ⊕a) H r1 ×Hr2


as a union of the h pieces Sη ’s. That is, we have the following:
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 100

100 Rank Two Zeta Functions

Theorem 4.7. There is a decomposition for the fundamental domain given by

[
SL(OK ⊕ a) H r1 × Hr2 = iη Γη Sη ,
 

η∈CSL(OK ⊕a)

where CSL(OK ⊕a) denotes the set of SL(OK ⊕ a) cusps.

Similar to the totally real case, these h pieces of Sη ’s glue together along the
so-called ‘pants’ of their boundaries. Note that the action of Γη on H r1 × Hr2 is
free. Therefore, all fixed points of SL(OK ⊕ a) on H r1 × Hr2 lie on the boundaries
of Sη , characterized by the conditions that µ(η, τ) = µ(η0 , τ) for some η0 ∈ P1 (K).
Denote, as before, by η1 , . . . , ηh all the inequivalent cusps for the action of
SL(OK ⊕ a) on H r1 × Hr2 , and choose Aηi ∈ SL(2, K) to satisfy conditions listed
shortly before Definition 4.8, so that, in particular, Aηi (∞) = ηi , i = 1, 2, . . . , h.
Moreover,
1 +r2
n o
(i) Let S be the norm-one hyper-surface S := y ∈ Rr>0 : N(y) = 1 , and denote
by SUK2 a fundamental domain of the action of U K2 on S.
(ii) Let T be a fundamental domain
n for the translation actions of elements in ab−2o
on R × C , and let E := τ ∈ H r1 × Hr2 : ReZ (τ) ∈ T , ImJ (τ) ∈ R>0 · SUK2
r1 r2

be a fundamental domain for the action of A−1 η Γη Aη on H × H .

r1 r2

Then what we just said proves the following:

Theorem 4.8. For the decomposition of fundamental domain above, we have

(1) The space A−1η E ∩ Sη is a fundamental domain for the action of Γ η

on Sη .

(2) There exist elements α1 , . . . , αh of SL(OK ⊕ a) such that hi=1 αi A−1
S
ηi E ∩ Sηi
is a fundamental domain for the action of SL(OK ⊕ a) on H r1 × Hr2 .

To end this discussion, finally, we give an equivalent description for the funda-
mental domain constructed above. For this, we first introduce a natural geometric
truncation within this fundamental domain. Let T > 0 be a constant.
n o
(iii) We let S T := SL(OK ⊕ a) τ ∈ H r1 × Hr2 : µ(η, τ) ≤ T ∀η ∈ CSL(OK ⊕a) . And,
n o
(iv) We set W(η, T ) := τ ∈ H r1 × Hr2 : µ(η, τ) ≤ T .

Assume that T is sufficiently large satisfying the conditions that, for all Γ-cusps
η ∈ P1 (K), W(η, T ) ⊆ Sη and that W(η, T ) W(η0 , T ) = 0
T
 ∅ if [η] , [η] . Ac-
cordingly, the boundary ∂S T consists of h components iη Γη ∂W(η, T ) , each of
dimension 2r1 + 3r2 − 1. Moreover,
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 101

Distances to Cusps and Fundamental Domains 101

 Yr1 Yr2 
1 +r2
(v) We let Σ := (t1 , . . . , tr1 : s1 , . . . , sr2 ) ∈ Rr>0 : ti s2j = 1 , and in-
i=1 i=1
troduce a semi-product E := Rr1 × Cr2 n Σ.

There are actions of Σ on Rr1 × Cr2 and of E on H r1 × Hr2 , defined by component-

wise multiplication and (ui , v j ), (ti , s j ) zi : P j ) := µi zi + ui : s j P j + v j , re-




spectively. Since the boundary ∂W(∞, T ) is a partial homogeneous space of E,
we may and hence will view A−1 η Γη Aη ∂W(∞, Y) as the quotient of E by the dis-

crete subgroup A−1 η Γη Aη . Therefore, the quotient space Aη Γη Aη ∂W(∞, Y) admits
−1 
a natural (r1 + 2r2 )-torus bundle structure over the compact space U K2 Σ, whose

fibers are identified with the torus (Rr1 × Cr2 ) A−1 η Γη Aη . Finally,


n o
(vi) We set Fei (Yi ) := τ ∈ H r1 × Hr2 : ReZ(τ) ∈ Σ, ImJ(τ) ∈ R>T · SUK2 , and let
Fi (Yi ) = Ai · Fei (Yi ).

Clearly, the Fi (Yi )’s are disjoint from each other if the Yi ’s are sufficiently large.
All in all, what we have just discussed proves the following main theorem of
this chapter.

Theorem 4.9. For the action of the modular group SL(OK ⊕ a) on the space
H r1 × Hr2 , we have

(1) Topologically, there is a natural homeomorphism

SL(OK ⊕ a) H r1 × Hr2 ' S T ∪∂S T ∂S T × [0, ∞) .


In particular, SL(OK ⊕a) H r1 ×Hr2 is topologically a manifold with h ‘ends’



of the form T r1 +2r2 -bundle over T r1 +r2 −a × [0, ∞).
(2) Really analytically, there is a canonical isomorphism
SL(OK ⊕ a) H r1 × Hr2 ' S Y ∪ F1 (Y1 ) t · · · t Fh (Yh ) .


We end this chapter by pointing out that our constructions here are compatible
with more general theories of Baily-Borel [8, 9], Borel-Serre [15], Satake [94–96]
and Mumford et al. [1, 54]. We leave the details to the reader.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 103

Chapter 5

Rank Two Zeta Functions

5.1 Distance to Cusp and Stability

Let K be an algebraic number field of degree n (over Q). Denote by OK its ring
of integer, and by ∆K the norm of its discriminant. Let CLK be the ideal class
group of K and let a1 , . . . , ah be the OK -ideals such that CLK = [a1 ], . . . , [ah ] .

Denote by r1 , resp. r2 , the number of real embeddings and complex embeddings
of K. Let S ∞ be the set of Archimedean places of K, and for each σ ∈ S ∞ , set
Nσ = 1, resp. 2, if σ is real, resp. complex. For an element a of K, write (aσ ) =
(aσ )σ∈S∞ , or simply (a(i) ), its image under the canonical Minkowski embedding
K ,→ K ⊗Q R = Rr1 × Cr2 .

5.1.1 Parameter Space of Semi-Stable Rank Two Lattices

By definition, a rank two OK -lattice Λ = (P, ρ) consists of a projective OK -module

P of rank two and a metric ρ on the R-space P ⊗Z R. By Proposition 1.4, P is
isomorphic to the OK -module OK ⊕ a where a = ai for a certain i = 1, . . . , h.
Moreover, since P ⊗Z R ' (R2 )r1 × (C2 )r2 , the metric ρ may and will be iden-
tified with (ρσ )σ∈S∞ , where, for each σ ∈ S ∞ , ρσ is a metric on the associated
σ-component Kσ2 .
Let U K+ be the collection of units ε of OK such that for all real places σ ∈ S ∞ ,
εσ > 0, and let F+K be the fundamental domain of U K+ in the norm one hyper-
surface of (R∗+ )r1 +r2 . Then we have the following:

Theorem 5.1. All isometric classes of rank two OK -lattices are parametrized by
the modular points in the space
G
F+K × SL(OK ⊕ a) H r1 × Hr2 × R∗+ .



(5.1)
a∈{a1 ,...,ah }

103
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 104

104 Rank Two Zeta Functions

This is simply Theorem 1.1 of Part 1. More precisely, the factors in (5.1)
characterize the following inner structures of rank two OK -lattices.

(1) The factor R∗+ parameterizes the volumes of rank two OK -lattices.
(2) The factor F+K takes care of the actions of lattice automorphisms in the form
εI2 with ε ∈ U K and I2 the identity matrix of size 2.
(3) The factor SL(OK ⊕ a) H r1 × Hr2 parameterizes all isometric classes of rank


two OK -lattices with volume N(a)∆K .

Note that, by scaling a lattice with a constant factor t, the associated volume
for the lattice is changed by a factor tn . Hence, we may equally view the factor
SL(OK ⊕ a) H r1 × Hr2 as a parameter space for all isometric classes of rank two


OK -lattices with volume one.

Next we shift our attention to the stability of OK -lattices. By applying the

definition to rank two OK -lattices, we have that a rank two OK -lattice Λ is semi-
stable if and only if
vol(Λ1 )2 ≥ vol(Λ) ∀Λ1 ⊆ Λ. (5.2)
Obviously, when dealing with stability for rank two OK -lattices, the automor-
phisms εI2 with ε ∈ U K can be neglected, since the ε’s have norm one. More-
over, by scaling Λ with a constant factor, the stability for the original lattice and
the resulting lattice remains the same. Therefore, to study stability of rank two
OK -lattices, it suffices to concentrate on the lattices with a fixed volume, say,
N(a)∆K . Consequently, if we denote by SL(OK ⊕ a) H r1 × Hr2 ss the subspace



of SL(OK ⊕ a) H × Hr2 parametrizing rank two semi-stable OK -lattices of vol-
 r1

ume N(a)∆K with underlying projective OK -module OK ⊕ a, then what we have
just said gives the following:

Proposition 5.1. The isometric classes of all rank two semi-stable OK -lattices
are parametrized by
G
F+K × SL(OK ⊕ a) H r1 × Hr2 ss × R∗+ .



(5.3)
a∈{a1 ,...,ah }

There is a natural metric on the space (5.3). Indeed, the invariant Haar measure
dt
gives a canonical metric on the volume part R∗+ . Moreover, starting from the
t
classical Euclidean metric on Rr1 +r2 −1 , through Dirichlet’s unit theorem, there is
an induced flat metric on F+K . Finally, the product of the hyperbolic metrics on
H r1 × Hr2 induces a natural quotient metric on SL(OK ⊕ a) H r1 × Hr2 , since


canonical hyperbolic metric on H and on H are SL(OK ⊕ a)-invariant.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 105

Rank Two Zeta Functions 105

5.1.2 Rank One Sub-Lattices

Let Λ = (OK ⊕ a, ρ) be a rank two OK -lattice of volume N(a) · ∆K . There is the

following classification for rank one sub OK -lattices of Λ by Hayashi:

Lemma 5.1. ( [38]) Let Λ = (OK ⊕ a, (ρσ )) be a rank two OK -lattice of volume
N(a)∆K . We have

(1) For! every rank one sub OK -lattice Λ1 of Λ, there ! exists a non- zero vector
x x
∈ K 2 such that Λ1 is contained in Λ b−1 , where b := OK x + a−1 y.
T
y y
!
x
(2) Λ is semi-stable if and only if, for all nonzero elements ∈ K2,
y
Y x ! 2
σ
≥ N ab2 .


σ∈S
yσ ρσ
∞

Proof. (1) By Lemma 1.3 in Part 1, for a rank one ! sub OK -module Λ1 of OK ⊕ a, !
x T x
there exist a fractional ideal c and an element ∈ K such that Λ1 = Λ c .
2
y y
(See the discussion before Definition 1.7 in Part 1 for the meaning of this notation.)
Accordingly, set b = OK x + a−1 y. By definition, c · x ⊆ OK and c y ⊆ a. Hence
b · c ⊆ OK x + a−1 y · c = c · x + a−1 (c · y) ⊆ OK + a−1 · a = OK .

Therefore, c ⊆ b−1 . This then proves (1).

!
x
(2) We first note that, for rank one sub OK -lattices Λ1 and Λ b−1 of Λ
T
y
!!
x
in (1), by definition, vol Λ1 ≥ vol Λ b−1

T
. Thus, it suffices to check the
y
semi-stable condition
! for all rank one sub OK -lattices Λ1 induced from the sub-
−1 x
modules b . Now, by Example 1.1 in Part 1, a special version of Arakelov
y
Y x !
σ
Riemann-Roch theorem, we have vol(Λ1 ) = N(c) · ∆K · 1/2
. Therefore,
σ
y σ ρσ
 ! 2

1/2
Y xσ 
Λ is semi-stable if and only if  N b ∆K · −1
 ≥ N(a) · ∆K . This is
σ∈S ∞
yσ 
equivalent to
Y x ! 2
σ
≥ N ab2 = N a(OK x + a−1 y) · b



σ∈S ∞
yσ ρσ
= N (OK y + ax)b = N ax + OK y · N OK x + a−1 y ,




as desired. 
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 106

106 Rank Two Zeta Functions

5.1.3 Stability and Distances to Cusps

Let Λ = OK ⊕ a, (ρσ ) be a rank two OK -lattice of volume N(a) ∆K . Denote τΛ

its corresponding modular point in SL(OK ⊕ a) H r1 × Hr2 . Also, for our own


convenience, we define IJ := (i, . . . , i, j, . . . , j) ∈ H r1 × Hr2 . That is,
 √
i = −1 = (0, 1) ∈ H σ : real

(IJ)σ = 

(5.4)
 j = (0, 0, 1) ∈ H
 σ complex.
!
aσ bσ
Let gσ := be an element of SL(2, R), resp. of SL(2, C), such that ρσ is
cσ dσ
the metric on R2 , resp. on C2 , defined by the positive definitive symmetric matrix
gσ gtσ , resp. by the positive definitive Hermitian metric gσ gtσ . For simplicity, we
! !!
ab aσ bσ
write for the collection , g for the collection (gσ ), and denote the
cd cσ dσ
associated metrics (ρσ ) of Λ by ρg . Finally, we define
aσ (IJ)σ + bσ
! !
ab
(IJ) := . (5.5)
cd bσ (IJ)σ + dσ
! !!
ab aσ bσ
Lemma 5.2. For a (r1 + r2 )-tuple of matrices g = = with
cd cσ dσ σ∈S
! ∞

aσ bσ
∈ SL(2, Kσ ), let ρ = ρg = (ρgσ )σ∈S∞ be the associated metrics on the
cσ dσ
space (R2 )r1 × (C2 )r2 . We have
! ! !−1
2
Y
2 Nσ ∗∗ ab
(x, y) ρ = |aσ xσ + cσ yσ | + |bσ xσ + dσ yσ |
2
=N (IJ) .
Λ
σ∈S ∞
xy cd
(5.6)

Proof. Indeed, by definition, we have

2 2 Y Nσ
(x, y) ρ = (x, y)gΛ = |aσ xσ + cσ yσ |2 + |bσ xσ + dσ yσ |2 ,
Λ
σ∈S ∞
! ! ! ! ! (5.7)
∗∗ ab ∗ ∗ ∗ ∗
(IJ) = (IJ) = ((IJ)σ ) .
xy cd ax+cy bx+dy aσ xσ +cσ yσ bσ xσ +dσ yσ
On the other hand,

(a) When σ is a real place, for a point z = X + Yi ∈ H with X, Y ∈ R, if we set

Y
gσ (X + iY) =: Xσ∗ + Yσ∗ i with Xσ∗ , Yσ∗ ∈ R, then, Yσ∗ := .
(cσ X + dσ )2 + c2σ Y 2
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 107

Rank Two Zeta Functions 107

Thus, for z = (IJ)σ = i, we have X = 0, Y = 1, and the corresponding Yσ∗ is

simply given by:
1
(aσ xσ + bσ yσ ) · 0 + (bσ xσ + dσ yσ ) 2 + (aσ xσ + cσ yσ )2 · 12

1
= .
(bσ xσ + dσ yσ ) + (aσ xσ + cσ yσ )2
2

This latest expression is simply equal to the factor associated to the real place
σ on the right hand side of (5.7).
(b) When σ is a complex place, for a point P = Z + V j ∈ H with Z ∈ C and
V ∈ R, if we set M(Z + V j) =: Z ∗ + V ∗ i with Z ∗ ∈ C and V ∗ ∈ R, then
V V
V ∗ := = . Thus, for P = (IJ)σ = j, we have
|cσ Z + dσ | + |σ | V
2 2 2 kcσ P + dσ k2
Z = 0, V = 1, and the corresponding (V ∗ )2 is simply given by:
!2
1
|(aτ xτ + cτ yτ ) · 0 + (bτ xτ + dτ yτ )|2 + |aτ xτ + cτ yτ |2 · 12
!2
1
= .
|bτ xτ + dτ yτ |2 + |aτ xτ + cτ yτ |2
This latest expression is obviously equal to the factor associated to the com-
plex place σ on the right hand side of (5.7).

To complete the proof, it suffices to combine both real and complex places
discussed. 

We are now ready to state the following one of the main results in this Part.

Theorem 5.2. A rank two lattice Λ = OK ⊕ a, ρ of volume N(a)∆K is semi-

stable if and only if, for every cusp η ∈ P 1 (K) of SL(OK ⊕ a), the distance from the
associated modular point τΛ to the cusp η is not strictly less than one. That is,
Λ = (OK ⊕ a, ρ) is semi-stable ⇐⇒ d(η, τΛ ) ≥ 1
∀η ∈ P 1 (K).
!
x
Proof. By Lemma 5.1, it suffices to show that, for all nonzero elements ∈ K2,
y
Y x ! 2
σ
≥ N ab2 .


(5.8)
σ∈S ∞
y σ ρσ
!
x
Fix such an element ∈ K 2 , and let α = y, β = −x. Then, we obtain a cusp
y
α
" #
∈ P 1 (K) of SL(OK ⊕ a). Set b0 = OK β + a−1 α. Then β ∈ b0 and α ∈ ab0 . Hence
β
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 108

108 Rank Two Zeta Functions

if we define b = ab0 , then α ∈ b, β ∈ a−1 b, and

OK α + aβ ⊆ b = ab0 = a · OK β + a−1 α ⊆ OK · aβ + a−1 · aα = aβ + OK α.

Therefore, b = OK α + aβ, and ab20 = a · (a−1 b)2 = a−1 b2 . Consequently, by Lem-

mas 5.2 and 5.1, we conclude that, the rank two OK -lattice Λ = (OK ⊕ a, ρg ) of
α
" #
volume N(a)∆K is semi-stable if and only if, for every cusp η = ∈ P 1 (K),
β
! !
1 ∗ ∗
= µ(η, τΛ ) = N τΛ · N(a−1 b2 ) ≤ 1.
d(η, τΛ ) −β α
This is simply (5.8). 

5.1.4 Example: Truncated Fundamental Domain

As an example, we here reproduce the classical reduction theory with our lan-
guage. Clearly, the moduli space of isometric classes of rank two Z-lattices can
 
be proved to be identified with the double coset space SL(2, Z) SL(2, R) SO(2),

or better, with the quotient space SL(2, Z) H for the natural fractional actions. A
standard fundamental domain D ∈ H is given by:
( ) 
1 1 
D := z ∈ H : |z| ≥ 1, − ≤ x < r {y = 1} ∪ {|z| = 1 : x ≥ 0} . (5.9)
2 2
Obviously, for a rank two Z-lattice Λ of volume 1 in the standard Euclidean space
R2 , there exists an element x ∈ Λ\{0} such that its length kxk is equal to the first
Minkowski successive minimum λ1 of Λ, defined by λ1 := minv∈Λ { kλk}. Thus,
with a rotation if necessary, we may and hence will assume that x = (λ1 , 0).
Accordingly, by the classical reduction theory, there exists a vector ω := x0 + iy0
1
in D such that Λ is a lattice of volume λ−2 1 =: y0 generated by (1, 0) and ω.
λ1
To see when these lattices are semi-stable, we introduce, for a fixed real number
T ≥ 1, the subset DT of D by
DT := z ∈ D : y = =(z) ≤ T .

(5.10)
This, in terms of distance to cusps can be simply characterized by
( )
1
DT = z ∈ D : d(∞, z) ≥ . (5.11)
T
Indeed, (since K = Q,) there is one and only one cusp class, namely, ∞, for the
modular SL(2, Z). Consequently, for the lattices corresponding to the points in
DT , since λ−2
1 = y0 ≤ T , we have λ1 ≤ T . In other words,
−2

1
vol(Λ1 )2 ≥ ∀ rank one Λ1 ⊂ Λ. (5.12)
T
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 109

Rank Two Zeta Functions 109

To write this down in a more theoretical way, we let M≥T

 
Q,2 1 be the moduli space
of rank two Z-lattices Λ of volume 1 for which the volumes of all rank one sub-
lattices are ≤ 12 log T . Then, what we have said proves the following:

Lemma 5.3. (Geometric Truncation = Algebraic Truncation) Let T ≥ 1. The

moduli space DT is naturally identified with the moduli space M≥T
 
Q,2 1 . That is,
there is a canonical isomorphism

Q,2 1 ' DT .
M≥T
 

In particular, for rank two semi-stable Z-lattices of volume one, we have

Q,2 1 =: MQ,2 1 ' D1 .

M≥1
   

In other words, the moduli space of rank two semi-stable lattices of volume
one is given by the space D1 , the part of D under the line y = 1.

5.1.5 Moduli Space of Rank Two Semi-Stable Lattices

Back to the general situation. For a rank two OK -lattice Λ, denote by τΛ ∈ H r1 ×

Hr2 the corresponding moduli point. Then, by Theorem 5.2, Λ is semi-stable if
and only if d(η, τΛ ) ≥ 1 for all cusps η ∈ P 1 (K). This then leads to the following
truncation for the fundamental domain D of SL(OK ⊕ a) H r1 × Hr2 introduced


in §4.4.4.2.

Definition 5.1. Let T ≥ 1, and let a be a fixed OK -ideal.

(1) For a fixed cusp η ∈ P 1 (K), we define

n o
Xη (T ) := τ ∈ H r1 × Hr2 : d(η, τ) < T −1 .
n o
(2) We define DT := τ ∈ D : d(η, τΛ ) ≥ T −1 , ∀cusp η ∈ P 1 (K) .

To understand these spaces, we first establish the following:

Proposition 5.2. Let T ≥ 1 and η1 , η2 ∈ P 1 (K) be two cusps of SL(OK ⊕ a). Then
the following two statements are equivalent:

(1) The intersection Xη1 (T ) and Xη2 (T ) is not empty.

(2) The cusp η1 is equal to the cusp η2 .

This result is an effective version of Lemma 4.14.

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 110

110 Rank Two Zeta Functions

Proof. (2) implies (1) is clear. Hence, it suffices to prove the converse that, if
there exists an element τ ∈ H r1 × Hr2 satisfying both d(τ, η1 ) < 1 and d(τ, η2 ) < 1,
then η1 = η2 .
α α
" # " #
Write η1 = 1 , η2 = 2 with α1 , β1 , α2 , β2 ∈ OK . Set b1 := OK α1 + aβ1
β1 β2
and b2 = OK α2 + aβ2 . Then, the conditions that ηi ∈ Xηi (T ) for i = 1, 2 simply
mean that
! ! ! !
∗ ∗ ∗ ∗
N τ · N(a b1 ) > 1,
−1 2
N τ · N(a−1 b22 ) > 1.
−β1 α1 −β2 α2
Moreover, by definition, we have
α2 α∗2 β∗2 −α∗2
! ! ! ! !! !
∗ ∗ ∗ ∗
N (τ) · N(a b1 ) = N
−1 2
· · (τ) · N(a−1 b21 )
−β1 α1 −β1 α1 β2 β∗2 −β2 α2
! !
∗ ∗
N (τ)
−β2 α2
! ! !
∗∗ ∗ ∗
=N · (τ) · N(a b1 ) =
−1 2
· N(a−1 b21 ).
c d −β2 α2 2
!
∗ ∗
c· (τ) + d
−β2 α2
Here c := α1 β2 − β1 α2 and d := −β1 α∗2 + α1 β∗2 .
!
∗ ∗
Hence, if we set τ0 := (τ), it suffices to prove the following
−β2 α2

Lemma 5.4. With the same notation as above, assume that the following three
conditions are satisfied:

(i) N(τ0 ) · N(a−1 b22 ) > 1,

(ii) N(τ0 ) · N(a−1 b21 ) > kcτ0 + dk2 , and
(iii) α1 , β1 , α2 , β2 ∈ OK ,

Then, α1 β2 − β1 α2 = 0. That is, η1 = η2 .

Proof. First, note that α1 ∈ b1 , β1 ∈ a−1 b1 and α2 ∈ b2 , β2 ∈ a−1 b2 . Hence, we

have c = α1 β2 − β1 α2 ∈ a−1 b1 b2 . This implies that

N(c) ≥ N(a−1 b1 b2 ). (5.13)

Claim 5.1. We have kcτ0 + dk2 ≥ N(c)2 · N(τ0 )2 .

Proof. Indeed, much stronger results, namely, the corresponding inequality for
each Archimedean place, can be verified by direct calculations. We only check
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 111

Rank Two Zeta Functions 111

local inequalities for real places, since complex cases are similar. Thus, let σ be a
real place. Then, by definition,
2
cσ zσ + dσ = (cσ xσ + dσ )2 + c2σ y2σ ≥ c2σ y2σ ,
as desired. 

Hence, by condition (ii) (of the lemma), N(τ0 ) · N(a−1 b21 ) > N(c)2 · N(τ0 )2 .
This means that N(a−1 b21 ) > N(c)2 · N(τ0 ). Therefore, by (i), we further have
N(a−1 b21 ) · N(a−1 b22 ) > N(c)2 , or better, N(a−1 b1 b2 ) > N(c). This contradicts with
(5.13). This completes our proof of the lemma and the proposition. 

Definition 5.2. Let T ≥ 1 and let η1 , η2 , . . . , ηh be all the inequivalent cusps of

SL(OK ⊕ a).

(1) We define Fi (T ) (i = 1, . . . , h) to be the neighborhood of ηi consisting of

τ ∈ D whose distance to ηi is strictly less than T −1 .
(2) We define the truncated domain DT within D by
 
DT := D \ F1 (T ) ∪ . . . ∪ Fh (T ) .
In particular,
 
D1 := D \ F1 (1) t . . . t Fh (1) .

Then, together with Theorem 5.2, what we have just proved in Proposition 5.2
implies the following:

Theorem 5.3. There is a natural identification between the following two spaces:

(1) The moduli space of rank two semi-stable OK -lattices of volume N(a)∆K with
OK ⊕ a as the underlying projective module.
(2) The truncated compact domain D1 consisting of modular points in the funda-
mental domain D whose distances to all cusps are not less than 1.

5.2 Rank Two Non-Abelian Zeta Function

5.2.1 Definition

Let MF,2 be the moduli space of semi-stable OK -lattices of rank 2, and, for a fixed
T > 0, let MK,2 [T ] denote the moduli space of rank two semi-stable OK -lattices
of volumes T . By Theorem 5.1 (and Theorem 5.3), we have
G
MK,2 ' F2K × Da,1 × R∗+ (5.14)
a=a1 ,...,ah
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 112

112 Rank Two Zeta Functions

where F2K is the torus U K2 \Rr1 +r2 −1 , and Da,1 ⊆ H r1 × Hr2 is the truncated funda-
mental domain associated to SL(OK ⊕ a). Moreover, as mentioned before, induces
from the standard flat metric on Rr1 +r2 −1 , resp. hyperbolic metrics dhyp on H and
H, there is a natural metric dx on F2K , resp. dµss on D1 . These two metrics, to-
dt
gether with the canonical Haar measure on R∗+ , induce a natural metric dµ on
t
MK,2 . Symbolically, this means
dt
dµ := dx × dµss × . (5.15)
t
By an abuse of notations, sometimes, we also write dµss simply as dµ, if there is
no confusion.
For each rank two semi-stable OK -lattice Λ = (OK ⊕ a, (ρσ )) ∈ MK,2 , by Defi-
nition 2.2(2), the associated 0-th geo-arithmetical cohomology h0 (K, Λ) is defined
by
 
X  X X 
h (K, Λ) := log  exp − π
0 2
kxσ kρσ − 2π kxσ kρσ  ,
2 
(5.16)
x∈Λ σ:R σ:C
where x ∈ OK ⊕a and (xσ )σ∈S∞ denotes its image under the Minkowski embedding
OK ⊕ a ,→ K 2 ,→ (R2 )r1 × (C2 )r2 .

Furthermore, by Chapter 3 of Part 1, we have the following:

Definition 5.3. Let K be a number field. The rank two non-abelian zeta function
ζK,2 (s) of K is defined by:
b

Z  0 
ζK,2 (s) :=
b eh (K,Λ) − 1 · vol(Λ) s dµ(Λ), <(s) > 1. (5.17)
Λ∈MK,2

Denote by deg(Λ) be the Arakelov degree of Λ. Then, by the Arakelov-

Riemann-Roch Formula (see e.g. Theorem 1.5), the rank two non-abelian zeta
function b ζK,2 (s) can be equivalently, written as
Z  0 
ζK,2 (s) := (∆K ) s eh (K,Λ) − 1 · e−s deg(Λ) dµ(Λ), <(s) > 1. (5.18)
b

Λ∈MK,2
1
Moreover, by replacing Λ with its constant scalar Λ, we have
T
Z Z !
 0  dT
bζK,2 (s) = eh (K,Λ) − 1 · vol(Λ) s dµ(Λ)
R∗+ Λ∈MK,2 [T ] T
X Z Z Z ! (5.19)
 0  dT
= dx × eh (K,Λ) − 1 · vol(Λ) s dhyp µ T −ns .
a=a ,...,a FK2,+ R∗+ τΛ ∈Da,1 T
1 h
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 113

Rank Two Zeta Functions 113

By §3.2.2, the integrations above are all convergent. Hence, b ζK,2 (s) is well-
defined. More generally, we have the following results on basic zeta properties
ζK,2 (s).
of b

ζK,2 (s) of K, we have

Theorem 5.4. For the rank two zeta function b

(1) (Meromorphic Continuation) Rank two non-abelian zeta function b ζK,2 (s) is a
well-defined holomorphic function when <(s) > 1, and admits a meromor-
phic continuation, denoted also by b ζK,2 (s), to the whole complex s-plane.
ζK,2 (s) satisfies the standard functional equation
(2) (Functional Equation) b
ζK,2 (1 − s) = b
b ζK,2 (s).
(3) (Singularities and Residues) b ζK,2 (s) admits only two singularities. More pre-
cisely, it admits two simple poles, namely, at s = 0, 1, with the residue
ζK,2 (s) = vol MK,2 ∆K .
 

Res s=1b

5.2.2 Relations with Epstein Zeta Functions

For a fixed OK -ideal a = a1 , . . . , ah , let MK,2;a ⊆ MK,2 be the moduli space of

rank two semi-stable OK -lattices whose underlying projective modules are equal
to OK ⊕ a, and let MK,2;a [T ] ⊂ MK,2;a be the moduli space of rank two semi-
stable OK -lattices of volume T (with underlying projective modules OK ⊕ a). By
Theorem 1.4,

(a) There is a natural decomposition

h
G
MK,2 = MK,2;ai .
i=1
(b) There are natural isomorphisms
MK,2;ai ' F2K × Dai × R∗+ .

Definition 5.4. Let a be a fixed OK -ideal. Assume that <(s) > 1.

(1) We define an associated rank two partial zeta function by

Z  0 
ζK,2,a (s) :=
b eh (K,Λ) − 1 · vol(Λ) s dµ(Λ).
MK,2;a
(2) Let Λ be a rank two OK -lattice. We define a completed Epstein zeta function
ÊΛ (s) of Λ by
 rs rs r1 
r2  r
X 1 
ÊΛ (s) := π− 2 Γ( ) · 2π −rs Γ(rs) · N(a) · ∆K2 s · .
2 x∈(Λ\{0})/U +
kxkrs
Λ
r,F
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 114

114 Rank Two Zeta Functions

By specifying the discussions in Section 3.2 of Part 1 to rank 2 zeta functions,

the main results there imply the following:

ζK,2 (s) of K, we have

Proposition 5.3. For the rank two zeta function b

ζK,2 (s) admits a natural decomposition

(1) (Decomposition) The zeta function b
h
X
ζK,2 (s) =
b ζK,2:ai (s).
b
i=1
(2) (Non-Abelian Zeta Function = Integration of Epstein Zeta Function) The rank
ζK,2:a (s) of K is related to the completed
2 non-abelian partial zeta function b
Epstein type zeta function by
Z
ζK,2:a (s) = R2K
b bΛ (s) dµ,
E <(s) > 1.
MK,2;a [N(a)·∆K ]

Here R2K denotes the volume of F2K .

5.3 Epstein Zeta Function and Its Fourier Expansion

5.3.1 Automorphic Function in One Variable

Let Γ ⊂ GL(2, R) be a reduced Fuchsian group of first kind with cusps κ1 , . . . , κh .

For our own convenience, we assume that κ1 = ∞. By definition, a function f (z)
on H is called an automorphic function with respect to Γ if f (γz) = f (z) for all
γ ∈ Γ. For a cusp κ of Γ, say κi for a fixed i, denote Γκ the stabilizer group of κ
in Γ, and
( let !A ∈ GL(2, ) R) satisfying A(∞) = κ.! By §4.3.1, A Γκ A = Γ∞ . Since
−1

1n 1n
Γ∞ = : n ∈ Z , a typical element of Γ∞ acts on z ∈ H by a shift
01 01
z 7→ z + n. Hence, if f (z) is an automorphic function, f (Az) is a periodic function
in z of period 1, i.e. it satisfies the condition that f (A(z + 1)) = f (Az). Therefore,
if we define the m-th Fourier coefficient of f by
Z 1 √
am (y) := f (Az) e(−mx) dx, z = x + −1y (5.20)
0
√
where e(x) = exp(2π −1x), we obtain a Fourier expansion of f at the cusp κ
X
f (Az) = am (y) e(mx), (5.21)
m∈Z
provided that f satisfies the following growth conditions:

(∗) The function f (x + iy) is of polynomial growth as y → ∞. That is, as y → ∞,

f (x + iy) = O(yk ) for some constant k uniformly with respect to x ∈ R.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 115

Rank Two Zeta Functions 115

Define, as usual, the Eisenstein series E(z, s) of Γ (associated to the cusp ∞)

by
X
E(z, s) := y s (γz), <(s) > 1 (5.22)
γ∈Γ∞ \Γ

where y(z) := y if z = x + iy ∈ H with x ∈ R and y ∈ R>0 . It is well known

(see e.g. [62]) that E(z, s) satisfies the growth condition (∗) and is automorphic.
Consequently, we obtain a Fourier expansion of E(z, s) at ∞
X
E(z, s) = am (y, s) e(mx), (5.23)
m∈Z

where
Z 1 Z ∞ X
am (y, s) = E(z, s) e(−mx) dx = y(γz) s e(−mx) dx. (5.24)
0 0 γ∈Γ∞ \Γ

To write down am (y, s)’s explicitly, we use the following decomposition.

Lemma 5.5. For the double coset decomposition Γ∞ Γ Γ∞ , we have

 

!
GG  ab
Γ∞ Γ Γ∞ = Γ∞ Γ∞ γΓ∞ , c > 0, d (mod c), γ = ∈ Γ.
 
c,d
cd

Moreover, for a fixed c > 0,

c,d (Γ∞ γΓ∞ )

F
(1) is a disjoint union of finitely many double cosets.

(2) If γ < Γ∞ , Γ∞ γΓ∞ γ−1 = 1 .

T 

!
ab
Proof. Write Γ∞ Γ Γ∞ = γ Γ∞ γΓ∞ with γ = ∈ Γ. Since Γ does not
  F
cd
!
t b
contain any hyperbolic element , t > 0, the double coset Γ∞ γΓ∞ is equal
0 t−1
!
1b
to Γ∞ Γ∞ if c = 0. On the other hand, if c , 0, we may assume c > 0,
01
a0 b0
!
by replacing γ with −γ if necessary. Moreover, for γ0 = 0 0 with c0 > 0,
c d
Γ∞ γΓ∞ = Γ∞ γ0 Γ∞ if and only if c = c0 and d ≡ d0 (mod c). This gives the
decomposition in the lemma. To prove (1), it sufficient to note that, for a given
c > 0, there exist only finitely many d’s which are incongruent mod c, such that
(c, d) is the second row of some γ ∈ Γ. To prove (2), we use the fact that Γ∞ γγ0 and
Γ∞ γγ00 are different cosets in Γ∞ \Γ, if γ and γ0 are different elements of Γ∞ . 
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 116

116 Rank Two Zeta Functions

Consequently, we have
Z ∞ X
am (y, s) = δ0m y s + 
a
 y(γz) s e(−mx) dx
−∞ b
γ= ∈Γ∞ \Γ/Γ∞ , c,0

c d
∞
XZ ys
= δ0m y s + e(−mx) dx
−∞ |cz + d|
2s
c,d
 
X 1 X md ! Z ∞ ys
= δ0m y +
s 
 e  · e(−mx) dx
|c|2s  d c  −∞ |x2 + y2 |2s

c
X 1 X md ! Z ∞ 1
= δ0m y + y
s 1−s
· e · e(−mt)dt,
c
|c|2s
d
c −∞ (1 + t2 ) s
!
∗∗
where ∈ Γ and c > 0, d (mod c). For later use, we set
cd

1 Γ(s − 2 )
X 1 X md ! 1
φm (s) = · e and φ(s) := π 2 φ0 (s). (5.25)
c
|c|2s d c Γ(s)

To state the Fourier expansion for Eisenstein series, as usual, we define the
modified Bessel K-function K s (z) by
∞
π I−s (z) − I s (z) X ( 21 z) s+2m
K s (z) := with I s (z) := . (5.26)
2 sin sπ m=0
m!Γ(s + m + 2)

From the theory of special functions, we have the following beautiful:

Lemma 5.6. (see e.g. [117]) We have

 s
2π 1
· |n| s− 2 K s− 12 (2π |n|) n ∈ R\{0}


Γ(s)


Z ∞ 
1


e(−nt)dt = 


−∞ (1 + t )
2 s
1 Γ s − 2


 1

π n = 0.


 2
Γ(s)


Consequently, what we have said proves the following:

Proposition 5.4. (see e.g. [62]) The m-th coefficient of the Fourier expansion of
E(z, s) is given by
 s
2π 1 1
φm (s) · |m| s− 2 y 2 K s− 12 (2π |m|y) m , 0


 Γ(s)



am (y, s) = 




y s + φ(s)y1−s

 m = 0.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 117

Rank Two Zeta Functions 117

More generally, we define the Eisenstein series Ei (z, s) of Γ at κi by

X
Ei (z, s) := i γz),
y s (A−1 <(s) > 1. (5.27)
γ∈Γi \Γ

Then the Fourier expansion of Ei (z, s) at κ j is given by

X Z 1
Ei (A j z, s) = ai j,m (y, s) e(mx) where ai j,m (y, s) := Ei (A j z, s) e(−mx) dx.
m∈Z 0

Moreover, similar to Lemma 5.5, there is a double coset decomposition

! for
10
Γ∞ Ai ΓAi Γ∞ . Since, for i , j, κi / κ j , the coset Γ∞ Γ∞ does not ap-
 −1

01
pear in the above! decomposition unless i = j. Therefore, if we set, for c > 0, d
∗∗
(mod c),
cd i ΓAi ,
∈ A−1

1 Γ(s − 2 )
X 1 X md ! 1
φi j,m (s) = · e , and φ ij (s) := π 2 φi j,0 (s),
c
|c|2s d c Γ(s)
with a similar calculation as above, we obtain the following:

Proposition 5.5. (see e.g. [62]) The m-th coefficient of the Fourier expansion of
Ei (z, s) with respect to the j-th cusp κ j is given by
 s
2π 1 1
φi j,m (s) · |m| s− 2 y 2 K s− 12 (2π|m|y) m , 0


 Γ(s)



ai j,m (y, s) = 




δi j y s + φi j (s)y1−s

 m = 0.

For later use, we set Φ(s) := (φi j (s)). This matrix is symmetric, since it is
invariant under the involution map x 7→ x−1 .
To end this discussion, we mention that there is an alternative approach to
obtain the Fourier expansion above by using the fact that the Eisenstein series E
is a solution of the partial differential equation ∆E = λE with λ = s(s − 1). This
implies that the coefficients ai j,m (y, s) satisfy the condition
d2 ai j,m λ

2
− 4π2 m2 + 2 ai j,m = 0. (5.28)
dy y
Here ∆ denotes the hyperbolic Laplacian in §4.1.1. Using the solutions of this
well-known second order ordinary differential equation, based on the growth be-
haviors of the Eisenstein series near the cusps, we would also arrive at the conclu-
sions that the constant terms in the Fourier expansion are some combinations of
y s and y1−s and that the non-constant terms are essentially the Bessel K-functions.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 118

118 Rank Two Zeta Functions

5.3.2 Upper Half Space

5.3.2.1 Eisenstein Series over 3-Dimensional Hyperbolic Space

We start with the following analogue of Proposition 5.4 for functions over H. Let
Λ be a full rank lattice in C and denote Λ∨ the dual lattice of Λ defined by Λ∨ :=
µ ∈ C : hµ, zi ∈ Z, ∀z ∈ Λ , where h , i denotes the usual Hermitian product on

C. Let f : H → C be a Λ-invariant C 2 -function, satisfies the differential equation

∆( f ) = 2s(2 − 2s) f. (5.29)

Here ∆ denotes the hyperbolic Laplacian on H. (See e.g. §4.1.2 for the precise
definition.) Assume that f (z + r j) is of slow growth as r → ∞, that is, as r → ∞,
f (z + r j) = O(rk ) for a constant k uniformly with respect to z ∈ C.

Proposition 5.6. (See e.g. [25]) The Fourier expansion of f (z + r j) is given by

X
a0 r + b0 r2−2s +
 2s





 aµ · r K2s−1 2π|µ|r e hµ, zi s,0

 0,µ∈Λ ∨
f (z + r j) = 


X
2
+ 2 2
+ aµ · r K2s−1 2π|µ|r · e hµ, zi s = 0.



a r b r log r




 0 0
 ∨
0,µ∈Λ

Proof. Since the function z 7→ f (z + r j) is Λ-invariant, real analytic, and of slow

growth as r → ∞, it admits a Fourier expansion of the form
X
f (z + r j) = q (r) · e hµ, zi .


∨ µ
(5.30)
µ∈Λ

Write the hyperbolic Laplacian ∆ in terms of the coordinates z, r. Then the con-
dition (5.29) implies that the qµ (r)’s satisfy the following second order ordinary
differential equations with λ = 2s(2 − 2s)
2
!
2 d d
r 2 − r + λ − 4π |µ| r · qµ (r) = 0.
2 2 2
(5.31)
dr dr

When µ = 0, (5.31) becomes

d2
!
d
r2 − r + λ · q0 (r) = 0. (5.32)
dr2 dr
By the standard theory of ordinary differential equations,

(i) If s , 0, (5.32) admits a fundamental system of solutions r2s and r2−2s .

(ii) If s = 0, (5.32) admits a fundamental system of solutions r2 and r2 log r2 .
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 119

Rank Two Zeta Functions 119

This gives the constant terms of the Fourier expansion.

When µ , 0, we are led to the ordinary differential equation
d2 w
dw
u2 + 1 − 2α u + (βu)2 + α2 − r2 w = 0. ∀α, β ∈ R.


(5.33)
du2 du
By a well-known theory for special functions, (5.33) admits a fundamental system
of solutions K s and I s . In fact, if we let Zn (u) be an arbitrary solution of Bessel’s
differential equation of order n, then qµ (r) = rZ2s−1 (2πi|µ|r) and the function w =
uα · Zn (βu) satisfies (5.33), and hence can be written as a linear combination of K s
and I s . To apply this to our problem, let α = 1, β = 2πi|µ| and n = s. Then, there
are constants aµ , bµ satisfying
qµ (r) = aµ · r K2s−1 (2π|µ|r2 ) + bµ · r I2s−1 (2π|µ|r2 ). (5.34)
On the other hand, by definition,
Z
1
qµ (r) = f (z + r j) · e − hµ, zi dx dy


(5.35)
vol(P) P

where vol(P) denotes the Euclidean area of a fundamental parallelepiped P of Λ.

By our assumption, f is of slow growth. Hence, by (5.35), so is qµ (r). Thus, if
we use the fact that K s (x), resp. I s (x) , decreases exponentially, resp. increases
exponentially, as x → ∞, by (5.34), we conclude that the factor r I2s−1 (2π|µ|r2 )
does not appear. This implies that bµ = 0 and qµ (r) = aµ · r K2s−1 (2π|µ|r2 ). 

By definition, if Γ ⊂ SL(2, C) is an arithmetic subgroup and η = A(∞) ∈ P 1 (C)

is a cusp of Γ, the Eisenstein series of Γ at η is defined by
X
E A (P, s) := r(AMP)2s . (5.36)
M∈Γ0η \Γ

It is well known that, see e.g. [25], the series on the right hand converges for
<(s) > 1, is of slow growth and satisfies the partial differential equation
∆E = −λE. (5.37)
Moreover, if η = B (∞) ∈ P (C) is a distinct cusp of Γ,
−1 1

1ω
( ! )
(BΓB )∞ = BΓη B =
−1 0 0 −1
:ω∈Λ , (5.38)
01
for a certain rank two lattice Λ ⊂ C. Since the function P 7→ E A (B−1 P, s) is
Λ-invariant, the Eisenstein series E A (B−1 P, s) admits a Fourier expansion:
X
E A (B−1 P, s) = a0 r2s + b0 r2−2s + aµ r · K2s−1 (2π|µ|r2 ) · e hµ, zi . (5.39)


0,µ∈Λ∨
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 120

120 Rank Two Zeta Functions

To go further, motivated by the proof of Proposition 5.5, we consider the dou-

ble cosets

/
.
AΓ0ζ A−1 AΓB−1 BΓ0η B−1 .


(5.40)
!
∗∗
Let AMB−1 = . There exists an M ∈ Γ in the same coset such that c = 0 if and
cd
only if AMB−1 fixes ∞ if and only if η = B−1 ∞ and ζ = A−1 ∞ are Γ-equivalent.
Assume now that η !and ζ are Γ-equivalent, and fix L0 ∈ Γ satisfying L0 (η) = ζ. !
∗ ∗ ∗∗
Let AL0 B−1 = . Then, for all M ∈ Γ such that AMB−1 has the form ,
0 d0 0d
we have |d| = |d0 |. Moreover, there are exactly [Γζ : Γ0ζ ] different elements in Γ0ζ \Γ
with this property.
!
∗∗
Denote by R a system of representatives of the double cosets in
cd

/
.
AΓ0ζ A−1 AΓB−1 BΓ0η B−1 ,


(5.41)
such !that c , 0. If η and ζ are Γ-equivalent, set d0 characterized by the condition
∗ ∗
= AL0 B−1 for a certain L0 ∈ Γ satisfying that L0 η = ζ.
0 d0

Theorem 5.5. ([25]) For <(s) > 1, the Fourier expansion of the Eisenstein series
E A (B−1 P, s) is given by
 
 
π  X 
E A (B−1 P, S ) = δη,ζ [Γζ : Γ0ζ ]|d0 |−4s r2s +    |c|−4s  r2−2s


vol(P) s  ∗ ∗∈R 

cd
 

 
 
2s d
1 2π X  X e(−hµ, i) 
+ 2s−1 2π|µ|r · e hµ, zi .
 c 
|µ|2s−1 ·    2

  · rK

vol(P) Γ(2s) 0,µ∈Λ∨ ∗ ∗
  ∈R |c|
 4s 


c d 

Here, for cusps η, ζ of Γ, we set δη,ζ := 1 if η ∼ ζ, 0, otherwise.

1 X Z
aµ (r, s) = r AMB−1 (z + r j) 2s · e − hµ, zi dx dy.



Proof. Set
vol(P) M∈Γ0 \Γ P
ζ
Then
X
E A (B−1 P, S ) = aµ (r, s) e hµ, zi , ∀P = z + r j ∈ H.


(5.42)
µ∈Λ∨

We calculate the integration in aµ (r, s) case by case.

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 121

Rank Two Zeta Functions 121

(1) When µ , 0, the contribution of the corresponding terms to the integration is

zero, since exponential is periodic and its average over [0, 1] is 0; and
(2) When µ = 0, obviously, the contribution is δηζ [Γζ : Γ0ζ ]|d0 |−4s r2s .

Hence, it suffices to calculate the summation

Z !
X r
2s
  · e hµ, zi dx dy, (5.43)
∗
AMB−1 =
 ∗
, c,0 P kc(z + r j) + dk2
c d
where the summation extends over M ∈ Γ0ζ \Γ such that c , 0, or the same, over
cosets in AΓ0ζ A−1 \AΓB−1 . Therefore, the summation in (5.43) is equal to
Z !
X X r
2s
  · e hµ, zi dx dy
∗ ∗  ω∈Λ P kc(z + ω) + dk2 + |c|2 r 2
∈R
 
cd


Z !2s
X r
=  


  · e hµ, zi dx dy
∗ ∗  |cz + d| + |c| r
2 2 2
 ∈R C
c d
!2s
e(hµ, dc i)
Z
X r
=  


 
2 + r2
· e |µ| · x dx dy,
∗ ∗  |c| 4s
C |z|
 ∈R
c d 

where we have applied an orthogonal linear transformation of R2 sending µ to

(|µ|, r).
Thus, to complete the proof, it suffices to carry out the following elementary
calculations.

(i) When µ = 0, we have

Z !2s Z 
r −2s
dx dy = r 2−2s
|z|2 + 1 dx dy
C |z| + r
2 2
C
ρdρ πr2−2s
Z ∞
= 2πr 2−2s
= .
0 (ρ + 1)
2 2s 2s − 1
(ii) When µ , 0, we have, by Lemma 5.6,
Z !2s Z ∞Z ∞
r dy
· e |µ|x dx dy = r 2−2s



· e |µ|x dx
C |z| + r
2 2 2s
−∞ −∞ y + (1 + x )
2 2


Z ∞ Z ∞
dt e(r|µ|x) 2π2s 2s−1
= r2−2s dx = |µ| rK2s−1 (2π|µ|r2 ).
−∞ (1 + t 2 )2s
−∞ (1 + x 2 ) 2s−1/2 Γ(2s)
Combining all these, we obtain the formula in the proposition. 
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 122

122 Rank Two Zeta Functions

Remark 5.1. The factor [Γζ : Γ0ζ ] takes care of elliptic points, which are factored
out in our discussion. In the case of SL(OK ⊕ a), these elliptic points are induced
from units, since the fields involved now are imaginary quadratic fields (hence by
Dirichlet’s unit theorem, there are only finitely many of them.) We thank Elstrodt
for explaining this to us.

5.3.2.2 Eisenstein Series for Imaginary Quadratic Field

We next apply the results above to obtain Fourier expansion of Eisenstein series
associated to Γ := SL(2, OK ), where K is an imaginary quadratic field, following
closely [25].

Definition 5.5. (see e.g. [25]) Fix a, b two fractional ideal of K.

(1) We define the Eisenstein series Ea (P, s) and E

ba (P, s) associated to a by:
!2s
X r
Ea (P, s) :=N(a)2s <(s) > 1,
c,d∈a, O c+O d=a
kcP + dk2
K K

!2s
X r
ba (P, s) :=N(a)2s
E <(s) > 1.
c,d∈ar{(0,0)}
kcP + dk2
(2) We define the partial zeta functions ζa,b (s) associated to a, b, and ζ[a] (s) asso-
ciated to [a] ∈ CLK by:
X X
ζa,b (s) := N(ab−1 ) s N(λ)−s , ζ[a] (s) := N(b)−s <(s) > 1.
λ∈ab−1 r{0} b∈[a],b⊆OK

Tautologically, we have the following well-known facts, whose proof we omit.

Proposition 5.7. (see e.g. [25])

ba (P, s), when <(s) > 1, we have
(1) For the Eisenstein series Ea (P, s) and E
(i) The series defining Ea (P, s) and Eba (P, s) converge uniformly on compact
sets of P and admit meromorphic extensions to the whole complex s-plane.
ba (P, s) depend only on the class of a in CLK .
(ii) The functions Ea (P, s) and E
ba (P, s) are SL(2, OK )-invariant.
(iii) The functions Ea (P, s) and E
(iv) The functions Ea (P, s) and Eba (P, s) in P satisfy the differential equations
∆Ea (P, s) = 2s(2s − 2)Ea (P, s) and ∆Eba (P, s) = 2s(2s − 2)E
ba (P, s).
(2) For the partial zeta functions ζa,b (s) and ζ[a] (s), when <(s) > 1, we have
(i) The series defining ζa,b (s) and ζ[a] (s) converge uniformly on compact sets
of P and admit meromorphic extensions to the whole complex s-plane.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 123

Rank Two Zeta Functions 123

(ii) The function ζa,b (s) depends only on the ideal classes [a], [b] ∈ CLK .
(iii) We have
ζa,b (s) = ζa·b−1 ,OK (s) and ζa,OK (s) = |U K | · ζ[a−1 ] (s).
In particular, ζa,b (s) = |U K | · ζ[mn−1 ] (s).

ba (P, s), Eb (P, s), ζa,b (s) and ζ[a] (s) are
As to be expected, the functions E
closely related.

Lemma 5.7. (see e.g. [25]) We have

X
ba (P, s) =
|U K | · E ζa,b (2s)Eb (P, s) <(s) > 1.
[b]∈CLK

Proof. Let b run through a representative system R of CLK . We introduce, for a

pair (γ, δ) of generators of an arbitrary element b and λ ∈ ab−1 , the assignment
λ, (λ, δ) 7→ (c, d) := (λγ, λδ) ∈ a ⊕ a r {(0, 0)}.

This is well-defined and gives a surjective map. Easily, every (c, d) admits pre-
cisely |U K | different inverse images. 

Consequently, to obtain the Fourier expansions for the Eisenstein series above,
it suffices to treat Eb (P, s). For this, we next point out that the function Eb (P, s)
αβ
!
is essentially the Eisenstein series E A (P, S ). Indeed, for A = ∈ SL(2, K),
γδ
let aA = OK γ + OK δ, bA = OK α + OK β. In this way, we obtain two maps from
SL(2, K) to the group of fractional ideals of K, namely, the assignments A 7→ aA
and A 7→ bA , respectively. By Lemma 4.1, these maps are surjective.

δ αβ
" # !
Lemma 5.8. (see e.g. [25]) Let ζ = ∈ P 1 (K) be a cusp of Γ and A = ∈
−γ γδ
SL(2, K) be the corresponding matrix as in Lemma 4.2. Then
1
E A (P, s) = (NaA )−2−2s EaA (P, s) <(s) > 1.
2
2
" #(c, d) ∈ K which generate aA as an OK -module.
Proof. Let L be the set of pairs
δ
!
d
For each (c, d) ∈ L, since A = ∞, there exists an M ∈ Γ such that M =
−γ −c
δ
!
. Consequently, there exists an natural assignment
−γ
φ : L → Γ0ζ \Γ, (c, d) 7→ Γ0ζ M.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 124

124 Rank Two Zeta Functions

Easily, φ is well-defined. Moreover, if (c, d) ∈ L and φ (c, d) = Γ0ζ M with M ∈ Γ,

!
∗∗
then AM = . In terms of the Eisenstein series, this gives the factor aA on
cd
!
∗∗
the right hand. On the other hand, for Γζ M ∈ Γζ \Γ, if we write AM =
0 0
, the
cd
φ-inverse image of Γζ M consists exactly two elements, namely, (c, d) and −(c, d).
0

This gives the factor 1/2. 

With all these, by Theorem 5.5, we can obtain the Fourier expansion of Ea (P, s)
and hence that of E ba (P, s). In the sequel, till the end of this discussion for imag-
inary quadratic field, we let η ∈ P 1 (K) be a cusp of Γ, let B ∈ SL(2, K) be a
matrix satisfying !B(η) = ∞,)and let a be a fractional ideal of K. For later use, set
ω
(
1
BΓ0η B−1 = : ω ∈ Λ , where Λ ⊂ C is a rank two OK -lattice. Let R0 be a
01
maximal system of representatives of (c, d) of a ⊕ a B−1 /BΓ0η B−1 with c , 0.

Corollary 5.1. ( [25]) The Fourier expansion of the Eisenstein series E ba (B−1 P, s)
is given by

ba (B−1 P, s) = N(aB )2s ζ(a, aB , 2s) · r2s + πN(a)

2s  X 
E 2s−1
|c|−4s r2−2s
|Λ| (c,d)∈R0
(5.44)
2s
2π N(a) 2s X X e −hω , c i
0 d

+ 0 2s−1
rK2s−1 2π|ω |r e hω , zi .
0 2
0

|ω |
|Λ| · Γ(2s) 0,ω0 ∈Λ∨ (c,d)∈R
|c|4s
0

ba (B−1 P, s) is Λ-invariant. Write its

Proof. Indeed, by a direct calculation, E
Fourier expansion as
X
ba (B−1 P, s) =
E aω0 (r, s) · e hω0 , zi ,


ω0 ∈Λ∨
with Fourier coefficients
N(a)2s
Z
X r
2s
· e − hω0 , zi dx dy.


aω0 (r, s) :=
|Λ|
P kcP + dk2
(c,d)∈ a⊕a B−1 ,(c,d),(0,0)

Decomposing the right hand side into two parts, depending on whether c = 0 or
not, we get
N(a)2s
Z
X r
2s
aω0 (r, s) = · e hω0 , xi dx dy


|Λ| kdk2
(0,d)∈(a⊕a)B−1 ,d,0 P
(5.45)
N(a)2s
Z
X r
2s
+ 0
,


· e − hω zi dx dy.
P kcP + dk
|Λ| 2
(c,d)∈(a⊕a)B ,c,0
−1
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 125

Rank Two Zeta Functions 125

Observe that the first term on the right hand side of this equation vanishes term-
wise when ω0 , 0, and moreover, when ω0 = 0, it is equal to
 X 
N(a)2s |d|−4s r2s . (5.46)
(0,d)∈(a⊕a)B−1 ,d,0

If d , 0, then (0, d) ∈ a ⊕ a B−1 if and only if (0, d)B ∈ a ⊕ a if and only if

B ζa,aB (2s) · r . This is the first term

daB ⊆ a. Hence, the sum of (5.46) is equal to a2s 2s

of the right hand side in the corollary. As for other terms, it suffices to apply the
same method of the proof for Theorem 5.5 to the second sum in (5.45). we leave
details to the reader. 

The coefficients appeared in (5.44) admit some arithmetic interpretations. The

up-shot is the following:

ba (B−1 P, s) is given by
Theorem 5.6. ([25]) The Fourier expansion of E

ba (B−1 P, s) = N(aB )2s ζa,aB (2s)r2s + √ 2π

E N(aB )2−2s ζa,a−1 (2s−1)r2−2s
∆K (2s−1) B

(5.47)
21+2s π2s N(aB ) X 2s−1 4π|ω|r2
! !
2ω
+ |ω| σ a,aB (ω, 1−2s)rK 2s−1 √ e h √ , zi .
∆Ks Γ(2s) ∆K ∆K
0,ω∈a2

Here, for fractional ideals a, b of K, s ∈ C and ω ∈ K ∗ , we have set

X
σa,b (ω, s) := Na−s Nλ s .
λ∈ab, ω∈λa−1 b

For a proof of this result, please refer to [25] for details.

5.3.3 Eisenstein Series for Rank Two Lattice

Let K be a number field of degree n with OK its ring of integer and ∆K the norm of
its discriminant. Denote by S the set of Archimedean places of K, and denote by
r1 , resp. r2 , the number of real embeddings, resp. complex embeddings, of K. We
have n = r1 + 2r2 . Moreover, for an element a ∈ K, denote by aσ its image in the
σ-completion Kσ of K. Similarly, we write its image of Minkowski embedding
as (a(i) ). Denote by U K the unit group of K, and by U K+ the collection of ε ∈ U K
such that εσ > 0 for all real σ ∈ S .

5.3.3.1 Epstein Zeta Function and Eisenstein Series

Guided by our discussion on non-abelian zeta functions for number fields in

§ 5.2.2, we introduce the following:
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 126

126 Rank Two Zeta Functions

Definition 5.6. Let a be a fractional ideal of K and let Λ = Pa = OK ⊕ a, ρ be

a rank two OK -lattice. We define the complete Epstein type zeta function E ba:Λ (s)
associated to Λ by

bΛ (s) := π−s Γ (s)
r1 2π−2s Γ(2s)
r2 · N(a)∆K
s ·

X 1
E 2s
x∈O ⊕a/U 2 r{(0,··· ,0)}
kxk Λ
K K

n o
where U K2 := ε2 : ε ∈ U K .

For later use, set

s
 s
ΓR (s) := π− 2 Γ and ΓC (s) := 2π−s Γ(s). (5.48)
2
! !!
ab aσ bσ
Write Λ = (OK ⊕ a, ρΛ (g)) with g = = =: (gσ )σ∈S . Recall
cd cσ dσ σ∈S
that, by (4.67), N(τ) := N(ImJ(τ)) for τ ∈ H r1 × Hr2 . Thus, for a non-zero vector
(x, y) ∈ OK ⊕ a, its Λ-norm is given by
! 2
2 ab
(x, y) Λ = (x, y)
cd
Y
Y
= (aσ xσ + cσ yσ )2 + (bσ xσ + dσ yσ )2 · |aτ xτ + cτ yτ |2 + |bτ xτ + dτ yτ |2 2


σ:R τ:C
 −1
 
N(gΛ (ImJ)) 
=   ,

2
x · gΛ (ImJ) + y 
 

r −times r −times
z1 }| { z2 }| {
where ImJ := (i, . . . , i, j, . . . , j) ∈ H r1 × Hr2 , and k · k is the same as in Defini-
tion 4.8. Since, for ε ∈ U K , kεk = 1, we have
 s
 

s X  N(ImJ(τΛ )) 
EΛ (s) := ΓR (2s) ΓC (2s) N(a)∆K
b r1 r2
 2  . (5.49)
(x,y)∈OK ⊕a/U K r{(0,0)}  x · τΛ + y 

Definition 5.7. Let a be a fixed OK -ideal.

(1) We define an Eisenstein series E2,a (τ, s) of a as a function of τ ∈ H r1 × Hr2 by

 s
 
X  N(ImJ(τ)) 
E2,a (τ, s) :=  2
 <(s) > 1.
(x,y)∈OK ⊕a/U K ,(x,y),(0,0)  x · τ + y 
 
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 127

Rank Two Zeta Functions 127

(2) We define a complete Eisenstein series of a by

 s
 
X  N(ImJ(τΛ )) 
b2,a (τ, s) := ΓR (2s)r1 ΓC (2s)r2 N(a)∆K
s
E   .
2
(x,y)∈OK ⊕a/U K r{(0,0)}  x · τ + y 
 

That is to say,
b2,a (τ, s) = ΓR (2s)r1 ΓC (2s)r2 N(a)∆K
s E2,a (τ, s).
E (5.50)

We have the following:

Lemma 5.9. For a rank two OK -lattice Λ = (OK ⊕ a, ρΛ ), denote by τΛ the

corresponding moduli point in the moduli space SL(OK ⊕ a) H r1 × Hr2 .


b2,a (τ, s) are SL(OK ⊕ a)-invariant,

(1) Tautologically, E2,a (τ, s) and hence also E
and are of slow growth when τ approaches to ∞.
(2) Two types of Eisenstein series are related by the follows:
bΛ (s) = E
E b2,a (τΛ , s).

Proof. (1) is rather standard whose proof can be found in any textbook on classi-
cal Eisenstein series. (2) is simply (5.49). 

5.3.3.2 Constant Term of Fourier Expansion

α
" #
To simplify our notation, write Γ = SL(OK ⊕ a). For a cusp η = ∈ P 1 (K), by
β
Lemma 4.2,! we may and hence will choose a (normalized) special transformation
α α∗
A= ∈ SL(2, K) such that, if we set b := OK α + aβ, then OK β∗ + aα∗ = b−1 .
β β∗
Consequently, A(∞) = η, and, by Theorem 4.4, the stabilizer group of η in Γ is
simply
1ω
( ! )
A−1 Γ0η A = : ω ∈ ab−2 . (5.51)
01
Since E2,a (τ, s) is Γ-invariant, it is Γ0η -invariant as well. Hence, E2,a (A(τ), s) is
invariant under the parallel transforms by the elements of ab−2 . This, together
with the slow growth property mentioned above, implies that there is a natural
Fourier expansion for E2,a (Aτ, s) as follows:
X
aω0 ImJ(τ), s · e2πihω ,ReZ(τ)i ,
0
E2,a (Aτ, s) =


(5.52)
ω0 ∈(ab−2 )∨
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 128

128 Rank Two Zeta Functions

where (ab−2 )∨ denotes the dual lattice of ab−2 . Thus, if we denote by P a funda-
mental parallelogram of ab−2 in Rr1 × Cr2 , then, by definition,
1
aω0 (ImJ(τ), s) :=
vol(ab−2 )
!s
(5.53)
Z
X N(ImJ(τ)) −2πihω0 ,ReZ(τ)i Y Y
× e dx σ dxτ dyτ .
(c,d)∈(O ⊕a)A/U P kcτ + dk2
σ:R τ:C
K K
(c,d),(0,0)

To compute these Fourier coefficients in more details, we break the summation

of 5.53 into two parts, depending on whether c = 0 or not.

1
(1) When c = 0, up to the factor , the corresponding contribution is
vol(ab−2 )
given by
Z !s
X N(ImJ(τ)) Y Y
e−2πihω ,ReZ(τ)i
0
dxσ dxτ dyτ
(0,d)∈(OK ⊕a)A/U K P kdk2 σ:R τ:C
d,0
! sZ (5.54)
X N(ImJ(τ)) Y Y
e−2πihω ,ReZ(τ)i
0
= dxσ dxτ dyτ .
(0,d)∈(OK ⊕a)A/U K
kdk2 P σ:R τ:C
d,0
So depending on whether ω0 = 0 or not, this may be further divided into two
subcases.
(1.a) When ω0 , 0, Z
we have
Y Y
e−2πihω ,ReZ(τ)i
0
dxσ · dxτ dyτ = 0. (5.55)
P σ:R τ:C
That is to say, the corresponding Fourier coefficient aω0 is zero.
(1.b) When ω0 = 0, sinceZ Y Y
dxσ · dxτ dyτ = vol(ab−2 ),
P σ:R τ:C
we have !s
X N(ImJ(τ))
a0 ImJ(τ), s =


(0,d)∈(OK ⊕a)A/U K ,d,0
kdk2
X (5.56)
= N(ImJ(τ)) s N(d)−2s .
(0,d)∈(OK ⊕a)A/U K ,d,0
X
To understand this latest summation N(d)−2s , we first note
(0,d)∈(OK ⊕a)A/U K ,d,0
αα α α∗ ∗
! !
that, by definition, (OK ⊕ a)A = (OK ⊕ a) , where is a matrix in
β β∗ β β∗
SL(2, F) such that, if b = OK α + aβ, then OK β∗ + aα∗ = b−1 .
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 129

Rank Two Zeta Functions 129

Claim 5.2. We have

OK ⊕ a A U K = OK α + aβ, OK α∗ + αβ∗ U K = b ⊕ ab−1 U K .





Proof. Indeed, by definition, OK α + aβ = b. So it suffices to prove that

OK α∗ + aβ∗ = ab−1 .
Clearly, α∗ ∈ ab−1 , β∗ ∈ b−1 . Hence OK α∗ + αβ∗ ⊆ ab−1 . On the other hand, as
showed before b−1 = OK β∗ + a−1 β∗ . Therefore,
ab−1 = a · (OK β∗ + a−1 β∗ ) ⊆ aβ∗ + OK α∗ ,
as desired. 

Furthermore, for ab−1 \{0} /U K , we have the following

Claim 5.3. Let a be a fractional ideal of K. We have

(1) there is a natural bijection

n o
a r {0} /U K −→ b ∈ [a−1 ] : b ⊂ OK an ideal


.
a 7→ b := aa−1
X
(2) Let ζ[a−1 ] (s) := N(b)−s . Then we have
b∈[a−1 ]:b⊂OK an ideal
X
ζ[a−1 ] (s) = N(a) s · N(a)−s .
a∈(a\{0}/U K
Proof. These are standard facts. Indeed, (1) may be found in [82], and (2) is a
direct consequence of (1). 

By combining (5.55), (5.56) and Claims 5.2, 5.3, we obtain the following:

Lemma 5.10. When c = 0, the corresponding Fourier coefficient is given by

a0 ImJ(τ), s = N(a−1 b 2s · ζ [a−1 b], 2s · N(ImJ(τ)) s .

(2) When c , 0, we have

1
aω0 ImJ(τ), s =


vol(ab−2 )
X Z
N(ImJ(τ))
!s Y Y (5.57)
−2πihω0 ,ReZ(τ)i
· e dxσ · dxτ dyτ .
(c,d)∈(OK ⊕a)A/U K ,c,0 P
kcτ + dk2 σ:R τ:C
!
∗∗
To compute this, we examine the action of A−1 Γ0η A on .
cd

Claim 5.4. Let (c, d) ∈ (OK ⊕ a)A and ω ∈ ab−2 . If c , 0, (c, cω + d) ∈ (OK ⊕ a)A.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 130

130 Rank Two Zeta Functions

α α∗
!
Proof. It suffices to deal the component cω + d. By Lemma 4.2, for A = ∈
β β∗
SL(2, F), we have α ∈ b, β ∈ a−1 b, α∗ ∈ ab−1 , β∗ ∈ b−1 . Hence,

c ∈ OK · b + a · a−1 b = b and d ∈ OK · ab−1 + a · b−1 = a b−1 .

Consequently, cω + d ∈ b · ab−2 + a b−1 = a b−1 . 

Easily, for (c, d) ∈ (OK ⊕ a)A/U K and ω ∈ ab−2 , if c , 0, we have

∗∗ 1ω
! ! !
∗ ∗
= . (5.58)
cd 01 c cω + d
!
∗∗
Denote by R a system of representatives of modulo the right action of
cd
A−1 Γ0η A, or the same, let R be a system of representatives of (c, d) modulo the
relation (c, d) ∼ (c, cω + d), for (c, d) ∈ (OK ⊕ a)A/U K and ω ∈ ab−2 when c , 0.
Since, for τ ∈ H r1 × Hr2 , we have ImJ(τ + ω) = ImJ(τ). This implies that


aω0 ImJ(τ), s
X Z Z !s
1 N(ImJ(τ)) −2πihω0 ,ReZ(τ)i
Y Y
= e dx σ dxτ dyτ
vol(ab−2 )   ω∈ab−2 P kc(τ + ω) + dk2 σ:R τ:C
∗ ∗ 
∈R
 
c d


X Z !s
1 N(ImJ(τ)) Y Y
· e−2πihω ,ReZ(τ)i
0
= dxσ · dxτ dyτ
Rr1 ×Cr2 kcτ + dk
−2
vol(ab )   2
∗ ∗  σ:R τ:C
∈R
 
c d


N(ImJ(τ)) s −2πihω0 ,ReZ(τ)+ d − d iY

Z  
1 X 1 Y
=   e c c dx σ dxτ dyτ
vol(ab−2 )  N(c)2s Rr1 ×Cr2 kτ + dc k2


∗ ∗  σ:R τ:C
 ∈R

c d

Therefore,


aω0 ImJ(τ), s
1 X e2πihω0 , dc iZ !s
N(ImJ(τ)) −2πihω0 ,ReZ(τ)iY Y
= −2 2s 2
e dxσ dxτ dyτ .
vol(ab )  N(c) Rr1 ×Cr2 kτk σ:R τ:C
∗ ∗ 
 ∈R
c d 

Similarly, depending on whether ω0 = 0 or not, we have

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 131

Rank Two Zeta Functions 131

(2.a) When ω0 = 0,
Z !s
1 X 1 N(ImJ(τ)) Y Y
a0 ImJ(τ), s = dxτ dyτ .


k dxσ k
vol(ab−2 )  N(c)2s Rr1 ×Cr2 kτk2 σ:R τ:C
∗ ∗ 
∈R
 
c d


To evaluate the integration, we calculate the contribution from every Archimedean

factor.
(2.a.i) Over each Archimedean place R, we have
Z !s Z  s
y 1  1  x
dx = s  x 2  d y
R x +y y R (y) + 1 y
2 2
(5.59)
Γ(s − 1
)
Z
dt 1
= y1−s = π 2 y1−s 2
.
R (1 + t )
2 s Γ(s)
(2.a.ii) Over each Archimedean place C, we have
Z !2s Z  2s
r 1 1  d x · rd y · r
=
 
2 + r2
dx dy
C ( ) +1
2s |z| 2

C |z| r  r r
Z r (5.60)
dx dy π
= r2−2s = r2−2s · .
C (1 + |z| )
2 2s 2s − 1
(2.b) When ω0 , 0,
1 X e−2πihω0 , dc i
aω0 ImJ(τ), s =


vol(ab−2 )   N(c)2s
∗ ∗ 
∈R
 
c d


Z !s
N(ImJ(τ)) Y Y
· e−2πihω ,ReZ(τ)i
0
× 2
dxσ · dxτ dyτ .
Rr1 ×Cr2 kτk σ:R τ:C

Similarly, to evaluate the integration, we calculate the contribution from every

Archimedean factor.
(2.b.i) Over each Archimedean place R, we have
Z !s Z  s
y −2πi|ω0 |·x 1  1  −2πi|ω0 |· yx y x
e dx = s  e d ·y
R x +y y R  ( yx )2 + 1 
2 2

y
Z
1 0
= y1−s e−2πi|ω |yt dt
R (1 + t )
2 s
(5.61)
s− 12
s 0 s− 12 y
= y · 2π |ω |
1−s
· 0

K 1 (2π|ω |y)
Γ(s) s− 2
1
2π s |ω0 | s− 2 1
= · y 2 K s− 21 (2π|ω0 |y).
Γ(s)
January 26, 2018 9:4 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 132

132 Rank Two Zeta Functions

(2.b.ii) Over each Archimedean place R, we have, by using the calculation over R
in (5.61),
Z !2s Z !2s
r −2πi|ω0 |·x r 0

2 + r2
e dx dy = 2 + y2 + r 2
e−2πi|ω |·x dx dy
C |z| R2 x
Z !2s
1 0 x x y
= y 2 r−2s e−2πi|ω |· r ·r d d · r2
R2 ( r ) + ( r ) + 1
x 2 r r
0
e−2πi|ω |·x·r
Z
=r 2−2s
dx dy
2 (x + y + 1)
2 2 2s
ZR Z
dy 0
= r2−2s 2 + x2 + 1)2s
· e−2πi|ω |rx dy dx
R R (y
 
Z Z √ 
1 y x + 1  −2πi|(ω0 )|rx
2
=r 2−2s

d√  e dx (5.62)
x2 + 1 (x + 1) 
  2 2 2s 
R  R ( √ y + 1) 2s
x2 +1
0
e−2πi|ω |rx
Z Z !
dt
=r 2−2s
· dx
R (1 + t )
2 2s 1
R (x2 + 1)2s− 2
 1 Γ(2s − 21 ) 
  Z 0
e−2πi|ω |rx
=r 2−2s 
· π · 2
· dx
Γ(2s)
  1
R (x2 + 1)2s− 2

 1 Γ(2s − 21 )   0 2s−1 (2s−1) 2π2s− 2

   1 
= r2−2s · π 2 · 0 

 · |ω | r K 2s−1 (2π|ω |r)
Γ(2s) Γ(2s − 2 )1


2π2s |ω0 |2s−1

= · rK2s−1 (2π|ω0 |r).
Γ(2s)
Or, more directly,
Z !2s
r 0

2 + r2
e−2πi|ω |·x dx dy
C |z|
Z  2s
1  de−2πi|ω0 |· rx ·r x · rd y · r
=r −2s  
C ( ) +1
 |z|
2 r r
Z r
1 0 (5.63)
= r2−2s e−2πi|ω |·x·r · dx dy
C (1 + |z| )
2 2s

2π2s |ω0 |2s−1 2s−1

= r2−2s · r K2s−1 (2π|ω0 |r)
Γ(2s)
2π2s |ω0 |2s−1
= rK2s−1 (2π|ω0 |r).
Γ(2s)
Therefore, by Lemma 5.10, and the calculations (5.59), (5.60), (5.61), (5.62)
or (5.63), we obtain the following:
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 133

Rank Two Zeta Functions 133

Theorem 5.7. The Fourier expansions for the Eisenstein series E2,a (Aτ, s) around
the cusp A(∞) is given by:

E2,a (Aτ, s) = ζ([a−1 b], 2s) · N(ab−1 )−2s · N(ImJ(τ)) s

r1
 Γ(s − 21 )   π r2

1 X 1 1
+ r
· (π 2 ) 
1  · · N(ImJ(τ))1−s
vol(ab−2 )   N(c)2s Γ(s) 2s − 1
∗ ∗ 
 ∈R
c cω + d


e2πihω , c i
0 d
1 X 1 1
+   · N(ImJ(τ)) 2 · N(ω0 ) s− 2
vol(ab−2 ) ∗ ∗∈R N(c)2s
 

c d
 

!r !r
2π s 1 Y 0 2π2s |ω0 |2s−1 2 Y
× K 1 (2π|ω |σ yσ ) K2s−1 (2π|ω0 |τ rτ ).
Γ(s) σ:R s− 2 Γ(2s) τ:C

Using technics in [25], we can write down all the Fourier coefficients explic-
itly, and give them arithmetic interpretations. However, for our application to the
rank two zeta functions in this book, this theorem on the constant term is already
sufficient, we omit the details to the reader.

5.4 Explicit Formula of Rank Two Zeta Function

5.4.1 Classical Rankin-Selberg Method

5.4.1.1 Classical Rankin-Selberg Method

Let Γ be a congruence subgroup of SL(2, Z). Induced from the canonical action
of SL(2, R), there is a natural action of Γ on H. Denote by κ1 = ∞, κ2 , . . . , κh
the inequivalent cusps of Γ. For each i = 1, . . . , h, let Γi := Γκi be the isotropy
group of κi , and fix an Ai ∈ GL2 (Q) satisfying A1 = I2 and Ai (∞) = κi . Obviously,
i Γi Ai = Γ∞ . In addition, we assume that
A−1
!
Γ is reduced at infinity, i.e. the
11
stabilizer group Γ∞ of ∞ is generated by .
01
Recall that the Eisenstein series Ei (z, s) of Γ at κi is defined by
X
Ei (z, s) := i γz),
y s (A−1 (5.64)
γ∈Γi \Γ

where y(x + iy) := y. For example, E1 (z, s) = E(z, s) is the Eisenstein series at ∞.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 134

134 Rank Two Zeta Functions

Set
 
E1 (z, s)
E (z, s)
 
E(z, s) :=  2 . (5.65)
 · · · 
Eh (z, s)
By Proposition 5.5, there is a matrix Φ(s) of functions such that
E(z, s) = Φ(s)E(z, 1 − s), Φ(s)Φ(1 − s) = Ih (5.66)
where
X 1 X
Φ(s) = (φi j )h×h and φi j,0 (s) = 2s

∗ ∗ 
 1. (5.67)
|c|
∈Ai ΓA j , d (modc)
 −1
c>0

cd


Let F(z) be a Γ-modular function. Recall that F(z) is of slow growth at infinity,
i.e. it satisfies the condition that
F(Ai z) = ψi (y) + O(y−N ), as y = =(z) → ∞, ∀N. (5.68)
In the sequel, till the end of this subsection, we assume, in addition, that, as z → 0,
the ψi (y)’s admit the following asymptotic expansions
l
X ci j αi j
ψi (y) = y logni j y, where ci j , αi j ∈ C, ni j ∈ Z≥0 . (5.69)
j=1
ni j !

As usual, write the Fourier expansion for F(z) at κi as

X
F(Ai z) = am (y, s)e(mx). (5.70)
n∈Z

We then define a normalized transform Ri (F, s) of F(z) at the cusp κi by

Z ∞ Z ∞Z 1 h i dx ∧ dy
a0 (y)−ψi (y) y dy =
i
. (5.71)

s−2
Ri (F, s) := F(Ai z)−ψi (y) y s
0 0 0 y2
Since the slow growth part has been truncated, the integration in (5.71) makes
perfect sense when <(s) is sufficiently large. Set
 c1 j 
 (1 − α1 j − s)n1 j + 1 
 
 
 1 R (F, s)   c2 j 
l 
R2 (F, s)
  X  
R(F, s) :=   and h(s) := − 
 (1 − α2 j − s)n 2 j + 1  . (5.72)
 · · · 
 
 
j=1  ··· 
Rh (F, s) ch j
 
 
(1 − αh j − s)nh j + 1

Proposition 5.8. ([37, 137]) We have

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 135

Rank Two Zeta Functions 135

(1) The function vector R(F, s) is well-defined for <(s) sufficiently large.
(2) (Functional Equation) R(F, s) = Φ(s)R(F, 1 − s).
1
(3) Each component of R(F, s) = h(s) + Φ(s)h(1 − s) + is entire.
ζ(2s)
s(s − 1)b

Proof. Let D be the standard fundamental domain for the action of SL(2, Z) on
H, and let DΓ be a fundamental domain of Γ with |x| ≤ 21 . Set

S ∞ (T ) := z ∈ H : =(z) > T, |x| ≤ 1/2 ,


n o
i z ∈ S ∞ (T ) = Ai S ∞ (T ).
S κi (T ) := z ∈ H : A−1
S

Introduce a truncated domain DT := DΓ \ i S κi (T ) , for sufficiently large T .
Then, by §4.4.1, DT is the fundamental domain for the action of Γ on
[ ( )
HT := γDT = z ∈ H : max =(A−1 i δz) ≤ T
δ∈Γ, i≥1
γ∈Γ
[ [
= z ∈ H : =(z) ≤ T \ S a/c ,


c≥1 a∈Z,(a,c)=1

where, for i = 1, . . . , h, S a/c := δ−1 Ai =(z) > T is an open disk in H tangent to


x-axis at the point (a/c, 0) and we take δ ∈ Γ satisfying δ(a/c) = κi . Consequently,
[ [
Γ∞ HT = { x + iy : 0 ≤ x ≤ 1, 0 ≤ y ≤ T } \ S a/c .


c≥1 a (modc), (a,c)=1

Since F is Γ-invariant, using a unfolding trick, we have

Z Z
F(z)E(z, s)dµ = F(z)y s dµ
DT Γ∞ \Γ
Z T Z 1 ∞
X X " (5.73)
= F(z)y s dµ − F(z)y s dµ.
0 0 c=1 a(modc) S a/c
(a,c)=1

To compute the summation on the right hand side, we divide it into two parts.
!
ab
(1) When a/c ∼ ∞, let γ0 = ∈ Γ. Then γ0−1 S a/c = z ∈ H : =(z) > T .


cd
Hence we have
" Z ∞Z ∞
F(z)y s dµ = F(z)=(γ0 z) s dµ
S a/c T −∞
Z ∞ Z 1/2 ∞
X " X
= F(z) =(γ0 (z + n)) s dµ = F(z)   =(γz) s dµ,
T −1/2 S ∞ (T )
a ∗
n=−∞ γ= ∈Γ

c ∗
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 136

136 Rank Two Zeta Functions

! !
a∗ 1n
where the sum runs over all γ = ∈ Γ, in the form γ0 for some n ∈ Z.
c∗ 01
Therefore, by an obvious relation
 
 ! G
G G G a ∗ 
Γ∞ \Γ/{±1} =  I2 ,
 

a ∗
∗

 c>0 a modz, a/c∼∞  ∈Γ c 
c∗
 

we have
∞
X X " "
F(z)y dµ =
s
F(z) · E(z, s) − y s dµ.


c=1 a(modc),(a,c)=1,a/c∼∞ S a/c S ∞ (T )

(2) When a/c ∼ κi≥2 / ∞, we have δ(a/c) = κi . Hence

∞
X X " "
F(z)y s dµ = F(z)E(z, s)dµ
c=1 a(modc),(a,c)=1,a/c∼κi S a/c S κi (T )
" "
= F(z)E(z, s)dµ = F(Ai z)E(Ai z, s)dµ.
Ai S ∞ (T ) S ∞ (T )

Therefore,
Z Z T h "
X h i
F(z)E(z, s)dµ = a∞
0 (y)y
s−2
dy − F(Ai z) E(Ai z, s) − δi∞ y s dµ,
DT 0 i=1 S ∞ (T )
Z T Z 1 Z T
since we have easily F(z)y s dµ = a∞
0 (y)y
s−2
dy.
0 0 0
Z 1
To go further, let e j∞ := e j∞ (y, s) := E(A j z, s)dµ be the constant term in
0
the Fourier expansion of E(z, s) at κ j . By Proposition 5.4, ei∞ = δi∞ y s + φi∞ y1−s .
Thus, for <(s) sufficiently large, we have
Z Z T Xh " h i
F(z)E(z, s)dµ = a∞0 (y)y s−2
dy − F(Ai z) E(Ai z, s) − ei∞ dµ
DT 0 i=1 S ∞ (T )
h "
X
− F(Ai z)φi∞ y1−s dµ
i=1 S ∞ (T )
Z T h "
X h i
= a∞
0 (y)y
s−2
dy − F(Ai z) E(Ai z, s) − ei∞ dµ
0 i=1 S ∞ (T )
h
X Z ∞
− φi∞ ai0 (y)y−1−s dy.
i=1 T
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 137

Rank Two Zeta Functions 137

h i
The difference E(Ai z, s) − ei∞ is an entire function of s and is of rapid decay with
respect to y. Hence we have
Z h "
X h i
F(z)E(z, s)dµ + F(Ai z) E(Ai z, s) − ei∞ dµ
DT i=1 S ∞ (T )
(5.74)
Z T h
X Z ∞
= a∞
0 (y)y
s−2
dy − φi∞ ai0 (y)y−1−s dy.
0 i=1 T

l
ci j ∂ni j
Z T
T s+αi j −1
X !
Set hiT (s) := ψi (y)y dy =
s−2
and hT (s) = (hiT (s)).
0 j=1
n i j ! ∂sni j s + α − 1
i j
We have
Z T Z T Z T
ai0 (y)y s−2 dy = ai0 (y) − ψi (y) y s−2 dy + ψi (y)y s−2 dy


0 0 0
Z ∞
= Ri (F, s) − ai0 (y) − ψi (y) y s−2 dy + hiT (s).


T
Z ∞ Z T
On the other hand, ψi (y)y dy = − −s−1
ψi (y)y−s−1 dy = −hiT (1 − s),
Z ∞ T 0

since ψ(y)y−s−1 dy = 0. This implies that

0
Z ∞ Z ∞ Z ∞
ai0 (y)y−s−1 dy = ai0 (y) − ψi (y) y−s−1 dy + ψi (y)y−s−1 dy


T T T
Z ∞
= (ai0 (y) − ψi (y))y−s−1 dy − hiT (1 − s).
T

Consequently, by (5.74), we have

Z h "
X h i
F(z)E(z, s)dµ + F(Ai z) E(Ai z, s) − ei∞ dµ
DT i=1 S ∞ (T )
Z ∞
= R∞ (F, s) − 0 (y) − ψ∞ (y) y
a∞ dy + h∞

s−2
T (s) (5.75)
T
h
X "Z ∞ #
− φi∞ (ai0 (y) − ψi (y))y−1−s dy − hiT (1 − s) .
i=1 T

Moreover, since
" X h Z ∞ X Z ∞
F(Ai z)ei∞ dµ = a∞
0 (y)y
s−2
dy + φi∞ ai0 (y)y−s−1 dy,
S ∞ (T ) i=1 T i T
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 138

138 Rank Two Zeta Functions

we have
h
X
R∞ (F, s) + h∞
T (s) + φi∞ hiT (1 − s)
i=1
Z h "
X h i
= F(z)E(z, s)dµ + F(Ai z) E(Ai z, s) − ei∞ dµ
DT i=1 S ∞ (T )
Z ∞ X h Z ∞
+ a∞ − ψ∞ (y) y s−2 dy + φi∞ ai0 (y) − ψi (y) y−1−s dy



0 (y)
T i=1 T
Z h "
X
= F(z)E(z, s)dµ + F(Ai z)E(Ai z, s) − ψi (y)ei∞ dµ.


DT i=1 S ∞ (T )
Similarly, by working with other cusps, we obtain
Z h "
X
F(z)Eκ (z, s)dµ + F(Ai z)Eκ (Ai z, s) − ψi (y)eiκ dµ


DT i=1 S ∞ (T )
(5.76)
h
X
= Rκ (F, s) + hκT (s) + φiκ hiT (1 − s).
i=1
Put all these in a vector form, what we just said then yields
Z Xh "
F(z)E(z, s)dµ + F(Ai z)E(Ai z, s) − ψi (y)ei (y, s) dµ


DT i=1 S ∞ (T )

= R(F, s) + hT (s) + Φ(s)hT (1 − s).

 l 
X ci j ∂ni j T s+αi j −1 − 1

Obviously, hT (s) − h1 (s) =    is entire in s. 
ni j ! ∂sni j s + αi j − 1 
j=1

Take the spacial case when F ≡ 1 is the constant function, this gives

Corollary 5.2. We have

Z Z T h
X Z ∞
Eκ (z, s) dµ = a∞
0 (y)y
s−2
dy − φi∞ ai0 (y)y−1−s dy
DT 0 i=1 T
h
X
= hκT (s) + φiκ hiT (1 − s).
i=1

Proof. The first equality is a direct consequence of (5.74). To prove the second
equality, it suffices to note that, for F ≡ 1,
(i) F(Ai z)Eκ (Ai z, s) − ψi (y)eiκ = Eκ (Ai z, s) − eiκ , whose integration over S ∞ (T ) is
simply zero:
(ii) Rκ (F, s) is also zero by definition. 
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 139

Rank Two Zeta Functions 139

5.4.1.2 Example: Rank Two Non-Abelian Zeta Function of Q

As an example of the classical Rankin-Selberg method, we here give an explicit

formula for the rank two non-abelian zeta function of Q. Recall that, by §5.1.4,
the points of the truncated fundamental domain DT := x ∈ D : y = =(z) ≤ T

(for the action of SL(2, Z) on H) are in one-to-one correspondence to the rank two
Z-lattices of volume one (in R2 ) whose first Minkowski successive minimums λ1
≤ 12 log T  
satisfy the condition λ1 ≥ T −1/2 . Hence, for the moduli space MQ,2 1 of
rank two Z-lattices Λ of volume 1 whose rank one sub Z-lattices have degrees
≤ 1 log T  
≤ 12 log T , we have, as stated in §5.1.4, MQ,22 1 ' DT . In particular, for the
moduli space of rank two semi-stable lattices of volume one, we have
M≤0Q,2 1 = MQ,2 1 ' D1 .
   
(5.77)

Definition 5.8. Fix T ∈ R>0 . We define a T -version of rank two zeta function
ζQ,2
b T
(s) of Q by
Z
ζQ,2
b T
(s) := b s) dx ∧ dy
E(z, <(s) > 1.
≤ 21 log T  
MQ ,2 1 y2

ζQ,2
This T -version of rank two zeta function b T
(s) can be calculated explicitly
in terms of the classical Riemann zeta function.

ζQ,2
Proposition 5.9. For the generalized zeta function b T
(s), we have

ζ(2s) s−1 b ζ(2s − 1) −s

ζQ,2 (s) = T .
T
b
b T −
s−1 s
ζQ,2 (s) of Q is given by
In particular, the rank two non-abelian zeta function b
ζ(2s) bζ(2s − 1)
ζQ,2 (s) = b
ζQ,2 (s) = <(s) > 1.
1
b
b −
s−1 s
Proof. This is a direct consequence of Proposition 5.8 and Corollary 5.2. Indeed,
by Proposition 5.8, the Fourier expansion of E(z,
b s) is given by
b s) = b
E(z, ζ(2s)y s + b
ζ(2s − 1)y1−s + non−constant term.
Hence, by Corollary 5.2,
Z T Z ∞
s
dy
dy
ζQ,2 (s) =
b T
ζ(2s)y 2 −
b ζ(2s − 1)y1−s 2
b
0 y T y
(5.78)
ζ(2s) s−1 b ζ(2s − 1) −s
= T ,
b
T −
s−1 s
as desired. 
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 140

140 Rank Two Zeta Functions

5.4.2 Generalization of Rankin-Selberg Method

In this subsection, we give a generalization of the Rankin-Selberg method for

the Eisenstein series on the three dimensional hyperbolic space. This should be
known to experts. But as we can hardly find any details in literatures, we decide
to write it down to motivate our later discussions.

Let H be the three dimensional hyperbolic upper half space. For the hyperbolic
dx2 + dy2 + dr2
metric ds2 := , its volume form, resp. its Laplace operator, is
r2
given by

2 ∂ ∂2 ∂2 ∂
2
!
dx ∧ dy ∧ dr
dµ := resp. ∆ := r + + − r . (5.79)
r3 ∂x2 ∂y2 ∂r2 ∂r

Let D ⊂ H be a domain with the boundary ∂D. If A, B, C are C 1 -functions on

D, by the Stokes formula,
$ " 
∂ ∂ ∂
! 
A + B + C dx∧dy∧dr = Ady∧dr + Bdr ∧dx+Cdx∧dy .
D ∂x ∂y ∂r ∂D
(5.80)
Moreover, if f and g are two C 2 -functions on D,
∂ ∂ ∂ ∂ ∂2 ∂ ∂ ∂ ∂ ∂2
! !
f ·g − f · g = 2 f · g, f· g − f · g = f · 2 g.
∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x
∂ ∂
And similar formulas hold for the functions and . Consequently,
∂y ∂r
∂2 ∂2 ∂2 ∂2 ∂2 ∂2
! !
+ + 2 f ·g− f · + + g
∂x2 ∂y2 ∂r ∂x2 ∂y2 ∂r2
∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂
! ! !
= f · g− f · g + f · g− f · g + f · g− f · g .
∂x ∂x ∂x ∂y ∂y ∂y ∂r ∂r ∂r
Clearly, the left hand side of the above equation is simply
1 ∂ 1 ∂  1 ∂f ∂g
! ! !
1 1 1
∆ + f · g − f · ∆ + g = ∆ f · g − f ∆g + · g − f · .
r2 r ∂r r2 r ∂r r2 r ∂r ∂r
Hence, we have
1  ∂ ∂ ∂
!
∂ ∂ ∂
!
∆ f · g − f ∆g = f · g − f · g + f ·g− f · g
r2 ∂x ∂x ∂x ∂y ∂y ∂y
(5.81)
∂ ∂ ∂ ∂f ∂g
! !
1
+ f ·g− f · g − ·g− f · .
∂r ∂r ∂r r ∂r ∂r
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 141

Rank Two Zeta Functions 141

This implies that

$   dx ∧ dy ∧ dr
∆ f · g − f ∆g
D r3
$ "
∂ ∂ ∂ ∂ ∂ ∂
! !
= f ·g− f · g + f ·g− f · g
D ∂x ∂x ∂x ∂y ∂y ∂y
∂ ∂ ∂ 1 ∂f ∂g dx ∧ dy ∧ dr
! !#
+ f ·g− f · g − ·g− f ·
∂r ∂r ∂r r ∂r ∂r r
$ "
∂ 1 ∂ ∂ ∂ 1 ∂ ∂
!! !!
= f ·g− f · g + f ·g− f · g
D ∂x r ∂x ∂x ∂y r ∂y ∂y
∂ 1 ∂ ∂
!!
+ f ·g− f · g
∂r r ∂r ∂r
1 ∂f ∂g 1 ∂f ∂g
! !#
+ 2 ·g− f · − 2 ·g− f · dx ∧ dy ∧ dr
r ∂r ∂r r ∂r ∂r
$
∂ 1 ∂ ∂ ∂ 1 ∂ ∂
" !! !!
= f ·g− f · g + f ·g− f · g
D ∂x r ∂x ∂x ∂y r ∂y ∂y
∂ 1 ∂ ∂
!! #
+ f · g − f · g dx ∧ dy ∧ dr
∂r r ∂r ∂r
"
1 ∂ ∂ 1 ∂ ∂
! !
1 1
= f · g − f · g dy ∧ dr + f · g − f · g dr ∧ dx
∂D r ∂x r ∂x r ∂y r ∂y
1 ∂ ∂
!
1
+ f · g − f · g dx ∧ dy,
r ∂r r ∂r
where, in the last equality, we have used the Stokes formula. This then proves the
following:

Lemma 5.11. (Stokes’ Formula for Hyperbolic Geometry) With the same nota-
tions as above, we have
$
dx ∧ dy ∧ dr
((∆ − 2s(2s − 2)) f · g − f (∆ − 2s(2s − 2)g)
D r3
"
∂ ∂ ∂ ∂
! !
dr dr
= f · g − f · g dy ∧ + f ·g− f · g ∧ dx
∂D ∂x ∂x r ∂y ∂y r
1 ∂  1 ∂  !
+ f ·g− f · g dx ∧ dy.
r ∂r r ∂r

Example 5.1. As a useful example, take f = 1 and g = E(s) b = E(P,

b s), the
Eisenstein series associated to the modular group SL(2, OK ). Here K denotes
an imaginary quadratic field with OK its ring of integers. Let D ⊂ H be the
fundamental domain D for the group Γ := SL(2, OK ) constructed in §4.4.4.2. As
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 142

142 Rank Two Zeta Functions

usual, write P = x + iy + jr ∈ D with x, y ∈ R and r ∈ R>0 . By applying Lemma

5.11, since ∆E(P,
b s) = 2s(2s − 2)E(P,
$
b s),
dx ∧ dy ∧ dr
− 2s(2s − 2) E(P,
b s)
r3
$  D

b s) − 1 · 0 dx ∧ dy ∧ dr

= −2s(2s − 2) · E(P,
r3
$D 
b s) dx ∧ dy ∧ dr

= (∆ − 2s(2s − 2))1 · E(P,
b s) − 1 · (∆ − 2s(2s − 2))E(P,
D r3
"
∂ b ∂ b 1 ∂ b
!
dr dr
= − E(P, s) dy ∧ − E(P, s) ∧ dx − E(P, s) dx ∧ dy.
∂D ∂x r ∂y r r ∂r
For a fixed T > 0, introduce a truncated domain D = DT ⊂ DT , a compact
subset of D, by cutting off the neighborhoods of cusps defined by r > T for the
cusp ∞, and its Ai -transform for other cusps κi , where Ai (∞) = κi , i = 1, . . . , h. In
terms of the notations in §4.4.4.2,
[h
DT := D r F
fη (T ),
i
(5.82)
i=1
and, we may and hence will assume that T is sufficiently large to guarantee
i=1 Fηi (T ) =
Sh f Fh f
i=1 Fηi (T ). Then, the boundary of DT may be classified into two
different types. Namely, the first, denoted as the Pi (T )’s, which are compact sur-
faces obtained from the intersection of the cuspidal neighborhood for ∞ with the
hyperplane defined by r = T , and its association for other cusps; and the second,
denoted by S σ,µ ’s, which are defined by the conditions µ(σ, P) = µ(τ, P) for pairs
of inequivalent Γ-cusps (σ, µ). It is well known that, see e.g. [25], since the group
Γ is finitely generated, we can and hence will pair the boundary surfaces S σ,µ ’s
to satisfying the condition that, within each pair, the oriented norm directions for
their components are opposite to each other, using the fact that the angles mea-
sured in hyperbolic geometry and in Euclidean geometry are the same. Therefore,
when calculating the integrations in Example 5.1, after cancelling the contribu-
tions from two boundary surfaces in every pair appeared in the S σ,µ ’s, what is left
on the right hand side of Example 5.1 is merely the part of the integrations over
the first type of the boundaries Pi (T )’s. Furthermore, for the induced integrations
over Pi (T )’s, we have the following:

Lemma
" 5.12. With the same notation as " above,
∂ b dr ∂ b dr
E(P, s) dy ∧ = 0 and E(P, s) ∧ dx = 0. (5.83)
Pi (T ) ∂x r Pi (T ) ∂y r
In particular,
$ h "
∂ b
!
dx ∧ dy ∧ dr 1 X
E(P, s)
b = r E(P, s) dx ∧ dy. (5.84)
D r3 2s(2s−2) i=1 Pi (T ) ∂r
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 143

Rank Two Zeta Functions 143

Proof. We use the Fourier expansion for E(P, s) given in Theorem 5.7. to divide
these integrations into two parts, one for the constant terms one and the other for
all the non-constant terms.
(1) The part resulting from the constant terms is simply zero. Indeed, by The-
orem 5.7, the constant terms are functions in r (and s) and hence are independent
of x and y. Therefore, the corresponding derivatives of these constant terms in the
x-direction and hence the integrations vanish.
(2) The part resulting from all the non-constant terms is zero so well, even be-
cause of totally different reasons. Indeed, for the non-constant terms, the averages
of exponential functions e2πihω ,zi over z = x + iy ∈ Pi (T ) are all zero.
0

This verifies (5.83) and hence also (5.84). 

To finally evaluate the integrals (on the right hand side) of (5.84), by (5.47) in
Theorem 5.6, the constant term of E ba (P, s) at the cusp ηi = B(∞) is given by
2π
N(aB )2s ζa,aB (2s)r2s + √ N(aB )2−2s ζa,a−1 (2s − 1)r2−2s
∆K · (2s − 1) B

For simplicity, denote it by ai · r2s + bi · r2−2s . Consequently, we have

" "
1 ∂ b 1 ∂ b
! !
1
Ea (P, s) dx ∧ dy = E(P, s) r=T · dx ∧ dy
2s(2s − 2) Pi (T ) r ∂r r ∂r P (T )
! i
ai 2s−2 bi −2s
= T + T vol(Pi (T )).
s−1 −s
Put all these together, we obtain the following:

Proposition 5.10. With the same notation as above, we have

$ h !
b s) dx ∧ dy ∧ dr = ai bi
X
2s−2 −2s
E(P, · T − · T · vol(Pi (T )).
DT r3 i=1
s−1 s

5.4.3 Rank Two Non-Abelian Zeta Function

Let K be an algebraic number field of degree n over Q with OK its ring of integers
of K and DK the absolute value of its discriminant of K.1 Denote by r1 , resp. r2 ,
the number of real, resp. complex, places of K and denote by S ∞ the collection of
all Archimedean places of K. For convenience, we use σ, resp. τ, for real, resp.
complex places in S ∞ . By an abuse of notation, let ∆K be the hyperbolic Laplace
operator for the space H r1 × Hr2 . That is,
X X
∆K := ∆σ + ∆τ , (5.85)
σ:R τ:C
1 Differentfrom the rest of this book, here we use DK instead of ∆K , for the purpose of saving ∆ to
represent Laplacian operators below.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 144

144 Rank Two Zeta Functions

where
∂2 ∂2 ∂2 ∂2 ∂2 ∂
! !
∆σ := y2σ + and ∆τ := rτ2 + + −r . (5.86)
∂xσ2 ∂y2σ ∂xτ2 ∂y2τ ∂rτ2 ∂rτ
Let a1 , . . . , ah be OK -ideals such that their associated ideal classes gives the ideal
group CLK of K. In this case, we use a as a running symbol for the ai ’s. Let
Γ := SL(OK ⊕ a). For the action of Γ on H r1 × Hr2 , denote by CΓ the set of Γ-
cusps, and write CΓ := [η1 ], . . . , [ηh ] for some η1 , . . . , ηh ∈ P 1 (K).


Let T ≥ 1 be a fixed constant. Following Definition 5.1, for a cusp ηi , within

the fundamental domain D ⊂ H r1 × Hr2 of SL(OK ⊕ a) constructed in §4.4.4.2, we
introduce an open neighborhood Xi (T ) of ηi by

Xi (T ) := τ ∈ D ⊂ H r1 × Hr2 : d(η, τ) < 1/T .


(5.87)

By Proposition 5.2, Xi (T ) ∩ X j (T ) = ∅ if i , j. Moreover, by Theorem 5.3, D1 (of

Definition 5.2) coincides with the moduli space of rank two semi-stable lattices of
volume one. Accordingly, as in Definition 5.2, set
h
G
DT := D \ Xi (T ). (5.88)
i=1

By §5.2.2 and §5.3.3, to obtain an explicit formula for the rank two non-abelian
ζK,2 (s) of K, it suffices to calculate the following integration
zeta function b
$
b2,a (τ, s)dµ(τ).
E (5.89)
DT

Here Eb2,a (τ, s) denotes the complete Eisenstein series in Definition 5.7.
Motivated by what we have done for totally real fields in the previous subsec-
tion, we use the following facts to treat the integration (5.89).

(1) The Eisenstein series Eb2,a (τ, s) is an ‘eigenfunction’ of the associated Laplace
operator.
(2) The Stokes’ formula would transform the integration domain from DT to its
boundary ∂DT .
(3) We divide the boundary ∂DT into two groups. Namely, the one consisting of
the so-called horo-spheres and the others for the rest. For the first group, since
horo-spheres boundaries are paired with opposite normal directions, there are
no contributes to our integration from such boundary components because of
cancellation within each pair.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 145

Rank Two Zeta Functions 145

(4) On the remaining boundary components, the Fourier expansion of the Eisen-
stein series Eb2,a (τ, s) can be used to simplify the calculation. Indeed, in this
case, there are two parts of the variable directions involved, namely, one in
the ReZ direction and another in the ImJ direction. It is easy to see that the
averages of higher degree terms (in the Fourier expansion for the Eisenstein
series E
b2,a (τ, s)) on the boundary ReZ directions are zero.

To be more precisely, since

∆σ yσs = s(s − 1) · yσs ∆τ rτ2s = 2s(2s − 2) · rτ2s ,



and (5.90)
by SL-invariance of hyperbolic metrics pointed out in §4.1, we conclude that
∆K Eb2,a (τ, s)
= r1 · s(s − 1)
+ r2 · 2s(2s − 2)

· E
b2,a (τ, s) <(s) > 1. (5.91)
This implies that
$ $
b2,a (τ, s) dµ(τ) = r1 + 4r2
E ∆K E
b2,a (τ, s) dµ(τ). (5.92)
DT s(s − 1) DT

Furthermore, using the Stokes Formula, we have

$
∆K Eb2,a (τ, s)dµ(τ)
DT
$   $
= ∆K E2,a (τ, s) · 1dµ(τ) −
b b2,a (τ, s) · ∆K 1
dµ(τ)
E
D DT
$ T  (5.93)
= ∆K E b2,a (τ, s) · ∆K 1
dµ(τ)


b2,a (τ, s) · 1 − E
DT
" "
 ∂E ∂1 ∂E
 
b2,a (τ, s) b2,a (τ, s)
= b2,a (τ, s) ·  dµ =

·1−E dµ.
∂ν ∂ν ∂ν

∂D(T ) ∂D(T )

∂
Here denotes the outer normal derivative and dµ denotes the volume element
∂ν
of the boundary ∂DT , by an abuse of notations.
∂ b
To go further, we next examine the integrand E 2,a (τ, s) in the ImZ directions
" ∂ν
∂ b
and study how the integration E2,a (τ, s) dµ can be simplified. For this,
∂D(T ) ∂ν
as usual, we first proceed our discussions by transforming the cusps η to ∞ using
some special transformations introduced in Lemma 4.2. Clearly, they do not really
change the integrations, since special transformations are elements of SL(2, K).
Thus, without loss of generality, we may assume that η = ∞.
To calculate our integrations more effectively, we need to simply write down

(a) The outer normal direction of the boundary, and

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 146

146 Rank Two Zeta Functions

(b) The fundamental domain of the stabilizer group of cusps2 .

For this purpose, we next introduce some variable changes for the ImJ directions.
Recall that the ImJ directions are simply the yσ -directions for the real places
σ’s, and the vτ -directions for the complex places τ’s. (Here to avoid notational
confusion, we use P = z + v j, instead of P = z + r j, to denote point P ∈ H,
since r may be confused with r1 and r2 for number of real and complex places.)
Moreover, for a point (z1 , . . . , zr1 , P1 , . . . , Pr2 ) in H r1 × Hr2 , its ImJ coordinates
(yσ1 , . . . , yσr1 , vτ1 , . . . , vτr2 ) ∈ Rr+1 +r2 yield a natural norm
Y Y
N(yσ1 , . . . , yσr1 , vτ1 , . . . , vτr2 ) = yσ1 · . . . · yσr1 · vτ1 · . . . · vτr2 2 = v2τ .



yσ ·
σ: R τ: C

Accordingly, let ε1 , . . . , εr1 +r2 −1 be a generator of the unit group U K of K (mod-

ulo the torsion). By Dirichlet’s unit theorem, see e.g. §V.1 of [66], the matrix

 1 log |ε(1) (1)

 
1 | · · · log |εr1 +r2 −1 |

 
· · · ··· ··· ···  (5.94)
 (r1 +r2 ) (r1 +r2 ) 
1 log |ε1 | · · · log |εr1 +r2 −1 |


r1 +r2 −1,r1 +r2

is invertible. Let (e(i)
j )i=0, j=1 be its inverse. Then, by definition, we have

1
j =
(i) The entries of the first row is given by e(0) , j = 1, 2, . . . , r1 + r2 :
r1 + r2
1 +r2
rX
(ii) j = 0 for all i = 1, . . . , r1 + r2 − 1: and
e(i)
j=1
1 +r2
rX
( j)
(iii) j log |εk | = δik for all i, k = 1, . . . , r1 + r2 − 1.
e(i)
j=1

This implies that

 1 1 1 
 r1 +r2 r1 +r2 ··· r1 +r2 
 
 (i)  
e j =  e(1)
1 e(1)
2 ··· e(1)
r1 +r2  .

(5.95)
 ··· ··· ··· · · · 
1 +r2 −1) 1 +r2 −1) 1 +r2 −1)

e(r e(r · · · e(r

1 2 r1 +r2

2 By the construction of the fundamental domain D in § 4.4.4.2, near cusps, D coincides with funda-

mental domain of the associated stabilizer groups.

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 147

Rank Two Zeta Functions 147

Define then a change of variables by:

Y Y
, . . . , , , . . . , = v2τ ,




 Y0 := N(y σ1
y σ r 1
vτ 1
v τ r 2
) y σ ·
σ:R  τ:C



 r r2
1


 1 X X 
Y1 :=  ei log yσi + (1) (1) 2



  er1 + j log vτ j  


 2 i=1 j=1







 ··· ··· ··· ··· ···
 r r2

1
1 X

+r +r
 X 
(r −1) −1)
+ (r 2  .
 1 2 1 2
Y := e log y e log v

+r σ

r −1 r1 + j τj 

2  i=1 i

 1 2

  i 
j=1

Obviously, by inverting the above relations, we have

1 Y
r1 +r2 −1 (i) 2Yq

r1 +r2
y = Y εq i = 1, . . . , r1 ,

σ


 i
 0 q=1
(5.96)

 1
r1 +r2 −1
 2
1 + j) 2Yq
Y
r +r

 v2τ j = Y0 1 2 ε(r j = 1, . . . , r2 .


 q
q=1

Set accordingly
1 +r2 −1
r1Y
r +r2 1 + j) 2
2Yq
t j := v2τ j = Y0 1 ε(r
q j = 1, . . . , r2 . (5.97)
q=1

Note here, for the complex places τ, the power 2 is added since Nτ = 2.
With respect to this change of variables, from the precise construction in
§ 4.3.4.2, the fundamental domain for the action of Γη ⊂ SL(OK ⊕ a) on H r1 × Hr2
becomes simply the regular domain characterized by
1 1
0 < Y0 < ∞, − ≤ Y1 , . . . , Yr1 +r2 −1 ≤ ,
2 2 (5.98)
(xσ1 , . . . , xσr1 : zτ1 , . . . , zτr2 ) ∈ Fη (ab ),
−2

where Fη (ab−2 ) denotes a fundamental parallelepiped associated with the lattice

ab−2 in Rr1 × Cr2 .
Certainly, with respect to this change of variables, the volume form is changed
accordingly. To calculate this change, recall that the hyperbolic metric on H r1 ×
Hr2 is given by
!
g 0
g = ImJ (5.99)
0 gReZ
where the metrics for the ImJ and ReZ directions are given by
 
 1 1  1 2  1 2 
gImJ = gReZ = diag  2 , . . . , 2 , 2 , . . . , 2  . (5.100)
y1 yr1 v1 vr2
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 148

148 Rank Two Zeta Functions

First we treat the ReZ directions. Since, under the above change of variables,
the changes in such directions are given by a parallel transformation along the
lattices ab−2 , there is no change of volume here. Next we treat the ImJ directions.
Recall that, in general, to find the matrix (g̃i j ) obtained from (gi j ) by the change
of variables, we need to calculate the partial derivatives so as to get g̃i j from the
formula
X ∂xα ∂xβ
g̃i j = gαβ . (5.101)
α,β
∂ x̃i ∂ x̃ j

(Here, in terms of gi j and g̃i j , the variables are assumed to be renumbered as

x1 , x2 , · · · , xr1 +r2 and x̃1 , x̃2 , · · · , x̃r1 +r2 respectively.) In the case at hands, by defi-
nition, originally, (gi j ) = gImJ . And, by a direct calculation, we have
∂yσi 1
= yσ , i = 1, · · · , r1 ,
∂Y0 (r1 + r2 )Y0 i
∂yσi
= 2 log ε(i)
q · yσi , i = 1, · · · , r1 , q = 1, · · · , r1 + r2 − 1
∂Yq
(5.102)
∂tτ j 1
= tτ , j = 1, · · · , r2 ,
∂Y0 (r1 + r2 )Y0 j
∂tτ j 2
= 2 log ε(qj) · tτ j , j = 1, · · · , r2 , q = 1, · · · , r1 + r2 − 1.
∂Yq
Since the metric matrices are always symmetric, by (5.101), we are lead to calcu-
late the following three types of products of matrices:
 
 1 1  1 2  1 2 
X0 · diag  2 , · · ·, 2 , 2 , · · ·, 2  · X0t ,

y1 yr1 v1 vr2
 
 1 1  1 2  1 2 
X0 · diag  2 , · · ·, 2 , 2 , · · ·, 2  · Xqt , (5.103)
y1 yr1 v1 vr2
 
 1 1  1 2  1 2 
X p · diag  2 , · · ·, 2 , 2 , · · ·, 2  · Xqt

y1 yr1 v1 vr2

where
yσr1 tτr2
!
yσ1 tτ1
X0 := ,··· , , ,··· ,
(r1 + r2 )Y0 (r1 + r2 )Y0 (r1 + r2 )Y0 (r1 + r2 )Y0
and, for p, q = 1, 2, · · · , r1 + r2 − 1,
 
2 (r1 ) 2 (r1 +1) 2 2 (r1 +r2 ) 2 2
X p = log ε(1)
p yσ 1
, · · · , log ε p yσr1
, 2 log ε p vτ1 , · · · , 2 log ε p vτr .
2
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 149

Rank Two Zeta Functions 149

r1 r2
log ε(rp 1 + j) = 0, we have
2
X X
This is rather direct. Indeed, since log ε(i)
p +
i=1 j=1

1
g̃11 = , g̃1 j = g̃ j1 = 0, j = 2, · · · , r1 + r2 . (5.104)
(r1 + r2 )Y02
And, for i, j = 1, . . . , r1 + r2 − 1,
r1 1 +r2
rX
2 2
X
g̃(i+1)( j+1) = 4 log ε(p)
i log ε(p)
j + 4 log ε(p)
i log ε(p)
j . (5.105)
p=1 p=r1 +1
 r1 +r2 −1
Consequently, if we set R := log ε(p)
q be the regulator of K, then
p,q=1

4r1 +r2 −1 2
det(g̃i j ) = R. (5.106)
(r1 + r2 )Y02
Therefore, by combining this with contributions in the ReZ directions, we have
s 
 1  Y Y
dω =  det(g̃i j ) · 2  dY0 dY1 · · · dYr1 +r2 −1
 dxσ dzτ
Y0 σ:R τ:C
(5.107)
2r1 +r2 −1 dY0 Y Y
= √ R dY1 · · · dYr1 +r2 −1 dxσ dzτ .
r1 + r2 Y02 σ:R τ:C

To make all this useful, we need to find the induced volume form on the bound-
ary ∂DT as well. This is now rather direct. Indeed, by our constructions, the
boundary ∂DT of DT consists of

(1) The corresponding parts of the boundary of D, and

(2) The hyperplane of D defined by the condition Y0 = T 0 := N(ab−2 ) · T 3 .

Therefore, if we set dµ be the volume element of this hyperplane, then

√
1 r1 + r2 r1 +r2 −1 Y Y
dµ = √ dω Y =T 0 = 0
2 R dY1 · · · dYr1 +r2 −1 dxσ dzτ .
g̃11 0 T σ:R τ:C

Furthermore, if we let ν be the unit normal to the hyperplane, since

D ∂ ∂ E
, = g̃11 Y =T 0 = (r1 + r2 )T 0 2 , (5.108)
∂Y0 ∂Y0 0

we have
!
1
ν= √ , 0, . . . , 0 . (5.109)
r1 + r2 T 0
3 The factor N(ab−2 ) is added here, because, in the definition of D , what we used is the distance to
T
cusp, not simply Y0 .
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 150

150 Rank Two Zeta Functions

In particular, the outer normal derivative of a function f is given by

∂f ∂f
!
1 √
= √ , 0, . . . , 0 · grad f = r1 + r2 · T 0 . (5.110)
∂ν r1 + r2 T 0 ∂Y0

With the volume form issue settled, we now examine in more details the
boundary ∂DT . By (the fact that the group SL(OK ⊕ a) is finitely generated and)
our concrete construction of the fundamental domain D for Γ, the boundary ∂DT
of DT consists of finitely many of surfaces which are either the parts of the horo-
spheres or the parts Xi (T ) of the planes cutting out by the equation Y0 = T i0 , where
T i0 = N(ab−2i ) · T (with bi the fractional ideal associated to the cusp ηi ). Moreover,
besides the hyperplanes associated to the Y0 = T 0 ’s, the part of the horospheres
(appeared on the boundary) is divided into the sets of inequivalent pairs, within
each of which, the integration of the outer normal derivative along one surface is
equal to the integration of the inner normal derivative along the remaining sur-
face. (In terms of Y p≥1 , in each of the pairs, the two members are determined
by the equations Y p = ± 12 .) Thus, after such calculations, we are left with the
integrations on the Xi (T )’s, where Xi (T ) denotes the part of the boundary of DT
coming from the pull back of the intersection of the hyper-surface Y0 = T i0 with
Fηi , i = 1, 2, . . . , h. That is,

$ $
b2,a (τ, s)dµ(τ) = 1 1
E ∆K E b2,a (τ, s)dµ(τ)
DT r1 + 4r2 s(s − 1) DT
"
∂E
b2,a (τ, s)
= dµ (5.111)
∂DT ∂ν
h "
1 1 X ∂Eb2,a (τ, s)
= dτ.
r1 + 4r2 s(s − 1) i=1 Xi (T ) ∂ν

(Here we used the fact that for T ≥ 1, Xi (T ) are disjoint from each other, proved
in Proposition 5.2.)
Finally, to complete our calculation, we use the Fourier expansion of the
Eisenstein series Eb2,a (τ, s) listed in Theorem 5.7. First, we consider the non- con-
stant terms there. Easily, the average for e2πit (together with its derivative) over an
! ∂Eb (τ,s)
interval of length 1 is zero. Thus, in X (T ) 2,a∂ν dτ, contributions from higher
i
degree terms (in the Fourier expansion) is simple zero. Hence, in the integration
of (5.111), we are only left with contributions resulting from the constant terms
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 151

Rank Two Zeta Functions 151

in the Fourier expansion for E(s).

b Consequently, with T i0 = N(ab−2
i ) · T , we have,
$
b2,a (τ, s)dµ(τ)
E
DT
h "
1 1 X ∂E
b2,a (τ, s)
= dτ
r1 + 4r2 s(s − 1) i=1 Xi (T ) ∂ν
h "
1 1 X ∂
= A0i Y0s + B0i Y01−s dµ


r1 + 4r2 s(s − 1) i=1 Xi (T ) ∂ν
h "
1 1 X √ ∂
= r1 + r2 · T i0 A0i Y0s + B0i Y01−s


r1 + 4r2 s(s − 1) i=1 Xi (T ) ∂Y 0
√
r1 + r2 r1 +r2 −1 Y Y
× 0 2 · r dY1 . . . dYr1 +r2 −1 · dxσ · dzτ
Ti σ:R τ:C
h "
r1 + r2 1 X
= · 2r1 +r2 −1 · r s A0i Y0 s−1 − (s − 1) B0i Y0 −s


r1 + 4r2 s(s − 1) i=1 Xi (T )
Y Y
× dY1 . . . dYr1 +r2 −1 · dxσ · dzτ
σ:R τ:C
h
r1 + r2 1 X
= 2r1 +r2 −1 R · s A0i T i0 s−1 − (s − 1) B0i T i0 −s


r1 + 4r2 s(s − 1) i=1
" Y Y
× dY1 . . . dYr1 +r2 −1 · dxσ · dzτ
Xi (T ) σ:R τ:C
h
r1 + r2 r1 +r2 −1 1 X  A
0i B0i 0 −s 
= 2 · r DK 2
N(ab−2
i )· · T i0 s−1 − T .
r1 + 4r2 i=1
s−1 s i

Here, in the last equality, we have used the fact that the lattice corresponding to
αi
cusp ηi = i with bi = OK αi + aβi and Y p ∈ − 2 , 2 . Therefore,
 1 1
is given by ab−2
βi
with the precisely formula we have obtained for A0i (s) and the functional equation
ζ(s), we have proved the following:
for b

Theorem 5.8. We have

$
b2,a (τ, s)dµ(τ) = ζK (2s) T s−1 − ζK (2 − 2s) T −s .
b b
E
DT s − 1 s

Proof. Indeed, by the functional equation, we only need to calculate the coeffi-
T s−1
cient of . By Theorem 5.7, the partial constant term A0,i for the completed
s−1
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 152

152 Rank Two Zeta Functions

Eisenstein series Eb2,a (τ, s) is given by



ΓR (2s)r1 ΓC (2s)r2 N(a)∆K s · ζ[a−1 bi ] (2s) · N(ab−1
i )
−2s
. (5.112)

T s−1
Hence, up to a constant factor depending only on K, the coefficient of in the
$ s−1
integration b2,a (τ, s)dµ(τ) is simply the summation hi=1 of N(ab−2 )A0,i
E
P
i
DT
twisted by a multiple factor N(ab−2 i )
s−1
, resulting from the discrepancy between T
T s−1
and T i0 . That is to say, the coefficient of is nothing but
s−1
h
X
ΓR (2s)r1 ΓC (2s)r2 N(a)∆K s ζ[a−1 bi ] (2s)N(ab−1 −2s
· N(ab−2 −2 s−1


i ) i ) · N(abi )
i=1
h
X
= ∆Ks · ΓR (2s)r1 ΓC (2s)r2 ζ[a−1 bi ] (2s)
i=1

= ∆Ks · ΓR (2s)r1 ΓC (2s)r2 ζK (2s) = ∆Ks · b

ζK (2s),
h
X
since ζ[a−1 bi ] (2s) = ζK (2s), resulting from the facts that, for a fixed a, the h ideal
i=1
classes [a−1 bi ] run over all elements of the class group of K, and that the total
Dedekind zeta function decomposes into a summation of partial zeta functions
associated to h-inequivalent ideal classes. 

As a direct consequence, we have the following:

ζK,2 (s) of a number K is

Theorem 5.9. The rank two non-abelian zeta function b
given by
ζK (2s) bζK (2s − 1)
ζK,2 (s) = <(s) > 1.
b
b −
s−1 s
Proof. By Theorem 5.3, the moduli space of rank two semi-stable lattices of vol-
ume N(a)∆K with underlying projective module OK ⊕ a is given by D1 . Hence, by
the previous theorem,
$
b2,a (τ, s)dµ(τ) = ζK (2s) − ζK (2 − 2s) .
b b
E
D1 s−1 s
That is to say,
ζK (2s) bζK (2 − 2s)
ζK,2:a (s) = .
b
b −
s−1 s
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 153

Rank Two Zeta Functions 153

Therefore, by §5.2.2 and §5.3.3,

ζK (2s) bζK (2 − 2s)
ζK,2 (s) = ,
b
b −
s−1 s
as desired. 

We end this discussion by pointing out that the method proving Theorem 5.8
is first adopted in [24] to establish the trace formulas related to totally real fields.

5.5 Zeros of Rank Two Non-Abelian Zeta Functions

In this section, following Suzuki ([111]), we show that all zeros of rank two zeta
functions lie on the central line.

5.5.1 Product Formula for Entire Function of Order 1

A function f (z) on C is called an entire function of order one if

(i) The function f (z) is analytic on C, and

(ii) As |z| → ∞, f (z) = O exp(|z|1+ε ) ∀ε > 0.

Let n(R) denote the number of zeros of f (z) inside CR , the circle of the radius
R centered at the origin. It is well known that, for an entire function f (z) of order
one, we have, see e.g. [116],
α
(1) n(R) = O(RX ), ∀α > 1,
(2) The series |ρn |−α converges. In particular,
ρn , f (ρn )=0
! !  z 2 !
z z
1− exp =1+O n → ∞. (5.113)
ρn ρn ρn
! !
Y z z f (z)
(3) The infinite product P(z) := 1− · exp converges, and is an
ρn
ρ n ρ n P(z)
entire function of order 1 without zeros, hence is of the form exp(A + Bz) for
some constants A, B.

All these then imply the following famous:

Theorem 5.10. (Hadamard Product Theorem) Let f (z) be an entire function of

order one on C. There exist constants A, B such that
! !
Y z z
f (z) = eA+Bz · 1− · exp . (5.114)
ρ
ρ ρ
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 154

154 Rank Two Zeta Functions

ζ(s), we
Example 5.2. (See e.g. [23]) For the complete Riemann zeta function b
have
! !
1 Y z z
ζ(s) = eA+Bz ·
s(s − 1) · b 1− · exp . (5.115)
2 ρ
ρ ρ

γ
!
1 1 1
Here A = − log 2, B = − − 1 + log 4π with γ = lim 1 + + · · · + − log n
2 2 n→∞ 2 n
denotes the Euler constant.

5.5.2 Zeros of Rank Two Non-Abelian Zeta Function of Q

We start with the following conditional statement of Suzuki.

Proposition 5.11. ([111]) Assume that the Riemann Hypothesis holds for the Rie-
ζQ,2 (s) of Q lie on
mann zeta function. All the zeros of the rank two zeta function b
the central line <(s) = 21 .
 
Proof. Let ξ(s) := s(1 − s)b
ζ(s) and F(z) := −ξ i
2 + 2iz .

Claim 5.5. We have

 i  i i 
F z+ −F z− = iz(1 + 4z2 ) b
ζQ,2 + iz .
4 4 2

Proof. Indeed, by definition,

F z + i/4 = −ξ i/2 + 2i(i/4 + z) = −ξ i/2 − i/2 + 2iz = −ξ(2iz).

So F z − i/4 = −Z 1 + 2iz and

F(z + i/4) − F(z − i/4) = (1 + 2iz)(−2iz)b ζ(1 + 2iz) − 2iz(1 − 2iz)b ζ(2iz)
ζ(1 + 2iz) b ζ(2iz) 
 
= 2iz(1 − 2iz)(1 + 2iz) · 
 b
−
1 + 2iz

2iz − 1
 ξ 2(i/2 + iz) ζ(2(i/2 + iz) − 1) 


= iz(1 + 4z ) ·   = iz(1 + 4z2 ) b
ζQ,2 (i/2 + iz),
2
b
−
(i/2 + iz) − 1 i/2 + iz
as desired. 

Claim 5.6. Assume that the  Riemann hypothesis holds for the Riemann zeta func-
i  i
tion. Then, all zeros of F z + −F z− are real. In particular, the Riemann
4 4
hypothesis holds for bζQ,2 (s).
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 155

Rank Two Zeta Functions 155

Proof. By Example 5.2, F(z) is an entire function of order 1. Hence, there are
constants A, B such that ! !
Y z z
F(z) = eA+Bz · 1− · exp .
ρ:F(ρ)=0
ρ ρ
Since the ρ’s are the zeros of complete Riemann zeta function, by the RH for ξ(s),
all the ρ’s are real.
Moreover, since F(z) = −ξ(i/2 + 2iz), we have, for x ∈ R,
F(x) = −ξ 1/2 + 2ix = −ξ 1/2 + 2ix = −ξ(i/2 − 2ix .

Thus, by the standard functional equation for ξ(s), we have

F(x) = −ξ 1 − (i/2 − 2ix) = −ξ i/2 + 2ix = F(x).

That is to say, F(x) takes only real values on x-axis. This implies that both the
constants A and B are real.
If z0 = x0 + iy0 be a zero of F(z + i/4) − F(z − i/4) = iz(1 + 4z2 ) b ζQ,2 (i/2 + iz),
z0 = 0 and/or z0 is a zero of b ζQ,2 (i/2 + iz) since b
ζQ,2 (i/2 + iz) admits simple poles at
z = ±1/2i. Consequently, F(z0 + i/4) = F(z0 − i/4). Hence, by taking the absolute
values for both sides, we arrive at
z0 + i/4 z0 + i/4
Y ! !
A+B(z0 +i/4)
e · 1− · exp
ρ ρ
! ! (5.116)
Y z0 − i/4 z0 − i/4
= eA+B(z0 −i/4) · 1− · exp .
ρ ρ
Since, from above, B ∈ R and ρ ∈ R, we get
Y∞
(x0 − ρn )2 + (y0 − 14 )2
1= .
n=1 (x0 − ρn ) + (y0 + 4 )
2 1 2

Thus, for y0 > 0, resp. for y0 < 0, the right hand side is < 1, resp. > 1. This is a
contradiction. Therefore, we must have y0 = 0. 
In this proof, the original RH for the Riemann zeta function
! is used to ensure
z0 + i/4
!
z0 − i/4
that ρ are real. This then forces the relation exp = exp .
ρ ρ
However, as pointed out by Lagarias, by the functional equation, we may pair
the roots ρ and 1 − ρ of the complete Riemann zeta function, or better, to
group the roots ρ, 1 − ρ, ρ and 1 − ρ altogether in the Hadamard product in-
volved and assume further that ! this Hadamard product is taken in the form
Y0 z
F(z) = eA+Bz · 1 − , where 0 means that ρ’s are paired or grouped
Q
ρ:F(ρ)=0 ρ
as above. In this way, we obtain the following unconditional:

ζQ,2 (s) of Q
Proposition 5.12. ([65]) All the zeros of the rank two zeta function b
lie on the central line <(s) = 12 .
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 156

156 Rank Two Zeta Functions

Proof. As above, we have

z0 + i/4 z0 + i/4
Y ! !
A+B(z0 +i/4)
e · 1− · exp
ρ ρ
! !
Y z0 − i/4 z0 − i/4
= eA+B(z0 −i/4) · 1− · exp .
ρ ρ
Arguing in the same way as above, we have B ∈ R. Now ρ, ρ̄, 1 − ρ, 1 − ρ̄
are regrouped altogether as single unit, we conclude that, for all such units,
1 1 2 <(ρ) 1 1 2 − 2 <(ρ)
+ = and + = are all reals. This then implies,
ρ ρ̄ |ρ|2 1 − ρ 1 − ρ̄ |1 − ρ|2
even without assuming the original RH for Riemann zeta function, that
Y ∞
(x0 − ρn )2 + (y0 − 14 )2
1= .
n=1 (x0 − ρn ) + (y0 + 4 )
2 1 2

Therefore, with the same arguments as in the proof for Proposition ref52318, we
have y0 = 0. 

5.5.3 A Simple Generalization

ζQ,2
Similar result holds for the functions b T
(s) as well, provided that T ≥ 1. Indeed,
by Theorem 5.8, we have
ζ(2s) ζ(2s − 1) −s
ζQ,2 (s) = ·T .
T
b b
b · T s−1 −
s−1 s
Consequently,
F(z + i/4) · T −i/2−iz − F(z − i/4) · T −i/2+iz = iz(1 + 4z2 ) b
ζQ,2
T
(i/2 + zi).
Therefore, using the same arguments as in the proof above, we have
Y z0 + i/4
z0 + i/4

eA+B(z0 +i/4) ·
1
1− · exp · T −iz− 2
ρ ρ
Y z 0 − i/4 z0 − i/4
1
= eA+B(z0 −i/4) · · T iz− 2 .


1− · exp
ρ ρ
This implies, similarly, that :
Y∞
(x0 − ρn )2 + (y0 − 41 )2 T −y0
1= · y .
n=1 (x0 − ρn ) + (y0 + 4 )
2 1 2 T 0
Or equivalently,
∞
Y (x0 − ρn )2 + (y0 − 14 )2
T 2y0
= . (5.117)
n=1 (x0 − ρn )2 + (y0 + 14 )2
Consequently, for T ≥ 1, we have
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 157

Rank Two Zeta Functions 157

(1) If y0 > 0, the left hand side of (5.117) is > 1. But the right hand side is < 1, a
contradiction.
(2) If y0 < 0, the left hand side of (5.117) is < 1. But the right hand side is > 1, a
contradiction.

This then proves the following:

ζQ,2
Theorem 5.11. Let T ≥ 1 be a fixed constant. All the zeros of b T
(s) lie on the
central line <(s) = 2 .
1

Obviously, the arguments above works even if the Riemann zeta function is
replaced by a classical zeta or L-function. For example, we have the following:

Theorem 5.12. Let K be a number field. All the zeros of rank two non-abelian
ζK,2 (s) of K lie on the central line <(s) = 12 .
zeta functions b

Proof. This is because

ζK (2s) bζK (2s − 1)
ζK,2 (s) = .
b
(1) The rank two zeta function for K is given by b −
s−1 s
ζK (s) is an entire function of order one (see e.g. [66]).
(2) The function s(s − 1)b

Hence, all the arguments in the proof of Theorem 5.12 work well. 

We end this elementary part by mentioning that, in the rest parts of this book,
the classical constructions and theories appeared here will be generalized (from
the group GL2 ) to all reductive groups, in particular, to the groups GLn . Unlike
the case for GL2 , where there is essentially only one proper parabolic subgroups,
for general reductive groups, the theories and constructions become very sophisti-
cated, because of the complications of the system of parabolic subgroups involved.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 159

PART 3

Eisenstein Periods and Multiple L-Functions

Eisenstein series are natural analytic invariants of arithmetic groups. Being

smooth, their integrations over compact domains make sense, and can be cal-
culated for lower rank groups in terms of the classical L-functions using the
Rankin-Selberg method. This is a special form of the trace formula and works
for general reductive groups. Indeed, the theory of Eisenstein series induced from
L2 -representations was developed by Langlands to obtain their meromorphic con-
tinuations and functional equations, and the analytic truncations with sufficiently
regular parameters were later introduced by Arthur in the trace formula. These
truncations are projections, and when applying to the constant functions, lead to
some compact domains, and hence the Eisenstein periods, namely, the integra-
tions of Eisenstein series over these compact domains. In this setting, the classi-
cal Rankin-Selberg method transforms to that of Jacquet-Lapid-Ragowski, based
on regularized integrations over cones. The resulting functions will be called the
multiple L-functions. In this part, we first give a notational review of the theories
above, then introduce the periods of reductive groups over number fields as an
example of the multiple L-functions, and finally present a joint work with Kim on
the volumes of the truncated compact domains above, a limit of which yields the
famous formula of Langlands on the volumes of fundamental domains.

159
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 161

Chapter 6

Multiple L-Functions

6.1 Reduction Groups

6.1.1 Parabolic Subgroups

Let F be a number field. By definition, the multiplicative group Gm over F is

an algebraic group such that, for any field extension E/F, the group Gm (E) (of
E-rational points of Gm ) is isomorphic to the multiplicative group E ∗ , and an F-
torus is an algebraic group defined over F which is isomorphic over the algebraic
closed field F of F to a finite product of copies of the multiplicative group. Write
∗
T(F) ' (F )r for some r ∈ Z≥1 , and say that T is split over a field extension E/F
if T(E) ' (E ∗ )r . We call r the rank or the absolute rank of the torus T. It is well
known that, for a fixed F-torus T, there is a unique minimal finite extension of F,
called the splitting field of T, over which T is split. Accordingly, the F-rank of T
is defined to be the maximal rank of a F-split sub-torus of T. In particular, a torus
is split over F if and only if its F-rank equals its absolute rank.
Opposite to the concept of torus is that of unipotent groups, in some sense.
By definition, an element g of an affine algebraic group G is called unipotent if
the associated right action Rg on the affine coordinate ring A[G] of G is locally
unipotent as an element of the ring of linear endomorphism of A[G], namely, its
restriction to any finite-dimensional stable subspace of A[G] is unipotent. Based
on this, an affine algebraic group is called unipotent if all its elements are unipo-
tent, and a subgroup of an algebraic group G is called the unipotent radical if it
is the unique largest normal unipotent connected closed subgroup of G. It is easy
to check that any unipotent algebraic group is isomorphic to a closed subgroup of
the group of upper triangular matrices with diagonal entries 1, and conversely any
such subgroup is unipotent. In particular, unipotent groups are nilpotent.
By definition, a smooth, affine algebraic group G over F is called reductive

161
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 162

162 Eisenstein Periods and Multiple L-Functions

if its unipotent radical over the algebraic closure F of F is trivial. And a sub-
group P of (a reductive group) G is called parabolic if P is closed in the Zariski
topology and the induced quotient space G/P is a complete algebraic variety. It
is well known that a closed subgroup of G is a parabolic subgroup if and only if
it coincides with the normalizer of its unipotent radical. In the sequel, let G be
a connected reductive group defined over F, and call all closed subgroups of G
defined over F as F-subgroups of G. Then, see e.g. [14] and [103], any two mini-
mal parabolic F-subgroups of G are conjugate over F, two parabolic F-subgroups
of G which are conjugate over some extension of the field F are conjugate over
F, and the set of conjugacy classes of parabolic subgroups, resp. of parabolic
F-subgroups, of G has 2r , resp. 2rF elements, where r, resp. rF , denotes the di-
mension of a maximal torus of the commutator subgroup [G, G] of G, resp. the
dimension of a maximal torus of the commutator subgroup [G, G] of G which
splits over F. In addition, each parabolic subgroup P of G admits a Levi decom-
position, namely, it can be represented as a semi-direct product of its unipotent
radical and an F-closed reductive subgroup, called a Levi subgroup of P, and two
Levi subgroups in P are conjugate by an element of P which is rational over F.
For a fixed minimal F-parabolic subgroup P0 of G, denote its unipotent radical
by N0 = NP0 and fix an F-Levi subgroup M0 = MP0 of P0 . Then P0 = M0 N0 is
a natural Levi decomposition of P0 . By definition, an F-parabolic subgroup P of
G is called standard if P contains P0 . For a standard parabolic subgroups P, there
exists a unique Levi subgroup M = MP of P which contains M0 . We call MP
the standard Levi subgroup of P. Similarly, there is a natural Levi decomposition
P = MP NP of P, where N = NP denotes the unipotent radical of P. From now
on, for simplicity, we use the terms parabolic subgroups and Levi subgroups to
denote standard F-parabolic subgroups and standard Levi subgroups respectively,
unless otherwise is stated.
Let P be a parabolic subgroup of G. Write T P for the maximal split torus in
the center ZP of MP and T P0 for the maximal split quotient torus of MP . Then
T P = Hom(Gm , ZP ) ⊗ Gm T P0 = Hom Hom(MPab , Gm ), Gm


and (6.1)
Here MPab denotes the maximal abelian quotient of MP . Moreover, the composition
T P ,−→ ZP ,−→ MP − MPab − T P0 (6.2)
is an isogeny, i.e. a morphism with finite kernel and finite cokernel. Consequently,
there is an injective morphism of free abelian groups of the same finite rank
X∗ (T P ) ,→ X∗ (T P0 ) (6.3)
where X∗ (T ) = Hom(Gm , T ) is the lattice of 1-parameter subgroups in the torus T .
Denote by X ∗ (T P ) := Hom(MP , Gm ) = Hom(MPab , Gm ) the space of characters of
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 163

Multiple L-Functions 163

MP . Easily,
X∗ (T P ) ' Hom(Gm , ZP ) X∗ (T P0 ) ' Hom X ∗ (T P0 ), Z .


and (6.4)
Following [3, 4], we set aP := X∗ (T P ) ⊗ R. Accordingly,
aP = X∗ (T P ) ⊗ R = X∗ (T P0 ) ⊗ R. (6.5)
For later use, denote the R-dimension of aP by d(P).
More generally, if P ⊆ Q are two standard parabolic subgroups of G, there are
canonical morphisms
T Q ,−→ T P ,−→ T P0 − T Q0 . (6.6)
The morphisms T Q ,→ T P and T P0  T Q0 induce a canonical embedding aQ ,→ aP
and a canonical retraction aP  aQ . Hence, there exists a canonical splitting
aP = aQ
P ⊕ aQ (6.7)
where aQ
P denotes the kernel of the restriction. Taking the dual R-vector spaces,
we have a∗P = X ∗ (MP ) ⊗ R = X ∗ (T P ) ⊗ R, and hence obtain a canonical splittings
a∗P = aQ∗
P ⊕ aQ .
∗
(6.8)
Similarly, if P ⊆ Q ⊆ R are standard parabolic subgroups of G, there are canonical
splittings
aP = aQ R
P ⊕ aQ ⊕ aR and a∗P = aQ∗
P ⊕ aQ ⊕ aR .
R∗ ∗
(6.9)

P , aQ
We shall denote by [·]Q , [·]RQ and [·]R the canonical projections of aP onto aQ R
P P
and aR respectively. In particular, for a0 := aP0 and a0 := aP0 ,

a0 = a0P ⊕ aQ
P ⊕ aQ and a∗0 = a0P∗ ⊕ aQ∗
P ⊕ aQ .
∗
(6.10)

6.1.2 Roots, Coroots, Weights and Coweights

Let P be a standard parabolic subgroup of G.

(1) Let ΦP ⊂ aG∗ ∗

P ⊆ aP be the set of non-trivial characters of T P , i.e. non-trivial
elements of Hom(T P , Gm ) which occur in the Lie algebra g of G.
(2) Let Φ+P ⊂ ΦP be the set of non-trivial characters of T P which occur in the Lie
algebra nP , the Lie algebra of the unipotent radical NP of P. Set Φ−0 := −Φ+P0 .

We have Φ0 := ΦP0 is a root system with Φ+0 := Φ+P0 the set of positive roots and
Φ0 = Φ+0 t Φ−0 .
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 164

164 Eisenstein Periods and Multiple L-Functions

(3) Let ∆0 := ∆P0 ⊂ Φ+0 be the set of simple roots, and let b
∆0 be the subsets of
simple weights in a∗0 .

+
Then ∆0 forms a basis of aG∗ α∈∆0 Z≥0 α, and the elements of ∆0 are non-
P
P0 , P0 ⊂
b
negative linear combinations of elements in ∆0 . In addition, for each α ∈ Φ0 , there
exists a corresponding coroot α∨ .

(4) Let ∆∨0 , resp. b

∆∨0 , be the basis of a0 dual to b
∆0 , resp. ∆0 .

Being the dual of the collection of simple weights and of simple roots, the sets ∆∨0
∆∨0 are the sets of simple coroots and of simple coweights, respectively, and
and b
are R-bases of the space aG0 .
For a standard parabolic subgroup P, ΦP may not be a root system, even many
constructions for a root system (Φ0 , Φ+0 , ∆0 ) can be parallel introduced.

(5) Let ∆P ⊂ Φ+P be the set of non-trivial restrictions to T P of simple roots (in ∆0 ),
or the same, be the set of non-trivial restrictions to aP of simple roots (in ∆0 ).
Namely, ∆P = α|aP ∈ a∗P : α ∈ ∆0 }r{0}.

(6) For each α ∈ ∆P , choose β ∈ ∆0 such that α = β|aP and set α∨ := [β∨ ]GP be the
projection of β∨ ∈ aG0 to aGP .

Since there exists a unique β ∈ ∆0 satisfying this property, α∨ is well-defined.

(7) Let ∆∨P := α∨ : α ∈ ∆P . It forms an R-basis of aGP . Let b

∆P := $Gα : α ∈ ∆P
 

P dual to ∆P , and let ∆P be the R basis of aP dual to ∆P .

∨
be the R-basis of aG∗ c∨ G

Example 6.1. There exists a canonical inner product (·, ·) on aG0 . Since the map
0 . Under this identification, for α ∈ Φ0 ,
a 7→ (a, ·) defines an isomorphism aG0 ' aG∗
2 4
α =
∨
α and (α , α ) =
∨ ∨
. In addition, for α ∈ ∆0 , the corresponding
(α, α) (α, α)
fundamental dominant weight $Gα is characterized by the following properties.

(α, α)
(a) $Gα , $Gα = .
4 !
2 2  G 
(b) For all β ∈ ∆0 , δαβ = $Gα , β = $ ,β .
(β, β) (β, β) α

More generally, in stead of (G, P), the above constructions also admit a relative
version for each pair (P, Q) of standard parabolic subgroups of G satisfying P ⊆ Q,
which are compatible with the natural decomposition aP = aQ P ⊕ aQ .
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 165

Multiple L-Functions 165

+
(8) Let ΦQP = ΦP∩MQ , resp. ΦP = ΦP∩MQ , resp. ∆P = ∆P∩MQ , be the set of
Q+ Q

+
α ∈ ΦP , resp. α ∈ ΦP , resp. α ∈ ∆P , which occurs in the Lie algebra mQ of
the Levi factor MQ of Q.1

Note that ∆Q P may also be understood as the subset of the elements α ∈ ∆P which
appear in the action of T P on the unipotent radical of P ∩ MQ . Indeed, MQ ∩ P is
a parabolic subgroup of MQ with the nilpotent radical NPQ := NP ∩ MQ . Thus ∆Q P
is simply the set of roots of the parabolic subgroup (MQ ∩ P, T P ).
As the notation suggested, ∆Q Q∗
P forms an R-basis of aP . Furthermore, being an
element of ∆P , the element α ∈ ∆P admits a companion α∨ , the projection of the
Q

coroot β∨ for a unique simple root β ∈ ∆0 satisfying β aG = α. By an abuse of

P

notation, we use the same notation α∨ to denote the projection of β∨ to aQ

P.
∨ Q∨
(9) Let ∆QPn := α : αo ∈ ∆P . Then ∆P forms an R-basis of aP . Denote by
Q Q
 ∨
Q∨
∆Q
b
P := $α : α ∈ ∆P the dual R-basis of aP of ∆P .
Q Q Q∗

It is well known that aP is the subspace of aQ annihilated by ∆PQ , and that

the map P 7→ ∆PQ gives a natural bijection between the parabolic subgroups P
satisfying P ⊇ Q and the subsets of ∆Q .

6.1.3 Example with Special Linear Group

For G = SLn (R), let P0 = B be the Borel subgroup of G consisting of all the
upper triangular matrices, and let M0 be the Levi component consisting of the
diagonal matrices. Easily, the unipotent radical N0 of P0 consists of the upper
triangular matrices whose diagonal entries are all 1. Let A0+ be the ‘positive’ part
of the split torus T 0 consisting of the diagonal matrices diag a1 , a2 , . . . , an , where


a1 , a2 , . . . , an ∈ R>0 and i=1 ai = 1. Then,
Qn
 n

 X 
a0 =  , , . . . , n
=
 
H := (H H H ) ∈ : H 0 (6.11)
 
 1 2 n R i 

 
i=1
and the logarithmic map in (6.35) is given by
g 7→ HP0 a(g) = log a1 + log a2 + . . . + log an ,


HP0 : P0 → a0 (6.12)
where, for an element g ∈ G, g = n · m · a(g) · k denotes
 its Iwasawa decomposition
with n ∈ N0 , m ∈ M01 , a(g) = diag a1 , a2 , . . . , an ∈ A0+ and k ∈ K = SO(n).
1 To understand (8) and (9), please refer to the following canonical maps A ←- A ,→ M ,→ P =
0 P P
MP NP ,→ Q = MQ NQ ←- MQ . Elements of ∆P are given as the restriction of simple roots of α’s
on A0 to T P . Or equivalently, they are non-trivial characters of T P appearing in nP . Since there are
naturally morphisms nP ,→ p ,→ q ←- mQ , hence nP ∩ mQ makes sense in q, and so does ∆Q P . Here,
as to be easily guessed, p is the Lie algebra of P, etc...
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 166

166 Eisenstein Periods and Multiple L-Functions

Write ei ∈ a∗0 for the element satisfying ei (H) = Hi . The set of the associated
simple roots becomes simply
∆0 = e1 − e2 , e2 − e3 , . . . , en−1 − en .

(6.13)
Sometimes, we identify ∆0 with the index set ∆ := {1, 2, . . . , n − 1} via the assign-
ment αi := ei − ei+1 7→ i (i ∈ ∆).
There is a well-known bijection between the ordered partitions of n and the
standard parabolic subgroups of G. More precisely, for an ordered partition
I := n1 , n2 , . . . , nk of n (so that n = n1 + n2 + . . . + nk and ni ∈ Z>0 ), the as-


sociated standard parabolic subgroup PI , or the same, Pn1 ,n2 ,...,nk , is defined to be
the one consisting of the blocked upper triangular matrices whose diagonal entries
are matrices A11 , A22 , . . . , Akk of sizes n1 , n2 , . . . , nk , respectively. Obviously, the
unipotent radical NI of PI is given by the blocked upper triangular matrices whose
diagonal blocks are the identity matrices of sizes n1 , n2 , . . . , nk . Namely, the ma-
trices of the type
 A11 ∗ . . . ∗ ∗   In1 ∗ . . . ∗ ∗ 
   

 0 A22 . . . ∗ ∗   0 In2 . . . ∗ ∗ 

   
. . .  for PI and . . .  for NI ,
 0 0 .. .. ..   0 0 .. .. .. 
 

0 0 . . . 0 Akk 0 0 . . . 0 Ink
   

where Ini denotes the identity matrix of size ni . Similarly, the (standard) Levi
component MI of PI consists of the blocked diagonal matrices whose diagonal
blocks are given by the block matrices of sizes n1 , n2 , . . . , nk , and the correspond-
ing AI+ consists of the matrices in the form diag a1 In1, a2 In2 , · · ·, ak Ink . Namely,


the matrices of the type
 A11 0 . . . 0 0   a1 In1 0 . . . 0 0 
   

 0 A22 . . . 0 0   0 a2 In2 . . . 0 0 

   
.  for MI and  ..  for AI+ ,
 0 0 .. 0 0 

 0 0 . 0 0 
0 0 . . . 0 Akk 0 . . . 0 ak Ink
   
0
Yk
where ai > 0 for i = 1, 2, . . . , k, and ani = 1.
i=1 i
Accordingly, if we write aI for the space aPI associated to PI , then aI is simple
the subspace of a0 defined by
H1 = H2 = . . . = HN1
 


 


 
HN1 +1 = HN1 +2 = . . . = HN2 


 

aI =  , , . . . ,
 
(H H H ) ∈ a (6.14)
 
1 2 k 0
.........,

 




 


 
=H = ... = H 
 
 H
Nk−1 +1 Nk−1 +2 Nk
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 167

Multiple L-Functions 167

where we denote Ni = n1 +n2 +. . .+ni for i = 1, 2, . . . , k. Moreover, by the Iwasawa

decomposition, the associated logarithmic map HI : AI+ → aI is characterized by
 
diag a1 In1 , a2 In2 , . . . , ak Ink 7→ (log a1 )(n1 ) , (log a2 )(n2 ) , . . . , (log ak )(nk ) . (6.15)

Therefore, the bijection between the parabolic subgroups of G and the subsets
of ∆0 , or better of ∆, is given explicitly as follows: The parabolic subgroup P = PI
corresponds to the subset I(P) of ∆ such that ∆ − I(P) = N1 , N2 , . . . , Nk . That is,

n oGn o
I(P) = 1, 2, . . . , N1 − 1 N1 + 1, N1 + 2, . . . , N2 − 1
G Gn o
··· Nk−1 + 1, Nk−1 + 2, . . . , Nk − 1 .
For simplicity, from now on, we will use I to indicate both the partition and its
corresponding subset I(PI ). For example, the subset I = ∅ corresponds to the
partition (1, 1, . . . , 1) of n = 1 + 1 + · · · + 1 (since ∆ − ∅ = {1, 2, . . . , n − 1}), and the
corresponding parabolic subgroup is simply P0 the minimal parabolic subgroup.
Let J be the set associated to the partition n = f1 + f2 + . . . + fm , where
f1 = n1 +n2 +. . .+nk1 , f2 = nk1 +1 +nk1 +2 +. . .+nk2 , . . . , fm = nkm−1 +1 +nkm−1 +2 +. . .+nkm .
Then, ∅ ⊆ I ⊆ J, and from ∆ − J ⊆ ∆ − I ⊆ ∆, we get P0 = P∅ ⊆ PI ⊆ P J . Then
a J is the subspace of a0 defined by
n  o
a J := H1( f1 ) , H2( f2 ) , . . . , Hm( fm ) ∈ a0 . (6.16)
Clearly a J is in aI since aI is a subspace of a0 defined by
n  o
aI := A(n 1 , A2 , . . . , An
1) (n2 ) (nk )
∈ a0 (6.17)
and we have the map
aJ → aI
   (nk ) (n )
 (6.18)
H1( f1 ) , H2( f2 ) , . . . , Hm( fm ) 7→ H1 , H1 , . . . , H1 1 , . . . , Hm km .
(n1 ) (n2 )

Moreover, by the Iwasawa decomposition, the logarithmic map H J : A J+ → a J

becomes simply the assignment
   
diag b1 I f1 , b2 I f2 , . . . , bm I fm 7→ (log b1 )( f1 ) , (log b2 )( f2 ) , . . . , (log bm )( fm ) . (6.19)

For each i ∈ ∆, the corresponding fundamental weight is given by

i n−1
n−iX i X
$i := $∆i = ` · α` + (n − `) · α` . (6.20)
n `=1 n `=i+1
As a Q-linear form on a∅ = a0 ,
i 
$i (H) = H1 + H2 + · · · + Hi − H1 + H2 + · · · + Hn = H1 + H2 + · · · + Hi
n
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 168

168 Eisenstein Periods and Multiple L-Functions

 
where H = H1 , H2 , . . . , Hn ∈ a0 . This implies that b
∆I = $i : i ∈ ∆ − I , since

n o
∆I = αi a : i ∈ ∆ − I . (6.21)
I

Note that = h ∆I iR and the elements of b

a∗I ∆I are non-degenerate positive
linear combination of elements in ∆I . Hence, both ∆I and b∆I generate a∗I , and
D E D E
a∗I = $i : i ∈ ∆ − I ⊇ $ j : j ∈ ∆ − J = a∗J . (6.22)
R R
In addition, the space aIJ is the subspace of aI annihilated by a∗J . So
n o n o
aIJ = H ∈ aI : α(H) = 0 ∀α ∈ a∗J = H ∈ aI : $ j (H) = 0 ∀ j ∈ ∆ − J
n1 H1 + n2 H2 + . . . + nk1 Hk1 = 0,
 
 
=
 (n1 ) (n2 )
, , . . . , (nk )
.
 
(H H H ) ∈ a

0
 1 2 k
nk1 +1 Hk1 +1 + nk1 +2 Hk1 +2 + . . . + nk2 Hk2 = 0, . . .
 

Since dim aIJ + dim a J = dim aI and aIJ ∩ a J = ∅, there is a natural decomposition
aI = aIJ ⊕ a J . (6.23)
Furthermore, we have
[k n o n o
I= N j−1 + 1, N j−1 + 2, . . . , N j − 1 = 1, 2, . . . , n − 1 − N1 , N2 , . . . , Nk−1 .

i=1
Motivated by (6.20), set, for 1 ≤ ` ≤ n j − 1,
` N j −1
Nj − ` X ` X 
$IN j−1 +` = l · αN j−1 +l + N j − l · αN j−1 +l . (6.24)
n j l=1 d j l=`+1
And, as usual, we define the coroot α∨i ∈ a0 by
` = i,




 1
(αi )` = 
∨
−1 ` = i + 1,

(6.25)




0

otherwise.
Then αi : i ∈ I is a basis of the subspace a∅I of a∅ and $i (α∨j ) = δi j for all i, j ∈ I.
 ∨
Consequently, $iI |a∅I : i ∈ I is its dual basis in (a∅I )∗ . In particular, we also get

n o
∆IJ = $kJ : k ∈ J − I = (∆ − I) − (∆ − J) .
b (6.26)
 ∗  ∗  ∗
To end this example, we mention that, within a∅J = a∅I ⊕ aIJ , the space
 ∗
a J is generated by ∆∅J = αl : l ∈ J − ∅ = J ⊆ ∆∅ , the root system. Similarly,

 ∅ ∗  ∗
a∅I is generated by ∆∅I = α` : ` ∈ I − ∅ = I ⊆ ∆∅ and aIJ is generated by


∆IJ which is not a subset of ∆∅ in general, even ∆IJ ⊂ ∆I consists of the restriction
of α` to aIJ with ` ∈ J − I = (∆ − I) − (∆ − J). This confirms the definition that
n o
∆IJ = α` aJ : ` ∈ J − I = (∆ − I) − (∆ − J) . In particular, if J = ∆, we recover the
I
relations a∆∅ = a∅ , a∆I = aI and ∆∆∅ = ∆∅ , ∆∆I = ∆I .
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 169

Multiple L-Functions 169

6.1.4 Logarithmic Map

Let F be a number field with A = AF its associated adelic ring. For a fixed
connected reductive group G defined over F, denote by ZG the center of G. Let
P0 be a minimal parabolic subgroup of G and denote a Levi decomposition of P0
by P0 = M0 N0 , where, as usual, we fix once and for all the Levi M0 , and N0 is the
unipotent radical of P0 . Recall that a parabolic subgroup P is G is called standard
if P ⊇ P0 , and that for such parabolic subgroups, there exists one and only one
Levi decomposition P = MN satisfying the condition that M ⊇ M0 and N is the
unipotent radical of P. We call M the standard Levi subgroup of P. Denote by
Rat(M) the group of rational characters of M, i.e. the morphism M → Gm , where
Gm denotes the multiplicative group. Obviously, Rat(M) = X ∗ (T P ). Set
a∗M := Rat(M) ⊗Z R = a∗P , a M := HomZ (Rat(M), R) = aP ,
(6.27)
a∗M,C := Rat(M) ⊗Z C, a M,C := HomZ (Rat(M), C).
For convenience, sometimes, we also write

a∗M := <a∗M,C and a M := <a M,C . (6.28)

For any χ ∈ Rat(M), we obtain a (real) character defined by

|mv |χv v
Y
|χ| : M(A) → R∗ m = (mv ) 7→ m|χ| := (6.29)
v∈S

where | · |v denotes the v-absolute values. Set

M(A)1 := ∩χ∈Rat(M) Ker|χ|. (6.30)

Obviously, M(A)1 is a normal subgroup of M(A).

Let X M be the group of complex characters which are trivial on M(A)1 . Define
a logarithmic morphism H M := log M : M(A) → a M by

hχ, log M (m)i := log(m|χ| ) ∀χ ∈ Rat(M) ⊂ a∗M . (6.31)

Clearly,

M(A)1 = Ker(log M ) and log M (M(A)/M(A)1 ) ' a M . (6.32)

Hence, in particular, there is a natural isomorphism κ : a∗M,C ' X M . Set

√
<X M := κ(a∗M ) and =X M := κ( −1 · a∗M ). (6.33)

In addition, we introduce a working space, a subgroup XGM ⊂ X M consisting of

complex characters of M(A)/M(A)1 which are trivial on ZG(A) .
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 170

170 Eisenstein Periods and Multiple L-Functions

More generally, let χi : i = 1, . . . , d(P) be a Z-bases of the character lattice


Xd(P)
X ∗ (MP ) of MP . Then, every χ = si χi (si ∈ C) in a∗P,C := a∗P ⊗ C determines a
i=1
morphism
Yd(P) si
P(A) → C∗ , p 7→ pχ := χi (p) . (6.34)
i=1
Consequently, there is a natural logarithmic map
HP : P(A) → aP characterized by h HP (p), χ i = pχ (∀χ ∈ a∗P ). (6.35)
1 1 1
We denote the kernel of HP by P(A) and set MP (A) := MP (A) ∩ P(A) .
Let also T+P be the set of t = (tv ; tσ ) ∈ T P (A) such that

(1) tv = 1 for all finite places v of F; and

(2) χ(tσ ) ∈ R>0 is independent of infinite places σ of F for all χ ∈ X ∗ (MP ).

Obviously, M(A) = T+P · M(A)1 .

6.1.5 Compatible Haar Measures

There is always a maximal compact subgroup K of G(A) such that, for all standard
parabolic subgroups P = MN of G,
   
P(A) ∩ K = M(A) ∩ K · N(A) ∩ K . (6.36)
In the sequel, fix such a maximal compact subgroup K of G(A). Accordingly, we
obtain the so-called Langlands decomposition
G(A) = M(A) · N(A) · K, (6.37)
and hence a natural map
mP : G(A) → M(A)/M(A)1 g = m · n · k 7→ M(A)1 · m (6.38)
where g ∈ G(A), m ∈ M(A), n ∈ N(A) and k ∈ K.
Moreover, we may hence will choose some Haar measures on M0 (A), N0 (A),
K such that

(a) The induced measure on M(F) is the counting measure and the volume of
the induced measure on M(F)\M(A)1 is 1. (Recall that, from the classical
reduction theory, M(F)\M(A)1 is of finite volume.)
(b) The induced measure on N0 (F) is the counting measure and the volume of
N(F)\N0 (A) is 1. (Recall that being unipotent radical, N(F)\N0 (A) is com-
pact.)
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 171

Multiple L-Functions 171

(c) The volume of K is 1.

Accordingly, There are induced Haar measures via log M on the spaces
a M0 , a∗M0 , etc... Furthermore, if we denote by ρ0 the half of the sum of the positive
roots of the maximal split torus T 0 (in the central Z M0 ) of M0 ,
Z
f 7→ f (mnk) dk dn m−2ρ0 dm (6.39)
M0 (A)·N0 (A)·K
defined for continuous functions with compact supports on G(A) induces a Haar
measure dg on G(A). This in turn yields induced measures on M(A), N(A) and
hence on a M , a∗M , P(A), etc..., for all parabolic subgroups P. It is well known that
the following compatibility condition holds
Z Z
f (mnk) dk dn m−2ρ0 dm = f (mnk) dk dn m−2ρP dm
M0 (A)·N0 (A)·K M(A)·N(A)·K
for all continuous functions K with compact supports on G(A), where ρP denotes
one half of the sum of all positive roots of the maximal split torus T P of the central
Z M of M.

6.1.6 Classical Reductive Theory

Fix an isomorphism T 0 ' GRm . Embed R∗+ ,→ A = Afin × A∞ by the assignment

t 7→ (1; t). Then we obtain a natural injection (R∗+ )R ,→ T 0 (A) which splits.
Denote by A M0 (A) the unique connected subgroup of T 0 (A) which projects onto
(R∗+ )R . More generally, for a standard parabolic subgroup P = MN, set, with Z∗
the center of the group ∗,
A M(A) := A M0 (A) ∩ Z M(A) . (6.40)
Clearly, A M(A) = T+P and hence M(A) = A M(A) · M(A)1 . For later use, set also
AGM(A) := a ∈ A M(A) : logG a = 0 .

(6.41)
Then A M(A) = AG(A) ⊕ AGM(A) .
Note that K and N(F)\N(A) are all compact, and M(F)\M(A)1 is of finite
volume. With the Langlands decomposition G(A) = N(A) M(A) K in mind, the
reduction theory for G(F)\G(A), or more generally, for P(F)\G(A), is reduced to
that for A M(A) since ZG (F) ∩ ZG(A) \ZG(A) ∩ G(A)1 is compact as well.2 As such,
for t0 ∈ M0 (A), set
A M0 (A) (t0 ) := a ∈ A M0 (A) : aα > t0α ∀α ∈ ∆0 .
n o
(6.42)
2 One of the main applications of the reduction theory is to control the integrations of moderate growth

functions, which are often divergent. For this, since the space M(F)\M(A)1 is of finite volume, what
should be considered is over the affine part A.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 172

172 Eisenstein Periods and Multiple L-Functions

Then, for a fixed compact subset ω ⊂ P0 (A), we can form an associated Siegel set
n o
S (ω; t0 ) := p · a · k : p ∈ ω, a ∈ A M0 (A) (t0 ), k ∈ K . (6.43)

In particular, for a sufficiently big ω and a sufficiently small t0 ∈ M0 (A) (in the
sense that t0α is very close to 0 for all α ∈ ∆0 ), the classical reduction theory
recalled in Chapter 4 may be generalized as

G(A) = G(F) · S (ω; t0 ). (6.44)

More generally, set

APM0 (A) (t0 ) := a ∈ A M0 (A) : aα > t0α ∀α ∈ ∆0P ,

n o

n o (6.45)
S P (ω; t0 ) := p · a · k : p ∈ ω, a ∈ APM0 (A) (t0 ), k ∈ K .

Similarly as above, for a sufficiently big ω and a sufficiently small t0 ∈ M0 (A),

G(A) = P(F) · S P (ω; t0 ). (6.46)

Here, as in §6.1.2, ∆0P denotes the set of positive roots for (P0 ∩ M, T 0 ).

6.2 Multiple L-Functions

6.2.1 Growth Condition

Let F be a number field with A its adelic ring, and let G be a connected reductive
group over F. Denote by G(A) the associated adelic group of G. Since every
algebraic reductive group is a linear algebraic group, we fix, once and for all, an
embedding iG : G ,→ SLn which sends g to (gi j ). Accordingly, introduce a height
function on G(A) by setting
Y
sup |gi j |v ∀i, j .

kgk := (6.47)
v∈S

It is well known that, up to O(1), these height functions are essentially unique.
Because of this uniqueness, all the growth conditions introduced in the sequel are
well-defined, i.e. they do not depend on the height function we use.
By definition, a function f : G(A) → C is said to have moderate growth if
there exist some constants c, r ∈ R such that

| f (g)| ≤ c · kgkr ∀g ∈ G(A). (6.48)

Similarly, for a standard parabolic subgroup P of G with the Levi decomposition

P = MN, a function f : N(A)M(F)\G(A) → C is said to have moderate growth
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 173

Multiple L-Functions 173

if there exist some constants c, r ∈ R, and a certain λ ∈ <X M0 such that, for all
a ∈ A M(A) and k ∈ K,
f (amk) ≤ c · kakr · mP0 (m)λ ∀m ∈ M(A)1 ∩ S P (ω; t0 ). (6.49)
In addition, a complex valued function f : G(A) → C is said to be smooth if, for
every g = gfin · g∞ ∈ G(Afin ) × G(A∞ ), there exist an open neighborhood Vfin , resp.
V∞ , of gfin , resp. g∞ , in G(A) and a C ∞ -function f1 : V∞ → C such that
f (g0fin · g0∞ ) = f1 (g0∞ ) ∀g0fin ∈ Vfin , g0∞ ∈ V∞ . (6.50)
By contrast, a function f : S (ω; t0 ) → C is said to be rapidly decreasing if
there exists a constant r > 0 and, for all λ ∈ <X M0 , there exists a constant c > 0
such that
|φ(ag)| ≤ c · kak · mP0 (g)λ ∀a ∈ A M(A) , g ∈ G(A)1 ∩ S (ω; t0 ). (6.51)
More generally, a function f : G(F)\G(A) → C is said to be rapidly decreasing if
the restriction f S (ω;t ) of f to a Siegel set S (ω; t0 ) in (6.44) is rapidly decreasing.
0

6.2.2 Automorphic Forms

Definition 6.1. A function φ : N(A)M(F)\G(A) → C is called automorphic if

(1) φ is of moderate growth, i.e. φ satisfies (6.49).

(2) φ is smooth, i.e. φ satisfies (6.50).
(3) φ is K-finite, i.e. the C-span of all φ(k1 · ∗ · k2 ) parametrized by (k1 , k2 ) ∈ K × K
is finite dimensional.
(4) φ is z-finite, i.e. the C-span of all δ(X)φ parametrized by all X ∈ z is finite
dimensional. Here z denotes the center of the universal enveloping algebra
U := U LieG(A∞ ) of the Lie algebra of G(A∞ ) and δ(X) denotes the deriva-


tive of φ along X.

For an automorphic function φ, set

φk : M(F)\M(A) → C m 7→ m−ρP φ(mk), ∀k ∈ K. (6.52)
It is well known that φk is an automorphic form on M. Introduce the space

n o
A N(A)M(F)\G(A) := automorphic forms on N(A)M(F)\G(A) . (6.53)

Definition 6.2.

(1) For a measurable locally L1 -function f : N(F)\G(A) → C, its constant term

along with the standard parabolic subgroup P = N M is defined by
Z
fP : N(A)\G(A) → C, g 7→ f (ng)dn. (6.54)
N(F)\G(A)
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 174

174 Eisenstein Periods and Multiple L-Functions

(2) An automorphic form φ ∈ A(N(A)M(F)\G(A)) is called a cusp form if φQ ≡ 0

for any standard parabolic subgroup Q properly contained in P.

Introduce the space of cusp forms by

n o
A N(A)M(F)\G(A) 0 := cusp forms on N(A)M(F)\G(A) .


(6.55)
Easily,

(a) All cusp forms are rapidly decreasing, and hence

(b) There is a natural pairing
h·, ·i : A0 (N(A)M(F)\G(A)) × A(N(A)M(F)\G(A)) −→ C (6.56)
defined by
Z
hψ, φi := ψ(g) φ(g) dg. (6.57)
Z M(A) N(A)M(F)\G(A)

To go further, for a (complex) character ξ : Z M(A) → C∗ of Z M(A) , within the

space of automorphic forms, set
φ(zg) = zρP · ξ(z) · φ(g) 
 

φ ∈ A(N(A)M(F)\G(A)) .
 
A(N(A)M(F)\G(A))ξ := 
 
∀z ∈ Z M(A) , g ∈ G(A) 

 


Similarly, within the spaces of cusp forms, set

A0 (N(A)M(F)\G(A))ξ := A0 (N(A)M(F)\G(A)) ∩ A(N(A)M(F)\G(A))ξ .
In addition, based on these ξ-pieces, introduce the total spaces
X
A(N(A)M(F)\G(A))Z := A(N(A)M(F)\G(A))ξ ,
ξ∈Hom(Z M(A) ,C∗ )
X (6.58)
A0 (N(A)M(F)\G(A))Z := A0 (N(A)M(F)\G(A))ξ .
ξ∈Hom(Z M(A) ,C∗ )

From the special structure of A M(A) -finite functions and the Fourier analysis over
the compact space A M(A) \Z M(A) , we obtain a natural isomorphism
'
C[a M ] ⊗ A(N(A)M(F)\G(A))Z −→ A(N(A)M(F)\G(A))
(6.59)
(Q, φ) 7→ g 7→ Q(log M (mP (g)) · φ(g)

and an isomorphism
'
C[a M ] ⊗ A0 (N(A)M(F)\G(A))Z −→ A0 (N(A)M(F)\G(A))ξ . (6.60)
(See below for a more concrete and useful form.) Accordingly, let Π0 (M(A))ξ be
isomorphism classes of irreducible representations of M(A) occurring in the space
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 175

Multiple L-Functions 175

A0 (M(F)\M(A))ξ . (More precisely, we should use the product group M(A f ) ×

(M(A) ∩ K, Lie(M(A∞ ))C )) instead of M(A).) Set accordingly
[
Π0 (M(A) := Π0 (M(A))ξ . (6.61)
ξ∈Hom(Z M(A) ,C∗ )

For a π ∈ Π0 (M(A))ξ , let

A0 (M(F)\M(A))π := isotypic component of type π of A0 (M(F)\M(A))ξ , (6.62)
i.e. the subspace of cusp forms of M(A) generating a semi-simple isotypic
M(A f ) × (M(A) ∩ K, Lie(M(A∞ ))C ))-module of type π. Accordingly, set
A0 (N(A)M(F)\G(A))π
n o (6.63)
:= φ ∈ A0 (N(A)M(F)\G(A)) : φk ∈ A0 (M(F)\M(A))π ∀k ∈ K .
Then
M
A0 (N(A)M(F)\G(A))ξ = A0 (N(A)M(F)\G(A))π . (6.64)
π∈Π0 (M(A))ξ

More generally, let V be an irreducible M(A f ) × (M(A) ∩ K, Lie(M(A∞ ))C ))-

module in A(M(F)\M(A)) and let π0 be the induced representation, which we call
an automorphic representation of M(A). Similar to (6.62) and (6.63), set
A(M(F)\M(A))π0 := isotypic subquotient of type π0 of A(M(F)\M(A)),
A(N(A)M(F)\G(A))π0 (6.65)
n o
:= φ ∈ A(N(A)M(F)\G(A)) : φk ∈ A(M(F)\M(A))π0 ∀k ∈ K .
Then, there is an isomorphism
'
V ⊗ Hom M(A f )×(M(A)∩K,Lie(M(A∞ ))C )) (V, A(M(F)\M(A))) −→ A(M(F)\M(A))π0 .
Naturally, if A(M(F)\M(A))π0 ⊂ A0 (M(F)\M(A)), we say that π0 is a cuspidal
representation.
Two automorphic representations π and π0 of M(A) are said to be equiva-
lent if there exists λ ∈ XGM such that π ' π0 ⊗ λ. This, in practice, means that
A(M(F)\M(A))π = λ · A(M(F)\M(A))π0 . That is, for any φπ ∈ A(M(F)\M(A))π ,
there exists a φπ0 ∈ A(M(F)\M(A))π0 such that φπ (m) = mλ ·φπ0 (m). Consequently,
A(N(A)M(F)\G(A))π = (λ ◦ mP ) · A(N(A)M(F)\G(A))π0 . (6.66)
Denote by P := [π0 ] the equivalence class of π0 , and we say (M, P) is a cuspidal
datum of G if π0 is cuspidal. Since P is an XGM - principal homogeneous space, and
it admits a natural complex structure. Accordingly, for each π ∈ P, we set <π :=
<χπ = |χπ | ∈ <X M , where χπ is the central character of π, and =π := π ⊗ (−<π).
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 176

176 Eisenstein Periods and Multiple L-Functions

6.2.3 Eisenstein Series and Their Constant Terms

Definition 6.3. Let π be an irreducible automorphic representation of M(A) and

let φ be an automorphic form in A(N(A)M(F)\G(A))π . The Eisenstein series
E(φ, π) associated to φ and π is defined as a function E(φ, π) : G(F)\G(A) → C
characterized by
X X
E(φ, π)(g) := φ(δg) = mP (δg)λ+ρ ψ(δg) := E(ψ; λ; g), (6.67)
δ∈P(F)\G(F) δ∈P(F)\G(F)

where ψ ∈ A(N(A)M(F)\G(A))π0 is induced from φ via (6.66).

It is well known that, see e.g. [75], we have the following:

Theorem 6.1. Let ψ ∈ A(N(A)M(F)\G(A))π0 be the automorphic form.

(1) There is an open cone C ⊂ <XGM such that, if <π ∈ C, the Eisenstein series
E(λ · ψ, π0 ⊗ λ)(g), namely E(ψ; λ; g), converges uniformly for g in a compact
subset of G(A) and λ in an open neighborhood of 0 in XGM . In particular,
nif P = [π] is cuspidal, we may eveno take C to be the positive Weyl cone
λ ∈ <XGM : hλ − ρP , α∨ i > 0, ∀α ∈ ∆GP .
(2) The Eisenstein series E(ψ; λ; g) is an automorphic forms on G(F)\G(A).
 
For an automorphic form ψ ∈ A N(A)M(F)\G(A) of P-level, we may write
the associated Eisenstein series as a convergence series
X
E ψ, λ, g := mP (δg)λ+ρ · φ(δg)


(6.68)
δ∈P(F)\G(F)

for λ ∈ a∗P,C
satisfying <λ belongs to a certain positive cone CP .
Set P = [π]. For the elements w ∈ W, the associated Weyl group of (G, P0 ),
we fix once and for all some representatives in G(F) of w, which we denote by w
as well, with an abuse of notation. Set M 0 := wMw−1 and denote by P0 = N 0 M 0
the associated parabolic subgroup. Then W acts naturally on the automorphic
representations, from which we obtain an equivalence classes wP of automorphic
representations of M 0 (A).

Definition 6.4. The associated intertwining operator M(w, π) is defined by

Z
M(w, π)φ (g) := φ(w−1 n0 g)dn0


∀g ∈ G(A). (6.69)
N 0 (F)∩wN(F)w−1 \N 0 (A)

Proposition 6.1. (§§ II.1.6-7 of [75]) Assume that h<π, α∨ i  0 for all α ∈ ∆GP .
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 177

Multiple L-Functions 177

(1) For an automorphic φ, M(w, π)φ depends only on double coset M 0 (F)wM(F).
In particular, M(w, π)φ is well-defined for w ∈ W.
(2) The integral (6.69) converges absolutely and uniformly for g varying in a
compact subset of G(A).
(3) M(w, π)φ ∈ A(N 0 (A)M 0 (F)\G(A))wπ . And if φ is L2 , which from now on we
always assume, so is M(w, π)φ.
(4) Assume that w = wαn · · · wα1 satisfying the condition that, for wi := wαi · · · wα1
and Mi = wi Mw−1 i , we have αi+1 is a simple root of ∆G , w−1 (αi+1 ) > 0 and
 −1 P i
the set of positive multiple roots of an element of wi (αi+1 ) : i = 0, . . . , n − 1
coincides with α ∈ ∆GP : α > 0, wα < 0 . Then


M(w, π0 ) = M(wαn , wn−1 π0 )M(wαn−1 , wn−2 π0 ) · · · M(wα1 , π0 ). (6.70)

(5) The constant term of E φ, λ, g with respect to a parabolic subgroup P0 =

M 0 N 0 is given by
X
E M /M M(w, λ)φ, wλ (g)
0
 
E P0 (φ, λ, g) := (6.71)
w∈W(M,M 0 )

where

w−1 (α) > 0 ∀α ∈ T + (T 0 , M 0 ) 

 

0
,
 
W(M, M ) :=  w∈W (6.72)
 
wMw ,→ M standard Levi 
−1 0

 


and E M /M denotes the induced relative Eisenstein series from M to M 0 .

0

For instance, if M 0 is conjugate to M, we have

X  
E P0 (φ, λ, g) := M(w, λ) φ (g). (6.73)
w∈W(M)

Example 6.2. Let G = SL3 and take φ to be the constant function 1 on M0 . Being
cuspidal, from above, the associated Eisenstein series E(1, λ, g) and its constant
term along with P0 are given by

E(1, λ, g) =
X
mP0 (δg)λ+ρ0 ∀< λ ∈ C0 ,
δ∈P0 (F)\S L3 (F)
(6.74)
E P0 (1, λ, g) =
X
M(w, λ) · mP0 (g)w(λ+ρ0 ) .
w∈W(M0 )

Moreover, since P0 is minimal, its Levi M0 is maximal. Thus, for P0 satisfying

M 0 ' M0 , wM0 w−1 6,→ M 0 , and hence E P0 (1, λ, g) ≡ 0.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 178

178 Eisenstein Periods and Multiple L-Functions

6.2.4 Stability Truncation on Adelic Space

Denote by S fin , resp. S ∞ , be the collection of (normalized, inequivalent) non-

Archimedean places, resp. Archimedean places, of F. For each v ∈ S := S fin tS ∞ ,
denote by Fv the v completion of F.Set also r1 , resp. r2 , be the number of real
places, resp. complex places, in S ∞ .
Let X = Spec OF be the associated arithmetic curve. For a vector bundle E of
rank n over X, i.e. a locally free OF -sheaf of rank n, denote by E F the fiber of E
at the generic point Spec F ,→ Spec OF of X. Then E F is an F-vector space of
dimension n. For each v ∈ S fin , set EOv := H 0 (Spec OFv , E) a free Ov -module of
rank n, where Ov denotes the integer ring of the non-Archimedean local field Fv .
In particular, we have a canonical isomorphism canv : Fv ⊗Ov EOv ' Fv ⊗F E F .
For a basis αF : F n ' E F of the generic fiber of E and a basis αOv : Onv ' EOv
(v ∈ S fin ), the elements gv := (Fv ⊗F αF )−1 ◦ canv ◦ (Fv ⊗Ov αOv ) ∈ GLn (Fv ) define
an element gA := (gv )v∈S fin of GLn (Afin ), since, for almost all v, gv ∈ GLn (Ov ).
By this construction, we obtain a bijection Φ from the set of isomorphism classes
of triples (E; αF ; (αOv )v∈S fin ) as above onto GLn (Afin ). Moreover, for γ ∈ GLn (F),
κ = (κv ) ∈ v∈S fin GLn (Ov ), if this bijection maps the triple (E; αF ; (αOv )v∈S fin ) onto
Q
gA , it then maps the triple (E; αF ◦ γ−1 ; (αOv ◦ κv )v∈S fin ) onto γgA κ. Therefore, we
have the following well-known:

Lemma 6.1. The morphism Φ induces a bijection between the set of isomorphism
classes of vector bundles of rank n on the curve Spec OF and the double coset
 Q
space GLn (F) GLn (Afin ) v∈S fin GLn (Ov ).

More generally, let n = n1 + · · · + nk be an ordered partition I = (n1 , . . . , nk )

of n. There is a natural bijection ΦP from the set of isomorphism classes of triple
(E∗ ; α∗,F : (α∗,Ov )v∈S fin ) onto PI (Afin ), where E∗ := (0) = E0 ⊂ E1 ⊂ · · · ⊂ Ek is a


filtration of vector bundles of rank (n1 , n1 + n2 , . . . , n1 + n2 + · · · + nk = n) over X,
i.e. each E j is a vector bundle of rank n1 + n2 + · · · + n j over X and each quotient
E j /E j−1 is torsion free, which is compatible with an isomorphism of filtrations of
F-vector spaces
α∗,F : (0) = F0 ⊂ F n1 ⊂ · · · ⊂ F n1 +n2 +···+nk =n ' (E∗ )F ,


(6.75)
and with an isomorphism of filtrations of free Ov -modules
α∗,Ov : (0) ⊂ Onv 1 ⊂ · · · ⊂ Onv 1 +n2 +···+nk =n ' (E∗ )Ov ,


(6.76)
for every v ∈ S fin . This then proves the following:

Lemma 6.2. Let I = (n1 , . . . , nk ) be an ordered partition of n and let PI be the

corresponding standard parabolic subgroup of GLn .
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 179

Multiple L-Functions 179

(1) The morphism ΦP induces a bijection between the set of isomorphism classes
of filtrations of vector bundles of successive ranks (n1 , n1 + n2 , . . . , n) over X
Q
and the double coset space PI (F)\PI (Afin )/ v∈S fin PI (Ov ).
(2) The natural embedding PI (Afin ) ,→ PI (A) admits a modular interpretation
(E∗ ; α∗,F : (α∗,Ov )v∈S fin ) 7→ (Ek ; αk,F : (αk,Ov )v∈S fin ). (6.77)
(3) The canonical projection PI (Afin ) → MI (Afin ) → GLn j (Afin ) for j = 1, . . . , k,
where MI denotes the standard Levi of PI , admits a modular interpretation
(E∗ ; α∗,F : (α∗,Ov )v∈S fin ) 7→ (gr j (E∗ ); gr j (α∗,F ), gr j (α∗,Ov )v∈S fin ), (6.78)
where gr j (E∗ ) := E j /E j−1 , gr j (α∗,F ) : F n j ' gr j (E∗ )F and gr j (α∗,Ov ) : Ov n j '
gr j (E∗ )Ov , v ∈ S fin are induced by α∗,F and α∗,Ov .

Furthermore, associated to g = (g f ; g∞ ) ∈ GLn (Afin ) × GLn (A∞ ) = GLn (A)

is first a rank n vector bundle Eg on Spec OF , which, through the Minkowski
embedding E F ,→ (Rr1 × Cr2 )n , gives a discrete subgroup, a free OF -module of
rank n. In particular, g∞ = (gσ ) induces a natural metric on Eg by twisting the
standard one on (Rr1 × Cr2 )n via the linear transformation induced from g∞ gt∞ :=
(gσ gtσ ). As a direct consequence, see e.g. [67], we have
degar (Eg , ρg ) = − log N(detg)


(6.79)
where N : GL1 (AF ) = IF → R>0 is the norm map of the idelic group IF of F.
Accordingly, for g = (gfin ; g∞ ) ∈ GLn (A) and a parabolic subgroup Q of GLn ,
denote by Eg;Q
∗ the filtration of the vector bundle Egfin induced by the parabolic
subgroup Q. This gives a filtration of hermitian vector bundles (Eg;Q g;Q
∗ , ρ∗ ) with
the hermitian metrics ρg;Q
j on Eg;Qj given by the restrictions of ρg∞ .

Definition 6.5. Let g = (gfin ; g∞ ) ∈ GLn (A) and let Q be a parabolic subgroup of
GLn . We define pgQ : [0, n] → R to be a polygon characterized by the conditions

(a) pgQ (0) = pgQ (n) = 0,

(b) pgQ is affine on the interval rankEg;Q g;Q 
i−1 , rankEi

; and
(c) For all indices i,
degar (Eg , ρg )
pgQ (rankEg;Q g;Q g;Q g;Q
i ) = degar (Ei , ρi ) − rankEi · .
n
Then, by Proposition 1.6, there is a unique convex polygon p̄g , the canonical
polygon associated to g, which bounds all the pgQ ’s from above for all parabolic
g
subgroups Q of GLn . Moreover there exists a parabolic subgroup Q such that
pg g = p̄g .
Q
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 180

180 Eisenstein Periods and Multiple L-Functions

Proposition 6.2. Let K = v∈S fin GLn (Ov )×O(n)r1 ×U(n)r2 . For any fixed polygon
Q
p : [0, n] → R satisfying p(0) = p(n) = 0 and d ∈ R, the subspace

M≤p
F,n (d) := g ∈ GLn (F)\GLn (A) K : degar (g) = d, p̄ ≤ p
g
 

is compact.

Proof. This is a restatement of Theorem 1.3 on the moduli spaces of semi- stable
OK -lattices, based on the classical reduction theory. More precisely, it consists of
two parts, namely, compactness at the lattice level, and compactness of fibers for
the canonical projection Φ.
For the compactness at the lattice level, note that with the degree fixed, by the
Arakelov-Riemann-Roch formula stated at 1.5, the volumes of OF -lattices of rank
n corresponding to the (Eg , ρg )’s for all g ∈ GLn (F)\GLn (A) K are fixed. Hence

the condition pg ≤ p implies that there is an upper bound for the volumes of all
the sublattices of (Eg , ρg ). This further implies that there is a lower bound for the
Minkowski successive minimums for the lattices involved. Therefore, as proved
in Theorem 1.3, at the lattice level, the corresponding space is compact.
On the other hand, fibers of the projection from GLn (F)\GLn (A)/GLn (OF )
to the space of isometry classes of OF -lattices are simply given by the following
product v∈S fin SLn (Ov ) × SOn (R)r1 × SUn (C)r2 , namely K. But it is well-known
Q
that K itself is compact. 

More generally, similarly to what L. Lafforgue does for the function fields
in [66], we have the following:

Proposition 6.3. Let n be a fixed positive integer.

(1) For a fixed parabolic subgroup P of GLn and an element g ∈ GLn (A), there
is a unique maximal element p̄gP among all the pgQ ’s, where Q runs over all
parabolic subgroups of GLn which are contained in P.
(2) For any fixed polygon p : [0, n] → R, a real number d ∈ R, and a standard
parabolic subgroup P of GLn , the subset
n o
MP;≤p
F,n (d) := g ∈ GLn (F)\GL n (A)/K : degar (g) = d, p̄g
P ≤ p, pg
P ≥ −p

is compact.

More generally, for a connected reductive reductive group over F, a similar theory
can be developed. We will leave this to Part 6.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 181

Multiple L-Functions 181

6.2.5 Multiple L-Functions

Eisenstein series are analytic invariants of reductive groups over number fields
and the associated theory for the so-called Eisenstein system in [70], see also [90]
and [75], is one of the highest achievements in mathematics. In this section, we
introduce multiple L-functions as integrations of these Eisenstein series over some
intrinsically defined geometric spaces.
For simplicity, we assume that our reductive group G is the general linear
group GLn . As in §6.1.3, we let P0 be the minimal parabolic subgroup P0 consist-
ing of upper triangular matrices corresponding to the partition (1, . . . , 1). Then its
(standard) Levi component M0 consists of diagonal matrices. Let P = PI = U I MI
be the standard parabolic subgroup associated to the partition I = (n1 , . . . , n|P| ) of
n, and denote by MI , resp. NI , the standard Levi subgroup, resp. the unipotent
radical, of PI . (See e.g. §6.1.3) For a fixed irreducible automorphic representation
π0 of MI (A), if π is equivalent to π0 , there exists a λ ∈ XGM such that π ' π0 ⊗ λ,
and π also can be realized as a submodule of the space of automorphic forms of
MI (A). Indeed, A(M(F)\M(A))π = λA(M(F)\M(A))π0 where λ is considered as
a function on M(A). Therefore,
A(N(A)M(F)\G(A))π = λ ◦ mP A(N(A)M(F)\G(A))π0
choose an automorphic form
φ ∈ A(NI (A)MI (F)\G(A))π ∩ L2 (NI (A)MI (F)\G(A)) := A2 (NI (A)MI (F)\G(A))π .
Here, as usual, L2 (NI (A)MI (F)\G(A)) denotes the space of L2 functions on
ZG(A) NI (A)MI (F)\G(A). Following Definition 6.3, we denote by E(φ, π)(g)
the associated Eisenstein series. By Theorem 6.1, E(φ, π)(g) belongs to
A(G(F)\G(A)), namely, E(φ, π)(g) is an automorphic form on G(A).
Being automorphic forms, Eisenstein series are of moderate growth. In par-
ticular, they are generally not integrable over G(F)\G(A)1 . On the other hand,
these series are smooth and hence are integrable over any compact subset of
G(F)\G(A)1 . To make the integration meaningful, we need to find intrinsically
defined compact domains in G(F)\G(A)1 . For this, we fix a convex polygon
p : [0, n] → R. Hence, it makes
h i sense to talk about the p- semi-stable OF -lattices
of rank n. Denote by M≤p F,n 1 the moduli spaces of p-semi-stable OF -lattices of
rank n and volume 1. Then, by Proposition 6.2, there is a natural identification
n o
M≤p
F,n [1] = g ∈ GLn (F)\GLn (A) : degar (g) = ∆F , p̄ ≤ p .
n/2 g
(6.80)

Q,n [1] = MQ,n [1]

For example, if we take p ≡ 0 to be the constant polygon zero, M≤0
is simply (the adelic inverse image of) the moduli space of rank n semi-stable Z-
lattices of volume one.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 182

182 Eisenstein Periods and Multiple L-Functions

Definition 6.6. The multiple L-function L≤p

F,n (φ, π) for number field F associated
to the L2 -automorphic form φ ∈ A2 (NI (A)MI (F)\G(A))π and the convex polygon
p is defined by
Z
L≤p
F,n (φ, π) := E(φ, π)(g) dg, <(π) ∈ C. (6.81)
M≤p
F,n [1]

More generally, for a standard parabolic subgroup P of a reductive group G

with a fixed maximal open compact subgroup K of G(A), and a fixed element ν
which belongs to the closure of the positive Weyl chamber C, as a generalization of
Proposition 6.3 above, which works only for GLn , by the main results in §16.3.3.1
below, there are natural compact moduli spaces (associated to (P, ν)-semi-stable
principal G-lattices of slope µ)
n o
F (µ) := g ∈ P(F)\G(A)/K : slope(Eg ) = µ, µ̄P (Eg ) ≤P ν, µ̄P (Eg ) ≥P −µ .
MP;≤ν

Here µ is a fixed element in X∗ (A0P ), Eg is the principal G-lattice associated to g,

and µ̄P (Eg ) denotes the P-canonical slope of Eg . On the other hand, analytically,
for a standard parabolic subgroup Q ⊇ P with the standard Levi decomposition
Q = NQ MQ , and an automorphic form φ ∈ A(N(A)M(F)\G(A))π with π an irre-
ducible automorphic representation of M(A), by the discussion after Definition 6.3
and Theorem 6.1, there is an open cone CQ Q Q
P ⊂ <(X M ) such that, if <(π) ∈ CP , the
relative Eisenstein series
X
E Q/M (φ, π)(g) := φ(δg), ∀g ∈ Q(F)\G(A), (6.82)
δ∈P(F)\Q(F)

is an automorphic form belonging to A(Q(F)\G(A)).

Definition 6.7. The multiple L-function (associated to µ; Q, ν; φ) is defined by

Z
Q, ≤ν
LF, µ (φ, π) := E Q/M (φ, π)(g) dg, <(π) ∈ CQ P. (6.83)
MQ;≤ν
F (µ)

Remark 6.1. (1) When defining multiple L-functions, we assume that φ comes
from a single irreducible automorphic representation. However, such a restriction
is artificial, even it simplifies constructions and particularly results below.
(2) Discussions below for the multiple L-functions L≤p F,n work well for general
Q, ≤ν
multiple L-functions LF, µ (φ, π). In the sequel, for simplicity, we only work with
≤p
LF,n .
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 183

Multiple L-Functions 183

6.2.6 Meromorphic Extension and Functional Equations

With the same notation as in §6.2.3, let G = GLn , set P = [π], and for
w ∈ W = Sn , set M 0 := wMw−1 , and denote by P0 = N 0 M 0 the associated
parabolic subgroup, and hence the intertwining operator M(w, π). By the fun-
damental work of Langlands, these intertwining operators not only satisfy nice
properties as explained in Proposition 6.1, but play an important role in estab-
lishing the function equations of Eisenstein series. Consequently. we obtain the
following basic properties of the multiple L-functions.

Theorem 6.2. The multiple L-functions satisfy the following properties.

(1) (Meromorphic Continuation) L≤p

F,n (φ, π) for <(π) ∈ C is well-defined and ad-
mits a unique meromorphic continuation to the whole space P.
(2) (Functional Equation) As meromorphic functions on P,
L≤p ≤p
F,n (φ, π) = LF,n (M(w, π)φ, wπ) ∀w ∈ W. (6.84)

Proof. (1), resp. (2), comes from Theorem IV.1.8, resp. Theorem IV.1.10 of [75].
This certainly explains why only L2 -automorphic forms are used in original defi-
nition of multiple L-functions. 

6.2.7 Holomorphicity and Singularities

Let π ∈ P and α ∈ ∆GM . We introduce a function h : P → C by the assignment

that
π ⊗ λ 7→ hλ, α∨ i ∀λ ∈ XGM ' aGM . (6.85)
Here, as usual, α∨ denotes the coroot associated to α. In addition, for an h as
above, introduce its associated root hyperplane by
H := π0 ∈ P : h(π0 ) = 0 .

(6.86)
Obviously, H depends on the base point π we choose. Accordingly, we also denote
h by hH , since it is determined by H. In addition, for a set D of root hyperplanes,
we use the following conventions.

(a) The singularities of a meromorphic function f on P is said to be carried out

by D if, for all π ∈ P, there exist a finitely supported function nπ : D → Z≥0
such that π0 7→ ΠH∈D hH (π0 )nπ (H) · f (π0 ) is holomorphic at π0 ;
(b) The singularities of f are said to be without multiplicity at π if nπ ∈ {0, 1};
(c) The set D is said to be locally finite, if, for any compact subset C ⊂ P, we
have that {H ∈ D : H ∩ C , ∅} is finite.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 184

184 Eisenstein Periods and Multiple L-Functions

Theorem 6.3. The multiple L-functions satisfy the following properties.

(3) (Holomorphicity)
(a) When <(π) ∈ C, L≤p F,n (φ, π) is holomorphic.
(b) L≤p
F,n (φ, π) is holomorphic at π where <(π) = 0.
(4) (Singularities) Assume that φ is a cusp form. Then
(a) There is a locally finite set of root hyperplanes D such that the singularities
of L≤p
F,n (φ, π) are carried out by D.
(b) All singularities of L≤pF,n (φ, π) are without multiplicities at π provided that
h<(π), α i ≥ 0 ∀α ∈ ∆GM .
∨

(c) There are only finitely many of singular hyperplanes of L≤p F,n (φ, π) which
n o
intersect π ∈ P : h<(π), α i ≥ 0, ∀α ∈ ∆ M .
∨

Proof. As above, this is a direct consequence of the fundamental results of Lang-

lands on function equations of Eisenstein series, done in [70]. Indeed, (3), resp.
(4), comes from Proposition IV.1.8, resp. Proposition IV.1.11, of [75]. 

We end this chapter with a comment that the above discussion works for gen-
eral reductive groups G as well, if we replace the semi-stable OK -lattices with the
semi-stable arithmetic principal G-torsors to be introduced in Part 6.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 185

Chapter 7

Periods of Reductive Groups

7.1 Arthur’s Analytic Truncation

In this section, we recall the constructions and basic properties of Arthur’s analytic
truncation following [3, 4, 6].

7.1.1 Positive Cone and Positive Chamber

Let G be a connected reductive group over F. Fix a minimal parabolic subgroup

P0 of G and denote by P0 = M0 N0 its Levi decompositions, where N0 , resp. M0 ,
denotes the unipotent radical, resp. the Levi subgroup, of P0 . Let P be a standard
parabolic subgroup withP = MP NP its associated standard Levi decomposition,
where N, resp. M, denotes the unipotent radical, resp. the standard Levi subgroup,
of P. From the previous chapter, there are naturally associated spaces a0 , aP a0P
and their duals, the sets of ∆0 , ∆P , b
∆0 , b
∆P , and the decompositions

a0 = a0P ⊕ aP and aP = aGP ⊕ aG . (7.1)

Following [3, 4], we introduce a partial order on aGP .

+ G
Definition 7.1. The (acute) Weyl chamber aG+
P and the (obtuse) Weyl cone aP of
G
aP are defined by

P := H ∈ aP : hα, Hi > 0 ∀α ∈ ∆P ,
aG+
 G

+ G
(7.2)
aP := H ∈ aGP : h$Gα , Hi > 0 ∀α ∈ ∆P .


In addition, denote by τGP and b

τGP the characteristic functions of the subsets aG+
P
+ G G
and aP in aP , respectively.

185
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 186

186 Eisenstein Periods and Multiple L-Functions

Lemma 7.1.
+ G
(1) (see e.g. [44]) When P = P0 , the spaces aG+
0 and a0 coincides with the usual
positive Weyl chamber and the positive cone. In particular, + aG0 is the dual
cone of aG+
0 and
[  
aG0 = 0 .
w aG+ (7.3)
w∈W

Here, W denotes the Weyl group of (G, P0 ).

(2) (see e.g. [3, 4]) Every fundamental weights is a certain non-negative linear
combination of simple roots. Consequently,
+ G
aG+
P ⊆ aP . (7.4)

More generally, let P ⊆ Q be two standard parabolic subgroups of G. By

the previous chapter, the spaces aQ P and the sets ∆P , ∆P make sense. Moreover,
Q bQ

aP = aQP ⊕ aQ . We extend the linear functionals in ∆P and ∆P to the elements of

Q bQ
the dual space a0 by means of the canonical projection from a0 to aQ
∗
P given by the
decomposition a0 = a0P ⊕ aQP ⊕ a Q .

Definition 7.2. ( [4]) Let P ⊆ Q be the standard parabolic subgroups of G. The

positive chamber and the positive cone associated to P ⊆ Q are defined by
n o n o
H ∈ a0 : (α, H) > 0 ∀α ∈ ∆PQ = a0Q ⊕ H ∈ aPQ : (α, H) > 0 ∀ α ∈ ∆PQ ⊕ aP ,
n o n o
H ∈ a0 : ($, H) > 0 ∀ $ ∈ b ∆PQ ⊕ aP ,
∆PQ = a0Q ⊕ H ∈ aPQ : ($, H) > 0 ∀ $ ∈ b

τPQ their associated characteristic function.

respectively. Denote by τPQ and b

Similarly, as above, since all elements of b∆PQ are non-negative linear combina-
tions of the elements in ∆PQ , we obtain the following:

Lemma 7.2. Let P ⊆ Q be the standard parabolic subgroups of G. For the char-
acteristic function τPQ , resp. b
τPQ , of the positive chamber, resp. the positive cone,
we have
τPQ ≥ τPQ .
b (7.5)

7.1.2 Preliminary Estimations

Let P0 be a minimal parabolic subgroup of G and let P be am associated standard

parabolic subgroup of G. Denote τGP and b τGP in §7.1.1 simply by τP and b τP ,
respectively.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 187

Periods of Reductive Groups 187

Lemma 7.3 (Preliminary Estimation). ([4]) Let k · k be a fixed Euclidean norm

on aP . There are constants c and N such that
X    N
bτP H(δx) − T ≤ c kxk ekT k ∀ x ∈ G(A)1 , ∀T ∈ aP . (7.6)
δ∈P(F)\G(F)
Moreover, there are only finitely many δ ∈ P(F)\G(F) such that the corresponding
summands are not zero.1

As a direct consequence, we have the following:

Corollary 7.1. ( [4]) For fixed N ≥ 0 and T ∈ a0 , there exist constants c0 and N 0
such that, for any function φ on P(F)\G(A)1 , and x, y ∈ G(A)1 ,
!
X  
N0 N0 |φ(u)|
φ(δx) ·b
τP H(δx) − H(y) − T ≤ c kxk · kyk · sup
0
N
. (7.7)
δ∈P(F)\G(F) u∈G(A)1 kuk

7.1.3 Langlands’ Combinatorial Lemma

Let P1 ⊆ P2 be two standard parabolic subgroups of G. Following Arthur ( [4]),

we define a function on a0 by
X 2
σ21 (H) := σPP21 (H) := (−1)|∆3 | τ31 (H) b
τ3 (H) ∀H ∈ a0 . (7.8)
P3 :P1 ⊆P3 ⊆P2

Here, as in the sequel, ∆23 , τ3 , denotes ∆PP23 , resp. τPP31 , resp. b

resp. τ31 , resp. b τP3 .

Lemma 7.4. (Lemma 6.1 in [3]) The function σ21 on a0 coincides with the char-
acteristic function of the following subset (of a0 )
α(H) > 0 ∀α ∈ ∆21
 


 


 
β(H) ∆ 2 
.

H ∈ a ≤ 0 ∀β ∈ \∆ (7.9)
 

 1 1 1 


 
$(H) > 0 ∀$ ∈ ∆2

 


b

Applying this to the special case P1 = P2 , we get the following important:

Corollary 7.2 (Langlands’ Combinatorial Lemma). Let P ⊆ Q be parabolic

subgroups of G. For all H ∈ a0 ,
Q
X
τRQ (H) = δPQ ,
(−1)|∆R | τRP (H) b
R:P⊆R⊆Q
X R
(7.10)
τRP (H) τRQ (H) = δPQ .
(−1)|∆P | b
R:P⊆R⊆Q

1 When δ runs over P(F)\G(F), the collection of all δPδ−1 ’s consists of all parabolic subgroups of G

conjugating to P.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 188

188 Eisenstein Periods and Multiple L-Functions

Let P ⊆ Q be parabolic subgroups of G and fix a vector Λ ∈ a∗0 . Set

Q ∨
εQ
P (Λ) := (−1)
#{α∈∆P :Λ(α )≤0}
,
φPQ (Λ, ·) := 1 (7.11)
$ (H) > 0, if Λ(α∨ ) ≤ 0


∈ a0 α

∀α ∈ ∆PQ .


H

 
$α (H) ≤ 0, if Λ(α ) > 0
∨




 


Here the $α ’s are the fundamental weights defined by h$α , βi = δαβ ∀α, β ∈ ∆0 ,
and, for a subset S ⊆ a0 , 1S denotes the characteristic function of S in a0 . As an
application of Langland’ combinatorial lemma, we have the following:

Lemma 7.5. (Lemma 6.3 in [3]) Fix a Λ ∈ a∗0 . As a function on a0 ,


0, if Λ(α ) ≤ 0, ∃α ∈ ∆P
X  ∨ Q
εP (Λ) · φP (Λ, ·) · τR (·) ≡ 
R R Q
.

(7.12)
1, otherwise

R:P⊂R⊂Q

7.1.4 Langlands-Arthur’s Partition: Reduction Theory

To facilitate our ensuing discussion, we here briefly recall a classical reduction

theory for the adelic spaces associated to reduction groups G and the so-called
Langlands-Arthur’s partition for the spaces P(F)\G(A) associated to parabolic
subgroups P of G.
Let ω ⊂ N0 (A) M0 (A)1 be compact and fix T 0 ∈ −a+0 . For a parabolic subgroup
P of G, the associated Siegel set sP (T 0 , ω) is defined to be the collection of all the
elements pak, where p ∈ ω, k ∈ K, and the a ∈ A0 (R)0 ’s satisfying the condition
that α H0 (a) − T 0 are positive for each α ∈ ∆0P . In other words,

n o
sP1 (T 0 , ω) := ω · a · K : α H0 (a) − T 0 > 0 ∀α ∈ ∆0P .


(7.13)

As usual, we call a T ∈ a+0 sufficiently regular, resp. a T ∈ −a+0 sufficiently small

if, for any α ∈ ∆0 , α(T )  0, resp. α(T )  0.

Proposition 7.1 (Classical reduction theory). For a sufficiently big compact

subset ω of N0 (A)M0 (A)1 and a sufficiently small T 0 ∈ a+0 ,

G(A) = P(F) · sP (T 0 , ω). (7.14)

 a fixed T ∈ a0 , let sP (T 0 , T, ω) be the set of x in s (T 0 , ω) such

P P
In addition, for
that $ H0 (x) − T ≤ 0 for each $ ∈ b ∆0 1 . Namely,
n   o
sP (T 0 , T, ω) := x ∈ sP (T 0 , ω) : $ H0 (x) − T ≤ 0 ∀$ ∈ b
∆0P . (7.15)
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 189

Periods of Reductive Groups 189

Let F P (x, T ) be the characteristic function of the set of x ∈ G(A) such that δx
belongs to sP1 (T 0 , T, ω) for some δ ∈ P1 (F). Namely,

F P (x, T ) := 1n o (7.16)
x ∈ G(A) : δx ∈ sP1 (T 0 , T, ω) ∃δ ∈ P1 (F) .

Lemma 7.6. ([5])

(1) The function F P (x, T ) is left AP (R)0 NP (A)MP (F)-invariant.

(2) Viewed as a function on the quotient space A1 (R)0 N1 (A)M1 (F)\G(A), the
function F P (x, T ) is the characteristic function of a compact subset defined
by the projection of sP (T 0 , T, ω) (in the space A1 (R)0 N1 (A)M1 (F)\G(A)).

Example 7.1. For (standard) parabolic subgroups P1 ⊆ P2 of G, let A∞ ∞

1 := AP1 be
0
the identity component AP1 (A) of AP1 (R), and set

1,2 := AP1 ,P2 := AP1 ∩ MP2 (A) .

A∞ ∞ 1
(7.17)

Then the logarithmic map HP1 maps A∞ 2

1,2 isomorphically onto a1 , the orthogonal
complement of a2 in a1 . And, for T 0 , T ∈ a0 , set
n     o
1,2 (T 0 , T ) := a ∈ A1,2 : α H1 (a) − T > 0, α ∈ ∆1 ; $ H1 (a) − T < 0, $ ∈ ∆1 ,
A∞ ∞ 2 b2

where ∆21 := ∆P1 ∩M2 and b

∆21 := b
∆P1 ∩M2 . Then we have the following:

Lemma 7.7. ( [5]) Let T 0 ∈ a0 be sufficiently small. Viewed as a function on

G(F)\G(A)1 , F(x, T ) := F G (x, T ) is the characteristic function of a compact sub-
set defined by the projection of N0 (A) · M0 (A)1 · A∞P0 ,G (T 0 , T ) · K.

Furthermore, by Lemma 7.5, we obtain the following:

Proposition 7.2 (Langlands-Arthur’s Partition). ( [4]) Let P be a standard

parabolic subgroup of G, and let T 0 ∈ a+0 be a sufficient regular. Then for
T ∈ T 0 + a+0 ,
X X  
F 1 (δx) · τ1P H0 (δx) − T = 1 ∀x ∈ P(F)\G(A). (7.18)
P1 :P0 ⊆P1 ⊆P δ∈P1 (F)\G(F)

7.1.5 Arthur’s Analytic Truncation

Definition 7.3. ( [3,4]) Let T ∈ a+0 be sufficiently regular and let φ be a continuous
function on G(F)\G(A)1 .
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 190

190 Eisenstein Periods and Multiple L-Functions

(1) The constant term φP (x) of φ along P is defined by

Z
φP (x) := φ(nx) dn. (7.19)
N(F)\N(A)

Here and in the sequel, the measure on N(A) is chosen to be the one satisfying
Z
dn = 1. (7.20)
N(F)\N(A)

(2) The analytic truncation Λ φ (x) of φ of Arthur is defined by

T


X X  
ΛT φ (x) := (−1)|∆P | φP (δx) · b
τP H(δx) − T ,


(7.21)
P δ∈P(F)\G(F)

where the sum is over all (standard) parabolic subgroups.

Since all parabolic subgroups of G can be obtained as conjugations of standard

parabolic subgroups by the elements from P(F)\G(F),
X  
ΛT φ (x) = (−1)|∆P | φP (x) · b
τP H(x) − T ,


(7.22)
P

where the sum runs over all parabolic subgroups, standard or not.
As it stand, Arthur’s analytic truncation is indeed a well-designed device, in
which all the constant terms are tackled in such a way that the contributions from
all different levels of the parabolic subgroups are not only counted at the cor-
responding cuspidal region, but there is no overdone totally for these parabolic
subgroups, since each parabolic subgroup is essentially contributed once and only
once.

7.1.6 Basic Properties

Arthur’s analytic truncations admit the following basic properties.

Theorem 7.1. ([4]) Let T ∈ a0 be sufficiently regular.

(1) Let φ : G(F)\G(A) → C be a locally L1 function. Then, for almost all g,

ΛT ΛT φ(g) = ΛT φ(g). (7.23)
Moreover, if φ is also locally bounded, then the equality holds for all g.
(2) Let φ1 , φ2 be two locally L1 functions on G(F)\G(A). Suppose that φ1 is of
moderate growth and φ2 is rapidly decreasing. Then
Z Z
ΛT φ1 (g) · φ2 (g) dg = φ1 (g) · ΛT φ2 (g) dg. (7.24)
ZG(A) G(F)\G(A) ZG(A) G(F)\G(A)
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 191

Periods of Reductive Groups 191

(3) Let Kfin be an open compact subgroup of G(Afin ), and let r, r0 be two positive
real numbers. Then there exists a finite subset Xi : i = 1, 2, . . . , N ⊂ U,

the universal enveloping algebra of g∞ , such that, for any right Kfin -invariant,
smooth function φ on G(F)\G(A),
N
X 0
n o
ΛT φ(ag) ≤ kgk−r sup |δ(Xi )φ(ag0 )| kg0 k−r : g0 ∈ G(A)1 , (7.25)
i=1

where a ∈ AG(A) , g ∈ G(A)1 ∩ S , and S is a Siegel domain with respect to

G(F)\G(A).

The key to prove of this result is that the nilpotent radicals of parabolic sub-
groups have relatively simple structures. Namely, they all admit filtrations of sub-
spaces with affine successive quotients. For details, please refer to Arthur’s origi-
nal papers mentioned above.

Corollary 7.3. [4]

(1) If φ is a cusp form, then ΛT φ = φ.

(2) If φ is of moderate growth, i.e. there exist some constants c, N > 0 such that
|φ(x)| ≤ ckxkN for all x ∈ G(A), then so is ΛT φ.

7.2 Analytic Arthur Periods and Geometric Eisenstein Periods

7.2.1 ΛT 1 as Characteristic Function of Compact Set

Definition 7.4. ( [4]) Let T ∈ a0 be a sufficiently regular element, and let φ be

a continuous function on G(A). Then, for a parabolic subgroup P, the level P-
Arthur’s analytic truncation of φ is defined by
X P
X  
ΛT,P φ(g) := (−1)|∆R | φR (δg) · b
τRP H(δg) − T . (7.26)
R:R⊆P δ∈R(F)\P(F)

Obviously, ΛT,G = ΛT .

Lemma 7.8. ([4]) Let T ∈ a+0 be sufficiently regular.

(1) (Inversion Formula) For a G(F)-invariant function φ,

X X  
φ(g) = ΛT,P φ(δg) · τP H(δg) − T . (7.27)
P δ∈P(F)\G(F)
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 192

192 Eisenstein Periods and Multiple L-Functions

(2) Assume that T ∈ CP and that φ is automorphic on N(A)M(F)\G(A). Then

ΛT,P φ(m) isZa rapidly decreasing function on M(F)\M(A)1 . In particular, the
integration ΛT,P φ(m) dm is well-defined.
M(F)\M(A)1

(1) comes from Langlands’ combinatorial lemma. For a detailed proof, please
refer [4].

Definition 7.5. Let T ∈ a+0 be sufficiently regular. The truncated domain Σ(T ) :=
 
ZG(A)G(F)\G(A) of the space G(F)\G(A)1 is defined by
T

Σ(T ) := ZG(A)G(F)\G(A) := g ∈ ZG(A)G(F)\G(A) : ΛT 1(g) = 1 . (7.28)

  n o
T

Proposition 7.3. ([5]) For a sufficiently regular T ∈ a+0 ,

ΛT 1(x) = F(x, T ). (7.29)
That is to say, the subset Σ(T ) coincides with the projection of compact subspace
N0 (A) · M0 (A)1 · A∞P0 ,G (T 0 , T ) · K to G(F)\G(A) introduced in Lemma 7.6(2). In
1

particular, Σ(T ) is compact.

Proof. By Proposition 7.2 on Arthur’s partition for G(F)\G(A),

X X    
F P δx, T · τP HP (δx) − T = 1 (7.30)
P δ∈P(F)\G(F)

where τP is the characteristic function of H ∈ a0 : α(H) > 0, α ∈ ∆P and


   
F P nmk, T = F MP m, T , n ∈ NP (A), m ∈ MP (A), k ∈ K. (7.31)
On the other hand, applying the inversion formula (1) to the function 1, we get
ΛT,P 1 (δx) · τP HP (δx) − T = 1.
X X    
(7.32)
P δ∈P(F)\G(F)

With this, the desired result is immediately obtained by induction. 

All these can be used to prove the following result, by applying the Langland
combinatoric lemma and the associated inverse formula. For details, please refer
to [5].

Corollary 7.4. ([5]) For a sufficiently regular T ∈ a+0 , the characteristic function
F Q (g, T ) introduced in (7.16) coincides with the partial truncation along P for the
constant function 1. That is to say,
F P (g, T ) = ΛP,T 1(g) =
X P
X  
(−1)|∆R | τRP H(δg) − T .
b
R:P0 ⊆R⊆P δ∈R(F)\P(F)
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 193

Periods of Reductive Groups 193

Note that the right hand side makes sense even when T is not sufficiently
regular. This then leads to the definition of F P (g, T ) for general T . Namely,

F P (g, T ) := ΛP,T 1(g) :=

X P
X  
(−1)|∆R | bτRP H(δg) − T ∀T  0.
R:P0 ⊆R⊆P δ∈R(F)\P(F)

Proposition 7.4. ( [3–5], see also [41]) The function F P (g, T ) on the space
N(A)M(F)\G(A)1 is bounded and compactly supported, uniformly for T varying
in a compact subset.

Proof. Following Arthur ([3–5]), let

Q
X
ΓQ
P (H, X) := (−1)|∆R | τRP (H)b
τRQ (H − X),
R:P⊆R⊆Q
X R
(7.33)
ΓQ
b
P (H, X) := (−1)|∆P | τRP (H − X)b
τRQ (H).
R:P⊆R⊆Q

By Landings’ combinatorial lemma,

Q
X
τQ
bP (H − X) = τRP (H)ΓRQ (H, X),
(−1)|∆R |b
R:P⊆R⊆Q
X R
(7.34)
τQ
P (H − X) = ΓRP (H, X)τRQ (H),
(−1)|∆P |b
R:P⊆R⊆Q

and
Q
X
(−1)|∆R | ΓRP (H, X)b
ΓRQ (H, X) = δPQ ,
R:P⊆R⊆Q
X R
(7.35)
ΓRP (H, X)ΓRQ (H, X) = δPQ .
(−1)|∆P |b
R:P⊆R⊆Q

Q
∆P R
ΓQ
Consequently, since, by definition, b P (H, X) = (−1) ΓP (H − X, −X),
Q
X
τQ
P (H − X) = (−1)|∆R | ΓRP (H − X, −X)τRQ (H),
R:P⊆R⊆Q
X X (7.36)
F (g, T + X) =
P
F R (δg, T )ΓRP (H(δg) − T, X).
R:P0 ⊆R⊆P δ∈R(F)\P(F)

In addition, by Lemma 2.1 of [5], ΓQ Q

P (H, X) is compactly supported in H ∈ aP ,
uniformly for X varying in a compact subset. Hence our assertion is a direct
consequence of the previous proposition. Indeed, by Lemma 7.6, for a sufficiently
regular T , the function F P (g, T ) is a characteristic function of a compact subset.

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 194

194 Eisenstein Periods and Multiple L-Functions

7.2.2 Geometric Eisenstein Period

Fix a sufficiently regular T ∈ a0 and let φ be an automorphic form of G as defined

in Definition (6.1) of §6.2.3. By Theorem 7.1(3), ΛT φ is rapidly decreasing, and
hence integrable over G(F)\G(A)1 .

Definition 7.6. Let φ be an automorphic form of G and let T ∈ a+0 be sufficiently

regular. The T -version of Arthur’s period of φ is defined by
Z
A(φ; T ) := ΛT φ(g) dg. (7.37)
G(F)\G(A)

By the unipotent property of Arthur’s analytic truncation ΛT established in

Theorem 7.1(1), namely ΛT ◦ ΛT = ΛT ,
Z Z  
A(φ; T ) = ΛT φ dµ(g) = ΛT ΛT φ (g) dµ(g). (7.38)
ZG(A) G(F)\G(A) ZG(A) G(F)\G(A)

Hence, by the self-adjoint property for Λ in Theorem 7.1(2),

T
Z Z
1(g) · ΛT ΛT φ (g) dµ(g) = ΛT 1 (g) · ΛT φ (g) dµ(g)
     
ZG(A) G(F)\G(A) Z G(F)\G(A)
Z G(A)
ΛT ΛT 1 (g) · φ(g) dµ(g),
 
=
ZG(A) G(F)\G(A)

where 1 takes constant one on G(A). Indeed, by Theorem 7.1(4), ΛT φ and ΛT 1

are rapidly decreasing. Hence, using ΛT ◦ ΛT = ΛT again, we arrive at
Z
A(φ; T ) = ΛT 1(g) · φ(g) dµ(g). (7.39)
ZG(A) G(F)\G(A)

Therefore, by Proposition 7.3,

Z Z Z
ΛT φ(g)dµ(g) = ΛT 1(g) · φ(g)dµ(g) = φ(g)dµ(g). (7.40)
ZG(A) G(F)\G(A) ZG(A) G(F)\G(A) Σ(T )

All these then complete a proof of the following:

Theorem 7.2. Let T ∈ a0 be sufficiently regular, and let φ be an automorphic

form on G(F)\G(A). Then,
Z Z
φ(g) dµ(g) = ΛT φ(g) dµ(g). (7.41)
Σ(T ) G(F)\G(A)1

The importance of this fundamental relation can hardly be over-estimated. In-

deed, as to be seen in Part 6, not only this theorem holds for all T ∈ C, but Σ(T )
admits a natural geometric interoperation in terms of the moduli space of semi-
stable arithmetic principal G-torsors.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 195

Periods of Reductive Groups 195

Definition 7.7. Let T ∈ a+0 be a sufficiently regular element, and let φ ∈

A2 (NI (A)MI (F)\G(A))π be a P-level of L2 -automorphic form with π an irre-
ducible automorphic representation of M(A). The (geometric) Eisenstein period
for φ is defined by Z
ωG;T
F (φ, π) := E(φ, π)(g) dg <(π) ∈ C. (7.42)
Σ(T )
Here E(φ, π)(g) denotes the Eisenstein series induced from φ.

Even, via the theory of the so-called Eisenstein systems, we may clarify some
preliminary structures of these Eisenstein periods ωG;T
F (φ, π), in general, it is very
difficulty to understand them completely, due to the complications of Eisenstein
series and their associated constant terms. However, when φ is a cusp form, the
situation changes dramatically as we will see below, essentially because the con-
stant terms of the associated Eisenstein series have very simple structures.

7.2.3 Regularized Integration over Cone

Let V be a real n dimensional vector space, denote its dual space by V ∗ . Then VC∗
is the space of complex linear forms on V. By a cone in V we shall mean a closed
subset of the form
n o
C := x ∈ V : hµi , xi ≥ 0 ∀i (7.43)
where µi }ni=1 is a basis of V ∗ . Let e j be the dual basis of V. We shall say that
 
λ ∈ VC∗ is negative, resp. non-degenerate, with respect to C if <(hλ, e j i) < 0, resp.
hλ, e j i , 0, for each j = 1, . . . , n.
Denote by S (V ∗ ) := Sym(V ∗ ) the symmetric algebra of V ∗ . In the sequel,
we view elements of this space as polynomial functions on V. By definition, an
exponential polynomial function on V is a function of the form
r
X
f (x) = ehλi ,xi Pi (x) (7.44)
i=1
where the λi are distinct elements of VC∗ and the Pi (x) are non-zero elements of
S (VC∗ ). We call λi ’s the exponents of f . Consider the integral
  Z
IC f ; λ := f (x)ehλ,xi dx, (7.45)
C
and set
Hk,i := λ : hλ, ek i = hλi , ek i
n o
(1 ≤ i ≤ r). (7.46)
r
X
Lemma 7.9. ([47]) Let f (s) = ehλi ,xi Pi (x) be an exponential polynomial func-
i=1
tion on V, and let C be a cone in V.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 196

196 Eisenstein Periods and Multiple L-Functions

(1) The λi ’s are uniquely determined. And f is integrable over C if and only if λi
are negative with respect
 to C for all i.
(2) The function IC f ; λ has analytic continuation to V ∗ with (only) hyperplane
singularities along Hi,k , 1 ≤ k, i ≤ r.  
(3) Suppose that f is absolutely integrable over C. Then IC f ; λ is holomorphic
  R
at 0 and IC f ; 0 = C f (x) dx.
 
(4) Function IC f ; λ is holomorphic at 0 if and only if for all i, λi are non-
degenerate with respect to C, i.e. hλi , ek i , 0 for all pairs (i, k).

Denote by τY the characteristic function of a set Y. For λ ∈ V ∗ such that λi − λ

are negative with respect to C for all i = 1, 2, . . . , n, set
 Z Xn
e−hλi −λ,T i IC Pi (∗ + T )ehλi ,∗i ; λ .
  
F λ; C, T :=
b f (x) · τC (x − T ) · e−hλ,xi dx =
V i=1
 
By the lemma above, this integral is absolutely convergent and F b λ; C, T extends
to a meromorphic function on V ∗. With this,  following [47], we call the function
f (x) · τC (x − T ) #-integrable if Fb λ; C, T is holomorphic at λ = 0, and we set
Z #  
f (x) · τC (x − T )dx := F
b 0; C, T . (7.47)
V
Obviously, the #-integral exists if and only if each exponent λ j is non-degenerate
with respect to C.
In addition, if we assume that V = W1 ⊕W2 is a decomposition of V, and let C j
be a cone in W j . Write T = T 1 + T 2 and x = x1 + x2 relative to this decomposition.
If the exponents λ j of K are non-degenerate with respect to C = C1 + C2 , the
function
Z #
w2 7→ f (w1 + w2 )τC1 1 (w1 − T 1 ) dw1 (7.48)
W1
is defined and is an exponential polynomial. In particular,
Z # Z # Z # !
f (x)τC (x − T )dx = f (w1 + w2 )τC1 1 (w1 − T 1 ) dw1 τC2 2 (w2 − T 2 ) dw2
V W2 W1
and, as a function of T , it is an exponential polynomial with the same exponents
as f .
More generally, let g(x) be a compactly supported function on W1 and consider
functions of the form g(w1 − t1 )τC (w2 − T 2 ), which we call functions of type (C).
It is clear that the integral
Z
F(λ) :=
b f (x) · g(w1 − t1 )τC (w2 − T 2 ) · e−hλ,xi dx (7.49)
V
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 197

Periods of Reductive Groups 197

converges absolutely for an open set of λ whose restriction to W2 is negative with

respect to C2 . Furthermore, F(λ) b has a meromorphic continuation to V ∗ . Follow-
ing [47], we say that the function f (x)g(w1 − t1 )τC (w2 − T 2 ) is #-integrable if F(λ)
b
is holomorphic at λ = 0. And in this case, we define
Z #
f (x) · g(w1 − t1 )τC (w2 − T 2 ) dx := F(0).
b (7.50)
VR
As a function of w2 , W f (w1 + w2 )g(w1 − T 1 ) dw1 is an exponential polynomial
1
on W2 , hence
Z #
f (x) g(w1 − t1 )τC (w2 − T 2 ) dx
V
Z # Z ! (7.51)
= f (w1 + w2 )g(w1 − T 1 ) dw1 τ2 (w2 − T 2 ) dw2 .
C
W2 W1

Example 7.2. We have

Z #
ehλ,T i
ehλ,xi τC (x − T ) dx = (−1)n vol e1 , e2 , . . . , en · Qn .


(7.52)
V j=1 hλ, e j i
a e : a j ≥ 0 and vol e1 , e2 , . . . , en is the volume of
 Pn

Here C is the cone
 Pn j=1 j j
the parallelepiped j=1 a j e j : a j ∈ [0, 1] .
In particular, when V = R, this coincides with
eλT
Z ∞
eλt dt = − (7.53)
T λ
and, in the case C ⊂ R is a cone,
Z #  Z #


ehλ,xi 1 − τC x − T dx = − ehλ,xi τC (x − T ) dx (7.54)
V V
since 1 − τC is the characteristic function of the cone −C.

7.2.4 Regularized Period of Automorphic Form

Let τk (X) be a function of type (C) on aP that depends continuously on k ∈ K, i.e.

there is a decomposition aP = W1 ⊕ W2 such that b τk has the form
gk (w1 − T 1 )τC2k (w2 − T 2 ) (7.55)
where the compactly supported function gk varies continuously in the L1 -norm
and linear inequalities defining the cone C2k vary continuously.
With applications to automorphic forms in mind, let f be a function on
M(F)N(A)\G(A) of the form
X s
φ j (m, k) · α j HP (a), k ehλ j +ρP ,HP (a)i
 
f (namk) = (7.56)
j=1
for n ∈ N(A), a ∈ AP , m ∈ M(A)1 and k ∈ K, where, for all j,
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 198

198 Eisenstein Periods and Multiple L-Functions

(a) φ j (m, k) is absolutely integrable on M(F)\M(A)1 × K,

(b) λ j ∈ a∗P and α j (X, k) is a continuous family of polynomials on aP such that,
for all k ∈ K, α J (X, k)ehλ j ,Xi τk (X) is #-integrable.

Remark 7.1. The assumptions above are motivated by the micro structure for the
constant terms of Eisenstein series. On the other hand, by the reduction theory,
the quotient M(F)\M(A)1 is of finite volume. Hence, when taking integration, we
should deal with the A-part, which is affine, over which the exponential polyno-
mial function can be controlled easily if the exponents are negative.

Following [47], we define the #-integration of f by

Z #  
f (g) · τk HP (g) dg :=
P(F)\G(A)
s Z
X Z ! Z # ! (7.57)
φ j (mk) dm · α j (X, k)τk (X) dX dk.
j=1 K M(F)\M(A)1 aP
R
In particular, there is no integration involving N(F)\N(A) dn here. This is because,
In practice, f is supposed to be the constant term of an Eisenstein series along P,
which is N(A)-invariant.
Let EP ( f ) be the set of distinct exponents {λ j } of f . Obviously, EP ( f ) is
uniquely determined by f , evenR the functions φ j and α j in (7.56) are not. How-
ever, since the function X 7→ M(F)\M(A)1 f (eX mk) dm is an exponential polynomial
on aP and the original #-integral is equal to
Z Z # Z !
f (eX mk) dm e−hρP ,Xi τk (X) dX dk,
K aP M(F)\M(A)1

hence the #-integral in (7.57) is independent of the choices of these functions. In

addition, if each of theZ exponents λ j is negative with respect to C2k for all k ∈ K,
the ordinary integral f (g) · τk (HP (g)) dg is absolutely convergent and
P(F)\G(A)1
equals to the #-integral by Lemma 7.9(3). This then implies the following:

Lemma 7.10. ([47]) Assume that f satisfies (7.56).

R#  
(1) The #-integral P(F)\G(A) f (g) · τk HP (g) dg of (7.57) is well-defined.
(2) If each of the exponents λ j of f is negative with respect to C2k for all k ∈ K,
Z  
f (g) · τk HP (g) dg
P(F)\G(A)1

is absolutely convergent and equals to the #-integral.

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 199

Periods of Reductive Groups 199

Example 7.3. For a sufficiently regular T ∈ a+0 , and a level P = MU automorphic

form Ψ ∈ AP (G) := A(N(A)M(F)\G(A)), the constructions  above can be applied
to ΛT,P Ψ(g) and the characteristic function τP HP (g) − T . Indeed, by the theory
of automorphic forms, see e.g. [75],
Xs  
Ψ(namk) = Q j HP (a) · ψ j (amk) (7.58)
j=1

where n ∈ N(A), a ∈ AP , m ∈ M(A)1 and k ∈ K, and the Q j ’s are polynomials

and ψ j ∈ AP (G) satisfies ψ j (ag) = ehλ j +ρP ,HP (a)i ψ j (g) for some λ j ∈ a∗P and for all
R m 7→ Λ ψ j (mk) is rapidly decreasing on
T,P
a ∈ AP . Moreover, since the function
M(F)\M(A)1 × K, the integration M(F)\M(A)1 ×K ΛT,P ψ j (mk)dm dk is well-defined.
Therefore, for τP the characteristic function of the cone spanned by the coweights
∆∨P , we have
b
Z #  
ΛT,P Ψ(g) · τP HP (g) − T dg
P(F)\G(A)1
exists if and only if
hλ j , $∨ i , 0 ∆∨P
∀$ ∈ b and λ j ∈ EP (Ψ).
The same holds for the product Λ Ψ(g)·τP (HP (gx)−T ) when x ∈ G(A). (Indeed,
T,P
−1
if x ∈ G(A), there is an element  k(g) ∈ K such that gk(g) ∈ P0 (A). Since
HP (gx) = HP (g) + HP (k(g)x), τP X − T − HP (k(g)x) is the characteristic function
of a cone depending continuously on g.) Set then
Z #  
IGP,T (Ψ) := ΛT,P Ψ(g) · τP HP (g) − T dg. (7.59)
P(F)\G(A)1

More generally, for an automorphic form φ ∈ A(G), denote by EP (φ) be the

set of exponents EP (φP ). Set
n o
A(G)∗ := φ ∈ A(G) : hλ, $∨ i , 0 ∀$∨ ∈ b ∆∨P , λ ∈ EP (φ), P , G . (7.60)
If φ ∈ A(G)∗ , then IP,T
G
(φP ) exists for all P. Following [47], we define a regular-
ized period by
Z ∗ X
φ(g) dg = IGP,T (φP ) := IGT (φ). (7.61)
G(F)\G(A) P

As above, for x ∈ G(A f ), let ρ(x) denote right translation by x defined by

ρ(x)φ(g) = φ(gx). Since A(G) is stable under right translation by G(Afin ) and
ρ(x)φ has the same set of exponents as φ, the space A(G)∗ is invariant under right
translation by G(A f ). Indeed, if
 X 
Q j HP (a) · ehλ j +ρP ,H(a)i φ j (mk),
 
φP namk = (7.62)
j
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 200

200 Eisenstein Periods and Multiple L-Functions

for k ∈ K, if we write the Iwasawa decomposition of kx as kx = n0 a0 m0 K(kx),

since amkx = n∗ aa0 mm0 K(kx) for some n∗ ∈ N(A), we have
 X 
Q j HP (a) + HP (a0 ) · ehλ j +ρP ,H(a)+H(a )i φ j mm0 K(kx) . (7.63)
0
   
φP namk · x =
j
 
This implies that EP ρ(x)φ = EP (φ). More generally, we have the following:

Theorem 7.3. ([47])

(1) IGT definesR a G(Afin )-invariant linear functional on A(G)∗ .

∗
(2) IGT (φ) := G(F)\G(A)1 φ(g) dg is independent of the choice of T .
(3) If φ ∈ A(G) is integrable over G(F)\G(A)1 , then φ ∈ A(G)∗ and
Z ∗ Z
φ(g) dg = φ(g) dg. (7.64)
G(F)\G(A)1 G(F)\G(A)1

7.2.5 Regularized Periods

Let f ∈ AP (G). Assume that

 Xl
φ j (mk) · α j H(a) ehλ j +ρP ,H(a)i
  
f namk = (7.65)
j=1

where n ∈ N(A), a ∈ AP , m ∈ M(A)1 , k ∈ K, and that, for all j, the α j (X)’s are
polynomials, and the φ j (g)’s are automorphic forms in AP (G) such that φ j (ag) =
φ j (g) for a ∈ AP . Following [47], we define
Z ∗  
f (g) · τP H(g) − T dg
P(F)\G(A)1
l Z
X Z ∗ Z #  (7.66)
:= φ j (mk) dm dk × α j (X)ehλ j ,Xi τP (X − T ) dX .
j=1 K M(F)\M(A)1 aP

From the discussions above, (7.66) is well-defined provided that

(a) hµ, $∨ i , 0 for all Q ⊂ P, $∨ ∈ (b∆∨ )PQ , and µ ∈ EQ (φ),

(b) hλ, α i , 0 for all α ∈ ∆P , λ ∈ EP (φ).
∨

Accordingly, let A(G)∗∗ be the space of φ ∈ A(G) such that (a) and hence also (b)
are satisfied for all P. As a direct consequence of the constructions and results in
the previous subsection, we have the following:

Theorem 7.4. ([JLR]) For automorphic forms φ in A(G)∗∗ ,

Z X Z ∗  
Λ φ(g) dg =
T
(−1)|∆P |
φP (g) · b
τP H(g) − T dg.
G(F)\G(A)1 P P(F)\G(A)1
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 201

Periods of Reductive Groups 201

7.3 Periods of Reductive Groups

7.3.1 Eisenstein Periods for Cusp Forms

As mentioned earlier, in general, the Eisenstein periods of Definition 7.7

Z
≤T
LG,F (φ, π) := E(φ, π)(g)dµ(g). (7.67)
Σ(T )
are very difficult to study due to the complications of the constant terms. How-
ever, by Langlands’ explicit formula for the constant terms of the Eisenstein series
associated to cusp forms recalled in §6.2.3, for π0 an irreducible cuspidal repre-
sentation in L2 M(F)\M(A)1 and φ ∈ A0 (N(A)M(F)\G(A))π0 is cuspidal, the


constant term of the associated Eisenstein series E(g, φ, λ) := E(φ, π0 ⊗ λ) along
the Borel subgroup of G is given by
X
E B (g, 1, λ) = ehwλ+ρ,H(g)i M(w, λ)(φ), (7.68)
w∈W
where W(w, λ) denotes the associated intertwining operator. In addition, we have
the following result of Bernstein:

Lemma 7.11. ([47]) If φ ∈ A0 (N(A)M(F)\G(A))π is a cusp form, then

Z ∗
E(φ, π)(g) dµ(g) = 0
G(F)\G(A)1
Z ∗
for all π such that both E(φ, π) and E(φ, π)(g) dµ(g) are well-defined.
G(F)\G(A)1

All these, together with Example 7.2, implies the following result on the Eisen-
stein periods associated to cusp forms.

Proposition 7.5. ([47]) Let φ ∈ A0 (N(A)M(F)\G(A)) be cuspidal. Then for a

sufficiently regular T ∈ a+0 ,
Z
1
ΛT E(g, φ, λ) dg
vol(Λb∨ ) G(F)\G(A)1
X 0
ehwλ−ρ,T i
Z
M(w, λ)φ (mk)dmdk P = P0




ρ, α

∨i
Q
= hwλ −

α∈∆0
 1
w∈W M0 (F)\M0 (A) ×K




 0 otherwise
Here Λ b∨ :=  Pα∈∆ aα $∨α : aα ∈ [0, 1) .
0 0

Here, the volume vol Λ

b∨
appears, because the measure on a0 is defined using
P 0
the co-weights lattice α∈∆0 Z$∨α . For example, when G = SLn , vol Λb∨
= n.
0
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 202

202 Eisenstein Periods and Multiple L-Functions

7.3.2 Gindikin-Karpelevich Formula

Taking an example, let G be a split, semi-simple group over F or rank n, and

choose φ ≡ 1 as the constant function 1 on the maximal subtorus of G. Since
we are working on a minimal parabolic subgroup, and we can apply Theorems 7.2
and 7.5 to this setting, consequently, for a sufficiently regular T ∈ a+0 ,

ehwλ−ρ,T i
Z
1
X
ωG;T ( M0 , λ) = vol Λ
b ∨

· M(w, λ) dm (7.69)
α∈∆0 hwλ − ρ, α i
F 0 Q ∨
w∈W T0 (F)\T0 (A)1

Indeed, since vol(K) = 1 and M(w, λ)φ (mk) is independent of m, k,

Z Z
M(w, λ)φ (mk)dmdk = M(w, λ)


dm
M0 (F)\M0 (A)1 ×K M0 (F)\M0 (A)1
Z (7.70)
= M(w, λ) = ζF (1) · M(w, λ).
b n
T0 (F)\T0 (A)1

Here in the last two equalities, we have used the fact that, for the nilpotent rad-
ical N0 of M0 , the volume of N0 (F)\N0 (A) is one (see e.g. §6.1.5), and that
vol T0 (F)\T0 (A)1 = vol F ∗ \I1F n = b ζF (1)r where bζF (1)r := Res s=1b
ζF (s) (see



e.g. [66], or [113] where different normalization of the volume form is used). On
the other hand, by the well-known Gindikin-Karpelevich formula, we can write
down the associated intertwining operator M(w, λ) explicitly.

Lemma 7.12. (See e.g. [La3]) For the Eisenstein series EG/M0 (1, λ, g),

ζF hλ, α∨ i
Y b

M(w, λ) =
. (7.71)
α>0,wα<0 ζF hλ, α i + 1
b ∨

7.3.3 Periods of Reductive Groups

By all these, particularly, Lemma 7.12 and the relation (7.69), we finally obtain

Theorem 7.5. Let T ∈ aG+ o be sufficiently regular. Up to a constant factor,

F (1 M0 , λ)
ωG;T ehwλ−ρ,T i ζF hλ, α∨ i
X Y b

= ζ
b F (1)n

. (7.72)
α∈∆0 hwλ − ρ, α i α>0,wα<0 b
∨
Q
vol Λ
b∨

0 w∈W ζ F hλ, α∨i + 1

Note that the right hand side of (7.72) makes sense for all T ∈ a0 , even the
right hand side is only defined for sufficiently positive T .

Definition 7.8. ( [126]) For a reductive group G over a number field F and T ∈ a0 ,
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 203

Periods of Reductive Groups 203

(1) The T -period ωG;T

F (λ) of G over F is defined by

ehwλ−ρ,T i ζF (hλ, α∨ i)
X Y !
ωG;T .
b
F (λ) := · (7.73)
α∈∆0 hwλ − ρ, α i α>0,wα<0 b
∨
Q
w∈W ζF (hλ, α∨ i + 1)

(2) The period ωGF (λ) of G over F is defined to be ωG;0

F (λ), namely,

ζF (hλ, α∨ i)
!
X 1 Y
ωGF (λ)
b
:= · (7.74)
α∈∆0 hwλ − ρ, α i α>0,wα<0 b
∨
Q
w∈W ζF (hλ, α∨ i + 1)

As they stand, these periods are several variable functions in λ ∈ a0,C .

7.3.4 Volumes of Truncated Fundamental Domains

Let G be a split, semi-simple group of rank r over F. In this subsection, following

a joint work of Kim and myself ( [57]), we calculate the volume of Σ(T ). This
then recovers Langlands’ famous formula ( [68]) on the volumes of fundamental
domains of G(F)\G(A)1 , by taking the limit T → ∞.
To state our result, for a w ∈ W, let
Aw := α ∈ ∆0 : wα < 0 Bw = α ∈ ∆0 : wα ∈ ∆0 .
 
and (7.75)
Then, Aw ∩ Bw = ∅. Introduce the set Wspa of the standard Weyl elements by
n o
Wspa := w ∈ W Aw t Bw = ∆0 , (7.76)
and let Pw be the standard parabolic subgroup associated to Bw . In addition, let
nw := #{α ∈ ∆0 : wα < 0}, (7.77)
and, for i > 1, set
ni := #{α > 0 : hρ, α∨ i = i} − #{α > 0, : hρ, α∨ i = i − 1},
ni,w := #{α > 0, wα < 0 : hρ, α∨ i = i} − #{α > 0, wα < 0 : hρ, α∨ i = i − 1}.

Proposition 7.6. ([57]) For a sufficiently positive T ∈ a+0 , the volume of the trun-
cated domain Σ(T ) ⊂ G(F)\G(A)1 is given by
vol Σ(T )

Y 
ζF (1)nw i>1 b
ζF (i)ni,w
 X Q
= ζ
· ehwρ−ρ,T i . (7.78)
 −n
b
 b F (i) i
vol Λ0 α ∨i − 1
  Q
b ∨

i>1

w∈Wspa α∈∆0 −wBw hwρ,

Proof. By Theorem 7.5,

Z
E(1 M0 ; λ; g) dg = vol Λ
  G;T
0 ζF (1) ωF (λ).
b∨
b n
(7.79)
Σ(T )
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 204

204 Eisenstein Periods and Multiple L-Functions

Our strategy is to take the residues for both side of this equation at λ = ρ, because,
by the result of Langlands ( [68]), the residue of E(g, 1, λ) at λ = ρ is constant.
Indeed, since Σ(T ) is compact for every sufficiently regular T ,
Z !
Resλ=ρ ∧ E(g, 1, λ) dg
T
Σ(T )
Z (7.80)
= Resλ=ρ E(g, 1, λ) dg = vol Σ(T ) · Resλ=ρ E(1 M0 ; λ; g).


Σ(T )
Hence it suffices to calculate the residues Resλ=ρ E(1 M0 ; λ; g) and Resλ=ρ ωG;T
F (λ).
We first treat Resλ=ρ E(1 M0 ; λ; g).
It is well-known that Resλ=ρ E(1 M0 ; λ; g) is obtained from the residue of con-
stant term along the Borel subgroup P0 , and, by Proposition 6.1, or better, (6.73),
the constant term of E(1, g, λ) along P0 of G is given by
X
E B (g, 1, λ) = M(w, λ)ehwλ+ρ,H(g)i . (7.81)
w∈W
Hence, from hρ, α∨ i = 1 for α ∈ ∆0 and hρ, α∨ i > 1 if α < ∆0 , α > 0,
 
 Y   
Resλ=ρ E(g, 1, λ) = lim  hλ, α i − 1 · M(w0 , λ)
∨
(7.82)
λ→ρ
α∈∆0
where w0 is the longest Weyl group element, which transforms all the positive
ζF hλ, α∨ i
Y b

roots to negative. Now, by Lemma 7.12, M(w0 , λ) =
, we have
α>0 ζF hλ, α i + 1
b ∨
Y
Resλ=ρ E(g, 1, λ) = b
ζF (1)n · ζF (i)ni .
b (7.83)
i>1
Therefore, by (7.80),
Z ! 
E(1 M0 ; λ; g) dg = b
Y 
ζF (1)n ζF (i)ni · vol Σ(T ) .


Resλ=ρ b (7.84)
Σ(T ) i>1
Next we treat the residue of the right hand side of (7.79) at λ = ρ. Since
hwλ, α∨ i = hλ, (w−1 α)∨ i, with nw = # α ∈ ∆0 : wα < 0 ,

 
 X Y hλ, α∨ i − 1 
Resλ=ρ ωG;T
F (λ) = λ→ρ
lim  ehwλ−ρ,T i M(w, λ) 
w∈W α∈∆0
hwλ, α ∨i − 1 
(7.85)
ζF (1)nw i>1 b ζF (i)ni,w
X Q
=
.
hwρ−ρ,T i
b
e
α∈∆0 ,w−1 α<∆0 hwρ, α i − 1
∨
Q
w∈W spa

Therefore, the volume of Σ(T ) is given by

 
vol Σ(T ) Y X
i>1 ζF (i)
Q b ni,w
= ζF (i)
b −ni
ζF (1) Q
b nw

ehwρ−ρ,T i ,
α∈∆0 ,w α<∆0 hwρ, α i − 1
∨
vol ∆0
b∨

i>1 w∈Wspa
−1

as desired. 
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 205

Periods of Reductive Groups 205

For many purposes, the formula (7.78) is sufficient. However, to expose the
hidden structures related to standard parabolic subgroups, we prove the following:

Lemma 7.13.

(1) There is a bijection between Wspa and the collections of standard parabolic
subgroups of G.
(2) If w ∈ Wspa , hwρ, α∨ i < 0 for all α ∈ ∆ − wBw .

Proof. (1) For each w ∈ Wspa , we obtain a subset Bw ⊆ ∆ and hence also an
associated standard parabolic subgroup Pw . Conversely, any standard parabolic
subgroup P of G is associated to a unique subset J ⊆ ∆. Set W J ⊆ W be the
subgroup of W generated by the reflections of α ∈ J, and introduce a subset
W(J) := w ∈ W/W J : w(J) ⊆ Φ+ ⊆ W/W J . It is known that the set W(J) gives

all the distinguished coset representatives of W/W J . Let w J be the element of the
maximal length in W(J). Then w J ∈ Wspa and in fact Wspa = w J : J ⊆ ∆ .


(2) To start with, we let w0 be the element of the maximal length in W and
let w0,J be the element of the maximal length in W J . By [20], for w J in (1), we
have w J = wl wl,J . Assume that α ∈ ∆ − wBw . Obviously, there exists a certain
β ∈ ∆ such that w0 (α) = −β. On the other hand, by definition, w0,Bw w0 = w−1 .
Consequently, w−1 (α) = −w0,Bw (β). This implies that β < Bw . Indeed, otherwise,
−w0,Bw (β) ∈ Bw and w−1 (α) ∈ Bw . Since α < wBw , we arrive at w−1 (α) < Bw , a
contradiction. But with β < Bw , we have wl,Bw (β) > 0, and hence w−1 (α) < 0. 

Consequently, we may rewrite (7.78) in terms of standard parabolic subgroups.

Theorem 7.6. ([57]) For a sufficiently regular T , we have

X
rk P J ζF (1)
rkP J Q b −ni +ni,wJ hw ρ−ρ,T i
i>1 ζF (i)
vol Σ(T ) = vol Λ .
b

b∨
(−1) e J
α∈∆0 −w J J (1 − hw J ρ, α i)
0 Q ∨
J⊂∆ 0

Proof. Obviously, the rational factors appeared in (7.78) may be rewritten as

Y   Y  
hwρ, α∨ i − 1 = (−1)rkPw 1 − hwρ, α∨ i .
α∈∆0 ,w−1 α<∆0 α∈∆−wBw

This, together with (7.78) implies that

ζF (1)nw i>1 b
ζF (i)ni,w
Y X Q
vol Λ ζF (i)−ni
· ehwρ−ρ,T i . (7.86)
b
b∨
b (−1)rkPw Q
0
i>1 w∈Wspa α∈∆0 −wBw 1 − hwρ, α ∨i

To finally write the summation above as one over standard parabolic subgroups,
we need to give a group theoretic interpretations for other terms. For this, we
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 206

206 Zeta Functions for Reductive Groups

first claim that, for w ∈ Wspa , nw + ni,w = 0, and − ni = n, the rank of

P P
G. Indeed, − ni is the number of positive roots such that hρ, α∨ i = 1. It is
P
exactly n. Similarly, − ni,w is the number of positive roots such that wα < 0,
P
and hρ, α∨ i = 1. It is exactly nw = rankPw . So nw + (−ni + ni,w ) = n for each
P
w ∈ Wspa . 

Note that Σ(∞) = G(F)\G(A)1 , and in the above formula, as T → ∞, only

the term corresponding to w = 1 survives. In this way, as in [57], we obtain
an alternative proof of Langlands’ formula on volumes of fundamental domains
associated to all split, semi-simple reductive groups.

Corollary 7.5. ([68], see also [57]) Let G be a split, semi-stable group on F. For
the volume of (a fundamental domain of) G(F)\G(A)1 , we have
  Y
vol G(F)\G(A)1 =vol Λ b∨

0 ζF (i)−ni .
b
(7.87)
i>1
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 207

PART 4

Zeta Functions for Reductive Groups

For reductive groups over number fields, new type of zeta functions are introduced
as the residues of the associated periods. They were initially discovered for SL3
as a by-product of the study of the rank three non-abelian zeta function. With a
result of Diehl on the residues of Siegel Eisenstein series associated to Sp2n , we
are able to locate the singular hyperplanes of the SLn and Sp2n -periods and fur-
ther discover the single variable zeta functions associated. To understand what
happens for other groups, G2 naturally becomes a target, being exceptional and
of rank two. The result is fascinating. Not only two different zeta functions are
discovered for G2 , the role played by the maximal parabolic subgroups is also
exposed. This finally leads to the zeta functions for reductive groups and their
maximal reductive groups, which are conjecturally to satisfy the functional equa-
tion and the Riemann Hypothesis. In this part, we first give a detailed account of
the above process, then present Suzuki’s proof for the Riemann Hypothesis of the
G2 -zeta functions based on our explicit formulas, and more importantly, Komori’s
beautiful Lie theoretic proof of the functional equations for all these new zeta
functions based on an involution of the Weyl elements involved. For the case of
(SLn , Pn−1,1 ), the functional equation was first jointly proved by Kim and myself.
Even our original proof is not produced, we instead use a joint work with Zagier
to write down the (SLn , Pn−1,1 )-zeta functions explicitly.

207
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 209

Chapter 8

Zeta Functions for Reductive Groups

8.1 Zeta Function for SL n: Genuine but Different

In the following two sections, we, following [126], explain how a study of high
rank non-abelian zeta functions leads to a new type of zeta functions associated to
reductive groups and their maximal parabolic groups. For simplicity, we assume
that the base field involved is the field of rationals.

8.1.1 Non-Abelian Zeta Function and Eisenstein Period

For a fixed n ∈ Z>0 , denote by b ζQ,n (s) the rank n non-abelian zeta function of
Q. By Proposition 3.2, b ζQ,n (s) can be calculated as an integration of Eisenstein
series over a moduli space of semi-stable lattices. Namely, (up to a constant factor
depending only on n),
Z
b Λ(g), n s dg.
 
ζQ,n (s) =
b E (8.1)
MQ,n [1] 2
Here MQ,n [1] ⊂ SLn (Z)\SLn (R)/SOn denotes the moduli spaces of semi-stable
(Z-)lattices of rank n and volume one, and

X 1
s) := π−s Γ(s) E(Λ(g), s) := π−s Γ(s)


E(Λ(g),
b
2s
(8.2)
x∈Λ(g)r{0}
kxk
denotes the complete Epstein zeta function for the lattice Λg = (Zn , ρ(g)), where
ρ(g) denotes the metric on Zn ⊗R = Rn induced by to the positive definitive matrix
gt · g for some g ∈ SLn (R).
Identify MQ,n [1] with the semi-stable part of a fundamental DSLn (Z) of
SLn (Z)\SLn (R)/SOn . Denote by 1MQ,n [1] the characteristic function of MQ,n [1].
By (8.1), we have
Z  n 
ζQ,n (s) =
b 1MQ,n [1] (g) · Eb Λ(g), s dg. (8.3)
SLn (Z)\SLn (R)/SOn 2

209
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 210

210 Zeta Functions for Reductive Groups

Up to this point, it is natural to ask whether 1MQ ,n [1] is related with Arthur’s
analytic truncation. This is indeed the case.

Theorem 8.1 (Theorem 14.2). Denote by Λ0SLn , or simply Λ0 , the Arthur analytic
truncation associated to SLn . Let 1 be the constant function of the value one on
the space SLn (Z)\SLn (R)/SOn . Then
1MQ,n [1] = Λ0SLn 1. (8.4)
In addition, Λ0SLn is unipotent and self-adjoint.

Recall that to make the original Arthur analytic truncation ΛT work, the pa-
rameter T has to be sufficiently regular T , a condition claiming that (T, α)  0
for all simple roots α. Our result significantly generalize this truncation theory of
Arthur. Indeed, as we will see in Theorem 14.2 below, it suffices for us to assume
that the positivity of the (T, α)’s.
We will prove this theorem in Part 6, where a much more general result will
be given. Instead, based on the basic properties of Arthur’s analytic truncation in
§7.1.6,1 we give the following direct consequence of Theorem 7.2.

Corollary 8.1 (Theorem 15.1). All non-abelian zeta functions are Eisenstein pe-
riods. More precisely, up to a constant factor depending only on n,
Z
b Λ(g), n s dg.
 
ζQ,n (s) =
b Λ0 E (8.5)
SLn (Z)\SLn (R)/SOn 2

This clearly indicates that to study non-abelian zeta functions, we should apply
the theory of Arthur periods, or better, Eisenstein periods. For this, we next expose
some explicit relation between the Epstein zeta function and Eisenstein series.

8.1.2 Epstein Zeta Function and Siegel-Eisenstein Series

Let SPn be the space of positive definitive matrices of determinant one and size
n, and let Γn := diag(±1, . . . , ±1) ∩ SLn (Z) SLn (Z). Denote by Qn1 ,...,nk the


subgroup of Γn induced from the standard parabolic subgroup Pn1 ,...,nk of SLn (R)
associated to the partition n = n1 + · · · + nk , that is, the subgroup of Γn consisting
 
H1 ∗ · · · ∗ 

 0 H2 · · · ∗ 

of matrices in SLn (Z) of the form  . . . .  for suitable H j ∈ SLn j (Z) (1 ≤
 .. .. . . .. 


0 0 · · · Hk
 

1 More precisely, we should use the properties listed in §15.1.1.

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 211

Zeta Functions for Reductive Groups 211

j ≤ k). Set s = (s1 , s2 , . . . , sk ) and N j = n1 +n2 +· · ·+n j for 1 ≤ j ≤ k. In addition,

for two matrices A and B, whenever it makes sense, we set A[B] := Bt · A · B; and,
for a matrix M = (mi j )ni, j=1 and k = 1, 2, . . . , n, denote by Mk the matrix (ai j )ki, j=1 .

Definition 8.1. Let Y ∈ SPn be a positive definitive matrix of size n and let
n = n1 +2 + . . . + nk be a fixed partition of n.

(1) The Siegel zeta functions Zn∗1 (Y; sn1 , . . . , sn−1 ) associated to Y and Qn1 ,1,...,1 is
defined by
n−1
−s j
X Y
Zn∗1 (Y; sn1 , . . . , sn−1 ) := Y[N] j ∀1 ≤ n1 ≤ n − 1. (8.6)
N∈Qn1 ,1,...,1 \Γn j=n1

(2) The Siegel Eisenstein series En1 ,n2 ,...,nk s; Y associated to Y and Qn1 ,...,nk is


defined by
X k
Y −s j n j +n j+1
En1 ,...,nk Y; s1 , . . . , sk := <(s j ) > . (8.7)


Y[A j ]
(A j ∗)=A∈Qn1 ,...,nk \GLn j=1
2
A j ∈Zn×N j

(3) For every m ≤ n, the Koecher zeta function associated to Y and Qm,n−m is
defined by
X −s n
Zm,n−m (Y, s) := Y[A] <(s) > . (8.8)
2
n×m
A∈Z /GLm (Z)
rankA=m

Lemma 8.1. (See e.g. [115]) With the same notation as above,

(1) Zn∗1 (Y; sn1 , . . . , sn−1 ), En1 ,n2 ,...,nk (Y; s1 , . . . , sk ) and Zm,n−m (X, s) are well-
defined, and admit meromorphic continuations to the whole parameter
spaces.
(2) ([22]2 ) There exists a constant c, depending only on n1 , such that
 n1 
Res sn = n1+1 Zn∗1 (Y; sn1 , . . . , sn−1 ) = cn1 Zn∗1 +1 Y; sn1 +1+ , sn1 +2 , . . . , sn−1 . (8.9)
1 2 2
In particular, by taking r = 1 and repeating this process, we obtain, up to a
constant factor,
  n−1
Res sn−1 =1 · · · Res s2 =1 Res s1 =1 Z1∗ (Y; s1 , s2 , . . . , sn−1 ) = |Y|− 2 . (8.10)
Yn−1
(3) Zn,0 (Y, s) = |Y|−s · ζ(2s − j), and
j=0
Z1,n−1 (Y; s)
|Y|−s · En−1,1 (Y −1 ; s) = E1,n−1 (Y; s) = . (8.11)
Z1,0 (I; s)
2 Please correct a misprint in [22] for (8.9).
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 212

212 Zeta Functions for Reductive Groups

When k = n and n j = 1 for all j, resp. k = 2 and n1 = n − 1, the associated

parabolic subgroup P1,...,1 , resp. Pn−1,1 , of SLn is minimal, resp. maximal.

Definition 8.2. Let Y ∈ SPn . Let s = (s1 , s2 , . . . , sn ), resp. (s1 , s2 ).

(1) The standard power function p−s (Y) of Y is defined by

Yn
p−s (Y) := |Y j |−s j . (8.12)
j=1
(2) The Siegel Eisenstein
Xseries associated to P1,...,1 and Y is defined by
E(n) (Y; s) := p−s (Y[γ]) <(s j ) > 1, 1 ≤ j ≤ n − 1. (8.13)
γ∈Q1,...,1 \Γn
(3) The Siegel Eisenstein series associated to Pn−1,1X and Y is defined by
En−1,1 (s1 , s2 ; Y) := En−1,1 (1; s1 , s2 ; Y) := |Y[A1 ]|−s1 |Y[A]|−s2 .
(A1 ∗)=A∈Qn−1,1 \Γn
A1 ∈Zn×(n−1)

By a rather direct computation we have the following:

Lemma 8.2. For s = (s1 , . . . , sn ), set s∗ := (sn−1 , sn−2 , . . . , s1 , −(s1 + s2 +· · ·+ sn )).

Then
E(n) (Y; s1 , s2 , . . . , sn ) = Z1∗ (Y; s1 , s2 , . . . , sn−1 ) and E(n) (Y −1 ; s) = E(n) (Y; s∗ ).
In particular,
Z1∗ (Y −1 ; t1 , t2 , . . . , tn−1 ) = Z1∗ (Y; tn−1 , . . . , t2 , t1 ). (8.14)

Consequently, since |Y| = 1, we haveX

En−1,1 (Y; s, t) = |Y|−t |Y[A1 ]|−s
(A1 ∗)=A∈Qn−1,1 \Γn
X (8.15)
= |Y| −t
|Y[A]n−1 |−s = Zn−1
∗
(Y, s).
A∈Qn−1,1 \Γn
This, together with Lemma 8.1, then establishes the following relations be-
tween the Epstein zeta functions are the Siegel Eisenstein series.

Proposition 8.1. Let g ∈ SLn (R) such that Y = gt g ∈ SPn . Denote by Λ(g) the
Z-lattices (Zn , ρg ). We have
1 X 1  s 
∗
Zn−1 (Y −1 ; s) = · |Y[x]|−s = · E Λ(g); , (8.16)
ζ(2s) x∈Zn \{0} ζ(2s) n/2
Z1∗ (Y; s1 , s2 , . . . , sn−1 ) = E(n) Y; s1 , s2 , . . . , sn .


(8.17)
In particular,
 n−2 
E Λ(g); s = Restn−2 =1,...,t2 =1, t1 =1 Z1∗ Y; ns − , tn−2 , tn−3 , . . . , t2 , t1 . (8.18)


2
That is to say, Epstein zeta functions are residues of Siegel Eisenstein series.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 213

Zeta Functions for Reductive Groups 213

8.1.3 Langlands’ Eisenstein and Siegel’s Eisenstein

We now rewrite Siegel’s Eisenstein series, introduced in classical language, as
Langlands’ Eisenstein series, introduced in a language which is more convenient
for some theoretical purpose. The point of course is about the power function ps
of (8.12) and the function mB in §6.1.5.
Let Y ∈ SPn be a positive definite matrix Y satisfying |Y| = 1. By the Gram-
Schmidt process, there exists an upper triangular unipotent matrix N and a diago-
nal matrix A = diag(a1 , a2 , . . . , an ) such that Y = A[N]. Obviously, ai = |Yi |/|Yi−1 |
for i = 1, 2, . . . , n. Hence, by definition,
Y n
p−s (Y) = |Y j |−s j
j=1
−(s2 +s3 +···+sn )  −(sn−1 +sn )  −sn
−(s1 +s2 +···+sn )

= |Y1 | |Y2 |/|Y1 | · · · |Yn−1 |/|Yn−2 | · |Yn |/|Yn−1 |
−sn
1 +s2 +···+sn ) −(s2 +s3 +···+sn ) n−1 +sn )

= a−(s
1 a2 · · · a−(s
n−1 · a1 a2 · · · an−1
1 +s2 +···+sn−1 ) −(s2 +s3 +···+sn−1 )
= a−(s
1 a2 n−1 .
· · · a−s n−1

Here, in the last step, we have used the fact that nj=1 a j = |Y| = 1.
Q

On the other hand, since Y is positive definite, there exists a matrix g such that
Y = gt g. Indeed, we may take g = T (g)N where T (g) = diag(t1 , t2 , . . . , tn ) with
a j = t2j . Consequently, by definition,
mB (g)λ+ρB = T (g)λ+ρB . (8.19)
Hence, if we write λ = (z1 , z2 , . . . , zn ) ∈ C , then j=1 z j = 0. In these coordinates,
n Pn
by §6.1.3, the Weyl vector ρ becomes
n − 1 n − 1 n − 1 n − 1
ρ = ρB = , − 1, . . . , 1 − ,− . (8.20)
2 2 2 2
Thus, by a direct calculation, we obtain
mB (g)λ+ρB = t1−[(n−1)+(2z1 +z2 +···+zn−1 )] t2−[(n−2)+(z1 +2z2 +···+zn−1 )] · · · tn−1
−[1+(z1 +z2 +···+2zn−1 )]
. (8.21)
Recall now that the relative Langlands Eisenstein series on SLn induced from
the constant function 1 on the Levi part of theX
Borel B = P1,1,...,1 is defined by
E(1; λ; g) := E SLn /B (1; λ; g) := mB (δg)λ+ρB . (8.22)
γ∈S Ln (Z)/P1,1,...,1 =B
Hence, if we make the coordinate changes from λ = (zi ) to s = (si ) by the linear
transformations 



 2s1 = 1 + (z1 − z2 )

= 1 + (z2 − z3 )

 2s2



 (8.23)




 ··· ···

= 1 + (z − z ),

 2s

n−1 n−1 n
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 214

214 Zeta Functions for Reductive Groups

what we have said then proves the following elementary relations between the
Epstein zeta functions and the Eisenstein series.

Proposition 8.2. Let Y ∈ SPn be a positive definite matrix with det Y = 1.

(1) For λ = (z1 , z2 , . . . , zn ) satisfying z j = 0 and s = (s1 , s2 , . . . , sn−1 ) satis-

Pn
j=1
fying (8.23), we have

E(1; λ; g) = E(n) (s; Y).

(2) Let s be a variable characterized by the condition 2ns − n + 1 := z1 − z2 . Then

E Λ(g); s = Resz2 −z3 =1 Resz3 −z4 =1 · · · Reszn−1 −zn =1 E(1; z1 , z2 , . . . , zn ; g).

8.1.4 Zeta Function for SL n

By Corollary 8.1 and Proposition 8.2, to obtain an explicit expression for the rank
ζQ,n (s), it suffices to evaluate the integration
n zeta function b
Z
Resz2 −z3 =1 · · · Resz3 −z4 =1 · · · Reszn−1 −zn =1 E(1; z1 , z2 , . . . , zn ; g) dµ(g).
 
MQ,n [1]

Moreover, by Theorem 8.2, f the moduli spaces MQ,n [1] (of semi-stable lattices
of rank n and volume one) is compact, we can freely interchange the orders of
R
(a) the operation of taking the integration on M [1] , and
Q,n
(b) the operation of taking the residues Resz2 −z3 =1 · · · Resz3 −z4 =1 · · · Reszn−1 −zn =1 .

Therefore, we are led to evaluate

Z !
Resz2 −z3 =1 · · · Resz3 −z4 =1 · · · Reszn−1 −zn =1 E(1; z1 , z2 , . . . , zn ; g) dµ(g) .
MQ,n [1]

Hence, by the theory of Eisenstein periods developed in Chapter, in particular, by

Theorem 8.2, we have 1MQ,n [1] = Λ0 1, and
Z
E(1; z1 , z2 , . . . , zn ; g) dµ(g)
MQ,n [1]
Z
= Λ0 1(g) · E(1; z1 , z2 , . . . , zn ; g) dµ(g) (8.24)
SLn (Z)\SLn (R)/SOn
Z
= Λ0 E(1; z1 , z2 , . . . , zn ; g) dµ(g).
SLn (Z)\SLn (R)/SOn

Here in the last step, we have used the fact that Λ0 is unipotent and self-adjoint.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 215

Zeta Functions for Reductive Groups 215

Up to this point, it is only natural to recall that, for T ∈ a+0 sufficiently regular,
we have, by Corollary 8.2 and Theorem 7.5,
Z
E(1; z1 , z2 , . . . , zn ; g) dµ(g)
Σ(T )
Z
= ΛT E(1; z1 , z2 , . . . , zn ; g) dµ(g) (8.25)
SLn (Z)\SLn (R)/SOn

ehwλ−ρ,T i ζ hλ, α∨ i
X Y b

=
.
α∈∆ hwλ − ρ, α iα>0,wα<0 b
∨
Q
w∈W ζ hλ, α∨ i + 1
Here Σ(T ) denotes the compact subset in (a fundamental domain of the space)
SLn (Z)\SLn (R)/SOn with characteristic function ΛT 1. Therefore, by Proposi-
tion 7.4 and Theorem 8.2, (8.25) holed for all T ∈ a+0 t {0}. All these then prove
the following:

Theorem 8.2. For all T ∈ a+0 t {0} and λ = (z1 , . . . , zn ) satisfying ni=1 zi = 0,
P

ehwλ−ρ,T i ζ hλ, α∨ i
Z

E(1; λ; g) dµ(g) =
X Y b

. (8.26)
α∈∆ hwλ − ρ, α i α>0,wα<0 b
∨
Q
Σ(T ) w∈W ζ hλ, α∨ i + 1

Motivated by this, we in a modification of the period of SLn over Q.

Definition 8.3. Let λ = (z1 , z2 , . . . , zn ) satisfies z1 + z2 + · · · + zn = 0.

(1) The T -period ωQ,T

SLn
(λ) of SLn over Q is defined by

ehwλ−ρ,T i ζ hλ, α∨ i
X Y b

ωQ,T
SLn
(z1 , z2 , . . . , zn ) =
.
α∈∆ hwλ − ρ, α i α>0,wα<0 b
∨
Q
w∈W ζ hλ, α∨ i + 1
SLn
(2) The T -version zeta function ZQ,T (z1 ) of SLn over Q is defined by
 SL 
SLn
ZQ,T (z1 ) := Resz2 −z3 =1 Resz3 −z4 =1 . . . Reszn−1 −zn =1 ωQ,T
n
(λ)
= Resz2 −z3 =1 Resz3 −z4 =1 . . . Reszn−1 −zn =1
(8.27)
ehwλ−ρ,T i ζ hλ, α∨ i
X Y b
!

.
α∈∆ hwλ − ρ, α iα>0,wα<0 b
∨
Q
w∈W ζ hλ, α∨ i + 1
In particular, ωQ,0
SLn SLn
(λ) and ZQ,0 (z1 ) are called the period and the zeta function,
of SLn over Q respectively, and denote by ωQSLn (z1 ) and ZQSLn (z1 ), respectively.

Normally, after taking residues in (8.27), there are still zeta factors left in the
denominators for terms corresponding to w ∈ W, despite of the cancelations of
the same zeta factors in both denominator and numerators (with possible uses of
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 216

216 Zeta Functions for Reductive Groups

the functional equation of Riemann zeta function). We want to eliminate all such
reminding zeta factors (in the denominators). For this, we can form the ‘least
common multiple’ of them, based on the following observations.

(i) There are finitely many of non-trivial zeta functions, say, I(SLn ) of them,
   SL   SL 
ζ a1SLn z1 + b1SLn , b
b ζ a2 n z1 + b2SLn , . . . , b
ζ aI(SLn
n)
z1 + bI(SL
SLn
n)
,
 QI(SL )  SL 
satisfying the condition that the product i=1
n
ζ ai n z1 + biSLn ·ZQSLn (z1 )
b
admits only finitely many singularities.
(ii) There are finitely many factors of special zeta values, say J(SLn ) of them,
   SL   SL 
bζ c1SLn , b
ζ c2 n , . . . , bζ c J(SL
n
n)
,

satisfying the condition that there  SL  ζ value

is no special factor left in any of the
b
 Q J(SL
n) b
denominators of the product i=1 ζ c i
n
· ZQ
SLn
(z1 ).

Definition 8.4. The zeta function bζQ;o

SLn
of SLn over Q is defined by
 
Yn ) 
  I(SL Yn ) 
 J(SL  SL  
bζQ;o s := 
SLn  ζ ai s + bi
b SLn SLn
· ζ c j  · ZQ n s .
b SLn 
(8.28)
i=1 j=1

These new zeta functions are extremely beautiful, since they satisfy the func-
tional equation and a weak Riemann Hypothesis.

Theorem 8.3 (Functional Equation). There exists a constant c SLn such that

ζQ;o
SLn
c SLn − s = b
ζQ;o s.
SLn



b (8.29)

This result in the case n = 2, 3, 4, 5 was first verified by the author in [126],
and was jointly proved by Kim and myself ([58]). In the next chapter, we will
ζQ;o
explicitly write done bSLn

s and consequently obtain a new proof.
 
ζQSLn s of SLn over Q is defined by
Definition 8.5. The zeta function b
 c SLn − 1 
ζQSLn (s) := b
b ζQ;o
SLn
s+ . (8.30)
2
Accordingly, the functional equation becomes

ζQSLn (1 − s) = b
b ζQSLn (s). (8.31)

Moreover, beyond everyone’s expectation, we have the following beautiful:

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 217

Zeta Functions for Reductive Groups 217

ζQSLn(s)
Theorem 8.4 (Weak Riemann Hypothesis). All but finitely many zeros of b
1
lie on the central line <(s) = .
2
This result will be proved in Part 6. In the case of n = 2, resp. 3, resp. 4, 5, the
Riemann Hypothesis for b ζQSLn (s) is proved by [65], resp. [109], resp. [55], earlier.

8.2 From SL n to Sp2n: Role of Periods

8.2.1 Sp2n-Periods

Motivated by the discussion in the previous subsection, we next use the periods
to introduce zeta functions for other types of reductive groups. We start with the
symplectic groups. Let G = Sp2n be the symplectic group of size n. Denote √ by Sn
to be the Siegel upper half space of rank n. For any Z ∈ Sn , write Z = X + −1Y
with X its real part and Y its imaginary parts. By definition, Y ∈ Sn if and only if
Y = Im Z > 0 and Z t = Z is symmetric. There is a natural action of Sp2n on Sn
defined as follows:
!
A B
(M, Z) 7→ MhZi := (AZ + B) · (CZ + D)−1 ∀M = ∈ Sp2n (R), ∀Z ∈ Sn .
CD
Set also Y(M) := ImMhZi. It is well known that the √ action of Spn (R) on Sn is
transitive and the stabilizer group of the element −1 I in Sp2n (R) is equal to
Sp2n (R) ∩ SO(2n). Consequently, there is a natural isomorphism
Sp2n (R) SO(2n) ∩ Sp2n (R) ' Sn .


(8.32)
By an abuse of notations, set Γn := diag(±1, ±1, · · · , ±1) ∩Sp2n (Z) Sp2n (Z)


( ! )
∗∗
be the Siegel modular group, and let P = Pn := ∈ Γn be the associated
0∗
maximal parabolic subgroup. Moreover, when Z ∈ Sn , we define the Siegel-Maaβ
Eisenstein series, or more correctly, the Siegel-Epstein zeta function by
X |Y|−s
En (Z; s) := . (8.33)
γ∈P\Γ
kCZ + Dk−2s
n

Motivated by our discussions on the periods and zeta functions associated to SLn ,
we introduce the following:

Definition 8.6. For a T in the positive cone of the root space associated to Sp2n ,
the T -version period of a single variable for Sp2n over Q is defined by
Z
Sp
zQ,T2n (s) := ΛT En (Z; s) dµ(Z). (8.34)
Γn \Sn
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 218

218 Zeta Functions for Reductive Groups

Sp Sp
In particular, we call zQ,02n (s) := ζQ,T2n (s) T =0 the principal zeta function for Sp2n
Sp
over Q and denote it simply by zQ 2n (s).

Similar to the case of SLn , to write down these integral periods explicitly, we
make some preparations.

8.2.2 Siegel Eisenstein Series

( ! )
A∗
As usual, for a fixed n1 ∈ Z>0 , denote by Pn1 ,1,...,1 := ∈ Γn the standard
0 B
parabolic subgroup of Sp2n associated to the partition n = n1 + 1 + 1 + · · · + 1,
 t   −1 
H 0 · · · 0 H ∗ · · · ∗

 ∗ 1 · · · 0
 
 0 1 · · · ∗

where A =  . . . .  and B =  . . . .  with det(H) = 1. Accordingly,
 .. .. . . ..   .. .. . . .. 
  
   
∗ ∗ ··· 1 0 0 ··· 1
we define the associated Siegel Eisenstein series by
X Y n
En1 (Z; sn1 , . . . , sn ) := |Y(γ)v )|−sv . (8.35)
γ∈Pr \Γ v=r

These Siegel Eisenstein series are closed related to the Siegel zeta functions asso-
ciated to the standard parabolic subgroup Qn1 ,1,...,1 of SLn .

Proposition 8.3. ([22]) We have

X  
En1 (Z; sn1 , . . . , sn ) = |Y(γ)|−sn · Zn∗1 Y(γ); sn1 , . . . , sn−1 .
γ∈P1,...,1 \Γn

Moreover, by Lemma 8.1, up to a certain constant depending only on n1 ,

n1
Res sn = r+1 Zn∗1 (Y; sn1 , . . . , sn−1 ) = cn1 Zn∗1 +1 (Y; sn1 +1 + , sn1 +2 , . . . , sn−1 ). (8.36)
1 2 2
Therefore, by taking the successive residues, up to constant factors, we have
  n−1
Res sn−1 =1 · · · Res s2 =1 Res s1 =1 Z1∗ (Y; s1 , s2 , . . . , sn−1 ) = |Y|− 2 (8.37)
and hence
 
Res sn−1 =1 · · · Res s2 =1 Res s1 =1 En1 (Z; sn1 , . . . , sn )
X  

= |Y(γ)|−sn Res sn−1 =1 · · · Res s2 =1 Res s1 =1 Zn∗1 Y(γ); sn1 , . . . , sn−1
γ∈P1,...,1 \Γn (8.38)
!
X
− n−1 n−1
= |Y(γ)|−sn · |Y(γ)| 2 = En Z; sn + .
γ∈P1,...,1 \Γn
2
January 18, 2018 9:46 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 219

Zeta Functions for Reductive Groups 219

8.2.3 Siegel Eisenstein Series and Langlands Eisenstein Series

To use our formulas for the classical Siegel Eisenstein periods above, similar to
the case for SLn , we next write them in terms of Langlands’ language.
√ λ = (z1 , z2 , . . . , zn ) be an element in the root space a0 of Sp2n . For Z =
Let
X + −1 Y ∈ Sn , set av = |Yv |/|Yv−1 | and a := (av ). Then
Yn
aλ (Z) = v = |Y1 |
a−z v −z1 +z2
|Y2 |−z2 +23 · · · |Yn−1 |−zn−1 +zn |Yn |−zn (8.39)
v=1
for s1 = z1 − z2 , s2 = z2 − z3 , . . . , sn−1 = zn−1 − zn , sn = zn . Accordingly, define
the associated power function by
n
∀s = (s1 , s2 , . . . , sn ).
Y
p−s (Y) := |Yµ |−sµ (8.40)
µ=1
Then, by (8.39) and (8.38), we obtain (1) and (2) below respectively.

Proposition 8.4. ([22]) With the same notation as above, we have

(1) E(1, λ, Y) = E1 (Z; s1 , s2 , . . . , sn ).

(2) Up to a constant factor,
n−1
) = Reszn−1 −zn =1 · · · Resz2 −z3 =1 Resz1 −z2 =1 E(1, z1 , z2 , . . . , zn , Y) .
 
En (Z, zn +
2
For example, when n = 2, i.e, for Sp(4), we have
1
Rest−s=1 E(1; t, s; Y) = En Z, s + .
  
(8.41)
2

8.2.4 Siegel-Maaβ-Eisenstein Period and Zeta Function of Sp2n

By Theorem 7.5 in Part 3, and the Gindikin-Karpelevich formula for the associated
intertwining operator, we arrive at
ehwλ−ρ,T i ζ hλ, α∨ i
Z

ΛT E(1; λ; M) dµ(M) =
X Y b
·
.
α∈∆ hwλ − ρ, α i α>0,wα<0 b
∨
Q
Sp(n,Z)\Sn w∈W ζ hλ, α∨ i + 1
Sp
This is nothing but the period ωQ,T2n (λ) of Sp2n introduced in Definition 7.7. Similar
to the group of SLn , by Proposition 8.4, we are led to the following:
Sp
Definition 8.7. The T -version zeta function ZQ,T2n (z1 ) of Sp2n over Q is defined by
Sp
ZQ,T2n (z1 ) := Resz1 −z2 =1 Resz2 −z3 =1 · · · Reszn−1 −zn =1
ehwλ−ρ,T i ζ hλ, α∨ i (8.42)
X Y b
!
·
.
α∈∆ hwλ − ρ, α i α>0,wα<0 b
∨
Q
w∈W ζ hλ, α∨ i + 1
Here λ = (z1 , z2 , . . . , zn ).
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 220

220 Zeta Functions for Reductive Groups

Sp
By Proposition 8.4, we have up to an affine change of variables, ZQ,T2n (z1 ) co-
Sp
incides with the period of single variable zQ,T2n (z1 ) of Sp2n in Definition 8.6.
Similar to the discussion after Definition
n  Sp  8.3, for Sp2n , we obtain finitely many
Sp
o
ζ ai 2n z1 + bi 2n : ai , 0, i = 1, . . . , I(Sp2n ) and finitely
factors of zeta functions b
n  Sp  o
many factors of special zeta values b ζ c 2n : j = 1, . . . J(Sp2n ) .
j

Sp
Definition 8.8. The zeta function bζQ;o2n of Sp2n over Q is defined by
Y2n )  Sp
  I(Sp Y2n )  Sp  Sp  
 J(Sp
 
Sp2n Sp2n
ζQ,T ;o s := 
b  ζ ai s + bi
b 2n
· ζ c j  · ZQ 2n s ,
b 2n 
<(s)  0.
i=1 j=1

Initial examples and results strongly suggest these zeta functions satisfy the
functional equation and the Riemann Hypothesis. In particular, we have

Theorem 8.5. ([110, 126]) Both zeta functions associated to Sp4 over Q satisfy
the functional equation and the Riemann Hypothesis.

8.3 Zeta Functions for G2

8.3.1 Looking for Zeta Functions of G2

8.3.1.1 Role of Maximal Parabolic Subgroups

To introduce new genuine zeta functions for a general reductive group G, we need
to take residues along some singular hyperplanes for the period
ζ hλ, α∨ i


X 1 Y b
ωQ (λ) :=
G
·
.
α∈∆ hwλ − ρ, α i α>0,wα<0 b
∨
Q
w∈W ζ hλ, α∨ i + 1
In the case when G = SLn or SP2n , this is done with the help of the so-called
Epstein zeta functions, a special type of single variable Eisenstein series. For a
general G, this proves to be too much to be expected. Above all, we should guar-
antee that the final singular variable zeta functions for reductive groups obtained
should satisfy the functional equation and the Riemann Hypothesis numerically.
Obviously, among all classical groups, we should start with the groups with
very lower ranks and smaller Weyl groups. By examining groups of types Bn , Dn ,
E6,7,8 , F4 and G2 , it is then only natural for us to focus on the exceptional and
interesting G2 , which is of rank two and admits only 12 Weyl elements.
Accordingly, in the sequel till the end of this subsection, let G be the ex-
ceptional group G2 . Fix a maximal split torus T in G and a Borel subgroup B
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 221

Zeta Functions for Reductive Groups 221

containing T . There are only two simple roots in ∆, namely, the short root αshort
and the long root αlong . For simplicity, denote them by α and β respectively. Then

Φ+ = α, β, α + β, 2α + β, 3α + β, 3α + 2β .
n o
(8.43)

Denote by Pα = P1 and Pβ = P2 the maximal standard parabolic subgroups

attached to α and β respectively, and by Mα , Mβ the corresponding Levi which are
known to isomorphic to GL2 . (See e.g., [43])
Fix a parametrization t : Q ∗ × Q∗ → T, (a, b) 7→ t(a, b) by

α(t(a, b)) = ab−1 and β(t(a, b)) = a−1 b2 . (8.44)

Then the actions of remaining positive roots are given by

(α + β)(t(a, b)) = b, (2α + β)(t(a, b)) = a,

(8.45)
(3α + β)(t(a, b)) = a2 b−1 , (3α + 2β)(t(a, b)) = ab.

And the corresponding coroot are given by

α∨ (x) = t(x, x−1 ), β∨ (x) = t(1, x), (α + β)∨ (x) = t(x, x2 ),

(2α + β)∨ (x) = t(x2 , x), (3α + β)∨ (x) = t(x, 1), (3α + 2β)∨ (x) = t(x, x).

Let X(T ) be the character group of T and a∗C = X(T )⊗C. Introduce coordinates
in a∗C with the basis 2α + β, α + β. Then the point (z1 , z2 ) ∈ C2 corresponds to
the character λ = z1 (2α + β) + z2 (α + β), so that λ(t(a, b)) = |a|z1 |b|z2 . As such,
ρ0 = 5α + 3β and C of the positive Weyl chamber in a∗C is characterized by
n o n o
C := λ ∈ a∗C : < hλ, γ∨ i > 0∀γ > 0 = z1 (2α + β) + z2 (α + β) : <z1 > <z2 > 0 .

For a positive root γ ∈ Φ+ , denote by wγ the reflection associated to γ, i.e. the

reflection on the space a∗C which reflects γ to −γ. And denote by σ(ω) the rotation
through ω with center at the origin. Then the Weyl group of G2 is given by

e, wα , wβ , w3α+β , w2α+β , w3α+2β , wα+β ,

 


 

W= .
 
π (8.46)
 
2π 4π 5π
 σ( ), σ( ), σ(π), σ( ), σ( ) 

 

 
3 3 3 3


By a direct calculation, we have the following table on wλ and {γ > 0 | wγ < 0}.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 222

222 Zeta Functions for Reductive Groups

w wλ; λ = (z1 , z2 ) {γ > 0 | wγ < 0}

e (z1 , z2 ) −
wα (z2 , z1 ) α
wβ (z1 + z2 , −z2 ) β
w3α+β (−z1 , z1 + z2 ) α, 3α + β, 2α + β
w2α+β (−z1 − z2 , z2 ) α, 3α + β, 2α + β, 3α + 2β, α + β
w3α+2β (−z2 , −z1 ) 3α + β, 2α + β, 3α + 2β, α + β, β
wα+β (z1 , −z1 − z2 ) 3α + 2β, α + β, β
σ( π3 ) (−z2 , z1 + z2 ) α + β, β
σ( 2π3 ) (−z1 − z2 , z1 ) 2α + β, 3α + 2β, α + β, β
σ(π) (−z1 , −z2 ) α, 3α + β, 2α + β, 3α + 2β, α + β, β
σ( 4π3 ) (z2 , −z1 − z2 ) α, 3α + β, 2α + β, 3α + 2β
σ( 3 )
5π
(z1 + z2 , −z1 ) α, 3α + β

Table 8.1: Actions of Weyl elements on positive roots for G2

In addition, for λ = (z1 , z2 ), from the definition, since

λ(t(x, x−1 )) = xz1 x−z2 = xz1 −z2 ,

λ(t(1, x)) = 1z1 xz2 = xz2 ,
λ(t(x, 1)) = xz1 1z2 = xz1 ,
(8.47)
λ(t(x2 , x)) = x2z1 xz2 = x2z1 +z2 ,
λ(t(x, x)) = xz1 xz2 = xz1 +z2 ,
λ(t(x, x2 )) = xz1 x2z2 = xz1 +2z2 ,

we have

hλ, α∨ i = z1 − z2 ,
hλ, β∨ i = z2 ,
hλ, (3α + β)∨ i = z1 ,
(8.48)
hλ, (2α + β)∨ i = 2z1 + z2 ,
hλ, (3α + 2β)∨ i = z1 + z2 ,
hλ, (α + β)∨ i = z1 + 2z2 .

Hence, by some tedious calculations, we have the following two tables.

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 223

Zeta Functions for Reductive Groups 223

w hwλ, α∨ i − 1 hwλ, β∨ i − 1
e z1 − z2 − 1 z2 − 1
wα z2 − z1 − 1 z1 − 1
wβ z1 + 2z2 − 1 −z2 − 1
w3α+β −2z1 − z2 − 1 z1 + z2 − 1
w2α+β −z1 − 2z2 − 1 z2 − 1
w3α+2β z1 − z2 − 1 −z1 − 1
wα+β 2z1 + z2 − 1 −z1 − z2 − 1
σ( π3 ) −z1 − 2z2 − 1 z1 + z2 − 1
σ( 2π3 ) −2z1 − z2 − 1 z1 − 1
σ(π) −z1 + z2 − 1 −z2 − 1
σ( 4π3 ) z1 + 2z2 − 1 −z1 − z2 − 1
σ( 3 )
5π
2z1 + z2 − 1 −z1 − 1

Table 8.2: Factors hwλ, α∨ i − 1 and hwλ, β∨ i − 1

Q ζ(hλ,γ∨ i)
b
w γ>0,wγ<0 b
ζ(hλ,γ∨ i+1)
e 1
ζ(z1 −z2 )
b
wα ζ(z1 −z2 +1)
b
bζ(z2 )
wβ bζ(z2 +1)
bζ(z1 −z2 ) b ζ(z1 ) b ζ(2z1 +z2 )
w3α+β bζ(z1 −z2 +1) b ζ(z1 +1) b ζ(2z1 +z2 +1)
ζ(z1 −z2 ) b
b ζ(z1 ) b ζ(2z1 +z2 ) b ζ(z1 +z2 ) b ζ(z1 +2z2 )
w2α+β ζ(z1 −z2 +1) b
b ζ(z1 +1) bζ(2z1 +z2 +1) b ζ(z1 +z2 +1) b ζ(z1 +2z2 +1)
ζ(z1 ) b
b ζ(2z1 +z2 ) b ζ(z1 +z2 ) b ζ(z1 +2z2 ) b ζ(z2 )
w3α+2β ζ(z1 +1) b
b ζ(2z1 +z2 +1) b ζ(z1 +z2 +1) b ζ(z1 +2z2 +1) b ζ(z2 +1)
bζ(z1 +z2 ) b ζ(z1 +2z2 ) b ζ(z2 )
wα+β bζ(z1 +z2 +1) b ζ(z1 +2z2 +1) b ζ(z2 +1)
ζ(z1 +2z2 ) b ζ(z2 )
σ( π3 )
b
ζ(z1 +2z2 +1) b
b ζ(z2 +1)
ζ(2z1 +z2 ) b ζ(z1 +z2 ) b ζ(z1 +2z2 ) b ζ(z2 )
σ( 2π
3 )
b
ζ(2z1 +z2 +1) b
b ζ(z1 +z2 +1) b ζ(z1 +2z2 +1) b ζ(z2 +1)
ζ(z1 −z2 ) b ζ(z1 ) b ζ(2z1 +z2 ) b ζ(z1 +z2 ) b ζ(z1 +2z2 ) b ζ(z2 )
σ(π)
b
ζ(z1 −z2 +1) b
b ζ(z1 +1) b ζ(2z1 +z2 +1) b ζ(z1 +z2 +1) b ζ(z1 +2z2 +1) b ζ(z2 +1)
ζ(z1 −z2 ) b ζ(z1 ) b ζ(2z1 +z2 ) b ζ(z1 +z2 )
σ( 4π
3 )
b
bζ(z1 −z2 +1) b ζ(z1 +1) b ζ(2z1 +z2 +1) b ζ(z1 +z2 +1)
ζ(z1 −z2 ) b ζ(z1 )
σ( 5π
3 )
b
ζ(z1 −z2 +1) b
b ζ(z1 +1)

Q ζ(hλ,γ∨ i)
Table 8.3: Factors
b
γ>0,wγ<0 b
ζ(hλ,γ∨ i+1)
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 224

224 Zeta Functions for Reductive Groups

Combining Tables 8.2 and 8.3, we obtain

1 1 Q ζ(hλ,γ∨ i)
·
b
w hwλ,α∨ i−1 hwλ,β∨ i−1 γ>0,wγ<0 b ζ(hλ,γ∨ i+1)
1 1
e z1 −z2 −1 z2 −1
1 1 ζ(z1 −z2 )
z2 −z1 −1 z1 −1 · b
b
wα ζ(z1 −z2 +1)
1 1 ζ(z2 )
z1 +2z2 −1 −z2 −1 · b
b
wβ ζ(z2 +1)
1 1 ζ(z1 −z2 ) b ζ(z1 ) b ζ(2z1 +z2 )
−2z1 −z2 −1 z1 +z2 −1 · b
b
w3α+β ζ(z1 −z2 +1) b ζ(z1 +1) bζ(2z1 +z2 +1)
1 1 ζ(z1 −z2 ) b ζ(z1 ) b ζ(2z1 +z2 ) b ζ(z1 +z2 ) b ζ(z1 +2z2 )
−z1 −2z2 −1 z2 −1 · b
b
w2α+β ζ(z1 −z2 +1) b ζ(z1 +1) bζ(2z1 +z2 +1) b ζ(z1 +z2 +1) b ζ(z1 +2z2 +1)
1 1 ζ(z1 ) b ζ(2z1 +z2 ) b ζ(z1 +z2 ) b ζ(z1 +2z2 ) b ζ(z2 )
z1 −z2 −1 −z1 −1 · b
b
w3α+2β ζ(z1 +1) b ζ(2z1 +z2 +1) b ζ(z1 +z2 +1) b ζ(z1 +2z2 +1) b ζ(z2 +1)
1 1 ζ(z1 +z2 ) b ζ(z1 +2z2 ) b ζ(z2 )
2z1 +z2 −1 −z1 −z2 −1 · b
b
wα+β ζ(z1 +z2 +1) b ζ(z1 +2z2 +1) b ζ(z2 +1)
ζ(z1 +2z2 ) b ζ(z2 )
σ( π3 ) 1 1
−z1 −2z2 −1 z1 +z2 −1 · b
b
ζ(z1 +2z2 +1) b ζ(z2 +1)
ζ(2z1 +z2 ) b ζ(z1 +z2 ) b ζ(z1 +2z2 ) b ζ(z2 )
σ( 2π
3 )
1 1
−2z1 −z2 −1 z1 −1 · b
b
ζ(2z1 +z2 +1) b ζ(z1 +z2 +1) b ζ(z1 +2z2 +1) b ζ(z2 +1)
ζ(z1 −z2 ) b ζ(z1 ) b ζ(2z1 +z2 ) b ζ(z1 +z2 ) b ζ(z1 +2z2 ) b ζ(z2 )
σ(π) 1 1
−z1 +z2 −1 −z2 −1 · b
b
ζ(z1 −z2 +1) b ζ(z1 +1) bζ(2z1 +z2 +1) b ζ(z1 +z2 +1) b ζ(z1 +2z2 +1) b ζ(z2 +1)
ζ(z1 −z2 ) b ζ(z1 ) b ζ(2z1 +z2 ) b ζ(z1 +z2 )
σ( 4π
3 )
1 1
z1 +2z2 −1 −z1 −z2 −1 · b
b
ζ(z1 −z2 +1) b ζ(z1 +1) bζ(2z1 +z2 +1) b ζ(z1 +z2 +1)
ζ(z1 −z2 ) b ζ(z1 )
σ( 5π
3 )
1 1
2z1 +z2 −1 −z1 −1 · b
b
ζ(z1 −z2 +1) b ζ(z1 +1)

1 1 Q ζ(hλ,γ∨ i)
Table 8.4: Terms ·
b
hwλ,α∨ i−1 hwλ,β∨ i−1 γ>0,wγ<0 b
ζ(hλ,γ∨ i+1)

By taking the summations of all terms in Table 8.4, we obtain the period
ωGQ2 (z1 , z2 ) for G2 over Q. However, it still looks very difficult to determine along
which line we should take the residue, since there are too many to determine.
Back to the general structure, recall that for G2 , the associated period is a two
variable function given by
ζ hλ, α∨ i


X 1 Y b
ωGQ2 (z1 , z2 ) = ·
, (8.49)
w∈W
hwλ − ρ, α∨short i hwλ − ρ, α∨long i α>0,wα<0 bζ hλ, α∨ i + 1
where (z1 , z2 ) = λ ∈ a0 , and ∆ = {αshort , αlong } consisting only two roots, i.e.
the short root α = αshort and the long root β = αlong . In particular, being two
variables, what we should find is merely one singular hyperplane in the (z1 , z2 )-
plane, namely, a line of the form az1 + bz2 + c = 0. To find the singular line,
we first concentrate on the rational factor hwλ−ρ,α∨ i1hwλ−ρ,α∨ i , up to a change of
short long
variables λ 7→ wλ, there are only two of them, namely,
hλ − ρ, α∨short i = 0 and hwλ − ρ, α∨long i = 0. (8.50)
January 18, 2018 9:46 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 225

Zeta Functions for Reductive Groups 225

At this point, by examining what we have done for the groups SLn and Sp2n ,
it is not difficult for us to detect that the singular hyper-planes along which the
residues are

(a) taken in the denominators of the rational functional factors in ωGQ (λ), i.e. in

1
, (8.51)
α∈∆ hwλ − ρ, α i
∨
Q

(b) concentrated on these associated to the identity Weyl element Id, i.e. on
1
. (8.52)
α∈∆ hλ − ρ, α i
∨
Q

This leads, for the group G2 , to the rational functions

1 1
i and , (8.53)
hwλ − ρ, α∨short hwλ − ρ, α∨long i

and hence the singular lines

(1) either the line hλ − ρ, α∨short i = 0, namely the line z1 − z2 − 1 = 0,

(2) or the line hλ − ρ, α∨long i = 0, namely the line z2 − 1 = 0.

In this way, we  obtain  two new zeta functions  for the group G2 , namely,
Reshλ−ρ,α∨short i=0 ωGQ2 (λ) , and Reshλ−ρ,α∨long i=0 ωGQ2 (λ) .
Since there is a one-to-one correspondence between the set of maximal (stan-
dard) parabolic subgroups and the set of simple roots, if we denote by Pshort and
Plong for the maximal parabolic subgroup of G2 corresponding to the (set of) sim-
ple root αlong and αshort , respectively, it is then only natural for us to name the
corresponding zeta functions (after the normalization process similarly to that
G /P
for zeta functions of SLn and Sp2n ) as b ζQ 2 long (s) and bζQG2 /Pshort (s), or the same,
/P
ζ G2 /Pα (s), respectively.
G
ζ 2 β (s) and b
b
Q Q

8.3.1.2 Zeta Function of G2 /Pα

From the period ωGQ2 (z1 , z2 ) of G2 over Q listed in Table 8.4, to obtain a zeta func-
ζ G2 /Pα (s) for G2 /Pα , we first take the residue along with the singular hyper-
tion b Q
plane hλ+ρ0 , α∨ i = 0 of ωGQ2 (z1 , z2 ), namely the line z1 −z2 −1 = 0. Set s = z2 (then
z1 = 1+s, z2 −z1 = −1, 2z1 +z2 = 3s+2, z1 +z2 = 2s+1, z1 +2z2 = 3s+1, z1 −1 = s
and z2 + 1 = s + 1). In such a way, we get the following (single variable) period
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 226

226 Zeta Functions for Reductive Groups

ωGQ2 /Pα (s) associated to G2 /Pα over Q (with the same order as in Table8.4).
ωGQ2 /Pα (s) := Resz1 −s−1=0 ωGQ2 (z1 , s)
1 1 1 1 1 1 1 b ζ(s + 1) bζ(3s + 2)
= + · +0+ ·
s − 1 −2 s b ζ(2) −3s − 3 2s ζ(2) ζ(s + 2) ζ(3s + 3)
b b b
1 1 1 ζ(s + 1) ζ(3s + 2) ζ(2s + 1) b ζ(3s + 1)
+
b b b
·
−3s − 2 s − 1 b ζ(2) bζ(s + 2) b ζ(3s + 3) bζ(2s + 2) b ζ(3s + 2)
1 ζ(s + 1) b ζ(3s + 2) b ζ(2s + 1) bζ(3s + 1) b ζ(s)
+ +0+0+0
b
·
−s − 2 b ζ(s + 2) b ζ(3s + 3) b ζ(2s + 2) bζ(3s + 2) b ζ(s + 1)
1 1 1 bζ(s + 1) b ζ(3s + 2) bζ(2s + 1) b ζ(3s + 1) b ζ(s)
+ ·
−2 −s − 1 b ζ(2) b
ζ(s + 2) b ζ(3s + 3) bζ(2s + 2) b ζ(3s + 2) b ζ(s + 1)
1 1 1 ζ(s + 1) ζ(3s + 2) ζ(2s + 1) 1 1 1 b ζ(s + 1)
+ + .
b b b
· ·
3s −2s − 2 b ζ(2) ζ(s + 2) ζ(3s + 3) ζ(2s + 2)
b b b 3s + 1 −s − 2 ζ(2) ζ(s + 2)
b b
Here, we have used the functional equation of b ζ(s) to change b ζ(az + b) with a < 0
or a = 0, b < 0 to ζ(−az − b + 1). Taking the cancellations of the common
b
factors in both the denominator and numerators, and multiplying with the factor
ζ(2) · b
b ζ(s + 2)bζ(2s + 2)b ζ(3s + 3), to clear out all the zeta factors in the determinators
of all terms in ωGQ2 /Pα (s), we arrive at
G2 /Pα 1 b
ζQ,o
b (s) := ζ(2) · b
ζ(s + 2)b ζ(2s + 2)b ζ(3s + 3)
s−1
1 b 1 b
− ζ(2) · bζ(s)b ζ(2s + 1)b ζ(3s + 1) − · ζ(s + 2)b ζ(2s + 2)bζ(3s + 3)
s+2 2s
1 1 1 b
+ ·bζ(s)b ζ(2s + 1)b ζ(3s + 1) − · ζ(s + 1)bζ(2s + 2)b ζ(3s + 2)
2(s + 1) 3s + 3 2s
1 1 b 1 1 b
− ζ(s + 1)b ζ(2s + 1)b ζ(3s + 2)− ζ(s + 1)bζ(2s + 1)bζ(3s + 1)
3s 2s + 2 3s + 2 s − 1
1 1 b
− · ζ(s + 1)bζ(2s + 2)b ζ(3s + 3).
3s + 1 s + 2
G2 /Pα G2 /Pα
One checks easily that b ζQ,o (−1 − s) = b ζQ,o (s).

Definition 8.9. The zeta function b ζQG2 /Pα (s) is defined by

bζ G2 /Pα (s) := b
Q ζ G2 /Pα (s − 1).
Q,o (8.54)
G /Plong
ζQ 2
Proposition 8.5. The zeta function b (s) satisfies the functional equation
G /P G /Plong
ζQ 2 long (1
b − s) = b
ζQ 2 (s). (8.55)
In ζQG2 /P1 (s)
addition, b admits only four singularities, namely, two simple poles and
two double poles located at s = −1, 2 and s = 0, 1, respectively.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 227

Zeta Functions for Reductive Groups 227

8.3.1.3 Zeta Functions of G2 /Pβ

Similarly, by taking the residue along hλ + ρ0 , β∨ i = 0, i.e. the line z2 = 1, we

G /P
obtain the period ωQ2 β (s) with s = z1 from

G /P G /P
ζQ,o2 β (s) := Reshλ+ρ0 ,β∨ i=0 ωQ2 β (z1 , z2 )
b
1 1 1 1
= +0+ · +0
s−2 s + 1 −2 b ζ(2)
1 ζ(s − 1) b ζ(s) b ζ(2s + 1) b
ζ(s + 1) b
ζ(s + 2)
+
b
·
−s − 3 ζ(s) ζ(s + 1) ζ(2s + 2) ζ(s + 2) ζ(s + 3)
b b b b b
1 1 ζ(s) b ζ(2s + 1) b
ζ(s + 1) b
ζ(s + 2) 1
+
b
·
s − 2 −s − 1 b ζ(s + 1) b
ζ(2s + 2) b
ζ(s + 2) b
ζ(s + 3) bζ(2)
1 1 ζ(s + 1) b
ζ(s + 2) 1 1 1 b ζ(s + 2) 1
+ +
b
· ·
2s −s − 2 b ζ(s + 2) b
ζ(s + 3) b
ζ(2) −s − 3 s b ζ(s + 3) b
ζ(2)
1 1 b ζ(s − 1) b ζ(s) b ζ(2s + 1) b
ζ(s + 1) b
ζ(s + 2) 1
+ · + 0 + 0.
−s −2 ζ(s) ζ(s + 1) ζ(2s + 2) ζ(s + 2) ζ(s + 3) b
b b b b b ζ(2)

ζ(2) · b
Multiplying with b ζ(s + 3)b
ζ(2s + 2), and shifting from s to s − 1 we arrive at
the following:

G /Pβ
ζQ 2
Definition 8.10. The zeta function b (s) for G2 /Pβ is defined by

G /P 1 b 1 b
ζQ 2 β (s) :=
b ζ(2) · b
ζ(s + 2)b
ζ(2s) − ζ(2) · b
ζ(s − 2)b
ζ(2s − 1)
s−3 s+2
1 1 b
+ ·bζ(s − 2)b
ζ(2s − 1) − · ζ(s + 2)b
ζ(2s)
2s − 2 2s
1 1
− ζ(s − 1)b
·b ζ(2s − 1) − ·bζ(s + 1)b
ζ(2s)
s(s − 3) (s − 1)(s + 2)
1 1
− ζ(s)b
·b ζ(2s) − ζ(s)b
·b ζ(2s − 1).
(2s − 2)(s + 1) (2s)(s − 2)

G /Pβ
ζQ 2
Proposition 8.6 (Functional Equation). The zeta function b (s) for G2 /Pβ
satisfies the functional equation
G /P G /P
ζQ 2 β (1 − s) = b
b ζQ 2 β (s).

G /P
In addition, bζQ 2 β (s) admits 4 singularities, namely, four simple poles located at
s = −2, 0, 1, 3.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 228

228 Zeta Functions for Reductive Groups

8.3.2 RH for Zeta Functions of G2 I: Preparations

8.3.2.1 Two Elementary Lemmas

The first is about a refine statement of the Hadamard products for entire functions.

Lemma 8.3 ([110]). Let F(s) be an entire function of genus zero or one. Assume

(a) F(s) is real on the real axis,

(b) there exists σ0 > 0 such that, except for finitely many zeros, all zeros of F(s)
lie in the vertical strip σ0 < <(s) < 1/2,
(c) there are only finitely many zeros of F(s) in the region <(s) ≥ 1/2,
(d) there exists C > 0 such that
N(T ) ≤ CT log T as T → ∞, (8.56)
where N(T ) denotes the number of zeros of F(s) in the region 0 ≤ =(ρ) < T ,
F(1 − σ)
(e) for large σ > 0, F(1 − σ)/F(σ) > 0 and, as σ → ∞, → 0.
F(σ)
Then F(s) admits a Hadamard product
" #
0
Y s Y  s  s
F(s) = Csm eB s 1− 1− 1−
0,ρ∈R
ρ ρ ρ̄
=(ρ)>0
0
with B ≥ 0. Moreover, the product in the right-hand side converges absolutely on
every compact set.

Proof. This is rather standard. We only treat the case that F(s) is of genus one,
since the genus zero case can be proved similarly. Obviously, F(s) is of genus one
if and only if the Hadamard product factorization
Y s
F(s) = eA+Bs sm 1− exp(s/ρ) (m ∈ Z≥0 ) (8.57)
ρ
ρ

converges absolutely and uniformly on any compact subsets of C if and only if

< ∞. In addition, (a) implies that F(ρ) = 0 if and only if F(ρ) = 0.
P −2
ρ |ρ|
Consequently, we may divide the set of zeros ρ = β + iγ, counted with multiplicity
into subsets B(ρ) consisting of {ρ, ρ} if γ > 0 and {ρ} if β , 0 and γ = 0. We fix a
unique zero in B(ρ) satisfying γ ≥ 0. Thus (b) implies that
 
m A+B0 s
Y  Y  s 
F(s) = s e  1 −  (8.58)
ρ

B(ρ) ρ∈B(ρ)

where the outer product on the right-hand side converges absolutely and uni-
formly on any compact subsets of C. Indeed, for each B(ρ), the convergence
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 229

Zeta Functions for Reductive Groups 229

factor exp(c(B(ρ))s) coincides with c(B(ρ)) = 2β|ρ|−2 for γ > 0. But (b) implies
|β − 1/2| < σ0 , and hence
X X X
|c(B(ρ))| ≤ |ρ|−1 + (2σ0 + 1) |ρ|−2 < ∞.
B(ρ) 0,ρ: real ρ

Therefore, the convergence factors exp(c(B(ρ))s)

X can be pulled out from the orig-
inal product. This gives (8.58) with B0 = B + c(B(ρ)).
B(ρ)

Claim 8.1. We have B0 ≥ 0.

Proof. We use conditions (c), (d) and (e) to verify it. Indeed, by (e) we have
F(1 − σ)
!
R 3 log → −∞ as σ → +∞. (8.59)
F(σ)
On the other hand, by (8.58) we have
F(1 − σ) 0
 σ − 1 m Y σ − 1 + β Y (σ − 1 + β)2 + γ2
= eB (1−2σ) .
F(σ) σ ρ=β∈R
σ − β ρ=β+iγ (σ − β)2 + γ2
γ>0

This implies that

 F(1 − σ)   1 X  1 − 2β 
log = B0 (1 − 2σ) + m log 1 − + log 1 −
F(σ) σ ρ=β∈R
σ−β
X  (1 − 2β)(2σ − 1)  (8.60)
+ log 1 − .
ρ=β+iγ
(σ − β)2 + γ2
γ>0

Note that, for β < 1/2,

 (1 − 2β)(2σ − 1) 
log 1 − < 0 for σ > 1/2, (8.61)
(σ − β)2 + γ2
     
and, for any fixed ρ = β + iγ, log 1 − σ1 , log 1 − 1−2β (1−2β)(2σ−1)
σ−β , log 1 − (σ−β)2 +γ2 →0
as σ → +∞. Hence, by (c), (8.61) holds except for finitely many zeros. Thus, if
we assume that B0 < 0, by (8.59) and (8.60), for sufficiently large σ > 1/2,
X  (1 − 2β)(2σ − 1) 
log 1 − ≥ 2|B0 |σ (8.62)
ρ=β+iγ
(σ − β) 2 + γ2
γ>0

since the number of real zeros is also finite by (b) and (c). But, for these same σ,
X  (1 − 2β)(2σ − 1)  X  (1 − 2σ0 )(2σ − 1) 
log 1 − ≤ log 1 −
ρ=β+iγ
(σ − β)2 + γ2 ρ=β+iγ
(σ − 1/2)2 + γ2
γ>0 γ>0
X 1
 (2σ − 1) ,
ρ=β+iγ
(σ − 1/2)2 + γ2
γ>0
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 230

230 Zeta Functions for Reductive Groups

Z ∞
dN(t)
which is nothing but the Stieltjes integral . Thus, by (d),
γ0 (σ − 1/2)2 + t2
∞ ∞
log(σ + γ0 )
Z Z
dN(t) (log t) dt
  .
γ0 (σ − 1/2)2 + t2 γ0 (σ − 1/2) + t
2 2 σ − 1/2

X  (1 − 2β)(2σ − 1) 
This implies that log 1 −  log(σ + γ0 ), which contra-
ρ=β+iγ
(σ − β)2 + γ2
γ>0
dicts with (8.62). Therefore we have verified the claim and the lemma. 

To end this discussion, we recall the following lemma for later use.

Lemma 8.4. ( [65]) Let f be an entire function of genus zero or one such that

(1) f (R) ⊆ R and f (1 − s) = ± f (s) for some choice of sign.

(2) There exists a constant a > 0 satisfying that all zeros of f (s) are contained in
the vertical strip |<(s) − 21 | < a.

Then, for any real constant c ≥ a,

f (s + c) 1 f (s + c) 1
> 1 <(s) > and < 1 <(s) < . (8.63)
f (s − c) 2 f (s − c) 2
In particular, for two real numbers satisfying a > b,
ξ(s + b) ξ(s + b)
! !
a+b−1 a+b−1
< 1 <(s) > − , = 1 <(s) = − . (8.64)
ξ(s + a) 2 ξ(s + a) 2

Proof. The first inequality, resp. the second equality, of (11.26) is a direct con-
sequence of (8.63), resp. the functional equation for ξ(s), it suffices to prove the
assertions in 11.26. The arguments before Claim 8.1 in the proof of the previous
lemma are valid for f (s) as well. Hence, we have the Hadamard product for f (s)
in the form
 
A+B0 s
Y  Y  s 
f (s) = e  1 − 
ρ

B(ρ) ρ∈B(ρ)

where, if we write ρ = β + iγ, B(ρ) consists of {ρ, 1 − ρ, ρ, 1 − ρ} if β , 12 ; {ρ, 1 − ρ}

of β = 21 and γ , 0 or β , 12 and γ = 0; and { 21 } if ρ = 12 . Each block is labeled
with the unique ρ for which β ≤ 21 and γ ≥ 0. Consequently, using the functional
equation condition in (1), we have B0 = 0 and hence
 
Y  Y  s 
f (s) = e A  1 −  . (8.65)
B(ρ) ρ∈B(ρ)
ρ
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 231

Zeta Functions for Reductive Groups 231

Indeed, with respect to the interchange s 7→ 1 − s, except the block B( 12 ), the

factors for other blocks B(ρ) permutes, and for the block B( 12 ), it merely yields
0 0
a sign change. This implies that eA+B s = eA+B (1−s) by the functional equation.
Therefore, B0 = 0. Consequently, we have
Y  Y 1 − s+c
 
f (s + c) ρ 

= 
 s−c 
 . (8.66)
f (s − c) B(ρ) ρ∈B(ρ)
1 − ρ


To complete the proof, it suffices to use the standard properties for the Möbius
s + (c − ρ)
transformation s 7→ τ = , which transforms the half plane <(s) > 21 ,
s − (c + ρ)
resp. <(s) < 12 , to the outside, resp. inside, of the circle |τ| = 1. 

8.3.2.2 Refined Hadamard Products

Following Propositions 8.5 and 8.6, introduce two entire functions by

Z1 (s) := 12s3 (s − 1)3 · (s + 1)(3s − 1)(2s − 1)(3s − 2)(s − 2) · bζQG2 /Pα (s),
G /P (8.67)
Z2 (s) := 4s2 (s − 1)2 · (s + 2)(s + 1)(2s − 1)(s − 2)(s − 3) · b
ζ 2 β (s). Q

Set ξ(s) = s(s − 1)b

ζ(s), A = 2b
ζ(2) − 1 > 0. Then, by Definitions 8.9 and 8.10, for
f1 (s) := (s − 1)2 (3s − 2)(As − A + 1)ξ(s + 1)ξ(3s)
− (s + 1)(s − 1)(s − 2)ξ(s)ξ(3s − 1) − 2(s − 1)2 (s − 2)ξ(s)ξ(3s),
(8.68)
f2 (s) := (As + 3)(s − 1)2 (s − 2)ξ(s + 2)
− 2(s − 1)(s − 2)(s − 3)ξ(s + 1) − (s + 2)(s − 2)(s − 3)ξ(s)
we obtain natural decompositions
Z1 (s) = ξ(2s) f1 (s) − ξ(2s − 1) f1 (1 − s),
(8.69)
Z2 (s) = ξ(2s) f2 (s) − ξ(2s − 1) f2 (1 − s).
As a key step to prove the Riemann Hypothesis of the zeta functions of G2 ,
we next expose a refined Hadamard product structures for fi (s) following Suzuki.

Proposition 8.7. ( [112]) There are constants B0i ≥ 0 (i = 1, 2) such that the
following Hadamard products for the entire functions fi (s) (i = 1, 2) in (8.68) are
convergent absolutely on any compact subset of C
! !
0 s Y s
fi (s) = ci smi eBi s 1 − 1−
β0,i βi <1/2 βi
0,βi ∈R
! ! Y " ! !#
s s s s
× 1− 1− 1− 1− .
ρ0,i ρ0,i <(ρi )<1/2
ρi ρi
=(ρ)i >0

Here, (i) c1 = f10 (0), c2 = f2 (0), m1 = 1, m2 = 0,

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 232

232 Zeta Functions for Reductive Groups

(ii) β0i = i and {βi } ⊂ R are finite,

(iii) ρ0,i < R, <(ρ0,i ) > 1/2, and the ρi ’s are not real.

By Lemma 8.3, to prove this proposition, it suffices to show that both f1 (s)
and f2 (s) satisfy all the five conditions in Lemma 8.3. Obviously, (a) is satisfied.
In addition, if (b) holds, (d) can be easily verified using well-known estimate
|ξ(s)| ≤ exp(C|s| log |s|)
and Jensen’s formula (see §4.1 of [109], for example). Hence it suffices to verify
the conditions (b), (c) and (e) for both f1 (s) and f2 (s), since
f1 (0) = 0, f1 (s) = f10 (0)s + O(s2 ), f10 (0) ' −2.176 , 0, f2 (0) ' −6.283 , 0.

8.3.2.3 Asymptotic Behaviors

Lemma 8.5. For i = 1, 2, we have

fi (1 − σ)
(1) For sufficiently large σ  0, is positive.
f1 (σ)
fi (1 − σ)
(2) As σ → +∞, = O(σ−2 ).
f1 (σ)
Proof. (1) By the functional equation of ξ(s), we have
f1 (1 − σ) = σ2 (3σ − 1)(Aσ − 1)ξ(σ − 1)ξ(3σ − 2)
+ σ(σ − 2)(σ + 1)ξ(σ)ξ(3σ − 1) + 2σ2 (σ + 1)ξ(σ)ξ(3σ − 2).
This implies
f1 (1 − σ) σ2 (3σ−1)(Aσ−1) ξ(σ−1)ξ(3σ−2) 1 + g1 (σ)
=
f1 (σ) (σ−1) (3σ−2)(Aσ−A+1) ξ(σ + 1)ξ(3σ) 1 − h1 (σ)
2
(8.70)
−1
ξ(σ − 1)ξ(3σ − 2) 1 + g1 (σ)
= 1 + O(σ ) · · ,
ξ(σ + 1)ξ(3σ) 1 − h1 (σ)
where
(σ − 2)(σ + 1) ξ(σ)ξ(3σ − 1) 2(σ + 1) ξ(σ)
g1 (σ) = · + · ,
σ(3σ − 1)(Aσ − 1) ξ(σ − 1)ξ(3σ − 2) (3σ − 1)(Aσ − 1) ξ(σ − 1)
(σ + 1)(σ − 2) ξ(σ)ξ(3σ − 1)
h1 (σ) = ·
(σ − 1)(3σ − 2)(Aσ − A + 1) ξ(σ + 1)ξ(3σ)
2(σ − 2) ξ(σ)
+ · .
(3σ − 2)(Aσ − A + 1) ξ(σ + 1)
Clearly, the numerator of (8.70) is positive for large σ > 0. On the other hand, by
Lemma 8.4, (8.63) to obtain
ξ(σ) ξ(3σ − 1)
< 1 (σ > 0), < 1 (σ > 1/3). (8.71)
ξ(σ + 1) ξ(3σ)
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 233

Zeta Functions for Reductive Groups 233

Thus, by A > 0, the denominator is positive for large σ > 0, and

h1 (σ) = O(σ−1 ) as σ → +∞. (8.72)
This verifies (1) for f1 (s).
Similarly, for f2 (s), we have
f2 (1 − σ) σ2 (σ + 1)(Aσ − A − 3) ξ(σ − 2) 1 + g2 (σ)
= · ·
f2 (σ) (σ − 1)2 (σ − 2)(Aσ + 3) ξ(σ + 2) 1 − h2 (σ)
(8.73)
−1
ξ(σ − 2) 1 + g2 (σ)
= 1 + O(σ ) · · ,
ξ(σ + 2) 1 − h2 (σ)
where
2(σ + 2) ξ(σ − 1) (σ + 2)(σ − 3) ξ(σ)
g2 (σ) = · + · ,
σ(Aσ − A − 3) ξ(σ − 2) σ2 (Aσ − A − 3) ξ(σ − 2)
and
2(σ − 3) ξ(σ + 1) (σ + 2)(σ − 3) ξ(σ)
h2 (σ) = · + · .
(σ − 1)(Aσ + 3) ξ(σ + 2) (σ − 1)2 (Aσ + 3) ξ(σ + 2)
Obviously, as σ  0, the numerator of (8.73) is positive. As for the positivity of
the denominator and the estimation
h2 (σ) = O(σ−1 ) as σ → +∞, (8.74)
the same arguments treating (8.70) work here as well if, instead of (8.71), we use
ξ(σ + 1) ξ(σ)
< 1 (σ > −1) and < 1 (σ > 0) (8.75)
ξ(σ + 2) ξ(σ + 2)
ξ(σ) ξ(σ + 1) ξ(σ)
and the trivial relation = .
ξ(σ + 2) ξ(σ + 2) ξ(σ + 1)
(2) Note that, for large σ > 0, by definition, we have
ξ(σ − 1)ξ(3σ − 2) ζ(σ − 1)b ζ(3σ − 2)
= 1 + O(σ−1 )

b
ξ(σ + 1)ξ(3σ) ζ(σ + 1)b
b ζ(3σ)
Γ((σ − 1)/2)Γ((3σ − 2)/2) ζ(σ − 1)ζ(3σ − 2)
= 1 + O(σ−1 ) · π2 ·


Γ((σ + 1)/2)Γ(3σ/2) ζ(σ + 1)ζ(3σ)
Γ((σ − 1)/2)
= 1 + O(σ−1 ) ·


· O(1).
Γ((σ + 1)/2)(3σ − 2)
By the Stirling r formula
2π  z z
Γ(z) = 1 + Oε (|z|−1 ) ( |z| ≥ 1, |arg z| < π − ε ),


(8.76)
z e
we have, as σ → +∞,
ξ(σ − 1)ξ(3σ − 2)
= O(σ−2 ) and g1 (σ) = O(1) + O(σ−1/2 ) = O(1). (8.77)
ξ(σ + 1)ξ(3σ)
Similarly,
ξ(σ − 2)
= O(σ−2 ) and g2 (σ) = O(σ−1/2 )+O(1) = O(1) as σ → +∞. (8.78)
ξ(σ + 2)
Therefore, by (8.72) and (8.77), resp. by (8.74) and (8.78), we obtain
f1 (1 − σ) f2 (1 − σ)
= O(σ−2 ) resp. = O(σ−2 ) as σ → +∞,
f1 (σ) f2 (σ)
as desired. 
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 234

234 Zeta Functions for Reductive Groups

8.3.2.4 Zero Free Regions

Lemma 8.6. There exist constants σ1 and σ1 such that all zeros of f1 (s) and f2 (s)
are contained in the strip σ1 < <(s) < σ2 .

Proof. We rearrange the functions f1 (s) and f2 (s) as

f1 (s) = − (s + 1)(s − 1)(s − 2)ξ(s)ξ(3s − 1) [ 1 + R11 (s) − R12 (s) ] ,
f2 (s) = − (s + 2)(s − 2)(s − 3)ξ(s) [ 1 + R21 (s) − R22 (s) ] ,
where
s−1 ξ(3s)
R11 (s) = 2 ·
s + 1 ξ(3s − 1)
(s − 1)(3s − 2)(As − A + 1) ξ(s + 1)ξ(3s)
R12 (s) = · ,
(s + 1)(s − 2) ξ(s)ξ(3s − 1)
(s − 1) ξ(s + 1) (As + 3)(s − 1)2 ξ(s + 2)
R21 (s) = 2 · , R22 (s) = · .
(s + 2) ξ(s) (s + 2)(s − 3) ξ(s)
Clearly the factor (s + 1)(s − 1)(s − 2)ξ(s)ξ(3s − 1), resp. (s + 2)(s − 2)(s −
3)ξ(s) has no zero in the left-half plane <(s) < −1, resp. <(s) < −2. Using the
functional equation, we have
√
6 πs(s − 1) Γ((1 − 3s)/2) ζ(1 − 3s)
R11 (s) = ,
(3s − 2)(s + 1) Γ((2 − 3s)/2) ζ(2 − 3s)
3πs(As − A + 1) Γ(−s/2)Γ((1 − 3s)/2) ζ(−s)ζ(1 − 3s)
R12 (s) = · · ,
(s − 2) Γ((1 − s)/2)Γ((2 − 3s)/2) ζ(1 − s)ζ(2 − 3s)
√ s + 1 Γ(−s/2) ζ(−s)
R21 (s) = 2 π ,
s + 2 Γ((1 − s)/2) ζ(1 − s)
(As + 3)(s − 1)(s + 1) Γ((−1 − s)/2) ζ(−1 − s)
R22 (s) = π · · .
s(s − 3) Γ((1 − s)/2) ζ(1 − s)
Therefore, for σ = <(s) < 0,
√ s(s − 1) Γ((1 − 3s)/2)
|R11 (s)| ≤2 π ζ(1 − 3σ)ζ(2 − 3σ),
(s − 3 )(s + 1) Γ((2 − 3s)/2)
2

s(s − 1 + A−1 ) Γ(−s/2) Γ((1 − 3s)/2)

|R12 (s)| ≤3πA
(s − 2) Γ((1 − s)/2) Γ((2 − 3s)/2)
× ζ(−σ)ζ(1 − σ)ζ(1 − 3σ)ζ(2 − 3σ),
√ s+1 Γ(−s/2)
|R21 (s)| ≤2 π ζ(−σ)ζ(1 − σ),
s + 2 Γ((1 − s)/2)
(s + 3A−1 )(s − 1)(s + 1) Γ((−1 − s)/2)
|R22 (s)| ≤πA ζ(−1 − σ)ζ(1 − σ).
s(s − 3) Γ((1 − s)/2)
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 235

Zeta Functions for Reductive Groups 235

In addition, since all |arg((1−3s)/2)| and |arg(2−3s)/2|; −s/2, (1− s)/2, (1−3s)/2
and (2 − 3s)/2; and arg(−1 − s)/2, arg(1 − s)/2 are strictly less than π/2. Hence
we can apply the Stirling formula for <(s) < 0, and then
r
Γ((1 − 3s)/2) 2 −1/2
= |s| (1 + O(|s|−1 )),
Γ((2 − 3s)/2) 3
Γ(−s/2) √ Γ((−1 − s)/2)
= 2 |s|−1/2 (1 + O(|s|−1 )), = 2 |s|−1 (1 + O(|s|−1 )).
Γ((1 − s)/2) Γ((1 − s)/2)
On the other hand, as σ → −∞,
ζ(1 − 3σ)ζ(2 − 3σ) → 1, ζ(−σ)ζ(1 − σ)ζ(1 − 3σ)ζ(2 − 3σ) → 1,
ζ(−1 − σ)ζ(1 − σ) → 1, ζ(−σ)ζ(1 − σ) → 1.
Therefore, if σ = <(s) < 0, and |s|, |σ| are both large,
r
8π √
|R11 (s)| ≤ · |s|−1/2 · (1 + O(|s|−1 )), |R12 (s)| ≤ 2 3πA · (1 + O(|s|−1 )),
3
√
|R21 (s)| ≤ 8π · |s|−1/2 · (1 + O(|s|−1 )), |R22 (s)| ≤ 2πA · (1 + O(|s|−1 )),
as desired. 

Lemma 8.7. There are at most finitely many zeros of the entire function f1 (s),
resp. f2 (s), in the region <(s) > 1/3, resp. <(s) > 0. In particular, there are at
most finitely many zeros of f1 (s) and f2 (s) in the right-half plane <(s) ≥ 1/2.

Proof. We have
f1 (s) = (s − 1)2 (3s − 2)(As − A + 1)ξ(s + 1)ξ(3s) [ 1 − Q11 (s) − Q12 (s) ] ,
(8.79)
f2 (s) = (As + 3)(s − 1)2 (s − 2)ξ(s + 2) [ 1 − Q21 (s) − Q22 (s) ] ,
where
(s + 1)(s − 2) ξ(s)ξ(3s − 1)
Q11 (s) = · ,
(s − 1)(3s − 2)(As − A + 1) ξ(s + 1)ξ(3s)
2(s − 2) ξ(s)
Q12 (s) = · ,
(3s − 2)(As − A + 1) ξ(s + 1)
2(s − 3) ξ(s + 1) (s + 2)(s − 3) ξ(s)
Q21 (s) = · , Q22 (s) = · .
(As + 3)(s − 1) ξ(s + 2) (As + 3)(s − 1) ξ(s + 2)
2

The factor (s−1)2 (3s−2)(As−A+1)ξ(s+1)ξ(3s), resp. (As+3)(s−1)2 (s−2)ξ(s+2),

has no zero in <(s) > 1/3, resp. <(s) > 0, except for s = 2/3, 1, resp. s = 1, 2.
Replacing 2s − 1 by 3s − 1 or s, resp. s + 1 or s, in (11.26), we obtain
ξ(3s − 1) 1 ξ(s + 1)
<1 (<(s) > ), <1 <(s) > −1 ,


ξ(3s) 3 ξ(s + 2)
(8.80)
ξ(s) ξ(s) ξ(s + 1) ξ(s)
<1, = < 1 (<(s) > 0).
ξ(s + 1) ξ(s + 2) ξ(s + 2) ξ(s + 1)
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 236

236 Zeta Functions for Reductive Groups

Let D1 and D2 be the region defined by

 1 


 <(s) ≥ 


 3 
,
 
D1 :=  s∈C
 
(s + 1)(s − 2) 2(s − 2) 
+

 

 ≥ 1
(s − 1)(3s − 2)(As − A + 1) (3s − 2)(As − A + 1)

 


(s + 2)(s − 3)
( )
2(s − 3)
D2 := s ∈ C <(s) ≥ 0, + ≥ 1 .
(As + 3)(s − 1) (As + 3)(s − 1)2
Then, by (8.79) and (8.80), f1 (s) , 0, resp. f2 (s) , 0, if s < D1 and <(s) ≥ 1/3,
resp. if s < D2 and <(s) ≥ 0. The region D1 is bounded, since
(s + 1)(s − 2) 2(s − 2)
+ <1
(s − 1)(3s − 2)(As − A + 1) (3s − 2)(As − A + 1)
for large |s|. Similarly, the region D2 is bounded. Therefore, there are at most
finitely many of zeros for f1 (s), resp. f2 (s), in <(s) ≥ 1/3, resp. in <(s) ≥ 0. 

8.3.2.5 Positivity

By the results above, we can apply Lemma 8.3 to the functions fi (s) (i = 1, 2).
Hence to complete the proof of Proposition 8.7, it suffices to prove the following:

Lemma 8.8. There are only three zeros for each of f1 (s) and f2 (s) in the right
half plane <(s) ≥ 1/2. More precisely, for f1 (s), resp. for f2 (s), these are given
by one real zero at s = 1, resp. s = 2, and other two non-real conjugate zeros
approximated located at s ' 0.927 ± i · 2.09, resp. s ' 1.17 ± i · 3.43.

Proof. The domain Di ∩ {<(s) ≤ 1/2} (i = 1, 2) is contained in the rectangle

R = [0.5, 5] × [−10, 10], where Di is the region in the proof of Lemma 8.7. By the
argument principle, the numbers of zeros of the fi (s)’s in R are given by
fi0
Z
1
(s)ds.
2πi ∂R fi
Since the values of these integrals are integers, computationally say using Math-
ematica, we can easily check that they are simply three. Similarly, we can locate
the above approximate values for non-real zeros as indicated in the lemma. 

8.3.3 Riemann Hypothesis for G2

In this subsection, we follow M. Suzuki to prove the following:

ζQG2 /Pα (s) and

Theorem 8.6 (Riemann Hypothesis). ([112]) All zeros of b
G /P 1
ζQ 2 β (s) lie on the central line < (s) = .
b
2
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 237

Zeta Functions for Reductive Groups 237

By (8.67), (8.68) and (8.69), this is equivalent to the following:

Theorem 8.7.

(1) All zeros of Z1 (s) lie on the line Re(s) = 1/2 except for four simple zeros
s = 0, 1/3, 2/3, 1.
(2) All zeros of Z2 (s) lie on the line Re(s) = 1/2 except for four simple zeros
s = −1, 0, 1, 2.

8.3.3.1 Zeros in the Right-Half Plane for Z

To prove Theorem 8.7, we first analysis the zeros of Zi (s) in the right half plane.

Proposition 8.8. There is no zero for the functions Z1 (s) and Z2 (s) in the right-half
plane <(s) ≥ 20.

Proof. We have

Z1 (s) = (s − 1)2 (3s − 2)(As − A + 1)ξ(s + 1)ξ(3s)ξ(2s)

× (1 − R11 (s) − R12 (s) − R13 (s) + R14 (s) + R15 (s)),
(8.81)
Z2 (s) = (s − 1)2 (s − 2)(As + 3)ξ(s + 2)ξ(2s)
× (1 − R11 (s) − R22 (s) + R23 (s) − R24 (s) − R25 (s)),

where
(s + 1)(s − 2) ξ(s) ξ(3s − 1)
R11 (s) = ,
(s − 1)(3s − 2)(As − A + 1) ξ(s + 1) ξ(3s)
2(s − 1)(s − 2) ξ(s)
R12 (s) = ,
(s − 1)(3s − 2)(As − A + 1) ξ(s + 1)
s2 (3s − 1)(As − 1) ξ(s − 1) ξ(3s − 2) ξ(2s − 1)
R13 (s) = ,
(s − 1) (3s − 2)(As − A + 1) ξ(s + 1) ξ(3s)
2 ξ(2s)
s(s + 1)(s − 2) ξ(s) ξ(3s − 1) ξ(2s − 1)
R14 (s) = ,
(s − 1)2 (3s − 2)(As − A + 1) ξ(s + 1) ξ(3s) ξ(2s)
2s2 (s + 1) ξ(s) ξ(3s − 2) ξ(2s − 1)
R15 (s) = ,
(s − 1) (3s − 2)(As − A + 1) ξ(s + 1) ξ(3s)
2 ξ(2s)
2(s − 3) ξ(s + 1)
R21 (s) = ,
(s − 1)(As + 3) ξ(s + 2)
(s + 2)(s − 3) ξ(s)
R22 (s) = ,
(s − 1)2 (As + 3) ξ(s + 2)
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 238

238 Zeta Functions for Reductive Groups

s2 (s + 1)(As − 3 − A) ξ(s − 2) ξ(2s − 1)

R23 (s) = ,
(s − 1)2 (s − 2)(As + 3) ξ(s + 2) ξ(2s)
2s(s + 1)(s + 2) ξ(s − 1) ξ(2s − 1)
R24 (s) = ,
(s − 1) (s − 2)(As + 3) ξ(s + 2) ξ(2s)
2

(s − 3)(s + 1)(s + 2) ξ(s) ξ(2s − 1)

R25 (s) = .
(s − 1)2 (s − 2)(As + 3) ξ(s + 2) ξ(2s)
Replacing 2s − 1 by s or 3s − 1, s − 1 or 3s − 2, and s − a (a = −1, 0, 1, 2) in
(11.26),
ξ(s) ξ(3s − 1)  1
< 1 (<(s) > 0), < 1 <(s) > ,
ξ(s + 1) ξ(3s) 3
ξ(s − 1) ξ(s) ξ(s − 1)
= < 1 (<(s) > 1),
ξ(s + 1) ξ(s + 1) ξ(s)
ξ(3s−2) ξ(3s − 1) ξ(3s − 2)  2
= < 1 <(s) > ,
ξ(3s) ξ(3s) ξ(3s − 1) 3
ξ(s + 1) ξ(s) ξ(s + 1) ξ(s)
< 1, (<(s) > −1), = < 1, (<(s) > 0),
ξ(s + 2) ξ(s + 2) ξ(s + 2) ξ(s + 1)
ξ(s − 1) ξ(s + 1) ξ(s) ξ(s − 1)
= < 1, (<(s) > 1),
ξ(s + 2) ξ(s + 2) ξ(s + 1) ξ(s)
ξ(s − 2) ξ(s + 1) ξ(s) ξ(s − 1) ξ(s − 2)
= < 1, (<(s) > 2).
ξ(s + 2) ξ(s + 2) ξ(s + 1) ξ(s) ξ(s − 1)
Hence |R1i (s)| ≤ C1i |s|−1 , resp. |R2i (s)| ≤ C2i (i = 1, 2, 4, 5) for <(s) > 1, resp.
<(s) > 2. Applying the Stirling formula to R13 (s) and R23 (s), we obtain |R13 (s)| =
(|s|−5/2 ) for <(s) > 1 as |s| → ∞ in the right-half plane, and |R23 (s)| = (|s|−5/2 )
for <(s)  0. Therefore Z1 (s) , 0, resp. Z2 (s) , 0, for some right-half plane
<(s) ≥ σ3 , resp. <(s) ≥ σ4 . Using the monotone decreasing property of ζ(σ) as
σ → +∞ and the effective version of Stirling’s formula ( [116])
! 21   s ( !)
2π s 1
Γ(s) = 1+Θ (<(s) > 1),
s e 8|s|
where the notation f = Θ(g) means | f | ≤ g, we have

|R11 (s)| ≤ 0.1, |R12 (s)| ≤ 0.3, |R13 (s)| ≤ 0.05 |R14 (s)| ≤ 0.1, |R15 (s)| ≤ 0.1,
|R21 (s)| ≤ 0.3, |R22 (s)| ≤ 0.13, |R23 (s)| ≤ 0.15 |R24 (s)| ≤ 0.2, |R25 (s)| ≤ 0.1

for <(s) ≥ 20 (in fact, these bounds already hold for <(s) ≥ 10). These estimates
imply Z1 (s) , 0 and Z2 (s) , 0 for <(s) ≥ 20 by (8.81), since (s − 1)2 (3s − 2)(As −
A + 1)ξ(s + 1)ξ(3s)ξ(2s), resp. (s − 1)2 (s − 2)(As + 3)ξ(s + 2)ξ(2s), has no zero in
the right-half plane <(s) ≥ 20. 
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 239

Zeta Functions for Reductive Groups 239

8.3.3.2 Zeros in the Strip for Z

Next, we analysis the zeros of the Zi (s)’s in a strip.

Proposition 8.9. There is no zero for Z1 (s), resp. Z2 (s), in the part of the strip
1/2 < σ < 20 characterized by |t| ≥ 25, resp. |t| ≥ 36.
√
Proof. Let ρi0 = βi0 + −1γi0 (i = 1, 2, γi0 > 0) be the complex zero of fi (s) in
Lemma 8.8. By Proposition 8.7, f1 (s) and f2 (s) has the factorization
0
 s  s 
fi (s) = fi0 (0) eBi s s(1 − s) 1 − 1− ·Πi (s) (i = 1, 2, B0i ≥ 0),
ρi0 ρi0
where
" #
Y s Y  s  s
Πi (s) = 1− 1− 1− .
0,β∈R
β ρ=β+iγ ρ ρ
β<1/2, γ>0

Note that all zeros of Πi (s) (i = 1, 2) lie in σi0 < <(s) < 1/2 for some σi0 . Hence
 g1 (1 − s) 
Z1 (s) =g1 (s) · 1 − (g1 (s) = f1 (s) · ξ(2s)),
g1 (s)
(8.82)
 g2 (1 − s) 
Z2 (s) =g2 (s) · 1 − (g2 (s) = f2 (s) · ξ(2s))
g2 (s)
and
g1 (1 − s) 0 Π1 (1 − s) s − 1 + ρ0 s − 1 + ρ0 ξ(2s − 1)
=eB1 (1−2σ) ,
g1 (s) Π1 (s) s − ρ0 s − ρ0 ξ(2s)
(8.83)
g2 (1 − s) 0 Π2 (1− s) s s−1+ρ0 s−1+ρ0 ξ(2s−1)
=eB2 (1−2σ) .
g2 (s) Π2 (s) 1− s s − ρ0 s − ρ0 ξ(2s)
Because B0i ≥ 0, we have
0
eBi (1−2σ) ≤ 1 (<(s) > 1/2). (8.84)

For the ratio Πi (1 − s)/Πi (s) in (8.83)(3 Lemmas), we have

Πi (1 − s) 1−s−ρ 1−s−ρ
Y !
= · < 1 (<(s) > 1/2), (8.85)
Πi (s) ρ=β+iγ
s−ρ s−ρ
β<1/2, γ>0

by term-by-term argument as in [65] by using β < 1/2 and

1−s−ρ 2 (2σ − 1)(1 − 2β)
=1− ,
s−ρ (σ − β)2 + (t − γ)2
where ρ = β + iγ is a zero of fi (s).
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 240

240 Zeta Functions for Reductive Groups

It remains to give an estimate for

s − 1 + ρ0 s − 1 + ρ0 ξ(2s − 1)
r1 (s) := · · ,
s − ρ0 s − ρ0 ξ(2s)
(8.86)
s s − 1 + ρ0 s − 1 + ρ0 ξ(2s − 1)
r2 (s) := · · .
1 − s s − ρ0 s − ρ0 ξ(2s)
For this, we use the following:
Using Lemma 8.2, we show the following:

Lemma 8.9. Let ρ10 = β10 + iγ10 ' 0.927 + i · 2.09, resp. ρ20 = β20 + iγ20 '
1.17 + i · 3.43, be the complex zero of f1 (s), resp. f2 (s), in Lemma 8.8. Let
s = σ + it be a point in the part of the strip 1/2 < σ ≤ 20 characterized by t ≥ 25,
resp. t ≥ 36. Then there exists at least two distinct zeros ρ = β + iγ of b
ζ(s) such
that 0 < β ≤ 1/2, |t − γ| ≤ 22,
s − 1 + ρ0 2s − 1 − (1 − ρ) s − 1 + ρ0 2s − 1 − (1 − ρ)
< 1, < 1, (8.87)
s − ρ0 2s − ρ s − ρ0 2s − ρ
and
s − 1 + ρ0 2s − 1 − (1 − ρ) s 2s − 1 − (1 − ρ)
· < 1, · < 1. (8.88)
s − ρ0 2s − ρ s−1 2s − ρ
√
Proof. By an abuse of notations, we write ρ0 = β0 + −1γ0 for ρi0 . By squaring
(8.87) and (8.88), we have
(σ + β0 − 1)2 + (t ± γ0 )2 (2σ + β − 2)2 + (t − γ)2
· < 1. (8.89)
(σ − β0 )2 + (t ± γ0 )2 (2σ − β)2 + (t − γ)2
Claim 8.2. ([64]) For any real value of t there exists at least three distinct zeros
ρ = β + iγ of b
ζ(s) such that 0 < β ≤ 1/2 and
|t − γ| ≤ 22. (8.90)

Proof. Suppose |t| ≥ 25. Then there exists at least three distinct zeros ρ = β + iγ
of bζ(s) satisfying 0 < β ≤ 1/2 and |t − γ| < 15.1 by applying Lemma 5 in [109]
to t + 10.1 and t − 10.1 (Lemma 5 in [109] is essentially Lemma 3.5 of [64]). For
|t| < 25, estimate (8.90) also holds for three distinct zeros because b
ζ(s) has zeros
at s = ±14.13, ±21.02, ±25.01. 

Therefore, it suffices to show that (8.89) holds for 0 < β ≤ 1/2, |t − γ| < 22,
1/2 < σ ≤ 20 and t ≥ 25, namely to show
(σ + β0 − 1)2 + (t ± γ0 )2 (2σ − 2 )2 + 222
3
· <1
(σ − β0 )2 + (t ± γ0 )2 (2σ − 12 )2 + 222
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 241

Zeta Functions for Reductive Groups 241

by the similar reasons in the later half of section 4.3 in [109]. Obviously, the
inequality (8.3.3.2) is equivalent to
 
(2σ − 1) 8(t ± γ0 )2 − P(σ) > 0, (8.91)
where P(σ) = 8(4β0 − 3)σ2 − 8(4β0 − 3)σ − 8β20 + 3890β0 − 1945.
Using the value β0 ' 0.927, resp. 1.17, we see that P(σ) < 3807, resp.
< 7777, for 1/2 < σ < 20. On the other hand, using the value γ0 ' 2.09, resp.
3.43, we see that 8(t ± γ0 )2 > 3872, resp. > 8192, for t ≥ 25, resp. ≥ 36 since
|t ± γ0 | = t ± γ0 > 22, resp. > 32, for t ≥ 25, resp. ≥ 36. Hence (8.91) hold, and it
implies (8.89). 

Lemma 8.9 and Zi (s) = Zi (s) implies

|ri (s)| < 1
for 1/2 < σ ≤ 20, |t| ≥ 25 (8.92)
by taking two distinct zeros of ζ(s) in that region. Indeed, the remaining terms in
b
r1 (s) are estimated by 2s−1−(1−ρ)
2s−ρ < 1 when <(s) > 1/2, where ρ is a zero of b
ζ(s).
And a similar estimation holds for the remaining terms in r2 (s). Estimates (8.84),
(8.85) and (8.92) show that, for 1/2 < σ ≤ 20,
g1 (1 − s) g2 (1 − s)
< 1 for |t| ≥ 25 and < 1 for |t| ≥ 36.
g1 (s) g2 (s)
By (8.82) this estimate implies Proposition 8.9, because g1 (s), resp. g2 (s), has
no zero in the part of the strip 1/2 < σ ≤ 20, characterized by |t| ≥ 25, resp.
|t| ≥ 36. 

8.3.3.3 Zeros in the Box for Z

Proposition 8.10 (Riemann Hypothesis). There are only two zeros for Z1 (s),
resp. Z2 (s), namely, the simple zeros at s = 2/3, 1, resp. s = 1, 2 in the box
1/2 < σ < 20, |t| ≤ 25, resp. 1/2 < σ < 20, |t| ≤ 36.

Proof. Because the boxes characterized by 1/2 < σ ≤ 20, |t| ≤ 25 or 36 are finite,
we can computationally check the assertions of Proposition 8.10 as in the proof of
Lemma 8.8, say using Mathematica. The outcome is as stated. 

Proof of Theorems 8.7 and 8.6. As a direct consequence of Propositions 8.8, 8.9,
8.10, and the functional equation of Z1 (s), resp. Z2 (s), it is easily to conclude that
all zeros of Z1 (s), resp. Z2 (s), lie on the line <(s) = 1/2 except for simple zeros
located at s = 0, 1/3, 1/2, 2/3, 1, resp. s = −1, 0, 1/2, 1, 2. This completes the
proof of Theorems 8.7 and 8.6. That is, the Riemann hypothesis holds for the zeta
functions associated to the exceptional group of type G2 . 
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 242

242 Zeta Functions for Reductive Groups

8.4 Zeta Functions for (G, P)

8.4.1 Singular Hyperplanes Located

From above, starting from the period ωGQ (λ) of G governed by huge symmetries

ζ hλ, α∨ i

!
X 1 Y b
ωGQ (λ) = ·
, (8.93)
α∈∆ hwλ − ρ, α i α>0,wα<0 b
∨
Q
w∈W ζ hλ, α∨ i + 1

to find the singular hyperplanes along which the residues are taken, our success in
introducing the zeta functions for SLn , Sp2n and G2 leads to the rational function
1
. (8.94)
α∈∆ hλ − ρ, α i
∨
Q

In other words, if we denote the rank of the groups by r, then

(a) There are exactly (r − 1)-singular hyperplanes along which we should take
residues, and
(b) all these (r − 1)-singular hyperplanes are of the forms hλ − ρ, α∨ i, taking from
the total r-factors in the denominator of (8.94).

This then exposes the hidden role played by the maximal parabolic subgroup
P of G. More precisely, for a standard maximal parabolic subgroup P of G, there
exists one and only one simple root αP such that, under the inverse order one-to-
one correspondence between parabolic subgroups of G and subsets of the set of
simple roots ∆, the maximal parabolic subgroup P of G is associated to the subset
∆ r {αP } of ∆. Therefore, we have the following interpretations of (a) and (b).

(?) The (r − 1) singular hyperplanes for taking the residues are given by

hλ − ρ, α∨ i = 0 α ∈ ∆ r {αP }. (8.95)

Upon this point, we are very sure how a new type of zeta functions for G, or
better, for the pairs (G, P), should be introduced.

8.4.2 Main Definition

Let G be a (split) reductive group of rank r over a number field F. Fix a mini-
mal parabolic subgroup P0 of G and let M0 be the Levi subgroup of P0 and let
T 0 be the maximal split torus of the center of M0 . Since T 0 contains only semi-
simple elements (which all commute with each other), it follows that, with respect
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 243

Zeta Functions for Reductive Groups 243

to the adjoint action Ad of G on the Lie algebra g of G, Ad(T 0 ) is diagonaliz-

able, and hence g admits a natural decomposition into the associated eigenspaces
parametrized by the characters α : T 0 → Gm (α ∈ X ∗ (T 0 )), namely,
M n o
g = g0 + ⊕ gα with gα := x ∈ g : Ad(t)(x) = α(t)x ∀t ∈ T 0 . (8.96)
α∈X ∗ (T 0 )

We denote the set of characters of T 0 which occur in the above decomposition by

Φ(G, T 0 ). It is well known that

(a) Φ(G, T 0 ) forms a root system.

(b) All F-split tori are conjugate in G(F), the set Φ(G, T 0 ) is essentially indepen-
dent of T 0 . For this reason, we sometimes denote Φ(G, T 0 ) by Φ(G).
(c) For each root α ∈ Φ(G, T 0 ), there is a naturally associated chroot α∨ , which is
a one parameter subgroup of T 0 . This induces a canonical pairing h·, ·i with Z
values between the subgroups X ∗ (T 0 ) of rational characters of T 0 and X∗ (T 0 )
of one-parameter subgroups of T 0 , and hence an identification
X ∗ (T 0 ) ⊗Z R =: a∗0 ' a0 := X∗ (T 0 ) ⊗Z R. (8.97)
(d) There is a natural Weyl group W of G, defined as the quotient
W := NormG(F) T 0 (F)/CentG(F) T 0 (F) (8.98)
where, NormG(F) T 0 (F) and CentG(F) T 0 (F) denote the normalizer subgroup
and centralizer subgroup of T 0 (F) in G(F), respectively. Since an element
in G(F) which normalizes, resp. centralizes, T 0 (F), also normalizes, resp.
centralizes T 0 , there is a natural action of W on T 0 and hence on a∗0 and a0 .
In particular, h·, ·i is W-invariant, and W is isomorphic to the Weyl group of
the root system Φ, generated by the reflections σα on the space a∗0 . Here, as
usual,
hα, xi
σα : a∗0 → a∗0 x 7→ x − 2 α. (8.99)
hα, α
 
Accordingly, denote simply by V, h·, ·i; Φ, Φ± , ∆, ∆∨ , b ∆∨ ; W; ρ the data asso-
ciated to the root system Φ(G, T 0 ) as in Part 3, and write ∆ =: α1 , . . . , αr and

∆∨ = λ1 , . . . , λr . We assume that hλi , α j i = δi j for all 1 ≤ i, j ≤ r. From the
b 
definition, ρ = 21 α>0 α = ri=1 λi . In addition, introduce a coordinate system on
P P
V by
X r X r
λ= si λi + ρ = (si + 1)λi . (8.100)
i=1 i=1

Let P be a maximal (standard) parabolic subgroup of G. By the theory of

algebraic groups, (the conjugation class of) P corresponds to a unique simple root
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 244

244 Zeta Functions for Reductive Groups

n o
αP ∈ ∆. Hence it makes sense to write ∆ r {αP } =: βP,1 , βP,2 , . . . , βP,r−1 . For
simplicity, we sometimes write αP as α p for a certain 1 ≤ p ≤ r. Accordingly, for
a meromorphic function f (λ), after taking the residues along (r − 1) hyperplanes
hλ, β∨P,k i = 0 (1 ≤ k ≤ r − 1), the resulting function


Reshλ−ρ,β∨P,r(G)−1 i=0 · · · Reshλ−ρ,β∨P,2 i=0 Reshλ−ρ,β∨P,1 i=0 f (λ)
or the same,
Res sr =0 . . . Res s p =0 . . . Res s1 =0 f (λ)
[

becomes a meromorphic function of the valuable s p . For convenience, we write

s := sP := s p .

Definition 8.11. ([126]) The period for (G, P) over F is defined by

 
ωG/P
F (s) := Reshλ−ρ,βP,r(G)−1 i=0 · · · Reshλ−ρ,βP,2 i=0 Reshλ−ρ,βP,1 i=0 ωF (λ) .
∨ ∨ ∨
G
(8.101)
Here, ωGF (λ) denotes the period for G over F, namely,
ζF (hλ, α∨ i)
!
X 1 Y
ωF (λ) := .
G
b
· (8.102)
α∈∆ hwλ − ρ, α i α>0,wα<0 b
∨
Q
w∈W ζF (hλ, α∨ i + 1)

Obviously, after taking the residues, there are still zeta factors left in the de-
nominators of the terms in ωG/P F (s), even after the cancelations of the common
zeta factors in both the numerators and the denominators. It is not too difficult to
see that there are a certain minimal integer I(G/P) ≥ 0 and some finitely many
non-trivial zeta function factors
     
ζF aG/P
b
1 s P + b1
G/P
, bζF aG/P
2 s P + b2
G/P
, ..., bζF aG/P
I(G/P) sP + bI(G/P) ,
G/P

QI(G/P)  G/P  G/P

such that the product i=1 ζF ai sP + bG/P
b
i · ωF (sP ) admits only finitely
many singularities. And similarly, there is a certain minimal integer J(G/P) ≥ 0
and some finitely many special zeta values factors
     
ζF cG/P , b
b
1 ζF cG/P , . . . , b
2 ζF cG/P , J(G/P)

such that there

Qare no factors of special b
 G/P  ζF values appearing in the denominators of
the product i=1 ζ
J(G/P) b
c
F i ·ωG/P
F (s P ). With all these, we are ready to introduce
the following main definition.

Definition 8.12. The zeta function b ζF;o

G/P
for (G, P) over F is defined by
  I(G/P)
Y   J(G/P)
Y  ! G/P  
ζF;o
b G/P
s := ζF aG/P
b
i s + bG/P
i · ζ
b F cG/P
j · ωF s . (8.103)
i=1 j=1

These zeta functions are called Weng zeta functions in literatures.

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 245

Zeta Functions for Reductive Groups 245

8.4.3 Basic Properties

By the standard properties of the Dedekind

 ζF (s), it is easy to con-
zeta function b
ζF;o s is a well-defined holomorphic function for
clude that the zeta function b G/P

<(s)  0, admits a unique meromorphic continuation to the whole complex s-

plane, and posses only finitely many singularities. Moreover, as to be expected,
we have the following:

Theorem 8.8 (Functional Equation). There exists a constant cG/P ∈ Q such that
ζF;o
G/P
− s + cG/P = b
ζF;o
G/P

s.


b (8.104)

This theorem was first verified for the lower rank groups SLn (n = 2, 3, 4, 5),
Sp4 and G2 , by myself, then proved for (G, P) = (SLn , Pn−1,1 ) by Kim-Weng
in [58], and finally for all (G, P) by Komori [59]. For details, please refer to the
next two chapters.
 
Definition 8.13. The zeta function b ζFG/P s for (G, P) over F is defined by
!
cG/P − 1
bζ s := ζF;o s +
G/P
b G/P
. (8.105)
2
 
By Theorem 8.8, b ζ G/P s admits the standard functional equation

ζFG/P 1 − s = bζFG/P s .



b (8.106)

ζFG/P s lie on the cen-

Conjecture 8.1 (The Riemann Hypothesis). All zeros of b
1
tral line <(s) = .
2
In this book, we will prove the following:

Theorem 8.9 (Weak Riemann Hypothesis). Let G be a Chevalier group over Q.

1
ζQG/P s lie on the central line <(s) = .


Then all but finitely many zeros of b
2
The proof consists of two parts, namely, the one on a numerical criterion based
on the refined symmetry involving the Lie structures by Ki-Komori-Suzuki ( [56])
in Part 5, and the other on our recent work about the stability of arithmetic princi-
pal torsors and the associated parabolic reduction in Part 6.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 247

Chapter 9

Zeta Function of (SL n, P n−1,1)

In this chapter, we first characterize all the special permutations following

a joint work with Zagier ([135]). Then we write down an explicit formula
SL /P
for bζQ n n−1,1 (s) mainly following [135], where the case for function field is
treated. Finally, as a direct consequence, we obtain the functional equation for
SL /P
ζQ n n−1,1 (s), which was first established jointly with Kim ( [58]) using a slightly
b
different method.

9.1 Special Weyl Elements

9.1.1 General Setting

Let (V, h·, ·i) be an Euclidean space of dimension n, let ∆ ⊂ Φ = Φ+ t Φ− ⊂ V

be a root system associated to a reductive group G and a fixed minimal parabolic
subgroup P0 , where ∆, resp. Φ+ , resp. Φ− , denotes the set of simple roots, resp.
positive roots, resp. negative roots. We have Φ− = −Φ+ . Write ∆ =: α1 , . . . , αn ,

then Φ ⊂ i=1 Z≥0 αi . Let λ1 , . . . , λr be the set of fundamental weights, charac-
Pn 
terized by hλ j , α∨i i = δi j i, j = 1, 2, . . . , r. Here, as usual, for an α ∈ Φ, denote by
2
α∨ the associated coroot α∨ := α. Finally, denote by ρ the associated Weyl
hα, αi
vector characterized by
n
1 X X
ρ := α= λ j. (9.1)
2 α∈Φ+ j=1

From the theory of reductive groups, there is a natural order-reversed corre-

spondence between standard parabolic subgroups P and subsets of ∆. In partic-
ular, for a fixed maximal standard parabolic subgroup P, denote by σP ∈ ∆ the
unique simple root such that P −→ ∆P := ∆ r {σP }. In particular, there is a
p, 1 ≤ p ≤ n such that αP = α p . For this reason, we sometimes write P in various

247
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 248

248 Zeta Functions for Reductive Groups

notations simply as p. In the sequel till the end of this chapter, unless otherwise is
stated, P stands for a maximal standard parabolic subgroup of (G, P0 ), and if there
is another maximal standard parabolic subgroup Q, we use the index q, 1 ≤ q ≤ n
instead.
For α ∈ Φ, denote by σα : V → V, v 7→ v − hv, αiα∨ the reflection with
respect to the α-perpendicular hyperplane in V to σ. Let W be the associated
Weyl group generated by n simple reflections σ j := σα j associated to simple roots
α j (1 ≤ j ≤ n). It is well known that W is also generated by reflections associated
to all the positive roots α ∈ Φ+ , or equivalently, to all the roots α ∈ Φ, and an
element w ∈ W admits a shortest decomposition as a product of σ j ’s. For a Weyl
element w ∈ W, let
Φw := Φ+ ∩ w−1 (Φ− ). (9.2)
Then, Φw is the set of positive roots whose w-image is negative, and |ΦW | co-
incides with the length of w, defined to be the number of elements in a shortest
decomposition of w interns of the σi ’s. Let w0 be the longest element of W. By
the theory of Coxeter groups, there is one and only one such an element w0 in W
such that w20 = Id, w0 (∆) = −∆ and hence w0 (Φ+ ) = Φ− .
Let P be a maximal standard parabolic subgroup of G. For simplicity, we
write ∆ p = β1 , . . . , βn−1 . Let Φ p be the collection of elements in Φ which are

perpendicular to λ p and set Φ±p := Φ p ∩ Φ± . It is not difficult to show that Φ p is
a root system. Accordingly, Φ+p is generated by ∆ p with Z≥0 coefficients, and, all
elements of the associated Weyl group W p , generated by σα for all σ ∈ Φ p , admit
shortest decompositions in terms of the σ j ’s with j , p. In particular, there exists
a unique longest element w p in W p such that
w p (∆P ) = −∆ p and w p (Φ+p ) = Φ−p . (9.3)
1 X X r
Similarly, set ρ p := α. In general, ρ p , λ j.
2 α∈Φ+ j=1,,p
p

Introduce a coordinate system (s1 , . . . , sn ) for V by

Xn Xn
λ =: (1 + s j )λ j = ρ + s jλ j. (9.4)
j=1 j=1
Then we obtain a ‘normalized’ single variable in p-direction characterized by
s := hwλ − ρ, α∨p i. (9.5)
Xn
Obviously, by definition, for α∨ = a j α∨j ∈ V,
j=1
n
X
hλ − ρ, α∨ i = a j s j. (9.6)
j=1
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 249

Zeta Function of (SLn , Pn−1,1 ) 249

Motivated by Definition 8.11, we introduce the following:

Definition 9.1. Let F a number field. The T -version period ωG/P

F (s; T ) of (G, P)
over F is defined by

ωG/P
F (s; T ) := Reshλ−ρ, β1 i=0 . . . Reshλ−ρ, βn−1 i=0 ωF (λ; T ).
∨ ∨
G
(9.7)

In particular, the period ωG/P

F (s) of (G, P) over F is defined by

ωG/P
F (s) := ωF (s; 0).
G/P
(9.8)

Here, ωGF (λ; T ) denotes the T -period of G over F in Definition 7.8, namely,
  
X Y 1   Y b ζ F (hλ, α ∨
i) 
ωF (λ; T ) :=
G
ehwλ−ρ,T i     . (9.9)
hwλ − ρ, α i α∈Φ b

∨
ζ (hλ, α ∨ i + 1) 

w∈W α∈∆ w F

For later use, we write

X
ωGF (λ) = ωGF (λ; 0) = Aw (λ), (9.10)
w∈W

where
  
Y 1   Y b ζF (hλ, α∨ i) 
Aw (λ) :=     . (9.11)
α∈∆
hwλ − ρ, α∨ i α∈Φ b ζF (hλ, α∨ i + 1)
w

9.1.2 Special Weyl Elements

Definition 9.2. An element w ∈ W is called (p-)special if

∆ p = α ∈ ∆ p : wα ∈ ∆ ∪ Φ− .

(9.12)

We set W 0 ⊆ W to be the subset consisting of all (p-)special Weyl elements in W.

Proposition 9.1. The T -period of (G, P) and hence also the period of (G, P) over
F are given by the summations over special Weyl elements. More precisely,
 
X  Y
hw(sλ p +ρ)−ρ,T i |Φw ∩∆ p |  1 
ωF (s; T ) :=
G/P
e cF  
, α ∨ is+hρ, α∨ i−1 

 hλ p 
w∈W 0 α∈w−1 (∆)r∆ p
   (9.13)
 Y   Y 1 
ζF hλ p , α∨ is + hρ, α∨ i  
 .


×  b  
α∈Φw r∆ p α∈(−Φw )
bζ F hλ p , α ∨ is + hρ, α∨ i 

ζF (s) at s = 1.
Here cF denotes the residue of b
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 250

250 Zeta Functions for Reductive Groups

The essentials for this proof come from [57], or better, [58] on functional
equation of group zeta functions associated to (G, P) = (SLn , Pn−1,1 ). See also
[59].

Proof. From the definition,

  
Y 1   Y b ζ F (hλ, α ∨
i) 
Aw (λ) =    .
ρ, α
 
∨
hwλ − i ζ (hλ, α ∨ i + 1) 

α∈∆ α∈Φw F
b
The group of factors in the first big bracket, resp. in the second big bracket, will be
called the rational factors, resp. the zeta factors, of Aw (λ). We want to know what
happens after taking successive residues along singular hyperplanes Reshλ−ρ, β∨k i=0
(1 ≤ k ≤ n − 1).
Since hλi , α∨ i = δi j ,
hλ − ρ, α∨j i = s j and hρ, α∨j i = 1. (9.14)
Consequently,
Reshλ−ρ, β∨1 i=0 . . . Reshλ−ρ, β∨n−1 i=0 Aw (λ)
= Res s1 =0 . . . Res[ s p =0 . . . Res sn =0 Aw (λ)
 n 
 Y 
= lim . . . [ lim . . . lim  sk  · Aw Σnj=1 (1 + s j )λ j


s1 →0 s p →0 sn →0
k=1,,p

 β∈∆ p hλ − ρ, β∨ i   Y b

 
ζF (hλ, α∨ i) 
Q
= lim . . . lim  .
α∈∆ hwλ − ρ, α i
 Q  
∨ 
α∈Φ ζF (hλ, α i + 1)
hλ−ρ, β∨1 i→0 hλ−ρ, β∨n−1 i→0 b ∨
w

Here Res or lim under b means the corresponding term is omitted as usual.
Therefore, to get a non-zero contribution for the particular w ∈ W, there should
be cancellations between either the common factors appeared in both numerator
and denominator in the first big bracket, or the rational factors in the numerator of
the first big bracket and the same number of poles from the zeta function factors
in the second big bracket. Recall that the complete Dedekind zeta function b ζF (s)
of F admits only two singularities, namely, two simple poles at s = 0, 1. Set
cF := Res  s=1 ζF (s). Then,  using (9.14) again, naturally, we should rewrite the
b
 Y 
product  hλ − ρ, β∨ i · Aw (λ) as
β∈∆ p
   
 Y   Y 1   Y 1 
β∨

 hλ, i−1     

β∈∆ p
 hλ, α∨ i−1   hλ, α∨ i−1  
α∈(w (∆)∪Φw )∩∆ p
−1 α∈(w (∆))r∆ p
−1
   (9.15)
 Y hλ, α∨ i − 1
b ζF (hλ, α∨ i)   Y ζ (hλ, α ∨
i) 
F  .
b
×     
α∈Φw ∩∆ p ζF (hλ, α∨ i + 1)
b 
α∈Φw r∆ p ζF (hλ, α i + 1)
b ∨ 
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 251

Zeta Function of (SLn , Pn−1,1 ) 251

In particular, when s j → 0 for all 1 ≤ j ≤ n, j , p, all the factors in the first big
bracket should be cancelled out with those in either the second and or the fourth
big brackets. Indeed, we have the following:

Lemma 9.1. Let s p = s and sk = 0 for all k , p.

(a) The second group of factors is given by

Y 1 Y 1
= .
hλ, α i − 1
∨ sk
−1α∈(w (∆)∪Φw )∩∆ p αk ∈(w−1 (∆)∪Φw )∩∆ p

(b) For α ∈ (w (∆)) r ∆ p , within the third group of factors,

−1

hλ, α∨ i − 1 = hλ p , α∨ i s + hρ, α∨ i − 1 . 0.
(c) For α ∈ Φw ∩ ∆ p , within the fourth group of factors,
hλ, α∨ i − 1 bζF (hλ, α∨ i)


cF
−→ + o(sk ).
ζF (hλ, α∨ i + 1)
b ζF (2)
b
ζF (hλ, α∨ i)
(d) For α ∈ Φw r ∆ p , within the fifth group of factors, the ratio
b
ζF (hλ, α∨ i + 1)
b
is a well-defined meromorphic function.

Proof. (a) This is a direct consequence (9.14).

(b) Since α ∈ (w−1 (∆)) r ∆ p , by (9.14), under the conditions sk = 0 for k , p,
hλ, α∨ i = hλ − ρ, α∨ i + hρ, α∨ i − 1 = hλ p , α∨ i s + hρ, α∨ i − 1. (9.16)
This proves the first equality. Moreover, since α ∈ w (∆) r ∆ p , hence hρ, α i , 1
−1 ∨

or hλ p , α∨ i , 0. This implies the last inequality.

(c) Since α ∈ Φw ∩ ∆ p , we write α = αk for 1 ≤ k ≤ n, k , p. Thus, by (9.14),
hλ, α∨ i − 1 bζF (hλ, α∨ i) ζF (sk + 1)


sk b
= .
hλ, α i + 1)
∨
ζF (sk + 2)
b
ζF (sk + 1)
sk b cF
Hence, when sk → 0 for αk ∈ ∆ p , we have → + o(sk ).
bζF (sk + 2) ζF (2)
b
(d) Similarly, as in (a), by (9.14), we have, for α ∈ Φw r ∆ p ,
ζF (hλ, α∨ i) ζF (hλ p , α∨ i s p + hρ, α∨ i)
b
= ,
b
(9.17)
ζF (hλ, α∨ i + 1)
b ζF (hλ p , α∨ i s p + hρ, α∨ i + 1)
b
under the conditions that sk = 0 for k , p. In addition,

(i) If hλ p , σ∨ i = 0, then α ∈ Φ+p r ∆ p and hence hρ, α∨ i ≥ 2. In this case, the

right hand side, and also the left hand side of (9.17) is meromorphic.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 252

252 Zeta Functions for Reductive Groups

(ii) If hλ p , σ∨ i , 0, it is clear, since the coefficients of s is not zero. 

Consequently,
 for a fixed w ∈ W, to have a non-zero constant contribution
 Y 
from  hλ − ρ, β∨ i · Aw (λ) in the case s p = s and all sk = 0 for k , p, all the
β∈∆ p
hλ − ρ, β∨ i’s in the first group of factors in (9.15) should be cancelled by either the
factors in (a) or the factors in (c) completely. This implies that, for this particular
w ∈ W, we must have that ∆ p ⊂ w−1 (∆) ∩ Φw , or equivalently, that w satisfies
the condition that ∆ p = α ∈ ∆ p : wα ∈ ∆ ∪ Φ− . Namely, w ∈ W 0 is special.

Moreover, by Lemma 9.1, we conclude that
 
X
w(sλ p +ρ)−ρ,T i 
 Y 1 
ωF (s; T ) =
G/P
e  
, α ∨ i s + hρ, α∨ i − 1 

 hλ p 
w∈W 0 α∈(w−1 (∆))r∆ p
  
 Y c F
  Y ζ
b F (hλ p , α ∨
i s + hρ, α∨
i) 
×    
α∈Φw ∩∆ p ζF (2) α∈Φw r∆ p ζF (hλ p , α i s + hρ, α i + 1)
b  b ∨ ∨ 
 
X  Y 1 
ew(sλ p +ρ)−ρ,T i cF w p 
|Φ ∩∆ |
= 
hλ p , α∨ i s + hρ, α∨ i − 1 
w∈W 0 α∈(w (∆))r∆ p
−1
   (9.18)
 Y   Y 1 
×  ζF (hλ p , α i s + hρ, α i) 
∨ ∨
 b   
α∈Φw ζF (hλ p , α is+hρ, α i+1)
∨ ∨ 
α∈Φw r∆ p
b
 
X  Y 1 
ew(sλ p +ρ)−ρ,T i cF
|Φ ∩∆ |
= w p
 
 hλ p , α∨ i s + hρ, α∨ i − 1 
w∈W 0 α∈(w (∆)r∆ p
−1
  
 Y   Y 1 
×  ζF (hλ p , α i s + hρ, α i) 
∨ ∨  .
 b  

α∈Φw r∆ p α∈(−Φw ) Fζ
b (hλ p , α ∨ i s+hρ, α∨ i) 

Hence, to complete the proof of the proposition, it suffices to apply the functional
ζF (1 − s) = b
equation b ζF (s) to the final group of factors. 

9.1.3 Special Permutations

From now on till the end of this chapter, we will specialize our discussion to
the case that G is the special linear group SLn and P is the maximal parabolic
subgroup P consisting of matrices whose final row vanishes except for its last
entry, corresponding to the ordered partition (n − 1) + 1 of n.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 253

Zeta Function of (SLn , Pn−1,1 ) 253

9.1.4 Working Site

Since we are working on SLn , let V e = (Rn , h·, ·i) denote the canonical Eu-
n
clidean space R , namely, the n-dimensional real vector space equipped with the
n
X
standard Euclidean metric h(xi ), (yi )i := xi yi . Denote the associated canon-
i=1
n
e1 , . . . , en so that Rn 3 (x1 , . . . , xn ) = xi ei . Then,
X
ical orthonormal basis by
i=1
by §6.1.3, for the semi-simple Lie group SLn , its root space is given by
Xn
V = (x1 , . . . , xn ) ∈ V xi = 1 , and the set ∆ of simple roots is given by
 e:
i=1
∆ = α1 , . . . , αn−1 , where


α1 = e1 − e2 , . . . , αn−1 = en−1 − en . (9.19)

Moreover, the set of roots Φ and the set of positive roots Φ+ can be described by
Φ = αi j := ei − e j : 1 ≤ i, j ≤ n, i , j and Φ+ = ei − e j : 1 ≤ i < j ≤ n ,
 

respectively. In addition, with an identification of V and its dual space V ∨ by h·, ·i,
the associated fundamental weights λ j nj=1 are characterized by


j n
j X
ek − ei
X
λj = 1 ≤ j ≤ n − 1. (9.20)
k=1
n i=1
In particular,
n−1
1 X X 1 
ρ= α= λ j = n − 1, n − 3, . . . , −(n − 1) (9.21)
2 α∈Φ+ j=1
2

and the (n − 1)-st fundamental dominant root (associated to Pn−1,1 ) is given by

1 
λn−1 = 1, . . . , 1, −(n − 1) . (9.22)
n
2
Obviously, α∨ = = α for all α ∈ Φ and hρ, α∨ i = 1 for all α ∈ ∆.
hα, αi
The Weyl group W is isomorphic to the symmetric group Sn on n-symbols,
with the actions of its elements σ ∈ Sn naturally described by
(x1 , . . . , xn ) 7→ (xσ(1) , . . . , xσ(n) ) ∀σ ∈ Sn . (9.23)
Form now on, when treating SLn , we will identify the Weyl group W with Sn .
Clearly, h·, ·i is Sn -invariant. That is,
hσ(x), σ(y)i = hx, yi, hσ(x), yi = hx, σ−1 (y)i ∀σ ∈ Sn , ∀x, y ∈ V. (9.24)
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 254

254 Zeta Functions for Reductive Groups

In particular, hwλ − ρ, α∨ i = hλ, w−1 (α)i − 1 for all α ∈ ∆.

Let P = Pn−1,1 be the standard maximal parabolic subgroup associated to the
partition n = (n−1)+1, corresponding to the simple root αn−1 = en−1 −en , or better,
to the subset α1 , . . . , αn−2 of ∆. Denote by P = MN its Levi decomposition.

Then M ' SLn−1 . We have
∆P = α1 , . . . , αn−2 := ∆0 ,
 




ΦP = ei − e j : 1 ≤ i, j ≤ n − 1, i , j ,
 
(9.25)



Φ+ = e − e : 1 ≤ i < j ≤ n − 1 .


 
P i j
Hence, in this case, (ΦP , ∆P ) does form a root system associated to SLn−1 . For
our own convenience, we view this system as a subsystem of that for SLn . For
example, under this, the associated ρP is given by
1 
ρP = n − 2, n − 4, . . . , −(n − 2), 0 ∈ Rn , (9.26)
2
1 
instead of a vector n − 2, n − 4, . . . , −(n − 2) of Rn−1 .
2
n−1
X
Recall that we have introduced a coordinate system λ = ρ + s j λ j . on V.
j=1
Xn−1 Xn−1
This is rather easy to use. Indeed, if a = ai αi = ai α∨i ∈ V,
i=1 i=1
Xn−1 n−1
X n−1
X
hλ − ρ, ai = h s jλ j, ai α∨i i = ai si .
j=1 i=1 i=1
This then verifies the following:

Lemma 9.2. We have

(1) For 1 ≤ i ≤ n − 1, hλ − ρ, α∨i i = si .

(2) Let α ∈ ∆. Then w−1 (α) ∈ ∆P if and only if hwλ − ρ, α∨ i = sk for a certain
1 ≤ k ≤ n − 2.

For later use, we will also write ∆P , Φ+P , ρP and λP simply as ∆0 , Φ0 + and ρ0 , $0
respectively. Easily, ∆0 = α1 , . . . , αn−2 , Φ0 + = {ei − e j : 1 ≤ i < j ≤ n − 1},


j=1 2 e j and $ = λn−1 = n j=1 e j − en . In addition, hρ, αi = 1

n−2 j
ρ0 = n−1 0 1 Pn
P
for all α ∈ ∆, and α∨ = α, hρ, αi = 1 for all α ∈ Φ+ . Hence ρ0 = ρ − n2 $0 .
Accordingly, for positive roots αi j := ei − e j ∈ Φ+ ,
hρ, αi j i = j − i, h$0 , αi j i = δ jn − δin , (9.27)
and, for λ s := s$ + ρ,
0




 j−i if i, j , n,
hλ s , αi j i =  s + n − i if j = n,

(9.28)



−s − n + j if i = n.


November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 255

Zeta Function of (SLn , Pn−1,1 ) 255

9.1.5 Special Permutations

Let S0n be the subset of Sn , which, under the identification Sn ' W, corresponds
to W 0 . We call elements of S0n special permutations. Since |∆P | = |∆| − 1 = n − 1,
by definition, σ ∈ S0n is special if and only if
α ∈ ∆P : wα < 0 ∪ α ∈ ∆ : σ(α) ∈ ∆P has cardinality n − 2.
 

Thus, by α ∈ ∆P : σ(α) < 0 ∪ β ∈ ∆P : σ(β) ∈ ∆ ⊆ ∆P , we have proved the

 
first statement of the following:

Lemma 9.3. Let σ ∈ Sn .

(1) The permutation σ is special if and only if σ satisfies the following condition:
{σ ∈ ∆P : σα < 0 or σα ∈ ∆ = ∆P . (9.29)
(2) ([58]) For special permutation σ ∈ S0n , if σ(ei − en ) > 0, then σ(ei+1 − en ) > 0.

Proof. It suffices to prove (2). Let σ(i) = p, σ(n) = q, σ(i + 1) = r. By our

assumption σ(ei − en ) > 0. Hence p < q. On the other hand, since σ ∈ S0n , we
have either σ(ei − ei+1 ) > 0 or σ(ei − ei+1 ) ∈ ∆. That is, either r < p or r = p + 1.
When r < p, we have r < q, which is equivalent to σ(ei+1 − en ) > 0. For the
remaining case of r = p + 1, clearly q cannot be p + 1 at the same time since σ is
a permutation. Hence q > p ≥ p + 2 = r + 1. This also implies that r < q, or the
same, σ(ei+1 − en ) > 0. 

To go further, we next follow Weng-Zagier ( [135]) to describe special permu-

tations explicitly. To start with, for σ ∈ Sn , set
Aσ := α ∈ ∆P : σ(α) ∈ ∆ Bσ := α ∈ ∆P : σ(α) ∈ Φ− . (9.30)
 
and
Then by Lemma 9.3(1), σ is special if and only if Aσ t Bσ = ∆P . This implies that
σ is special if and only if σ(i + 1) = σ(i) + 1 or σ(i + 1) < σ(i) for all 1 ≤ i ≤ n − 2
(or equivalently, since σ is a permutation, if and only σ(i + 1) ≤ σ(i) + 1 for all
1 ≤ i ≤ n − 2). Denote by t1 > t2 > · · · > tm the distinct values of σ(i) − i
for 1 ≤ i ≤ n − 2, and by Iν (1 ≤ ν ≤ m) the set of i ∈ {1, 2, . . . , n − 2} with
σ(i) − i = tν . Then σ maps Iν onto its image Iν0 = σ(Iν ) by translation by tν , and we
have Iν = {1, . . . , n − 1} and Iν0 = {1, . . . , n} r {a}, where a = σ(n) ∈ {1, . . . , n}.
F F

Lemma 9.4. ([135]) We have I1 < · · · < Im (in the sense that all elements of Iν
are less than all elements of Iν+1 if 1 ≤ ν ≤ m − 1) and I10 > · · · > Im0 (in the same
sense).
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 256

256 Zeta Functions for Reductive Groups

Proof. Indeed, let A denote the set of indices i ∈ {1, . . . , n − 2} with σ(i + 1) =
σ(i) + 1. Then σ(i) − i is constant when we pass from any i ∈ A to i + 1, so each
set Iν is a connected interval that is contained in A except for its right end-point i0 ,
which satisfies σ(i0 + 1) < σ(i0 ), so that i0 + 1 belongs to an Iµ satisfying tµ < tν
and hence µ > ν. But then Iµ contains a point that is bigger than one of the points
of Iν and that has an image under σ that is smaller than the image of that point,
and since all of these sets are connected intervals this means that all of Iµ lies to
the right of all of Iν and that all of Iµ0 lies to the left of all of Iν0 . 

These properties characterize special permutations and are illustrated in the

figure below, in which the lengths of the intervals Iν with Iν0 above (respectively
below) a are denoted by k1 , . . . , k p (resp. by `1 , . . . , `r ). We will denote the corre-
sponding special permutation by σ(k1 , . . . , k p ; a; l1 , . . . , lr ).

σ(i)

k1

k2
..
.
kp
a
l1

..
.

lr
i
k1 k2 · · · kp l1 ··· lr n
n−a a−1

Fig. 9.1: The special permutation σ(k1 , . . . , k p ; a; l1 , . . . , lr )

All these then prove the following:

Proposition 9.2. ( [135]) σ ∈ Sn is a special permutation if and only if σ is of

the form σ(k1 , . . . , k p ; a; l1 , . . . , lr ) for suitable a, k1 , . . . , k p and l1 , . . . , lr in Z≥0 .
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 257

Zeta Function of (SLn , Pn−1,1 ) 257

p
X r
X
Obviously, ki = n − a, ` j = a − 1, and p + r = m. Accordingly, for
i=1 j=1
later use, we introduce the numbers 0 = K0 < K1 < · · · < K p = n − a and
0 = L0 < L1 < · · · < Lr = a − 1 by
Ki = k1 + · · · + ki (1 ≤ i ≤ p), L j = l1 + · · · + l j (1 ≤ j ≤ r). (9.31)
We end this discussion on special permutations by determining the totality of
S0n . A direct computation shows |S02 | = 2, |S03 | = 5, |S04 | = 12. More generally,
we have the following:

Proposition 9.3. ([58], see also [135]) For n ≥ 3, an = 2n−3 (n + 2).

Proof. We first count the number of special permutations σ according to σ(1).

If σ(1) = 1, then σ = 1. If σ(1) = 2, then σ(2) = 3, . . . , σ(n − 2) = n − 1,
and σ(n − 1) = n or 1. If σ(1) = 3, then σ(2) = 4, . . . , σ(n − 3) = n − 1 and
σ(n − 2) = n or 1, 2. There are 4 such elements.
If σ(1) = p < n, then σ(2) = p + 1, . . . , σ(n − p) = n − 1, and σ sends
n − p, . . . , n − 1, n to 1, . . . , p − 1, n. By Proposition 9.2, we have σ(n − p + 1) = n
or σ(n) = n. If σ(n − p + 1) = n, then 1, . . . , p − 1 are arranged according to
that for the group SL p−1 . So there are exactly |S0p−1 | choices for these special
permutations. If σ(n) = n, σ sends n − p + 1, . . . , n − 1 to 1, . . . , p − 1. By
induction, the possibilities are 1 + 1 + 2 + 22 + · · · + 2 p−3 = 1 + 2 p−2 − 1 = 2 p−2 .
If σ(1) = n, then we can arrange 2, . . . , n according to SLn−1 case. So there
are |S0n−1 | elements.
Hence, the totality of S0n is given by
|S0n | = 1 + 2 + (2 + 2) + |S03 | + 22 + |S04 | + 23 + · · · + |S0n−2 | + 2n−3 + |S0n−1 |

= |S0n | = |S03 | + · · · + |S0n−1 | + 2n−2 + 3 = 2|S0n−1 | + 2n−3 .

By solving this recurrence relation, we have the result.
Alternatively, we count the special permutations σ according to σ(n).

Lemma 9.5. ([58]) We have


n o 2
 n−3
k = 2, . . . , n − 1
# w ∈ Sn : w(n) = k = 
0
. (9.32)
2n−2
 k = 1, n
Proof. We use two slight different methods to count the set S0n,a (1 ≤ a ≤ n) of
special permutations σ in Sn with σ(n) = a.
(i) ( [58]) When k , 1, n, by Lemma 9.3(2), σ sends 1, . . . , n − a to a +
 
1, . . . , n , and send n − a + 1, . . . , n − 1 to 1, . . . , a − 1 . By induction as in the
 
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 258

258 Zeta Functions for Reductive Groups

first proof, the number of arrangement of n − a + 1, . . . , n is 2a−1 , and the number


of arrangement of 1, . . . , n − a − 1 is 2n−a−2 . Hence the total number of special

permutation of this type is 2n−a−2 · 2a−1 = 2n−3 .
If k = 1, then σ should send 1, . . . , n − 1 to 2, . . . , n . The number of
 
arrangement is 2n−2 . Similarly, when k = n, the associated number is 2n−2 .
(ii) ([135]) From Proposition 9.2, particularly, by Figure 9.1 on the special per-
mutation σ(k1 , . . . , k p ; a; l1 , . . . , lr ) above, we find that S0n,a  Xn−a × Xa−1 where
XK for K ≥ 0 is the set of ordered partitions of K (decompositions K = k1 +· · ·+k p
with all ki ≥ 1). Clearly the cardinality of XK equals 1 if K = 0 (in which case
only p = 0 can occur) and 2K−1 if K ≥ 1 (the ordered partitions of K are in 1:1
correspondence with the subsets of {1, . . . , K − 1}, each such subset dividing the
interval [0, K] ⊂ R into intervals of positive integral length), so |Sn,a | equals 2n−2
for a ∈ {1, n} and 2n−3 for 1 < a < n. 

Consequently, the totality of S0n is 2 · 2n−2 + 2n−3 (n − 2) = 2n−3 (n + 2). 

Remark 9.1. (1) When n = 2, a2 = 2. Therefore, an is odd if and only if

n = 3. This can be used to give a meromorphic decomposition for the Riemann
zeta function bζ(s).
√ 1
(2) The cardinality of Sn is n!. By Stirling’s formula, n! ∼ 2π nn+ 2 e−n . Hence,
there is a significant portion of the special permutations in Sn .

9.2 Explicit Formula for Zeta Function of (SL n, P n−1,1 )

9.2.1 Rough Formula for Period of (SL n, P n−1,1 )

SL /P
To write down b ζQ n n−1,1 (s) explicitly, as in the proof of Proposition 9.1, we write
the multiple residues for (SLn , Pn−1,1 ) as a single limit. Indeed, since
hλ s − ρ, αn−1 i = s and lim hλ − ρ, αi ≡ 0 (∀ α ∈ ∆0 ) ,
λ→λ s
SL , P
Y  (9.33)
ωQ n n−1,1 (s) = lim (hλ − ρ, αi) · ωSLn
(λ) .
λ→λ s Q
Xα∈∆
0

Recall that, by (9.11), ωSLn

Q (λ) = Aw (λ). Accordingly, to pin down the non-
w∈W
zero contribution for the terms Qappearing in the limit, we should consider, for a
α∈∆0 hλ − ρ, αi · Aw (λ) , or equivalently, for an


w ∈ W, the function limλ→λs
σ ∈ Sn (' W), the function
0 hλ − ρ, αi ζ(hλ, αi)
Q !
Lσ (s) = lim Q α∈∆
Y
.
b
(9.34)
λ→λ s β∈∆ hσλ − ρ, βi α∈Φ+ , σ(α)<0 b
ζ(hλ, αi + 1)
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 259

Zeta Function of (SLn , Pn−1,1 ) 259

In order for Lσ (s) to be different from the constant function zero, by the first
two relations of (9.33), there should be a complete cancellation of all factors hλ −
ρ, αi in the numerator of the first term in (9.34) that vanish at λ = λ s with either

(i) the factors hσλ s −ρ, βi appearing in the denominator of the first term in (9.34),
or else
(ii) the poles at λ = λ s of the factors b
ζ hλ, αi appearing in the numerator of the


second term in (9.34) for which hλ s , αi = 1.

Since h· , ·i is σ-invariant, for α ∈ ∆0 , by (9.27), hσλ s − ρ, αi = hλ s , σ−1 αi − 1.

(S L , P )
Hence, for Lσ (s) to have a non-zero contribution to ωQ n n−1,1 (s), the union of
Aσ := α ∈ ∆0 : σα ∈ ∆ and Bσ := α ∈ ∆0 : σα < 0
 
(9.35)
must be of cardinality n − 2, or equivalently, ∆0 = Aσ t Bσ . This recovers special
permutations and also completes the proof of the first part of following:

Lemma 9.6. Let S0n be the collection of all special permutations.

SLn , Pn−1,1
(1) The period ωQ (s) is given by
SLn , Pn−1,1
X
ωQ (s) = Lσ (s) . (9.36)
σ∈S0n

(2) For a special permutation σ ∈ S0n ,

1
Lσ (s) = ζ [n] (s) · b
·b ζσ[<n] (s) , (9.37)
Rσ (s) σ
where
Y
Rσ (s) := hσλ s − ρ, αk i,
1≤k≤n−1
σ−1 αk <∆0
Y ζ(hλ s , αin i)
ζσ[n] (s) := ,
b
b
1≤i≤n−1 ζ(hλ s , αin i + 1)
b
σ(i)>σ(n)

ζ(hλ, αi j i)
Y Y b !
ζσ[<n] (s) :=
b 1−q −hλ−ρ,αk i

.
1≤k≤n−2 1≤i< j≤n−1 ζ(hλ, αi j i + 1)
b λ=λ s
σ(k)>σ(k+1) σ(i)>σ( j)

(3) For a special permutation σ ∈ S0n ,

Y b ζ(k) mσ (k) Y
ζσ[<n] (s) =
b = ζ(k)nσ (k)
b (9.38)
k≥1 ζ(k + 1)
b k≥1
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 260

260 Zeta Functions for Reductive Groups

ζ(1) := Res s=1 b

where b ζ(s) and

1 = # α ∈ Φ0 + : σα < 0, hρ, αi = k ,
X n o
mσ (k) :=
1≤i< j≤n−1
σ(i)>σ( j), j−i=k (9.39)
nσ (k) := mσ (k) − mσ (k − 1) , nσ (1) = mσ (1) = #Bσ .

Proof. To prove (2), we regroup the terms of (9.34) for special permutation σ ∈
S0n . First we cancel the terms in the numerator of the first factor in (9.34) for
α ∈ Aσ with the corresponding terms in the denominator for β = σα. The first
factor 1/Rσ (s) in (9.37) is the value at λ = λσ of the product of the remaining
terms β ∈ D r σAσ in this denominator. The second factor b ζσ[n] (s) in (9.37) is
the value at λ = λσ of the product of the terms in the second factor in (9.34) for
α < Φ0 + , i.e. α = ei − en > 0. The third factor b ζσ[<n] (s) in (9.37), which can also be
written as
ζ(hλ, αi)
Y Y b !
ζσ (s) =
b[<n]
(hλ − ρ, αi) · ,
α∈Bσ α∈Φ0 + ζ(hλ, αi + 1) λ=λ s
b
σ(α)<0

is obtained by collecting all the remaining zeta factors and rational factors appear-
ing in the numerator.
To prove (3), we note that the terms occurring in b ζσ[<n] (s) are of two types: for
ζ(hλ, αi j i)
b
α ∈ Bσ we must combine the quantities hλ−ρ, αk i and before taking
bζ(hλ, αi j i + 1)
the limit as λ → λ s because the first has a zero and the second has a pole, while
in the remaining zeta-quotients from the second term in (9.37), corresponding to
α ∈ Φ0 + r Bσ , we could simply substitute λ = λ s instead of taking a limit. Since
the residue of bζ(s) at s = 1 is 1, using the first equation in (9.28), we verify the
relation (3), whose right hand side, as one sees, does not depend on s at all. 

In (3), the numbers nσ (k) in (9.38) may well be negative. To obtain the zeta
functions, we should remove them according to §8.4.2 by a certain normalization
process. For details, please see the next subsection.

9.2.2 Applications of Special Permutations

In this subsection, following [135], we apply the results in §9.1.5 for special per-
mutations to calculate the rational factor Rσ (s) and the zeta factors b ζσ[n] (s) and
ζσ[<n] (s) appearing in Lemma 9.6(2) explicitly. We begin with Rσ (s).
b
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 261

Zeta Function of (SLn , Pn−1,1 ) 261

Lemma 9.7. Let σ = σ(k1 , . . . , k p ; a; l1 , . . . , lr ) be a special permutation. Then

Rσ (s) = (−1) p+r (k1 + k2 ) · · · (k p−1 + k p ) · (s + a + k p )
× (−s − a + l1 + 1) · (l1 + l2 ) · · · (lr−1 + lr ) .

Proof. By definition, Y Y
Rσ (s) = hσλ s − ρ, αk i = (hλ s , σ−1 αk i − 1).
1≤k≤n−1 1≤k≤n−1
σ−1 (αk )<∆0 σ−1 (αk )<∆0
For each k occurring in this product, write ei − e j =: αi j . Then the
σ−1 (αk ) =
condition αi j < ∆ says that the points (i, σ(i) = k) and ( j, σ( j) = k + 1) do not
0

belong to the same square block in Figure 9.1 of σ. Furthermore, we see that the
k’s occurring in the product, in decreasing order, together with the corresponding
values of i and j, are given by the first three colums of the following table

k i = σ−1 (k) j = σ−1 (k + 1) 1 − hλ s , αi j i

n − Kµ (1 ≤ µ < p) Kµ+1 Kµ−1 + 1 kµ + kµ+1
a n K p−1 + 1 s + a + kp
a−1 n − a + l1 n −s − a + l1 + 1
n − Lν (1 ≤ ν < r) Lν+1 Lν−1 + 1 lν + lν+1

while the fourth column follows from equation (9.28). 

ζσ[n] (s).
We next consider the zeta factor b

Lemma 9.8. Let σ = σ(k1 , . . . , k p ; a; l1 , . . . , lr ) be a special permutation. Then

ζ(s + a)
ζσ[n] (s) = .
b
b
ζ(s + n)
b

ζσ[n] (s) we at least need to

This lemma implies, in particular, that to normalize b
clear the denominator by multiplying by the zeta factor ζ(s + n).
b

Proof. This is much easier. From λ s = s$+ρ, we get hλ s , ei −en i = s+n−i. More-
over, from Figure 9.1, for the special permutation σ = σ(k1 , . . . , k p ; a; l1 , . . . , lr ),
we have
ei − en : 1 ≤ i < n, σ(i) > σ(n) = e1 − en , e2 − en , . . . , en−a − en .
n o n o

Therefore, by the definition of b ζσ[n] (s) given in Lemma 9.6(2), we have

n−a
Y ζ(hλ, αi) Y ζ(s + n − i) ζ(s + a)
ζσ[n] (s) = = =
b b b
b
α=ei −en , i≤n−1 ζ(hλ, αi + 1) i=1 ζ(s + n − i + 1) ζ(s + n)
b λ=λ s b b
σ(i)>σ(n)
as asserted. 
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 262

262 Zeta Functions for Reductive Groups

Before treating the zeta factor bζσ[<n] (s), recall that we need to make some nor-
malizations by clearing out the zeta factors appeared in the denominators. For
this, introduce


 # Bσ if k = 1,
rσ (k) := 

nσ (k) − n(k) if k ≥ 2,


where, in analogy with the numbers nσ (k) (equation (9.39), we have set
n o
m(k) = # α > 0 : hρ, αi = k and n(k) = m(k) − m(k − 1) . (9.40)
Clearly m(k) = n − k for 1 ≤ k ≤ n and n(k) = −1 for 2 ≤ k ≤ n. Moreover, by
Lemma 9.6(3),
Y Y Y
bζσ[<n] (s) · ζ(i)−n(i) = b
b ζ(1)#Bσ ζ(i)nσ (i)−n(i) =
b ζ(i)rσ (i) .
b (9.41)
i≥2 i≥2 i≥1
Y
We claim that ζσ[<n] (s)
b · ζ(i)
b −n(i)
would do this job. Indeed, if we set, for k ∈
i≥2
Z≥1 , the Siegel number b
vk by
k
Y
vk :=
b ζ(`),
b (9.42)
`=1

then we have the following:

Lemma 9.9. Let σ = σ(k1 , . . . , k p ; a; l1 , . . . , lr ) be a special permutation. Then

Y p
Y r
Y
ζ(i)rσ (i) =
b vki ·
b vl j .
b (9.43)
i≥1 i=1 j=1

In particular, rσ (k) ≥ 0.

Proof. This is based on a detailed analysis of rσ (k). Obviously,

n o n o
rσ (1) = # α ∈ ∆0 : σα < 0 = # (i, i + 1) : 1 ≤ i ≤ n − 2, σ(i) > σ(i + 1) .
If k ≥ 2, by definition,
m(k) − mσ (k) = # α > 0 : hρ, αi = k − # α ∈ Φ0 + : σα < 0, hρ, αi = k
n o n o

= # ei − en : hρ, αi = k + # α ∈ Φ0 + : σα > 0, hρ, αi = k

n o n o

= 1 + # α ∈ Φ0 + : σα > 0, hρ, αi = k ,
n o

since, by (9.27), ei − en : hρ, αi = k = en−k − en . Thus, by applying the

 
characterization graph in §4 for special permutation σ(k1 , . . . , k p ; a; l1 , . . . , lr ), we
conclude that α = αi j ∈ Φ0 + satisfying σα > 0 (or equivalently α = αi j satisfying
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 263

Zeta Function of (SLn , Pn−1,1 ) 263

i < j ≤ n − 1 and σ(i) < σ( j)) if and only if i and j belong to the same block,
say Iµ for some µ, associated to σ(k1 , . . . , k p ; a; l1 , . . . , lr ), and also σ( j) ∈ Iµ (or
equivalently j + 1 ∈ Iµ ), since otherwise σ(αi j ) < 0.
Denote by (m(k) − mσ (k))µ (resp. rσ,µ (k)) the contribution to m(k) − mσ (k)
(resp. to rσ (k)) of the block Iµ . With the discussion above, we have
X X
m(k) − mσ (k) = (m(k) − mσ (k))µ and rσ (k) = rσ,µ (k).
µ µ

Fix some µ and let Iµ := a+1, a+2, . . . , a+b with a, b ∈ Z>0 . Clearly, when

k = 1, rσ,µ (1) = # (a + b − 1, a + b) = 1, since, for other (i, i + 1), σ(i) < σ(i + 1).

Moreover, when k ≥ 2, by (9.27) and the characterization of the graph again,
n o
(m(k) − mσ (k))µ = # (i, j) : i, j + 1 ∈ Iµ , i < j, j = i + k
n o
= # (i, j) : a + 1 ≤ i < j < a + b, j = i + k .
Note that, for each fixed i (with a + 1 ≤ i < a + b),

n o 1 i + k < a + b

# (i, j) : a + 1 ≤ i < j < a + b, j = i + k =  .
0 i + k ≥ a + b


Hence, (m(k) − mσ (k))µ = b − (k + 1). This implies that, for all k ≥ 1, we have
rσ,µ (k) = (m(k − 1) − mσ (k − 1))µ − (m(k) − mσ (k))µ = 1. Consequently,
Y
ζ(k)rσ,µ (k) = b
b ζ(1) b
ζ(2) · · · b
ζ(b) .
i≥1

Equation (9.43) follows. 

9.2.3 Explicit Formula for Zeta Functions of (SL n, P n−1,1 )

Motivated by Lemmas 9.8, 9.9, let

n
SLn ,Pn−1,1
Y
NQ (s) = ζ(k) · b
b ζ(s + n). (9.44)
k=1

Lemma 9.10. In the case  of (G, P) = (SL , P ), the normalization factor

QI(G/P) b  G/P Q J(G/P) b  G/Pn  n−1,1
i=1 ζ F ai s P + bG/P
i and j=1 ζ F c j in §8.4.2 ζ(s + n) and
is given by b
Yn
ζ(k), respectively. In particular,
b
k=1
SL ,P SL ,P (SL ,P )
ζQ,on n−1,1 (s) = NQ n n−1,1 (s) · ωQ n n−1,1 (s) .
b (9.45)

Proof. This is a direct consequence of Lemmas 9.8, 9.9, since, by definition,

n(k) = −1 for k = 2, 3, . . . , n and b
ζ(1) = 1. 
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 264

264 Zeta Functions for Reductive Groups

Finally, we are ready to state the main theorem of this chapter.

SLn /Pn−1,1
ζF
Theorem 9.1. Let b (s) be the zeta function of (SLn , Pn−1,1 ) over F.

(1) We have
SL /P SL /P
ζF n n−1,1 (s) = b
b ζF,on n−1,1 (ns − n). (9.46)
Here
n
SL /P
X X X
ζF,on n−1,1 (s) =
b ζF (s+a) (9.47)
(−1) p+r−1Rk1 ,...,k p ;a;l1 ,...,lr (s) b
a=1 k1 ,...,k p >0 l1 ,...,lr >0
k1 +···+k p =n−a l1 +···+lr =a−1

with
vk1 . . .b
b vk p 1
Rk1 ,...,k p ;a;l1 ,...,lr (s) := ·
(k1 + k2 ) . . . (k p−1 + k p ) s + a + k p
(9.48)
1 vl1 . . .b
v lr
,
b
× ·
−s − a + 1 + l1 (l1 + l2 ) . . . (lr−1 + lr )
k
Y
ζF (1) = cF , and b
v1 := Res s=1b
where b vk = ζF (i) (k ≥ 2).
b
i=1
(2) (Functional Equation) We have
S L /P S L /P
ζQ,on n−1,1 (−n − s) = b
b ζQ,on n−1,1 (s), (9.49)
or equivalently,
S L /P S L /P
ζQ n n−1,1 (1 − s) = b
b ζQ n n−1,1 (s). (9.50)

SL /P SL /P
Proof. The formula for b ζQ,on n−1,1 (s) and hence for bζF,on n−1,1 (s), in (1) is a sim-
ple combination of Lemmas 9.6, 9.7, 9.8, 9.9 and 9.10. With the formula (9.47)
obtained, the functional equation (9.49) in (2) comes directly from the classical
ζ(1 − s) = b
relation b ζ(s). This also justifies (9.46) in (1) and the functional equation
(9.50) in (2). 

An analogue of (1) for the function field is founded in [135] jointly with Za-
gier, and (2) is first jointly proved with Kim in [58] with a different method. In the
next chapter, we follow Komori [59] to prove the functional equation for the zeta
functions associated to general reductive groups G.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 265

Chapter 10

Functional Equation

In §8.4.2, for connected reductive groups G defined over number fields F and
their maximal parabolic subgroups P, the zeta functions b ζFG/P (s) are introduced.
In this chapter, we shall establish functional equations for these zeta functions
following [59]. As mentioned earlier, in the case (G, P) = (SLn , Pn−1,1 ), the same
result was first proved by Kim and myself jointly in [58].

10.1 Preparations

The first step is to write down explicitly the zeta factors appeared in normalization
process in §8.4.2 to define the zeta function for (G, P)/F. We continuously use
the notations in the first section in §9.1.1.

10.1.1 Statement

We start with some notational preparations. For a pair of integer (k, h) ∈ Z2 , and a
special Weyl element w ∈ W 0 , set
n o
n p,w (k, h) := # α ∈ w−1 (Φ− ) : hλ p , α∨ i = k, hρ, α∨ i = h
n o (10.1)
M p (k, h) := max n p,w (k, h − 1) − n p,w (k, h) .
w∈W 0

Obviously, n p,w (k, h) = 0 for all but finitely many pairs (k, h), and the same holds
true for M p (k, h). Since Φ = Φ+ t Φ− , when k, h ≥ 1,

n p,w (k, h) = # α ∈ Φ+ ∩ w−1 (Φ− ) : hλ p , α∨ i = k, hρ, α∨ i = h .

n o
(10.2)

Denote by Γ the Dynkin diagram of Φ and by Aut(Γ) the automorphism group

of Γ, which we identify with a (sub)group of permutation of indices 1, . . . , n .

Denote by Aut(Φ) be the automorphism group of V which preserves Φ. Then

265
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 266

266 Zeta Functions for Reductive Groups

W ⊂ Aut(Φ) and Aut(Γ) ⊂ Aut(Φ) in a natural way. Indeed, from the simple
transitivity of W on positive roots and the fact that for γ ∈ Aut(Γ), γ(∆) = ∆, we
have Aut(Φ) = Aut(Γ) n W.
Finally, introduce the constant
cG,P := cP := c p := 2 hλ p − ρ p , α∨p i ∈ Z>0 . (10.3)

With all this, following [59], we are ready to state the normalization factor and
the functional equation of zeta function associated to (G, P) over F.

Theorem 10.1. ([59])

(1) The normalization factor in Definition 8.12 is characterized by

∞ ∞ 
Y Y
ζ (s; T ) = ω (s; T )  ζF (ks + h)  .
G/P G/P M (k,h)

b b p
F;o F
k=0 h=2

(2) The zeta ζF;o

function bG/P
(s; T ) satisfies the functional equation

ζF;o
G/P
− c p − s; w0 (T ) = bζF;o
G/P
(s; T ) = b
ζF;o
G/Q
(s; γ(T )).
b

Here γ ∈ Aut(Γ) with q = γ(p).1

Corollary 10.1. For the normalization factors in Definition 8.12, we have

X X
I(G/P) = M p (k, h), J(G/P) = M p (0, h)
(k,h)∈Z2 ,k,0 h∈Z

and
I(G/P)
Y   Y∞ Y
∞
ζF aG/P
b
i s + bG/P
i = ζF (ks + h) M p (k,h) ,
b
i=1 k=1 h=2
J(G/P)
Y   Y∞
ζF cG/P
b
j = ζF (ks + h) M p (0,h) .
b
j=1 h=2

ζFG/P (s) of (G.P)/F is given by

Accordingly, the zeta function b
ζFG/P (s) := b
b ζFG/P (−c p s; 0). (10.4)

Corollary 10.2 (Functional Equation). We have

ζFG/P (1 − s) = b
b ζFG/P (s). (10.5)
1 ByTheorem 9.1, cSLn ,P = n. Indeed, cSLn ,P = n can be explicitly calculated for all semi-simple
n−1,1
groups G. For more details, please refer to the final section of this chapter.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 267

Functional Equation 267

10.1.2 Lie Theoretic Relations (I)

In the sequel till the end of this chapter, we give a proof of Theorem 10.1. We start
with two (groups of) elementary relations on the Lie theoretic structures involved.

Lemma 10.1. Let w p be the longest element w p of W p . Then

w p (ρ) = c p λ p − ρ. (10.6)

This result will be used in the proof of micro functional equations in §10.2.1.

Proof. By the definition of w p , for α ∈ Φ+p , we have w p (α) ∈ Φ−p ⊂ Φ− . Moreover,

for α ∈ Φ+ rΦ+p , we have w p (α) ∈ Φ+ since α is of the form a p α p +. . . with a p > 0
and w p (α) = a p α p + . . . remains positive. Hence,
Φw p = Φ+ ∩ w−1 + +

p (Φ ) = Φ ∩ − w p (Φ ) = Φ p
− +

and
X X
w p (ρ) = ρ − α=ρ− α = ρ − 2ρ p . (10.7)
α∈Φw p α∈Φ+p

On the other hand, since wk (Φ+p ) = Φ+p r {αk } ∪ {−αk } for k , p, we have then


wk (ρ p ) = ρ p − αk = ρ p − hρ p , α∨k iαk , and hence hρ p , α∨k i = 1. Therefore,
n
X n
X
ρp = hρ p , α∨k iλk = λk + hρ p , α∨p iλ p = ρ + hρ p − λ p , α∨p iλ p .
k=1 k=1,,p

Combining this with (10.7), we have

c p λ p − w p (ρ) = ρ + c p + 2hρ p − λ p , α∨p i λ p = ρ.




To state the second group of relations, similar to n p,w , we introduce the count-
ing functions n p (k, h) for (k, h) ∈ Z2 by
n o
n p (k, h) := # α ∈ Φ : hλ p , α∨ i = k, hρ, α∨ i = h . (10.8)
Since Φ = Φ+ t Φ− , when k, h ≥ 1,
n p (k, h) := # α ∈ Φ+ : hλ p , α∨ i = k, hρ, α∨ i = h .
n o
(10.9)
These counting functions can be understood through a special generating function.
Indeed, if we define the character of the dual Lie algebra, by omitting the Cartan
sub-algebra contributions, by
X ∨
X(v) := eα (v) ∀v ∈ V, (10.10)
α∈Φ
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 268

268 Zeta Functions for Reductive Groups

where, as usual, eα (v) := ehv,α i , then

∨ ∨

X X
ehλ p ,α it+hρ,α i =
∨ ∨
X(tλ p + ρ) = n p (k, h)ekt+h .
α∈Φ k,h=−∞

Obviously, for v ∈ V and γ ∈ Aut(Φ), we have X(γ(v)) = X(v).

Lemma 10.2. Let w0 , resp. w p , be the longest element of W, resp. of W p , and for
γ ∈ Aut(Γ), set q ∈ γ(p). Then, for (k, h) ∈ Z2 , we have

nq (k, h) = n p (k, h),

n p (k, kc p − h) = n p (k, h),
(10.11)
nq,γwγ−1 (k, h) = n p,w (k, h),
n p (k, h) − n p,w0 ww p (k, kc p − h) = n p,w (k, h).

This result will be used to prove Theorem 10.1(1) on normalizations in §10.2.

Proof. All these can be deduced from some manuuplications of the generating
series X(v). We start with

X(tλq + ρ) = X γ(tλ p + ρ) = X(tλ p + ρ),

X X X
eα (tλ p + ρ) = eα γ(tλ p + ρ) = eα tλ p + ρ .
∨ ∨
∨

α∈w−1 (Φ− ) α∈(γw−1 γ−1 )(Φ− ) α∈(γw−1 γ−1 )(Φ− )

This implies the first and the third equality, respectively.

Similarly, by Lemma 10.1, we have

X (c p + t)λ p − ρ = X tλ p + w p (ρ) = X w p (tλ p + ρ) = X(tλ p + ρ).

(10.12)

This implies the second equality.

To verify the final one, first note that w0 (Φ± ) = Φ∓ for the longest element
w0 of W. Hence Φ = w−1 (Φ− ) t w−1 (Φ+ ) = w−1 (Φ− ) t w p (w p w−1 w0 ) (Φ+ ). This


implies
X X
eα (tλ p + ρ) + eα tλ p + ρ
∨ ∨
X(tλ p + ρ) =


α∈w−1 (Φ− ) α∈(w p (w0 ww p )−1 )(Φ− )
X
α∨
X
(10.13)
eα (c p + t)λ p − ρ ,
∨
= e (tλ p + ρ) +
α∈w−1 (Φ− ) α∈(w0 ww p )−1 (Φ− )

and hence the final equation. 

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 269

Functional Equation 269

10.2 Normalization and Micro Functional Equation

10.2.1 Overdone Normalization and Micro Functional Equation

In this section, we normalize the period ωG/P

F (s; T ) by multiplying an overdone
zeta factor and prove a functional equation for the resulting function.
Introduce a normalization for the period ωG/P
F (s; T ) by

Z p (s; T ) := HF,p (s) · ωG/P

F (s; T ) (10.14)
where HF,p (s) is defined by
Y
ζF hλ p , α∨ is + hρ, α∨ i .


HF,p (s) := H p (s) := b (10.15)
α∈Φ−

Obviously, by Proposition 9.1, since H p (s) is a product runs through all α ∈ Φ− , all
zeta factors appeared in the denominators of the Aw (s)’s in ωG/PF (s; T ) are cleared
out after multiplying by H p (s). On the other hand, as to be seen later, H p (s) is
an overdone factor, since it introduce some extra zeta function factors. Despite of
this, the function Z p (s) is very nice, since we can prove a functional equation for
it. To state it, we introduce an element γ0 ∈ Aut(Γ) characterized as follows: For
the longest element w0 of W, we have w0 (∆) = −∆. Hence, there is an element
γ0 ∈ Aut(Γ) such that −Id = γ0 w0 . Obviously, γ02 = Id.

Proposition 10.1. (Proposition 5.1 of [59]) Z p (s; T ) satisfies functional equation

Z p − c p − s; γ0 (T ) = Z p (s; T ) = Z p (s; γ(T )).


(10.16)
Here γ ∈ Aut(Γ) with q = γ(p).

Proof. Since Φ− r(−Φσ ) = Φ− r Φ− ∩w−1 (Φ+ ) = Φ− rw−1 (Φ+ ) = Φ− ∩w−1 (Φ− ),

 
X 
hw(sλ p +ρ)−ρ,T i |Φw ∩∆ p | 
Y 1 
Z p (s; T ) = e cF  
, α + α

hλ ∨ is hρ, ∨ i − 1 
 p
w∈W 0 α∈(w−1 (∆))r∆P
 
 Y 
ζF hλ p , α∨ is + hρ, α∨ i  .


×   b
α∈(Φw r∆ p )∪(Φ− ∩w−1 (Φ− )

Applying the relations

(Φw r ∆ p ) ∪ (Φ− ∩ w−1 (Φ− ) = (Φ+ ∩ w−1 (Φ− ) r ∆ p ∪ Φ− ∩ w−1 (Φ− )

= Φ+ ∩ w−1 (Φ− ) ∪ Φ− ∩ w−1 (Φ− ) r ∆ p





= (w−1 (Φ− )) r ∆ p .
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 270

270 Zeta Functions for Reductive Groups

we have
 
X  Y 1 
ehw(sλ p +ρ)−ρ,T i cF w p 
|Φ ∩∆ |
Z p (s; T ) = 
hλ p , α∨ is + hρ, α∨ i − 1 
w∈W 0 α∈(w (∆))r∆P
−1
 
 Y 
ζF hλ p , α is + hρ, α i 
∨ ∨

×   b
α∈(w−1 (Φ− ))r∆ p
X r
= eh p,w (s;T ) cF f p,w (s)g p,w (s),
p,w (10.17)
w∈W 0

where r p,w := |Φw ∩ ∆ p | and

Y 1
f p,w (s) := ,
hλ p , α∨ is + hρ, α∨ i − 1
α∈(w−1 (∆))r∆P
Y
ζF hλ p , α∨ is + hρ, α∨ i (10.18)


g p,w (s) := b
α∈(w−1 (Φ− ))r∆ p

h p,w (s; T ) := hw(sλ p + ρ) − ρ, T i.

Hence, to prove the functional equation, we fist need to establish functional equa-
tions for the functions f p,w (s), g p,w (s), h p,w (s; T ), and r p,w .

Proposition 10.2. (Micro Functional Equation) For w ∈ W, γ ∈ Aut(γ) and

q = γ(p), we have
r p,w = r p,w0 ww p , r p,γ−1 wγ = rq,w ,
f p,w (−c p − s) = f p,w0 ww p (s), f p,γ−1 wγ (s) = fq,w (s),
g p,w (−c p − s) = g p,w0 ww p (s), g p,γ−1 wγ (s) = gq,w (s),
− c p − s; γ0 (T ) = h p,w0 ww p (s, T ), h p,γ−1 wγ (s; T ) = hq,w (s; γ(T )).


h p,w

Proof. By definition, r p,w0 ww p = |Φw0 ww p ∩ ∆ p | = # α ∈ ∆P : (w0 ww p )(α) < 0 .


Since w p (∆ p ) = −∆ p , we have
α ∈ ∆P : (w0 ww p )(α) < 0 = α ∈ ∆P : (w0 w)(α) > 0 = α ∈ ∆P : w(α) < 0 .
  

Here in the last step, we have used w0 (ei − e j ) = ei j− ei . Therefore r p,w0 ww p = r p,w ,
the first equality.
Similarly, by definition,
r p,γ−1 wγ = |Φγ−1 wγ ∩ ∆ p | = # α ∈ ∆P : (γ−1 wγ)(α) < 0


= # γ(α) ∈ γ∆P : γ−1 (w(γ(α))) < 0



= # β ∈ ∆q : w(β) < 0 ,

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 271

Functional Equation 271

since γ(Φ± ) = Φ± , γ(∆) = ∆ and γ(∆ p ) = ∆ p . This gives the second equality
r p,γ−1 wγ = rq,w .
We next treat f p,w and g p,w . Fix w ∈ W and γ ∈ Aut(Γ). For a subset A ⊂ Φ
with A = ∆ or Φ− , we set
S p;w (A; s) := hλ p , α∨ is + hρ, α∨ i : α ∈ w−1 (A) r ∆ p


be a set of affine linear functionals of the form a s + b with a, b ∈ Z≥0 admitting

duplications. By definition, we have
Y 1
f p,w (s) :=
as+b∈S p,w (∆;s)
as + b − 1
Y
ζF as + b .


g p,w (s) := b
as+b∈S p,w (Φ− ;s)

Note that w0 (A) = −A. Hence,

−w p w−1 (A) r ∆ p = w p w−1 (−A) r w p (−∆ p ) = w p w−1 w0 (A) r ∆ p . (10.19)

Therefore, we have
S p,w (A; −c p − s)
= hλ p , α∨ i(−c p − s) + hρ, α∨ i : α ∈ w−1 (A) r ∆ p


= hλ p , −w p (α∨ )is + hc p λ p − w p (ρ), −w p (α∨ )i : α ∈ w−1 (A) r ∆ p



= hλ p , β∨ is + hρ, β∨ i : β ∈ (w p w−1 w0 )(A) r ∆ p



= S p,w0 ww p (A; s).

This then proves the third and fifth equalities.
Similarly, from γ(∆q ) = ∆q and γ(A) = A, we have
S p,γ−1 wγ (A; s) = hλ p , α∨ is + hρ, α∨ i : α ∈ (γ−1 w−1 γ)(A) r ∆ p


= hγ(λ p ), γ(α∨ )is + hρ, γ(α∨ )i : α ∈ (γ−1 w−1 γ)(A) r ∆ p



= hλq , β∨ is + hρ, β∨ i : β ∈ w−1 (A) r ∆q



= S q,w (A; s).

This then proves the fourth and sixth equalities.
Finally, for the seventh equality, we use the calculation
h p,w − c p − s; γ0 (T ) = hw((−c p − s)λ p + ρ) − ρ, γ0 (T )i

= h−w(λ p )s + w(−c p λ p + ρ) − ρ, γ0 (T )i
= h−(ww p )(λ p )s − (ww p )(ρ) − ρ, γ0 (T )i
= h(w0 ww p )(λ p s + ρ) − ρ, T i
= h p,w0 ww p (s, T ).
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 272

272 Zeta Functions for Reductive Groups

Here, in the 3rd, resp. the 4th, equality, we have used the relation w p (ρ) = c p λ p −ρ
in Lemma 10.1, resp. the relation γ0 w0 = −Id and the fact that h·, ·i is W-invariant.
Similarly, for the sixth equality, we have

hq,w (s; γ(T )) = hw(sλq + ρ) − ρ, γ(T )i

= h(γ−1 w)(sλq + ρ) − ρ, T i
= h(γ−1 wγ)(sλ p + ρ) − ρ, T i
= h p,γ−1 wγ (s; T ).

This then completes proof of the lemma. 

With all these, it is very easy to prove the functional equations for Z p . Indeed,
by (10.17), we have

X h p,w −c p −s;γ0 (T )
r p,w
Z p − c p − s; γ0 (T ) = e cF f p,w (−c p − s)g p,w (−c p − s)
w∈W 0
X
r p,w0 ww p
= eh p,w0 ww p s;T
cF f p,w0 ww p (s)g p,w0 ww p (s)
w∈W
w(∆ p )⊂(∆∪Φ− )
X

r
= eh p,v s;T
cFp,v f p,v (s)g p,v (s)
v∈W
(vw−1 −
p )(∆ p )⊂w0 (∆∪Φ )

= Z p (s; T ).

Here, in the second to the last, resp, the last step, we have used Lemma 10.2, resp.
the facts that w0 (A) = −A for A = ∆ or Φ± and that w p (∆ p ) = −∆ p .
Similarly,
X r
Z p s; γ(T ) = ehq,w (s;γ(T )) cFq,w fq,w (s)gq,w (s)


w∈W
w(∆q )⊂(∆∪Φ− )
X r
=
−1 wγ
eh p,γ−1 wγ (s;T ) cFp,γ f p,γ−1 wγ (s)g p,γ−1 wγ (s)
w∈W
∆q ⊂w−1 (∆∪Φ− )
X r
= eh p,v (s;T ) cFp,v f p,v (s)g p,v (s)
v∈W
∆q ⊂(γv−1 γ−1 )(∆∪Φ− )

= Z p (s; T ).

Here, in the last step, we have used the facts that γ(A) = A for A = ∆ or Φ± , and
that γ(∆ p ) = ∆ p . 
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 273

Functional Equation 273

10.2.2 Normalization

As mentioned earlier, even there is no zeta factor left in the denominator of each
w term for Z p (s; T ), it is overdone. In order to make a correction so as to arrive at
the zeta function for (G, P) over F, we need to single ut the extra factors used. For
this, following [59], we divide out extra zeta function factors appeared in Z p (s; T ).
For a fixed w ∈ W 0 , let
  
 Y   Y 1 
H p,w (s) :=  ζF (hλ p , α is + hρ, α i) 
∨ ∨
 b  
 
α∈Φw ζF (hλ p , α is + hρ, α i + 1)
∨ ∨
α∈Φw r∆ p
b
be the zeta factors appeared in (9.13) corresponding to w. In terms of pairing
hλ p , α∨ i, hρ, α∨ i = (k, h), and hence n p,w (k, h), we have


Y Y
ζF (hλ p , α∨ is + hρ, α∨ i) = b
b ζF (s + 1)n p,w (1,1) ζF (hλ p , α∨ is + hρ, α∨ i)
b
α∈Φw r∆ p α∈Φw r∆
∞ Y
Y ∞
=b
ζF (s + 1)n p,w (1,1) ζF (ks + h)n p,w (k,h) ,
b
k=0 h=2
Y ∞ Y
Y ∞
ζF (hλ p , α∨ is + hρ, α∨ + 1i) =
b ζF (ks + h + 1)n p,w (k,h)
b
α∈Φw k=0 h=1
Y∞ Y ∞
= ζF (ks + h)n p,w (k,h−1) .
b
k=0 h=2
This implies that
∞ Y
Y ∞
H p,w (s) = b
ζF (s + 1)n p,w (1,1) ζF (ks + h)n p,w (k,h)−n p,w (k,h−1) .
b (10.20)
k=0 h=2
Clearly, when n p,w (k, h) − n p,w (k, h − 1) < 0, the corresponding zeta function factor
should be understood as terms in denominator of H p,w (s). To get ride of these

a a > 0

factors, we introduce a test function δ(a) :=  , a (non-negative)

0 otherwise

counting function
n  o
N p (k, h) := max δ n p,w (k, h − 1) − n p,w (k, h) (k, h) ∈ Z2 (10.21)
w∈W 0
and an auxiliary function
∞ Y
Y ∞
DF,p (s) := D p (s) := ζF (ks + h)n p (k,h−1)−N p (k,h) .
b (10.22)
k=0 h=2

Lemma 10.3. The functions H p (s) and D p (s) are related by

∞ Y
Y ∞
D p (s) = H p (s) ζF (ks + h)−N p (k,h) .
b (10.23)
k=0 h=2
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 274

274 Zeta Functions for Reductive Groups

∞ Y ∞
H p (s) Y
Consequently, = ζF (ks + h)N p (k,h) is the minimal zeta factors
b
D p (s) k=0 h=2
which should be used in the normalization process of §8.4.2 so as to get the zeta
ζFG/P (s).
function b

Proof. Indeed, by definition,

Y
H p (s) = ζF (hλ p , α∨ is + hρ, α∨ i + 1)
b
α∈Φ+
Y∞ Y ∞ ∞ Y
Y ∞
= ζF (ks + h + 1)n p (k,h) =
b ζF (ks + h)n p (k,h−1) .
b
k=0 h=1 k=0 h=2

Hence, we get the formula by related definitions. 

Lemma 10.4. For (k, h) ∈ Z × Z≥1 , we have

M p (h, k) = N p (h, k). (10.24)

Proof. When l ≥ 0, we have

n p,Id (k, l) = # α ∈ Φ− : hλ p , α∨ i = k, hρ, α∨ i = l = 0.


Hence, n p,Id (k, h − 1) − n p,Id (k, h) = 0. Consequently, in N p (k, h), δ can be removed
without changing the outcome. 

Proof of Theorem 10.1(1). As a direct consequence of this lemma, we have

∞ ∞  ∞ ∞ 
Y Y Y Y
ω (s; T )  ζF (ks + h)  = ω (s; T )  ζF (ks + h)
G/P

M p (k,h)  G/P

N p (k,h) 
 b   b 
F F
k=0 h=2 k=0 h=2
!
H p (s) G/P (10.25)
= ω (s; T ) = b
ζQ;o
G/P
(s; T )
D p (s) F
since H p (s)/D p (s) is minimal. 

10.3 Functional Equation

10.3.1 Proof of Functional Equation

In this subsection, we prove the functional equation. Recall that, by (10.14) and
(10.25), we have
!
H p (s) G/P 1
ζQ;o
bG/P
(s; T ) = ω (s; T ) = Z p (s; T ). (10.26)
D p (s) F D p (s)
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 275

Functional Equation 275

Hence, to prove the functional equation stated in Theorem 10.1(2), by Proposi-

tion 10.1, it suffices to establish a functional equation for D p (s).

Proposition 10.3. We have

D p (−c p − s) = D p (s) = Dq (s),
where γ ∈ Aut(Γ) and q = γ(p).

Proof. The last equality comes directly form the definition. To prove the first, we
decompose D p (s) into two parts, one consisting of all special zeta values and the
other for the rest. That is,
D p (s) = D0p D1p (s), (10.27)
where
∞
Y
D0p := bζ(h)n p (0,h−1)−M p (0,h) ,
h=2
∞ Y
∞
(10.28)
Y
D1p (s) := ζF (ks + h)
b n p (k,h−1)−M p (k,h)
.
k=1 h=2
Since, for k ≥ 1 and h ≤ 1, n p,w (k, h − 1) = 0 and M p (k, h) = 0, we have
Y∞ Y ∞
D1p (s) = bζF (ks + h)n p (k,h−1)−M p (k,h) . (10.29)
k=1 h=−∞
Hence to complete our proof, it suffices to show that
D1p (−c p − s) = D1p (s). (10.30)
For this, we need to treat counting functions M p (k, h) and n p (k, h).

Lemma 10.5. For (k, h) ∈ Z2 , we have

(1) n p (k, kc p − h) − M p (k, kc p − h + 1) = n p (k, h − 1) − M p (k, h).

(2) Mq (k, h) = M p (k, h) for γ ∈ Aut(Γ) and q = γ(p).

Proof. (1) By the last equality of Lemma 10.2, we have

M p (k, kc p − h + 1)
= max n p,w (k, kc p − h) − n p,w (k, kc p − h + 1)

w∈W 0

= max n p (k, h) − n p (k, h − 1) − n p,w0 ww p (k, h) + n p,w0 ww p (k, h − 1)


w∈W 0

(By the fourth equality of Lemma 10.2)

= n p (k, h) − max n p (k, h) − n p (k, h − 1) − n p,w0 ww p (k, h) + n p,w0 ww p (k, h − 1)

w∈W 0

= n p (k, h − 1) − max n p,w0 ww p (k, h − 1) − n p,w0 ww p (k, h)


w∈W 0

= N p (k, h − 1) − M p (k, h).

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 276

276 Zeta Functions for Reductive Groups

(2) By the second equality in (10.1), it suffices to prove the equality for
n p,w (k, h). But this is a direct consequence of the third equality of Lemma 10.2
and the fact that ∆ p ⊂ w−1 (∆ ∪ Φ− ) if and only if ∆q ⊂ (γw−1 γ−1 )(∆ ∪ Φ− ),
which itself can be deduced from trivial relations γ(Φ± ) = Φ± , γ(∆) = ∆ and
γ(∆ p ) = ∆ p . 

Consequently,
∞ Y
Y ∞
D1p (−c p − s) = ζF (−kc p − ks + h)n p (k,h−1)−M p (k,h)
b
k=1 h=−∞
∞ Y
Y ∞
= ζF (kc p + ks − h + 1)n p (k,h−1)−M p (k,h)
b
k=1 h=−∞

ζF (1 − s) = b
(by the functional equation b ζF (s))
∞
Y ∞
Y
= ζF (ks + h)n p (k,kc p −h)−M p (k,kc p −h+1)
b
k=1 h=−∞
∞ Y
Y ∞
= ζF (ks + h)n p (k,h−1)−M p (k,h)
b (by Lemma 10.5) = D1p (s).
k=1 h=−∞

ζFG/P (s) as well.

This proves the functional equation for D1 (s) and hence for b 

10.3.2 Constant cG,P

To end this chapter, following [56], for each pair (G, P), we give the exact value
of the constant cG,P .

Lemma 10.6. (see e.g., Appendix 2 of [56]) Let G be a semi-simple Lie group.
Let α p be the simple root of G corresponding to the maximal parabolic group P.
Then

(A) cSLn ,P = n,

2n − p
 p,n
=

(B) cSO2n ,P (n ≥ 2),
2n
 p=n
(C) cSp ,P = 2n − p + 1 (n ≥ 3),
n


2n − p − 1
 p , n − 1, n
=

(D) cSO2n+1 ,P (n ≥ 4),
2n − 2
 p = n − 1, n
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 277

Functional Equation 277

p=1

  29
p=1

17
 

p=1 p=2
 
 

 12 
 
 19
p=2
  

 
 11 

p=2 p=3
  
9 14


 

 


p=3
  

 
 8 

p=3 p=4
  
7 11

 
 

= = p=4 =
  
(E) cE6 ,P cE7 ,P cE8 ,P

10

 9 p=4 
 
 9 p=5
p=5
  
13


 

 


p=5 p=6
  

 12 
 
 13
p=6
  
18

 
 

p = 6, p=7


11 
 


 
 23
p = 7,

14
 


p = 8,

17





 11 p = 1

p=2

7


(F) cF4 ,P = 





 5 p=3

p = 4,

8


5 p = long

(G) cG2 ,P = 

3 p = short.


This can be verified by writing down the room system in concrete terms.
We omit the details of the calculations, but offer the following numbering of the
Dynkin diagrams.

10.4 T-Version for SL3

In this section, we use a different coordinate system to calculate the T -version of

zeta functions for G = SL3 . As a by-product, we show that, unless T ∈ Cρ, resp.
T = 0, they do not satisfy the functional equation, resp. the Riemann hypothesis.
To start with, for the T -period ωG;T Q (λ) in Definition 7.8, when G = SL3 , we
set λ = (z1 , z2 , z3 ) satisfying z1 + z2 + z3 = 0, T = (x, y, −x − y), ρ = (1, 0, −1) and
W = S3 . Then, by taking residue of ωQ SL3 ;T
(λ) along z1 − z2 = 1 and making a
normalization to remove the zeta factors in the denominators, we get
SL /P ;T 1 1 1 b
ζQ,o3 1,2 (t) := b
b ζ(2)b
ζ(3t + 3)e3tx+3ty+4x+2y − ζ(3t + 3)e(3t+3)(x+y)
3t 2 3t + 1
1 1 b 1 b b
+ ζ(3t + 1)e−3tx + 0 − ζ(2)ζ(3t + 1)e−3ty+x−y
2 3t + 2 3t + 3
1 1 b
− ζ(3t + 2)e−3tx+x+2y .
3t 3t + 3
Here we write z2 = t and hence z1 = t + 1, z3 = −2t − 1, Similarly, by taking
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 278

278 Algebraic, Analytic Structures and the Riemann Hypothesis

residue of ωSL
Q
3 ;T
(λ) along z2 − z3 = 1 and making a normalization, we get
SL /P ;T 1 b
ζQ,o3 2,1 (t) := −
b ζ(2) · b
ζ(3s + 1) · e−3sx+x+2y + 0
3s + 3
1 1 1 1
− ζ(3s + 3) · e(3s+3)(x+y) +
·b ζ(3s + 1) · e−3sx
·b
2 3s + 1 2 3s + 2
1 1 1
− ζ(3s + 2) · e3sx+3sy+4x+2y + b
·b ζ(2)b
ζ(3s + 3)e−3sy+x−y .
3s 3s + 3 3s
Here we write z3 = s and hence z2 = s + 1, z1 = −2s − 1 Clearly, there is no
functional equation for these two functions. However, if we set y = 0 in T =
(x, y, −x − y) so that T = (x, 0, −x) ∈ C · ρ sits on the line spanned by ρ, then
SL /P ;xρ 1 1 1 b
bζQ,o3 1,2 (t) = b ζ(2)b
ζ(3t + 3)e3tx+4x − ζ(3t + 3)e(3t+3)x
3t 2 3t + 1
1 1 b 1 b b 1 1 b
+ ζ(3t + 1)e−3tx − ζ(2)ζ(3t + 1)e x − ζ(3t + 2)e−3tx+x ,
2 3t + 2 3t + 3 3t 3t + 3
SL /P ;xρ 1 b b 1 1 b
bζQ,o3 2,1 (t) = − ζ(2)ζ(3s + 1)e−3sx+x − ζ(3s + 3)e(3s+3)x
3s + 3 2 3s + 1
1 1 b 1 1 b 1
+ ζ(3s + 1)e−3sx − ζ(3s + 2)e3sx+4x + b ζ(2)bζ(3s + 3)e x .
2 3s + 2 3s 3s + 3 3s
It is rather easy to verify that these two functions satisfy the functional equation
SL3 /P1,2 ;xρ SL3 /P2,1 ;xρ
ωQ (−1 − s) = ωQ (x). (10.31)
To get the standard function equation, we introduce
SL /P 1 b b 1
ζQ;T3 1,2 (s) :=
b ζ(2)ζ(3s)T 3s+1 − b ζ(2)b
ζ(3s − 2) · T
3s − 3 3s
1 1 b 1 1 b 1 1b
− ζ(3s)T 3s + ζ(3s − 2)T −3s+3 − ζ(3t − 1)T −3s+4 ,
2 3s − 2 2 3s − 1 3s − 3 3s
SL )/P 1 1 b b
ζQ;T3 2,1 (t) := − b
b ζ(2)b
ζ(3s − 2) · T −3s+4 + ζ(2)ζ(3s)T
3s 3s − 3
1 1 b 1 1 b 1 1b
− ζ(3s)T 3s + ζ(3s − 2)T −3s+3 − ζ(3s − 1)T 3s+1 .
2 3s − 2 2 3s − 1 3s − 3 3s
Then the functional equation above becomes
SL )/P SL )/P
ζQ;T3 1,2 (1 − s) = b
b ζQ;T3 2,1 (s). (10.32)
On the other hand, if T , 0, it is easy to check with examples that there are zeros
of these functions off the central line, numerically. Since T = 0 corresponds to
semi-stable condition as we will see in Part 6, in some sense, we can say that the
Riemann hypothesis comes from the stability condition.
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 279

PART 5

Algebraic, Analytic Structures and

Riemann Hypothesis

The Riemann hypothesis for the 10 concrete zeta functions of SLn (n = 2, 3, 4, 5),
Sp4 and G2 over Q are proved by Suzuki-Lagarias, Suzuki and Ki. The common
features are the meromorphic decompositions M(s)+M(−c−s) for these zeta func-
tions determined by the refined symmetries, good controls for the zero free regions
of M(s), and an elementary but useful lemma of Ki. Following this strategy, the
author proves a weak Riemann hypothesis for zeta functions of (SLn , Pn−1,1 )/Q.
More generally, Ki-Komori-Suzuki show that a weak Riemann hypothesis would
hold for the zeta functions of Chevalley groups over Q, if their discriminants are
not vanished. In this part, we first give a new proof for the case (SLn , Pn−1,1 )/Q us-
ing explicit formulas for the related zeta functions obtained in Part 4, then present
the works of Ki-Komori-Suzuki on zeta functions of Chevalley groups over Q.
Their work is very technical, because of the complications of the algebraic and
analytic structures involved. We end this part with our confirmation of the weak
Riemann hypothesis for the zeta functions of the type E exceptional groups over
Q, by verifying that the associated discriminants are positive using Mathematica.

279
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 281

Chapter 11

SL n/P n−1,1
ζ
Conditional Weak RH for b (s)
Q

SL /P
ζQ n n−1,1(s), we first introduce its discriminant, then show that, if this discrim-
For b
inant is not zero, a weak Riemann hypothesis would hold.

11.1 Refined Symmetric Structure

11.1.1 Zeta Functions for (SL n, P n−1,1 )/Q

By Theorem 9.1, the zeta function of (SLn , Pn−1,1 ) over Q is given by

n
SL /P
X
ζQ n n−1,1(s) =
b ζ(s + a) · Ba−1 (s + a − 1)
Bn−a (−s − a) · b (11.1)
a=1
where B0 (s) ≡ 1, and, for k ≥ 1,
k
X X vk1 · · ·b vk p 1
Bk (s) = (−1) p−1 .
b
(11.2)
p=1 k1 ,...,k p >0
(k1 + k2 ) · · · (k p−1 + k p ) s − k p
k1 +...+k p =k
Here, for m = 1, 2, . . ., we have set
Y m
bvm := ζ(`)
b with ζ(1) = Res s=1b
b ζ(s) = 1. (11.3)
`=1
Easily,
k
X X vk1 · · ·b vk p−1 vk p 1
Bk (s) = (−1) p−1
b b
k p =1 k1 ,...,k p−1 >0
(k1 + k2 ) · · · (k p−2 + k p−1 ) k p−1 + k p s − k p
k1 +...+k p−1 =k−k p
k X
k−1
X X vk1 · · ·b vk p−1
= (−1) p−1
b
i=1 p=2 k1 ,...,k p−1 >0
(k1 + k2 ) · · · (k p−2 + k p−1 )
k1 +...+k p−1 =k−i
!
1 vi
.
b
×
k − (k1 + . . . + k p−2 ) s − i

281
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 282

282 Algebraic, Analytic Structures and Riemann Hypothesis

Therefore, for n ≥ 1,
n n−2
X vi X X vn1 · · ·b vnk
Bn (s) = (−1)k
b b
i=1
s − i k=1 n1 ,...,nk >0
(n1 + n2 ) · · · (nk−1 + nk )
n1 +...+nk =n−i
!
1
×
n − (n1 + . . . + nk−1 )
 
n−1
 n−2 
X vn−i X X vn1 · · ·b vnk 1 
=  (−1)k  .
b b
i=0
s − n + i  k=1 n1 ,...,nk >0
(n1 + n2 ) · · · (nk−1 + nk ) nk + n − i 
n1 +...+nk =i

Here in the last step, we have made a change i ↔ n − i. This yields

n−1 n
X vn−i X vk
Bn (s) = =
b b
an,i bn,k (11.4)
i=0
s − n + i k=1 s−k
n−2
X X vn1 · · ·b vnk 1
(−1)k =:
b
where an,i := bn,n−k .
k=1 n1 ,...,nk >0
(n1 + n2 ) · · · (nk−1 + nk ) nk + n − i
n1 +...+nk =i
Therefore,
n−1
SL /P
X −bn−1,k b
vk
ζQ n n−1,1(s) =
b ζ(s + 1)
b
k=1
s+1+k
n−1 X a−1
n−a X
X −bn−a,k b
vk ba−1,i b
vi b
+ ζ(s + a) (11.5)
a=2 k=1 i=1
s+a+k s+a−1−i
n−1
X −bn−1,i b
vi b
+ ζ(s + n).
i=1
s+n−1−i

In each of the summation, the (inverse of) rational factors is given by

(1) the product of s + 2, . . . , s + n when a = 1,

(2) the product of s + a + (n − a) = s + n, . . . , s + a + 1 and s + a − 2, . . . , s when
2 ≤ a ≤ n − 1,
(3) the product of s + n − 2, . . . , s when a = n.

ζ(s + a),
In parallel, for the remaining zeta factors b

(a) the product s(s + 1)b

ζ(s) is an entire function when a = 1,
(b) the product (s + a)(s + a − 1)bζ(s) is an entire function, when 2 ≤ a ≤ n − 1,
(c) the product (s + n)(s + n − 1)bζ(s + n) is an entire function, when a = n.
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 283

SLn /Pn−1,1
ζQ
Conditional Weak RH for b (s) 283

SLn /Pn−1,1
ξQ
Definition 11.1. We define the xi-function b (s) of (SLn , Pn−1,1 ) over Q by
n !
SL /P SL /P
Y
ξQ n n−1,1(s) :=
b (s + h) bζQ n n−1,1(s). (11.6)
h=0

What we have just said verifies the following:

SL /P
ξQ n n−1,1(s) is entire.
Lemma 11.1. The analytic function b

11.1.2 Refined Symmetric Structure

We begin with the following:

Definition 11.2.
n+1
≥ SL /P
ζQ 2 (s), called the upper half of b
(1) We define a function b ζQ n n−1,1(s), by
n−a X
a−1
≥ n+1 X0 X −bn−a,k b
vkba−1,i b
vi b
ζQ 2 (s) :=
b ζ(s + a) (11.7)
a≥ n+1
2
k=1 i=1
s+a+k s+a−1−i

where
P
X0 
 a≥ n+1
 n even
.
 2
:=  (11.8)
a≥ n+1
 
2  Pa≥ n+1 − 1 Pa= n+1

 n odd
2 2 2

n+1
≤ SL /P
ζQ 2 (s), called the lower half of b
(2) We define a function b ζQ n n−1,1(s), by
n−a X
a−1
≤ n+1 X0 X −bn−a,k b
vkba−1,i b
vi b
ζQ 2 (s) :=
b ζ(s + a) (11.9)
a≤ n+1
2
k=1 i=1
s+a+k s+a−1−i

where
P
X0 
 a≤ n+1
 n even
.
 2
:=  (11.10)
a≤ n+1
 
2  Pa≤ n+1 − 1 Pa= n+1

 n odd
2 2 2

n+1 n+1
≥ ≤
ξQ 2 (s) and b
(3) We introduce the functions b ξQ 2 (s) by
n ! n !
n+1
≥ n+1 n+1
≥ n+1
≥
Y ≤
Y
ξQ 2 (s) :=
b (s + h) bζQ 2 (s) ξQ 2 (s) :=
and b (s + h) bζQ 2 (s).
h=0 h=0
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 284

284 Algebraic, Analytic Structures and Riemann Hypothesis

Proposition 11.1 (Refined Symmetric Structure).

(1) (Decomposition) As meromorphic functions and entire functions,

n+1 n+1 n+1 n+1
SL /P ≥ ≤ SLn /Pn−1,1 ≥ ≤
ζQ n n−1,1(s) = b
b ζQ 2
ζQ
(s)+b 2
ξQ
(s) and b (s) = b
ξQ 2 (s)+b
ξQ 2 (s). (11.11)
(2) (FunctionalEquation)
n+1 n+1 n+1 n+1
≥ ≤ ≤ ≥
ζQ
b 2
(−n − s) = b
ζQ 2
(s) ζQ
and b 2
(−n − s) = b
ζQ 2
(s). (11.12)
In particular,
SL /P SL /P
ζQ n n−1,1(−n − s) = b
b ζQ n n−1,1(s). (11.13)

Proof. (1) is a direct consequence of Theorem 9.1, or better,(11.1), and Defini-

tion 11.2. To prove (2), it suffices to use the transformations that
Bn−a − (−n − s) − a Ba−1 (−n − s) + a − 1

= Bn−a (n + s − a) Ba−1 (−s − n + a − 1)

n−a↔a−1 (11.14)
←→ Ba−1 (s + a − 1)Bn−a (−s − a),
n−a↔a−1
ζ((−n − s) + a) =b
b ζ(s + n − a + 1) ←→ b ζ(s + a − 1 + 1) = b
ζ(s + a). 

11.2 Zero-Free Region of Upper Half Function

11.2.1 Dominant Term

We begin with the following:

Definition 11.3. For all 1 ≤ a ≤ n, we define the partial zeta functions bζQ(a) (s) and
the rational functions ra (s) by
 n−a
 X −bn−a,k b vk
a=1



s+a+k





 k=1
ζQ(a) (s)

 n−a a−1
 X X −bn−a,k b vk ba−1,i bvi
b 

1 < a < n . (11.15)

:= ra (s) := 
ζ(s + a)
b 


 k=1 i=1
s + a + k s + a − 1 −i

 n−1
−bn−1,i b
vi

 X
a=n



s+n−1−i




i=1

Accordingly, we define the partial xi-function b ξQ(a) (s) and the polynomial pa (s) by
 n 
Y
ξQ(a) (s) :=  (s + h) · b
 (a)
b ζQ (s) =: pa (s) ξ(s + a) 1≤a≤n (11.16)
h=0

where, as usual, ξ(s) := s(s − 1)b

ζ(s).
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 285

SLn /Pn−1,1
ζQ
Conditional Weak RH for b (s) 285

In terms of this definition,

n n
SL /P SL /P
X X
ζQ n n−1,1(s) =
b ζQ(a) (s)
b and ξQ n n−1,1(s) =
b ξQ(a) (s),
b
a=1 a=1 (11.17)
≥ n+1
X0 ≥ n+1
X0
ζQ 2 (s)
b = ζ (a) (s)
b and ξQ 2 (s)
b = ξ (a) (s).
b
a≥ n+1 Q
2 a≥ n+1 Q
2

Definition 11.4. We define the invariants Bk (1 ≤ k ≤ n), also called the discrim-
SL /P
ζQ n n−1,1(s), by
inants of b

k
X X vk1 · · ·b vk p
Bk = (−1) p−1 .
b
(11.18)
p=1 k1 ,...,b p >0
(k1 + k2 ) · · · (k p−1 + k p )
k1 +...+k p =k

In addition, set B0 = 1.

Proposition 11.2. If Bn−1 is not zero, the polynomial pn (s) dominates all the
polynomials pa (s) with a ≥ n+1
2 . Moreover, if Bn−a Ba−1 , 0,

n−1 a=n





deg pa (s)) =  n−2 n−1≥a≥2.

(11.19)



n − 1 a = 1



Proof. Denote by d(a) the right hand side of (11.19), and let La be the coefficient
of sd(a) in pa (s). By definition,
 n 
Y 1
pn (s) =  (s + h) · b
 (n)
ζQ (s)
h=0
ξ(s + n)
 n 
Y    1
=  (s + h) b ζ(s + n)Bn−1 (s + n − 1)
h=0
ξ(s + n) (11.20)
 
 n−2 
  n−1 

Y  X X vk1 · · ·bvk p 1
=  (s+h)  (−1)  .
 p−1
b 
 
+k +k +

h=0

 p=1
 k1 ,...,k p >0
(k1 2 ) · · · (k p−1 p ) s n−1−k p 

k1 +...+k p =n−1

This implies that d(n) ≤ n − 1 and

n−1
X X vk1 · · ·b vk p
Ln = (−1) p−1 = Bn−1 .
b
(11.21)
p=1 k1 ,...,k p >0
(k1 +k2 ) · · · (k p−1 +k p )
k1 +...+k p =n−1
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 286

286 Algebraic, Analytic Structures and Riemann Hypothesis

Hence, d(n) = n − 1 if and only if Bn−1 , 0. Furthermore, for n − 1 ≥ a ≥ n−1

2 ,

 n 
Y 1
pa (s) =  (s + h) · b
 (a)
ζQ (s)
h=0
ξ(s + a)
 n 
Y    1
=  (s + h) Bn−a (−s − a)b
ζ(s + a)Ba−1 (s + a − 1)
h=0
ξ(s + a)
 
 n  
n−a

Y X X v k · · ·b
v k 1 
(11.22)
=  (s+h)
  (−1) p−1 b 1 p 
+k +k

h=0

 p=1 k1 ,...,k p >0
(k1 2 ) · · · (k p−1 p ) −s−a+k p


k1 +...+k p =n−a
 
 
a−1
v ` · · ·b
v` 1  ζ(s + a) .
X X  b
1 q
×  (−1)q−1
b
 q=1 `1 ,...,`q >0
(`1 + `2 ) · · · (`q−1 + `q ) s + a − 1 − `q  ξ(s + a)
`1 +...+`q =a−1

ζ(s + a) 1
=
b
Since , we have d(a) ≤ n − 2 and
ξ(s + a) (s + a)(s + a − 1)
 
 
n−a
 X X vk1 · · ·b vk p 
La = − (−1) p−1
b 
 p=1 k1 ,...,k p >0
(k1 + k2 ) · · · (k p−1 + k p ) 
k1 +...+k p =n−a
 

a−1
 (11.23)
X X v`1 · · ·b v`q 
×  (−1)q−1
b 
 q=1 `1 ,...,`q >0
(`1 + `2 ) · · · (`q−1 + `q ) 
`1 +...+`q =a−1

= − Bn−a Ba−1 .

Hence d(a) = n − 2 if Bn−a Ba−1 , 0. This proves the proposition for a ≥ n+1
2 , and
hence all the cases by the functional equation. 

11.2.2 Zero Free Region on Right Half Plane

By definition, for any a satisfying n > a ≥ n+1

2 ,

ξQ(a) (s)
b pa (s) ξ(s + a) ra (s) ζ(s + a) ζQ(a) (s)
b
= = = (n) .
b b
(11.24)
ξQ(n) (s)
b pn (s) ξ(s + n)
b r n (s) ζ(s + n)
b ζQ (s)
b
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 287

SLn /Pn−1,1
ζQ
Conditional Weak RH for b (s) 287

Moreover,  
 
n−a
ra (s) X X vk1 · · ·b vk p 1 
=  (−1) p−1
b 
rn (s)  p=1 k1 ,...,b p >0
(k1 + k2 ) · · · (k p−1 + k p ) s + a + k p 
k1 +...+k p =n−a
 
 a−1 
X X v`1 · · ·b v`r 1 
×  (−1)r−1
b 

r=1 `1 ,...,`r >0
(`1 + `2 ) · · · (`r−1 + `r ) s + a − 1 − `r 
`1 +...+`r =a−1
 −1
 n−1 
X
k−1
X bvn1 · · ·b vnk 1 
×  (−1) 
(n1 + n2 ) · · · (nk−1 + nk ) s − nk 

n1 ,...,nk >0

k=1
n1 +...+nk =n−1


n−a (11.25)
1 X X vk1 · · ·b vk p
=  (−1) p−1
b
s  p=1 k1 ,...,b p >0
(k1 + k2 ) · · · (k p−1 + k p )
k1 +...+k p =n−a

 a−1
X X v`1 · · ·b v`r
×  (−1)r−1
b

r=1 `1 ,...,`r >0
(`1 + `2 ) · · · (`r−1 + `r )
`1 +...+`r =a−1

 n−1
X X vn1 · · ·b vnk
×  (−1)k−1
b

k=1 n1 ,...,nk >0
(n1 + n2 ) · · · (nk−1 + nk )
n1 +...+nk =n−1

(1 + (a + k p )/s)(1 + (a − 1 − `r )/s)
!−1 
 .

×
1 − nk /s
ra (s)
Hence, is bounded when |s| → ∞, if Bn−1 , 0. On the other hand, for the
rn (s)
zeta factors, by Lemma 8.4,
ξ(s + a) ξ(s + a)
! !
a+n−1 a+n−1
< 1 <(s) > − = 1 <(s) = − .
b b
&
ξ(s + n)
b 2 ξ(s + n)
b 2
(11.26)
Thus, for all a ≥ n+12 ,
ξ(s + a) n − 1 + n+1 3n − 1
<1 <(s) > − =− .
b 2
(11.27)
bξ(s + n) 2 4
3n − 1 5 n
Nevertheless, if n = 2, − = − < −1 = − . Hence, (11.27) implies that
4 4 2
ξ(s + a) n
<1 <(s) > − .
b
(11.28)
ξ(s + n)
b 2
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 288

288 Algebraic, Analytic Structures and Riemann Hypothesis

This then essentially proves the following:

Proposition 11.3. If Bn−1 is not zero, there are at most finitely many zeros
≥ n+1 n
ξQ 2 (s) lying in the right half- plane <(s) > − for all n ≥ 2, and
of b
2
3n − 1
<(s) > − when n ≥ 3.
4
Proof. Indeed, by definition,
ζ (a) (s) 
 X0 n−1 b 
n+1
≥
ξQ ξQ(n) (s) 1 +
(s) = b
2

b  (11.29)
a= n+1
2 bζ (n) (s)
where
X0 n−1 b ζ (a) (s) X0 n−1 ra (s) (s + n) (s + n − 1) b ξ(s + a)
= . (11.30)
n+1
ζ (s)
a= 2 b(n) a= n+1
2 r n (s) (s + a) (s + a − 1) ξ(s + n)
b
Therefore, by (11.28), for <(s) > − n2 , we have
X0 n−1 b ζ (a) (s) X0 n−1 ra (s) (s + n) (s + n − 1) b ξ(s + a)
≤
n+1
ζ (s)
a= 2 b(n) a= n+1
2 r n (s) (s + a) (s + a − 1) ξ(s + n) (11.31)
b
X0 n−1 ra (s) (s + n) (s + n − 1)
≤ .
a= n+1
2 rn (s)(s + a)(s + a − 1)
Set now
X0 n−1 ra (s) (s + n) (s + n − 1)
φn (s) = (11.32)
a= n+1
2 rn (s)(s + a)(s + a − 1)
by (11.25) and the conclusion followed, we conclude that, if Bn−1 , 0, for
|s| → ∞,
X0 n−1 ra (s)(s+n)(s+n−1) 1 X0 n−1 Bn−a Ba−1 1
|φn (s)| ≤ = +O 2 . (11.33)
a= n+1
2 rn (s)(s+a)(s+a−1) s a= n+1
2 Bn s
Let then Dn ⊂ C be the set consisting of points s such that
X0 n−1 ra (s) (s + n) (s + n − 1)
1> ≥ φn (s) . (11.34)
a= n+1
2 rn (s) (s + a) (s + a − 1)
By (11.33), obviously, Dn is a finite subset. Therefore, for the second factor in
(11.29), we have, by (11.28) and (11.34), if Bn−1 , 0,
X0 n−1 b ζ (a) (s) n
1+ n+1
>0 ∀s < Dn and <(s) ≥ − . (11.35)
a= 2 b(n)
ζ (s) 4
As discussed above, when n ≥ 3, we may take s from the region s < Dn and
3n − 1
<(s) > − . Finally, let us treat the first factor of (11.29). Note that
4
ξQ(n) (s) = pn (s) b
b ξ(s + n) (11.36)
where pn (s) is a polynomial degree n−1 provided Bn−1 , 0 and ξ(s) = s(s+1)b ζ(s).
Therefore, to complete the proof, it suffices to note that ξ(s + n) , 0 in the region
b
<(s) > −n + 1 and that there are at most n − 1 zeros for pn (s). 
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 289

SLn /Pn−1,1
ζQ
Conditional Weak RH for b (s) 289

As usual, a slight better result can be obtained. Indeed, with the same structure
as in the proof above, by a known result stated at page 135 of [116] obtained
by Vinogradov’s method applying to the factors of the Riemann zeta function
involved, we conclude the following:
n+1
≥
ξQ 2 (s)
Corollary 11.1. If Bn−1 is not zero, there are at most finitely many zeros of b
n
lying in the region <(s) ≥ − − δ(t), where s = σ + it and σ(t) is a positive
2
function on the real line satisfying δ(t) log |t| → ∞ as |t| → ∞.

11.2.3 Zero Free Region on Left Half Plane

In this section, we show the following

Proposition 11.4. If B[ n2 ] B[ n−1

2 ]
is not zero, there exists a positive number σn such
≥ n+1
ξQ
that b 2
(s) admits no zeros in the region <(s) ≤ −σn .

Proof. Recall that, by (11.29),

ζ(s +
 
≥ n+1
X0 n−1 a)
ξQ (s) =b ζ(s + n)  pn (s) +  .
 b 
2
b pa (s) (11.37)
a= n+1
2 ζ(s + n)
b

ζ(s + a)
b
We first treat the quotient . Since we are working on <(s) → −∞, we use
ζ(s + n)
b
ζ(s) to get
the functional equation for b
a−n Γ
−s−a+1

ζ(s + a) b ζ(−s − a + 1) ζ(−s − a + 1)
= =π .
b 2
2
−s−n+1
ζ(−s − n + 1)
(11.38)
ζ(s + n) ζ(−s − n + 1)
b b Γ 2
For the Γ factors, by Stirling’s formula, for arg(s) < π − ε,
√  s s 1 1 1  1 !
Γ(s) = 2πs 1+ + + . . . + + O . (11.39)
e 12s 288s2 am sm |s|m+1
So there are some real constants c j such that, for <(s) → −∞,
 
Γ −s−a+1 −s−a+1 −s−a+1
r
2 s+a−1( 2 ) 2 a−n
 = e 2
+ −s−n+1 −s−n+1

Γ −s−n+1
2
s n − 1 ( ) 22
c1 (a, n) cm (a, n)  1 !
× 1+ + ... + + O m+1 (11.40)
s sm |s|
! a−n  1 !
2 2 c1 (a, n) cm (a, n)
= − 1+ + ... + + O
s s sm |s|m+1
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 290

290 Algebraic, Analytic Structures and Riemann Hypothesis

−s − a + 1 π
!
since arg < < π − ε. Here, in above equalities, the c j ’s are
2 2
changed from line to line, with an abuse of notations. Next, we treat the quo-
ζ(−s − a − 1)
tient . Recall that for <(s) < 0,
ζ(−s − n + 1)
∞
X µ(k)
1
= , (11.41)
ζ(−s) k=1 k−s
where µ(k) denotes the Mobius numbers. So there exists real constants dk such that
ζ(−s − a + 1) X dk (a, n)
=1+ <(s) → −∞. (11.42)
ζ(−s − n + 1) k≥2
k−s

Therefore, as <(s) → −∞ and arg(−s) = π − ε,

ζ(s + a)  s  n−a
2 c1 (a, n) cm (a, n)  1 !
= − 1+ + ... + +
b
O
ζ(s + n)
b 2π s sm |s|m+1
  (11.43)
 X dk (a, n) 
× 1 +  .
k−s k≥2

X0 n−1 ζ(s + a)
Consequently, the second factor pn (s) +
b
pa (s) in (11.37)
a= n+1
2 ζ(s + n)
b
becomes
X0 n−1  s  n−a
2
pn (s)+ p (s) −
n+1 a
a= 2 2π
X dk (a, n)  (11.44)
 
c1 (a, n) cm (a, n)  1 ! 
× 1+ + ... + + O m+1 1 +

−s 
 .
s sm |s| k≥2
k 
1
Assume that Bn−a Ba−1 , 0. Then pa (s) is a polynomial in t = s 2 of degree 2n − 2
and 2n − 4 when a = n and a < n, respectively. Write
 s  n−a 2
qa (s) := pa (s) − . (11.45)
2π
Then
 n−( n22 +1)

 
 s
 p 2 +1 (s) − 2π n even

 n
q[ n2 +1] (s) = 

   n− n+1
2
(11.46)

 p n+1 (s) − 2πs 2

 n even,
2

1
and, as polynomials in z = s 2 ,
  n 
degz q[ 2n +1] (s) ≥ degt (qa (s)) + 1 ∀ + 1 < a ≤ n. (11.47)
2
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 291

SLn /Pn−1,1
ζQ
Conditional Weak RH for b (s) 291

This implies that, if B[ 2n ] B[ n−1

2 ]
, 0, (11.44) can be rewritten as
X dk ([ n + 1], n) !
q[ n2 +1] (s) 1 + ϕn (s) 1 + 2


(11.48)
k≥2
k−s
where the function ϕn (s) is defined by
c1 ([ 2n + 1], n) cm ([ n2 + 1], n)  1 !
1 + ϕn (s) := 1 + + ... + + O m+1
s sm |s|
n−1 !
X qa (s) c1 (a, n) cm (a, n)  1 
+ 1+ + ... + + O m+1 (11.49)
q[ n2 +1] (s) s sm |s|
a=n+1
2 +1

X dk (a, n) ! qn (s)
! X dk ([ n + 1], n) !−1
× 1+ + 1+ 2
.
k≥2
k−s q[ n2 +1] (s) k≥2
k−s

Since, as <(s) → −∞, the polynomial q[ n2 +1] (s) is dominating and the series
X dk (a, n)
1+ are always strictly positive and uniformly bounded, there exists a
k≥2
k−s
bounded subset D0n ⊂ C such that
|ϕn (s)| < 1 and hence 1 + ϕa (s) > 0 ∀s < D0n . (11.50)
Therefore, by (11.37) and (11.48), it suffices to prove the non-vanishing statement
ζ(s + n) q[ n2 +1] (s) of a zeta function and a poly-
in the proposition for the product b
nomial. But this is rather obvious, since it is well-known that, for <(s) < −n, the
1
ζ(s + n) has no zero, and that for a fixed polynomial of s 2 , its zeros
zeta function b
are always contained in a bounded domain, depending on the coefficients. 

≥ n+1
11.3 ξ
Refined Hadamard Product for b 2
(s)
Q

≥ n+1
ξ
11.3.1 Distributions of Zeros for b 2
(s)
Q

n
As usual, write s = σ + it, and let T > 1 and σ > . Denote by Nn (T ; σ) the
n+1
2
≥
number of zeros of bξQ 2 (s) in the region
n n o
− σ < <(s) < − and 0 < =(s) < T . (11.51)
2
Lemma 11.2. Assume that Bn−1 and B[ n2 ] B[ n−1 2 ]
are not zero. There exists real
constants c1 > 0, c2 and c3 such that, with the constant σn as in Proposition 11.4,
Nn (T ; +∞) = Nn (T ; σn ) =c1 (n) T log T + c2 (n) T + O(log T ). (11.52)
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 292

292 Algebraic, Analytic Structures and Riemann Hypothesis

Proof. This can be proved rather directly using some well-known techniques
in analytic number theory, based on Proposition 11.3 for the zero free region of
≥ n+1 ≥ n+1
ξ 2 (s) on the right, Proposition 11.4 for the zero free region of b
b
Q ξ 2 (s) on the Q
left, and the relations (11.29), (11.35) and (11.48), (11.50) in the proofs of these
two propositions. Indeed, by the Cauchy residue formula,
Z
1 ≥ n+1
Nn (T ; σn ) = ξQ 2 (s)
d log b
2πi RT
(11.53)
1  ≥ n+1 
= (∆1 + ∆2 + ∆3 + ∆4 ) arg ξQ (s)
b 2

2πi
where RT denotes a rectangle with vertices −σL + c i, σR + c i, σR + iT, −σL + iT ,
associated to some sufficient positive real numbers σL , σR and a certain small
enough positive constant c satisfying the conditions that
n+1
≥
ξQ
(i) there is no zeros of b 2
(s) on the boundary of RT , and

(ii) |φn (−σL + i t)| < 1 and |ϕn (σR + i t)| < 1 for |t| ≥ c

and ∆1 , ∆2 , ∆3 , ∆4 denotes the variations from σR + ci to σR + iT , from σR + iT

to −σL + iT , from −σL + iT to −σL + ci, from −σL + ci to σR + ci, respectively. By
 ≥ n+1

definition, for the bottom (oriented side), obviously, ∆4 arg b ξQ 2 (s) = O(1).
 ≥ n+1

To treat ∆1 arg b ξQ 2 (s) on the right, we deduce the calculations to either the
gamma functions, or π factors, or the shifted Riemann zeta functions, or a product
of a polynomial and a bounded function away from zero. Thus, with standard
calculations, thees contributions are of type T log T , T , O(log T ) and O(1), re-
 ≥ n+1

spectively. The same can be said for ∆3 arg b ξ 2 (s) on the left. Indeed, with
Q
an use of the functional equation for bζ(s), we may transform the calculation to a
new function in −s of similar structure, namely, a product of the gamma functions,
the π factors, the shifted Riemann zeta functions, and a polynomial times wth a
bounded function away from zero. Since <(s) is sufficiently negative, <(−s) be-
 ≥ n+1

comes sufficiently positive. Hence the same arguments for ∆1 arg b ξ 2 (s) on Q
the right above can be applied to this new function of −s so as to obtain similar
contributions of type T log T , T , O(log T ) and O(1), (for corresponding factors
of gamma functions, polynomials, shifted Riemann zeta functions, and bounded
 ≥ n+1

function away from zero,) respectively. To finally treat ∆2 arg b ξ 2 (s) on theQ
top, we use the following

Lemma 11.3 (Lemma in §9.4 of [116]). Let σ0 > 0, T  0 and take 0 ≤ α <
β < σ0 , which may depend on T . Assume that f (s) is an analytic function, taking
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 293

SLn /Pn−1,1
ζQ
Conditional Weak RH for b (s) 293

real value on real line, and regular for σ ≥ α, except possibly finitely many poles
on the real line, and takes no zero on the line t = T . Then, for σ ≥ β,
!−1
σ0 − α 3
arg f (σ + iT ) ≤ π log m Mα,T +2 log + π,



(11.54)
σ0 − β 2
where m, Mσ,t are positive constants satisfying
m ≤ < f (σ + it) f (σ0 + it0 ) ≤ Mσ,t (∀σ0 ≥ σ, 1 ≤ t0 ≤ t). (11.55)


and

Indeed, with an application of this lemma to the entire function

≥ n+1  ≥ n+1

f (s) = b
ξ 2 (s), we have ∆2 arg b
Q ξ 2 (s) = O(1). All this then proves the first
Q
equality of (11.52), and hence completes the proof since c1 (n) above does not
depending on σn . 

n+1
≥
ξ
11.3.2 Refined Hadamard Product for b 2
(s)
Q

≥ n+1
ξQ
There is a natural structure of Hadamard product for the function b 2
(s). To
explain this, set, for simplicity, that
 n 
≥ n+1 ≥ n+1
ξQ 2 (s) := b
e ξQ 2 − + i s . (11.56)
2
Proposition 11.5. Assume that Bn−1 and B[ n2 ] B[ n−1
2 ]
are not zero. The function
n+1
≥
ξQ
e 2
(s) admits a natural Hadamard product
n+1
(s) = κn eαn s Pn (s) Hn (s).
≥
ξQ
e 2
(11.57)
Here

(1) κn , 0 and αn are real constants,

(2) Pn (s) is a polynomial, admits no zeros in =(s) > 0 except for purely imaginary
zeros,
(3) Hn (s) is uniformly convergent on every compact subset of C in the form:
∞  !
Y s  s 
Hn (s) = 1− 1+ , (11.58)
m=1
ρm ρm

where, for all m, <(ρm ) > 0 and 0 < δ(t) < =(ρm ) < cn with δ and cn are the
function and constant in Corollary 11.1 and Proposition 11.4, respectively.
n+1
≥
Proof. Since ξ(ks + h) are entire functions of order one, e
ξQ 2
(s) is an entire
≥ n+1
function of order at most one. Moreover, ξQ 2 (s)
since b takes real values over real
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 294

294 Algebraic, Analytic Structures and Riemann Hypothesis

n+1
≥
axis, by the functional equation, if ρ is a zero of e
ξQ 2 (s), so is −ρ. Therefore, by
standard theory for entire functions of finite orders, using Propositions 11.3 and
≥ n+1
11.4, we conclude that eξ 2 (s) admits a Hadamard product in the form
Q
n+1
(s) = κn eβn s Pn (s) hn (s),
≥
ξQ
e 2
(11.59)
where κn , 0 and βn are complex numbers, Pn (z), is a non-zero polynomial having
no zero in =(z) > 0 except for purely imaginary zero, and
∞  !
Y z  z   z z 
hn (z) = 1− 1+ exp − , (11.60)
m=1
ρm ρm ρm ρm
where <(ρm ) > 0 for all m. and the infinite products converge uniformly on every
compact subset in C. Moreover, if set ρm = am + ibm with am ∈ R and bm ∈ R>0 ,
∞ ∞ ∞
X 1 1 X bm X 1
− ≤2 ≤ 2 cn . (11.61)
m=1 m
ρ ρm |ρ
m=1 m
|2 |ρ
m=1 m
|2
Here in the last step we have used Proposition 11.4. Since the sums on the
right hand sides of (11.60) are all finite, we can and hence will move the fac-
 z z 
tors exp − out of the infinite product. This then implies the existence of
ρm ρm
structural product in the statement of the proposition, if we set
∞ !
X 1 1
αn = βn + − . (11.62)
m=1
ρm ρm
Hence, to complete the proof, it suffices to prove the following

Lemma 11.4. With the same notation as above, αn is real.

≥ n+1
ξQ
Proof. By definition, e 2
(s) = κn eαn s Pn (s) Hn (z). Hence,
n+1
≥
ξQ
e 2
(−it) Pn (−it) X
∞
it + ρm
log = 2t=(αn ) + log + log . (11.63)
≥
ξQ
e
n+1
2
(it) Pn (it) m=1
it − ρm

Pn (−it)
Obviously, log = o(1) as t → + ∞. Moreover, by Propositions 11.3 and
Pn (it)
11.4, as t → +∞, we have 0 < bm < |cn | + 1, and hence
∞ ∞
∞
a2m + (t + bm )2 X

X 4bm t X t log m 
log 2 ≤ = O   = O(log t). (11.64)
m=1
a +(t−b )2
m m a2 +(t − b )2
m=1 m m2 +t2 
m m=1
≥ n+1  ≥ n+1
So, log ξQ 2 (−it) e
e ξQ 2 (it) = 2t=(αn ) + O(log t) as t → +∞. Consequently, to
prove =(αn ) = 0 and hence αn ∈ R, it is sufficient to prove the following
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 295

SLn /Pn−1,1
ζQ
Conditional Weak RH for b (s) 295

Lemma 11.5. As y → +∞, there exists a constant Cn , 0 and dn ≥ 0 such that

n+1
≥
ξQ
e 2
(−it)
= Cn tdn 1 + o(1) .


(11.65)
≥ n+1
ξQ 2 (it)
e

Proof. By the proofs of Propositions 11.3 and 11.4,

≥ n+1 n
n

 n

bξQ 2 − + t = b ζ n+ − +t 1+φ − +t
2 2 2
≥ n+1 n
b n

n

bξQ 2
− − t = ζ n + − − t q[ 2 +1] − − t
n
(11.66)
2 2 2
X dk ([ n + 1], n) ! n

× 1+ 2

1+ϕ − −t .
k≥2
n
k− − 2 −t 2

Here in the last step, we have used (11.48). Therefore

≥ n+1
 
ξQ 2 − n2 + t ζ n2 + t 1 + o(1) X dk ([ n + 1], n) !−1
b
b
= 1 + 2
k 2 +t
n
≥ n+1 ζ 2 − t q[ 2 +1] − 2 − t
n
n

ξQ n
2

n
b − 2 −t b k≥2

ζ 2n + t


b   (11.67)
= 1 + o(1)
ζ 1 − 2 + t q[ n2 +1] − 2 − t
n n
b


 t  n−1
2 1  
= n
1 + o(1) .
2π q[ 2 +1] − 2 − t
n

Here, in the last step, we have used (11.43). This proves Lemma 11.5, and hence
Lemma 11.4 and Proposition 11.5. 

SL n /P n−1,1
11.4 ζ
Conditional Weak Riemann Hypothesis for b (s)
Q

11.4.1 Elementary Lemma

We in this subsection prove an elementary lemma, as a consequence of the fol-

lowing well-known:

Theorem 11.1 (Hurwitz’s theorem). Let { fn } be a sequence of holomorphic

functions on a connected open subset D ⊂ C such that it converges to a holo-
morphic function f . 0 on D, uniformly on any compact subset of D. Assume
that f has a zero of order m at z0 ∈ D. Then there exists R > 0 such that, for
any 0 < r < R, fn has exactly m zeros in the disk |z − z0 | < r, counting with
multiplicities, for every n ≥ N for an N which may depend on r.
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 296

296 Algebraic, Analytic Structures and Riemann Hypothesis

We omit a proof of this result, as it can be found in literatures easily. Instead,

we state a simple corollary.

Corollary 11.2. Let {Fn } be a sequence of entire functions which converges to an

entire function F uniformly on any compact subset of C. Assume that there are at
most 2N roots outside the real axis for each Fn . Then, there are at most 2N roots
of F outside the real axis.

Proof. Otherwise, assume that, outside the real line, F = limn→∞ Fn has M(> 2N)
roots z1 , z2 , . . . , z` with multiplicities m1 , m2 , . . . , m` . Obviously, m1 + m2 + · · · +
m` = M. By Hurwitzs theorem, for each i = 1, 2, . . . , `, there exists R j > 0 such
that for any 0 < r j < R j , Fn has exactly m j roots in the disk |z − z j | < r j (0 <
r j < R j ) counting with multiplicities for all n ≥ N j , where N j may depend on r j .
Choose r > 0 such that 0 < r < R j and {z : |z − z j | ≤ r} are all away from the
real line, i.e. the closed disk |z − z j | ≤ r are entirely contained in the upper/lower
half plane for all 1 ≤ j ≤ `. Hence, for a sufficiently large n, Fn has exactly m j
zeros in the disk |z − z j | < r counting with multiplicities for all j. This implies
that, away from the real line, Fn admits at least M = m1 + m2 + · · · + m` zeros.
This is a contradiction. 

We will use this result to prove the following

Lemma 11.6 (Proposition 3.1 in [56]). Let W(z) be a function in C which admits
a Hadamard product of the form
∞ " ! !#
Y z z
W(z) = H(z)eαz 1− 1+ , (11.68)
k=1
ρk ρk

where H(z) is a nonzero polynomial, α ∈ R, =(ρk ) ≥ 0 for all k, and the product
in (11.68) is uniformly convergent in any compact subset of C. Assume that H(z)
has exactly N zeros, counted with multiplicities, in the lower half-plane =(z) < 0.
Then, away from the real line, there are at most 2N, counted with multiplicities
for the function W(z) + W(z), resp. W(z) − W(z)). In particular, there are at
most N zeros of W(z) + W(z) in the lower-half plane and in the upper-half plane,
respectively.

This lemma is first used by Ki in his study of our 10 concrete examples on

Weng zeta functions in [55]. The detailed proof presented here is provided by
Suzuki [111].

Proof. We start with the following result of de Bruijn.

December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 297

SLn /Pn−1,1
ζQ
Conditional Weak RH for b (s) 297

Claim 11.1. Let U(z) and V(z) be polynomials with real coefficients. Assume
that U(z) . 0, resp. V(z) . 0 and that there are exactly N zeros, counted with
multiplicities, of W(z) = U(z) + iV(z) in the lower half-plane =(z) < 0. Then,
outside the real line, there are at most 2N zeros, counted with multiplicities, for
U(z) and V(z), respectively.

Proof. We prove this following Lemma 2 at page 215 of [18]. We may assume
that U(z) and V(z) have no common real roots. This implies that W(z) admits
no real roots. We also assume that the degree m of W(z) satisfies m ≥ 2N, as
otherwise the result is an easy consequence of the relations 2U(z) = W(z) + W(z)
and 2iV(z) = W(z) − W(z).
By the argument principle and the assumption on the zeros of W(z), if z runs
over the real axis from −∞ to ∞, the argument of W(z) increases by (m − 2N)π.
Hence there are at least m − 2N different points on the real axis where W(z) be-
comes purely imaginary, resp. real. Hence, if z = ∞ is not one of these points,
there are m − 2N real roots for U(z), resp. V(z) at least, and otherwise there are
at least m − 2N − 1 real roots. But in the last case, the degree of U(z), resp. V(z)
must be < m. Therefore, in both cases, there are at most 2N non-real zeros for
U(z), resp. V(z). 

Since U(z) and V(z) are polynomials with real coefficients, W(z) = U(z) −
iV(z). Hence,
2U(z) = W(z) + W(z) and 2iV(z) = W(z) − W(z). (11.69)
Conversely, if W(z) is an arbitrary polynomial, by (11.69), the coefficients of U(z)
and V(z) must be real. Hence, we may rewrite Claim 11.1 as follows:

Claim 11.2. Let W(z) be a polynomial satisfying W(z) , −W(z), resp. W(z) ,
−W(z). Assume that there are exactly N zeros, counted with multiplicities, of
W(z) has in the lower half-plane =(z) < 0. Then, outside the real line, there are at
most 2N zeros, counted with multiplicities, of W(z) + W(z), resp. W(z) − W(z).

Back to the proof of the lemma. Following page 2379 of [55], we introduce
the wn (z)’s by
n "
αz n Y
! !#
 z z
wn (z) = H(z) 1 + 1− 1+ . (11.70)
n k=1 ρk ρk
Obviously, limn→∞ wn (z) = W(z) and wn (z) has exactly n zeros in the lower half-
plane =(z) < 0. By Claim 11.2, wn (z) + wn (z) has at most 2n zeros outside of the
real axes. Since wn (z) + wn (z) converges to W(z) + W(z) uniformly in any compact
subset of C, by Corollary 11.2, W(z) + W(z) has at most 2n zeros outside the real
axes. The same arguments work for W(z) − W(z). 
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 298

298 Algebraic, Analytic Structures and Riemann Hypothesis

SL n /P n−1,1
ζ
11.4.2 Conditional Weak Riemann Hypothesis for b (s)
Q

With all these, we are now ready to prove the following

Theorem 11.2. Assume that Bn−1 and B[ 2n ] B[ n−1 2 ]

are not zero. All but finitely
SL /P
ζQ
many zeros of SLn -zeta function b n n−1,1
(s) are simple and located on the central
n
line <(s) = − .
2
n+1
≥
ξQ
Proof. By Proposition 11.5, the partial zeta function b 2
(s) admits a Hadamard
SL /P
product. Moreover, by Proposition 11.1 for bζ n n−1,1(s),
Q
n+1
SL /P ≥ ≥ n+1
ζQ n n−1,1(s) = b
b ζQ 2
ζQ
(s) ± b 2
(−n − s). (11.71)
SL /P n
So, by Lemma 11.6, all but finitely many zeros ζQ n n−1,1(s)
of b lie on <(s) = − .
2
To prove these zeros are simple, we use the extra symmetry again. Indeed,
from (11.71), we have
SL /P n ≥ n+1 n
 
ξQ n n−1,1 − + it = b ξQ 2 − + it eiθ(t) ± eiθ(t) .


(11.72)
 n+1 2  2
≥
Here θ(t) := arg b ξQ 2 − n2 + it . To prove the simplicity of these zeros, it suf-

fices to show that θ(t) is strictly increasing as t → +∞. By Propositions 11.2 and
n
11.3, on the central line <(s) = − , we have, as |s| → ∞
2
≥ n+1
ξQ 2 (s) = pn (s)b ζ(s + n) 1 + o(1) .


b (11.73)
By taking logarithmic derivative, as t → +∞, Gamma factors give log t and
ζ0
!
log t
(1 + it) = O = o(log t). (11.74)
ζ log log t
Therefore, there exists a suitable constant c such that
d
θ − c p /2 + it = c log t 1 + o(1) .



(11.75)
dt
So θ(t) is strictly increasing as t → +∞ as desired. 

Hence, it suffices to show Bk , 0. This is indeed the case.

Theorem 11.3 (Theorem 15.2 and Proposition 15.1). Let n ≥ 1. For the moduli
space MQ,n [1] of semi-stable lattices of rank n and volume one,
n
  X X vn1 · · ·b vnk
0 < vol MssQ,n [1] = Bn+1 = (−1)k−1 .
b
k=1 n ,...,n >0
(n1 + n 2 ) · · · (nk−1 + nk )
1 k
n1 +...+nk =n

We will prove this and hence complete our proof of the weak Riemann hy-
SL ,P
ζQ n n−1,1 (s) (n ≥ 2) in Chapter 14.
pothesis for the zeta function b
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 299

Chapter 12

Algebraic and Analytic Structures and

Weak Riemann Hypothesis

The main purpose of this chapter is to prove a conditional weak Riemann hy-
pothesis for the zeta functions associated to reductive groups and their maximal
parabolic subgroups, essentially following [56]. We first expose some fine sym-
metric structures for the zeta functions associated to reductive groups and their
maximal parabolic subgroups, then, based on a detailed analysis of the Lie struc-
tures and some standard analytic number theoretic estimations involved, for the
Chevalley groups over Q, we show that, if the discriminants are not vanishing, all
but finitely many zeros of these zeta functions lie on the central line.

12.1 Criterion for Weak Riemann Hypothesis

12.1.1 Discriminant of (G, P)

Let G be a reductive group over a number field F. Fix a minimal parabolic sub-
group of G and let P be a standard maximal parabolic subgroup of G. Denote by
(V, h·, ·i); Φ, Φ± , ∆; W; ρ the associated root system. That is to say, (V, h·, ·i) is


the metrized root space, and Φ, resp. Φ± , resp. ∆, is the set of roots, resp. the
subset of ±-roots, Xresp. the set of simple roots, W = hσα : α ∈ ∆i is the Weyl
group, and ρ = α is the Weyl vector. For each w ∈ W, set Φw := Φ+ ∩ w−1 Φ−
α>0
be the set of positive roots such that their w-images become negative, and denote
by w0 ∈ W the longest element in W. It is well known that the length `(w) of
w coincides with Φw , w20 = id, and w0 (∆) = −∆ and hence w0 (Φ± ) = Φ∓ . For
later use, write ∆ =: α1 , . . . , αn and let λ1 , . . . , λn be the set of fundamental
 
X n
dominant weights. Then, hλ j , α∨i i = δi j for all 1 ≤ i, j ≤ n, and ρ = λ j . Here
j=1
as usual, for α ∈ Φ, denote by α∨ the associated co-root.

299
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 300

300 Algebraic, Analytic Structures and Riemann Hypothesis

Let α p ∈ ∆ be the unique element such that α p ⊆ ∆ is the subset natu-


rally corresponding to P, under the well-known order-reversing correspondence
between the set of standard parabolic subgroups of P and the subsets of ∆. Write
∆ p := ∆ r {α p } = β1 , . . . , βn−1 for convenience. Denote by Φ p ⊂ Φ the subset

consisting of roots which are perpendicular to λ p . Then, for Φ+p := Φ+ ∩ Φ p , we
1 X
have Φ p = Φ+p t Φ−p and Φ+p ⊂ n−1 k=1 Z≥0 βk . Set ρ p = α. Similarly, let W p
P
2 α∈Φ+
p
be the group generated by the reflections σα associated to elements α in ∆ p , or the
same in Φ p . Obviously, every element of W p admits a decomposition as a product
of the σβk ’s. Denote by w p the unique longest element of W p with respect to this
decomposition. Then w2p = id, w p (∆ p ) = −∆ p and w p (Φ±p ) = Φ∓p . For later use, set
c p := hλ p − ρ p , α∨p i. (12.1)
Recall that, w ∈ W is called special with respect to P if w(∆ p ) ⊂ ∆ ∪ Φ− .
Denote by WP0 , or simply W 0 , the subset of special Weyl elements in W.

Definition 12.1. Let G be a reductive group and let P be a maximal parabolic

subgroup of G, both defined over F.

(1) For w ∈ W 0 and (k, h) ∈ Z2 , the counting functions n p,w (k, h), n p (k, h) and
M p (k, h) are defined by
n p,w (k, h) := # α ∈ w−1 (Φ− ) : hλ p , α∨ i = k, hρ, α∨ i ,


n p (k, h) := # α ∈ Φ : hλ p , α∨ i = k, hρ, α∨ i ,

(12.2)
M p (k, h) := max n p,w (k, h − 1) − n p,w (k, h) .

w∈W 0
(2) The discriminant of (G, P) over F is defined by
X r
∆G/P
F := cFp,w c p,w d p,w . (12.3)
w∈W p
Here
W p := w ∈ W 0 : w(Φ+ r Φ+p ) ⊂ Φ+ ,


cF := Res s=1b ζF (s),

r p,w := w∆ p r Φ+ ,
+
−1
Y
c p,w := ξ(2)|∆ p ∩w (Φ )| ξ hρ, α∨ i + δα,w ,


α∈Φ+ r∆ p
(12.4)
−1 +
Y 1
d p,w := 2−|∆ p ∩w (Φ )|
hρ, α∨ i − 1
α∈w−1 (∆)∩(Φ p r∆ p )
Y 1
×
.
hρ, α∨ i + δα,w hρ, α∨ i + δα,w − 1


α∈Φ+p r∆ p
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 301

Algebraic and Analytic Structures and Weak RH 301

12.1.2 Criterion of Weak Riemann Hypothesis

Introduce a coordinate system λ = nj=1 (1 + s j )λ j = ρ + nj=1 s j λ j . Write s = s p .

P P

ζQG/P (s) for (G, P) over F is given by

By Theorem 10.1(1), the zeta function b
∞ ∞ 
Y Y
ζ (s) =  ζ(ks + h)  ω (s),
G/P
 G/P
M p (k,h) 
b
F
 b
F(12.5)
k=0 h=2

where ωG/P
F (s) is the period of (G, P) over F defined by
  
 X Y 1 Y b ζ F (hλ, α ∨
i) 
Res sn =0 . . . Res
[ s p =0 . . . Res s1 =0 
   .
 wλ − ρ, α i σ∈Φ b
∨
ζ (hλ, α ∨ i + 1) 
w∈W 0 α∈∆ w F

ζFG/P (s) satisfies the functional equation

By Theorem 10.1(2), b
ζFG/P (−c p − s) = b
b ζFG/P (s). (12.6)

In this chapter, we will prove the following main result of [56].

Theorem 12.1 (Criterion for weak Riemann Hypothesis). Let G be a Cheval-

ley group G over Q, i.e. a connected split semi-simple algebraic group over Q. If
cp
RG/P ζQG/P (s) lie on the central line <(s) = − .
, 0, all but finitely many zeros of b
Q 2

12.2 Refined Symmetries of Zeta Functions

12.2.1 Positively Oriented

Following §10.2.1, introduce an ‘overdone’ normalization for ωG/P

F (s) by

Z p (s) := F p (s) ωG/PF (s), (12.7)

ζF hλ p , α∨ i s + hρ, α∨ i . By the proof of Propositions 10.1,
Q

where F p (s) := α∈Φ− b
particularly, (10.17)
X r
Z p (s) = cFp,w f p,w (s)g p,w (s), (12.8)
w∈W 0

where
Y 1
f p,w (s) = ,
hλ p , α∨ is + hρ, α∨ i − 1
α∈(w−1 (∆))r∆ p
Y (12.9)
g p,w (s) = ζF hλ p , α∨ is + hρ, α∨ i .
b

α∈(w−1 (Φ− ))r∆ p
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 302

302 Algebraic, Analytic Structures and Riemann Hypothesis

Since the coefficients hλ p , α∨ i of the variable s appeared in g p,w (s) may well
be negative, by the functional equation b ζF (1 − s) = bζF (s),
Y
g p,w (s) = ζF 1−hλ p , α is−hρ, α i
∨ ∨
b

α∈((w−1 (Φ− ))∩Φ−
Y
ζF hλ p , α∨ is + hρ, α∨ i


× b
α∈((w−1 (Φ− ))∩Φ+ )r∆ p
Y
= ζF hλ p , α∨ is+hρ, α∨ i+1
b

(12.10)
α∈((w−1 (Φ+ ))∩Φ+
Y
ζF hλ p , α∨ is+hρ, α∨ i


× b
α∈((w−1 (Φ− ))∩Φ+ )r∆ p
Y
Y b
= ζF hλ p , α∨ is + hρ, α∨ i + 1 ζF hλ p , α∨ is + hρ, α∨ i .
b

α∈Φ+ rΦw α∈Φw r∆ p

1 w(α) > 0

Introduce, for w ∈ W , a characteristic function δα,w by δα,w := 
0

.
0 w(α) < 0

Accordingly, we can combine two products in (12.10) to write g p,w (s) as
+
−1
Y
g p,w (s) = b
ζF (2)|∆ p ∩w (Φ )| ζF hλ p , α∨ is + hρ, α∨ i + δα,w ,


b (12.11)
α∈Φ+ r∆ p

for which all the coefficients of s on the right hand are non-negative. Here the ad-
−1 +
ζF (2)|∆ p ∩w (Φ )| comes from the terms in the first product of (12.10)
ditional factor b
satisfying α ∈ ∆ p and w(α) ∈ Φ+ (so that hλ p , α∨ i = 0 and hρ, α∨ i = 1).

12.2.2 Refined Symmetries

To further understand Z p (s), we classify elements of W 0 using δα,w so as to ex-

pose a refined symmetry for Z p (s), which, as to be seen, is compatible with the
functional equation for Z p (s).

Definition 12.2.

(1) For an element w ∈ W 0 , we define a counting function `±p,w by

`±p,w := (Φ+ r Φ+p ) ∩ w−1 (Φ± ) . (12.12)
(2) We define the subsets of W 0 by
W p< := w ∈ W p0 : `+p,w < |Φ+ r Φ+p |/2 ,


W p> := w ∈ W p0 : `+p,w > |Φ+ r Φ+p |/2 ,


(12.13)
W p= := w ∈ W p0 : `+p,w = |Φ+ r Φ+p |/2 ,


W p≤ := W p< t W p= and W p≥ := W p> t W p= .

December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 303

Algebraic and Analytic Structures and Weak RH 303

Directly from the definition, we obtain the first two equalities below.

Lemma 12.1. Let w ∈ W 0 . We have X

`−p,w = (1 − δα,w ),
α∈Φ+ rΦ+p
(12.14)
`+p,w + `−p,w = Φ+ r Φ+p ,
`±p,w + `±p,w0 ww p = Φ+ r Φ+p .
Consequently, w0 W p≤ w p = W p≥ , and there is a natural decomposition:
W p0 = W p< t W p= t W p> . (12.15)

Proof. The first two come directly from the definition. Moreover, the second
equality implies that, for the third one, it suffices to treat `−p,w + `−p,w0 ww p . Easily,
(Φ+ r Φ+p ) ∩ (w0 ww p )−1 Φ− = w p (Φ+ r Φ+p ) ∩ w−1 Φ+ = (Φ+ r Φ+p ) ∩ w−1 Φ+ .
Hence, for all w ∈ W 0 ,
`−p,w + `−p,w0 ww p = (Φ+ r Φ+p ) ∩ w−1 (Φ− ) + (Φ+ r Φ+p ) ∩ (w0 ww p )−1 (Φ− )
= (Φ+ r Φ+p ) ∩ w−1 (Φ− ) + (Φ+ r Φ+p ) ∩ w−1 Φ+
= Φ+ r Φ+p ,
as desired. 

By Proposition 10.2, there exists a fixed ε p ∈ {±1} such that, for all w ∈ W 0 ,
there are the micro functional equations
r p,w0 ww p = r p,w ,
f p,w0 ww p (−c p − s) = ε p f p,w (s), (12.16)
g p,w0 ww p (−c p − s) = g p,w (s).
Hence, by Lemma 12.1, if we set1
 
 X 1 X 
E p (s) :=   +  cr p,w f (s) g (s), (12.17)
2  F p,w p,w
+ = w∈W p
w∈W p
then what we just said proves the following

Corollary 12.1. There exists a fixed ε p ∈ {±1}, depending only on p, such that
Z p (s) = E p (s) + ε p E p (−c p − s).

Note that all the terms in this decomposition of Z p (s) (in terms of E p (s)) are
meromorphic functions in s. This is very different from the decomposition for the
Riemann zeta function used in the study of Riemann-Siegel formula.
r p,w
1 When W p= is empty, the second summation
P
w∈W p= cF f p,w (s) g p,w (s) is defined to be zero.
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 304

304 Algebraic, Analytic Structures and Riemann Hypothesis

12.3 Dominate Term

12.3.1 Entire Function Oriented

To facilitate our ensuing discussion, we next introduce some rational functions to

transform their products with E p (s) and hence with Z p (s) into entire functions.
As usual, set ξF (s) := s(s − 1) b ζF (s). By replacing b ζ in g p,w in (12.11) with ξ,
we obtain an auxiliary entire function
+
−1
Y
g̃ p,w (s) =: ξF (2)|∆ p ∩w (Φ )| ξ hλ p , α∨ is + hρ, α∨ i + δα,w .


(12.18)
α∈Φ+ r∆ p

Hence, if we write the special ξ value factors separately,

g̃ p,w (s) = c p,w z p,w (s), (12.19)

where
(Φ+ )|
−1
Y
c p,w := ξ(2)|∆ p ∩w ξF hρ, α∨ i + δα,w ,


α∈Φ+ r∆ p
Y (12.20)
ξF hλ p , α∨ is + hρ, α∨ i + δα,w .


z p,w (s) :=
α∈Φ+ rΦ+p

g̃ p,w (s)
In addition, by a direct calculation, is a polynomial. given by
g p,w (s)
g̃ p,w (s) −1 +
Y   
= 2|∆ p ∩w (Φ )| hλ p , α∨ is+hρ, α∨ i+δα,w hλ p , α∨ is+hρ, α∨ i+δα,w −1 .
g p,w (s) α∈Φ+ r∆ p

Hence, if we set
Y g̃ p,v (s) 1 !
Q p (s) := and X p (s) := Q p (s) Z p (s), (12.21)
0
g p,v (s) f p,v (s)
v∈W

Q p (s) and Q p (s) f p,w (s) are polynomials and X p (s) is entire, since, by (12.9),
Y 1
f p,w (s) = . (12.22)
hλ p , α∨ is + hρ, α∨ i − 1
α∈w−1 (∆)r∆P

Consequently, by (12.8) and (12.19), we obtain the following decomposition of

X p (s) in terms of entire functions
X r
X p (s) = cFp,w c p,w a p,w (s) z p,w (s), (12.23)
w∈W 0
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 305

Algebraic and Analytic Structures and Weak RH 305

!
Y g̃ p,v (s) 1
where a p,w (s) := is the polynomial given by
g p,v (s) f p,v (s)
v∈W 0 r{w}

(Φ+ )|
Y −1
Y  
a p,w (s) = 2|∆ p ∩v hλ p , α∨ is + hρ, α∨ i − 1
v∈W p0 r{w} α∈(v−1 (∆)r∆ p
Y   !
× hλ p , α is + hρ, α i + δα,v hλ p , α is + hρ, α i + δα,v − 1 .
∨ ∨ ∨ ∨

α∈Φ+ r∆ p

In addition, by applying the micro functional equation (12.16), we have

Q p (−c p − s) =ε p Q p (s),
z p,w0 ww p (−c p − s) = z p,w (s), (12.24)
c p,w0 ww p a p,w0 ww p (−c p − s) =ε p c p,w a p,w (s).
Therefore, for
 
 X 1 X 
 cr p,w c a (s) z (s),
B p (s) :=  +  F p,w p,w p,w (12.25)
w∈W >
2 w∈W =
p p

by (12.17) and Lemma 12.1, what we have just said proves the following:

Proposition 12.1. With the same notation as above, we have

X p (−c p − s) = ε p X p (s) and X p (s) = B p (s) + ε p B p (−c p − s). (12.26)

In the sequel, we will use this refined symmetry systematically. As a prepa-

ration, we next tackle the problem that Q p (s) is not minimal. That is to say, even
Q p (s) Z p (s) is entire,
n Q p (s) is not a polynomial
o of the smallest degree satisfying
0
this property, since a p,w (s) : w ∈ W admits non-trivial common divisors. Sim-
ilarly, as noticed earlier, the F p (s) in (12.7) is ’overdone’ since it introduces extra
zeta factors. To overcome this, we make the following

Definition 12.3. We define the polynomial R p (s), the ξ-function factor ∆ p (s) and
the ξ-function ξG/P
F (s) associated to (G, P) over F by
n o
R p (s) := g.c.d a p,w (s) : w ∈ W 0 ,
∞ Y
Y ∞
∆ p (s) := ξF (ks + h)N p (k,h−1)−M p (k,h) ,
(12.27)
k=1 h=2
X p (s)
ξG/P
F (s) := ,
R p (s) ∆ p (s)
respectively. Here ‘g.c.d’ means the monic polynomial of maximal degree which
divides a p,w (s)’s in the polynomial ring C[s].
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 306

306 Algebraic, Analytic Structures and Riemann Hypothesis

Proposition 12.2. We have

(1) The function ξG/P

F (s) is an entire function, obtained as a product of the zeta
ζFG/P (s) with a suitable polynomial.
function b
(2) There is a natural decomposition
ξG/P
F (s) = A p (s) ± A p (−c p − s), (12.28)
where A p (s) is the entire function defined by
B p (s)
A p (s) := . (12.29)
R p (s) ∆ p (s)

Proof. (2) is a direct consequence of the previous proposition, and (1) comes
directly from the definition and the discussion in §10, in particular, §10.2.1 for
Theorem 10.1(1) on the normalization of our zeta function bζFG/P (s). 

Therefore, to prove Theorem 12.1, it suffices to prove the same statement for
the entire function ξG/P
F (s) for a Chevalley group G over Q.

12.3.2 Terms with Maximal Discrepancy

To go further, we study the terms appeared in B p (s), or better in A p (s), corre-

sponding to the special w ∈ W 0 with maximal `+p,w .

Definition 12.4. We define the subset W p of W 0 by

W p := w ∈ W 0 : `+p,w = Φ+ r Φ+p = w ∈ W 0 : Φ+ r Φ+p ⊆ w−1 (Φ+ ) . (12.30)
n o n o

By definition, if w ∈ W p , then δα,w = 1 for all α ∈ Φ+ r Φ+p . For this reason,

we call elements of W p as the special Weyl elements with maximal discrepancy.

Lemma 12.2.

(1) W p = w ∈ W 0 : `−p,w = 0 and contains the unit element id.


(2) For all w ∈ W p , z p,w (s) = z p,id (s).

Proof. (1) By definition (12.12), ` p,id = (Φ+ r Φ+p ) ∩ (Φ− ) = 0. Hence, id ∈ W p .

More generally, since Φ = Φ+ t Φ− and, for w ∈ W 0 , we have w(Φ) = Φ, we have
`−p,w = 0 ⇐⇒ (Φ+ r Φ+p ) ∩ w−1 (Φ− ) = ∅ ⇐⇒ Φ+ r Φ+p ⊆ w−1 (Φ+ ).
(2) By definition (12.20), we have, from (1), that
Y
z p,w (s) = ξF hλ p , α∨ is + hρ, α∨ i + 1 = z p,id (s),


α∈Φ+ rΦ+p
as desired. 
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 307

Algebraic and Analytic Structures and Weak RH 307

Motivated by this, we introduce the following

Definition 12.5. The functions z p (s) and a p (s) are defined by

X r
z p (s) := z p,id (s) and a p (s) := cFp,w c p,w a p,w (s). (12.31)
w∈W p

Lemma 12.3. With the same notation as above, we have

X r
a p (s) z p (s) = cFp,w c p,w a p,w (s) z p,w (s),
w∈W p
∞ Y
Y ∞ (12.32)
z p (s) = ξ(ks + h) N p (k,h−1)
.
k=1 h=2

Proof. By Lemma 12.2, we have

X r X r
cFp,w c p,w a p,w (s)z p,w (s) = z p (s) cFp,w c p,w a p,w (s).
w∈W p w∈W p

This establishes the first equality. To prove the second, note that, for α ∈ Φ+ ,
hρ, αi ≥ 1. Hence, with n p (k, h) = α ∈ Φ : hλ p , α∨ i = k, hρ, α∨ i = h ,


Y ∞ ∞

YY
z p (s) = ξ hλ p , α∨ is + hρ, α∨ i + 1 = ξ(ks + h)N p (k,h−1) ,
α∈Φ+ rΦ+p k=1 h=2

as desired. 

Definition 12.6. For each w ∈ W 0 , we define an invariant d p,w by

−1 +
Y 1
d p,w :=2−|∆ p ∩w (Φ )|
hρ, α∨i − 1
α∈w (∆)∩(Φ p r∆ p )
−1
(12.33)
Y 1
×
.
hρ, α∨ i + δα,w hρ, α∨ i + δα,w − 1


α∈Φ+p r∆ p

Note that, for each α ∈ w−1 (∆) ∩ (Φ p r ∆ p ), resp. α ∈ Φ+p r ∆ p , we have

hρ, α∨ i − 1 , 0, resp. hρ, α∨ i + δα,w hρ, α∨ i + δα,w − 1 , 0. Hence, this definition



makes sense.

Lemma 12.4. With the same notation as above, for a fixed w ∈ W 0 , as |s| → ∞,
Y 1
d p,v
hλ p , α∨ is + hρ, α∨ i − 1
a p,v (s) α∈v−1 (∆)rΦ p
= X . (12.34)
a p (s) r p,w
Y 1  !
cF c p,w d p,w 1+o(1)
hλ p , α∨ is+hρ, α∨ i−1
w∈W p α∈w (∆)rΦ p
−1
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 308

308 Algebraic, Analytic Structures and Riemann Hypothesis

Proof. We first write down the terms of a p,w (s) explicitly. From the definition, for
w ∈ W p0 ,
Y " −1 +
Y
a p,w (s) = 2|∆ p ∩v (Φ )| hλ p , α∨ is + hρ, α∨ i − 1


v∈W 0 r{w} α∈(v−1 (∆))r∆ p
Y  #


hλ p , α is + hρ, α i + δα,v
∨ ∨
hλ p , α is + hρ, α i + δα,v − 1
∨ ∨


×
α∈Φ+ r∆ p
Y " 1 Y
= hλ p , α∨ is + hρ, α∨ i − 1


d p,v
0
v∈W r{w} α∈(v−1 (∆))rΦ p
Y  #


hλ p , α is + hρ, α i + δα,v
∨ ∨
hλ p , α is + hρ, α i + δα,v − 1
∨ ∨


×
α∈Φ+ rΦ+p
" Y 1
= a∗p (s) d p,w
hλ p , α∨ is + hρ, α∨ i − 1
α∈w−1 (∆)rΦ p
#
Y 1
×
.
hλ p , α∨ is + hρ, α∨ i + δα,w hλ p , α∨ is + hρ, α∨ i + δα,w − 1


α∈Φ+ rΦ+p

Here a p (s)∗ is the polynomial defined by

Y" 1 Y
a∗p (s) := hλ p , α∨ is + hρ, α∨ i − 1


d p,v
0v∈W −1 α∈(v (∆))rΦ p
Y  #


hλ p , α is + hρ, α i + δα,v
∨ ∨
hλ p , α is + hρ, α i + δα,v − 1 .
∨ ∨


×
α∈Φ+ rΦ+p

Consequently,
X r
a p (s) = cFp,w c p,w a p,w (s)
w∈W p
X " r
Y 1
= a∗p (s) cFp,w c p,w d p,w
hλ p , α∨ is + hρ, α∨ i − 1
w∈W p α∈w−1 (∆)rΦ p
#
Y 1
×
.
hλ p , α∨ is + hρ, α∨ i + δα,w hλ p , α∨ is + hρ, α∨ i + δα,w − 1


α∈Φ+ rΦ+p

Since hλ p , α∨ i , 0 for all a ∈ Φ+ r Φ+p , we have, for v ∈ W p≥ and w ∈ M p , when

|s| → ∞,


hλ p , α∨ is + hρ, α∨ i + δα,v hλ p , α∨ is + hρ, α∨ i + δα,v − 1


Y

 = 1 + o(s).
hλ p , α∨ is + hρ, α∨ i + δα,w hλ p , α∨ is + hρ, α∨ i + δα,w − 1


α∈Φ+ rΦ+p
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 309

Algebraic and Analytic Structures and Weak RH 309

Therefore, as |s| → ∞,
Y 1
d p,v
hλ p , α∨ is + hρ, α∨ i − 1
a p,v (s) α∈v−1 (∆)rΦ p
= X " ,
a p (s) r
Y 1  #
cFp,w c p,w d p,w 1 + o(1)
hλ p , α∨ is + hρ, α∨ i − 1
w∈W p α∈w−1 (∆)rΦ p
as desired. 

12.3.3 Relation between W p and W p

We next show that, as |s| → ∞, a p (s) dominants a p,w (s) for w ∈ W ≥ r W p , using
(12.34). For this, we first examine the Lie structures of W p ⊂ W 0 .

Lemma 12.5. Let w ∈ W.

(1) If w ∈ W p , then w Φ+ r Φ+p ) = Φ+ r Φ+p .

(2) w Φ+ r Φ+p ) ⊂ Φ+ if and only if w ∈ W p .

Proof. (1) By definition, for 1 ≤ i ≤ n and i , p, σαi (α p ) = α p − 2hα p , αi )α∨i .

Hence, for w ∈ W p , w(α p ) − α p ∈ Φ p , since w is a finite product of σαi ’s (i , p).
Therefore, for α = ni=1 ai αi ∈ Φ+ r Φ+p , a p > 0 and w(α p ) = a p α p + ni=1,i,p a0i αi .
P P
This implies that w(α) ∈ Φ+ r Φ p , since w(α) ∈ Φ and a p > 0.
(2) By (1), it suffices to prove that w Φ+ r Φ+p ) ⊂ Φ+ implies w ∈ W p . Denote
by W = ti∈I wi W p be the coset decomposition of W with respect to W p . Assume
that w ∈ wi0 W p for a certain i0 ∈ I. From (1), W p (Φ+ r Φ+p ) ⊂ (Φ+ r Φ+p ). Hence
wi0 (Φ+ r Φ+p ) ⊂ Φ+ . Set σi := σαi be the reflection associated to simple roots
αi , and denote by wi0 = ki=1 σi j the shortest decomposition of wi0 in terms of the
Q
products of σi ’s. If wi0 < W p , there is an ik0 such that ik0 = p. We may and hence
will take ik0 to be the biggest one among i j ’s satisfying this condition. If k0 = k,
by Corollary of Lemma C in §10.2 of [44], we have σ1i · · · σik−1 σ p (α p ) < 0.


This contradicts with w(Φ+ r Φ p ) ⊂ Φ+ since α p ∈ Φ+ r Φ p . If ik , p, we write
wi = w0 w00 with w00 := σik0 +1 · · · σik ∈ W p and w0 := σi1 · · · σik0 . Accordingly,
w00 (Φ+ rΦ+p ) ⊂ Φ+ rΦ+p . This then implies that w0 (Φ+ rΦ p ) ⊂ Φ+ , a contradiction,
since w0 (α p ) = σi1 · · · σik0 −1 σ p (α p ) < 0. Therefore, for all 1 ≤ j ≤ k, i j , p.


This means wi0 and hence also w are elements of W p . 

Corollary 12.2. We have n o

W p = W 0 ∩ W p = w ∈ W p : w−1 (∆) r Φ p = 1 .


(12.35)
Fk(p)
More generally, if W = i=1 wi W p is the coset decomposition of W with respect
to W p , the level function ` p,w is constant on wi W p .
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 310

310 Algebraic, Analytic Structures and Riemann Hypothesis

Proof. By definition, W p = w ∈ W 0 : w(Φ+ r Φ+p ) ⊂ Φ+ . Therefore, from the


previous lemma, W p = w ∈ W 0 : w ∈ W p . This proves the first inequality. To

prove the second, note that all elements of W p belong to w ∈ W p . Thus, for
w ∈ W p , we have w(Φ p ) = Φ p and ∆ r w(Φ p ) = ∆ r Φ p = 1. Therefore,
w ∈ W p : |∆ r w(Φ p )| = 1 = w ∈ W p . Hence, by Lemma 12.5(1)
 

` p,w = w(Φ+ r Φ+p ) ∩ Φ+ = wi (Φ+ r Φ+p ) ∩ Φ+ = ` p,wi .

This means that ` p,w is constant on wi W p . 

The discussions so far can be used to study the denominator of (12.34) since
w−1 (∆) r Φ p = w−1 (α p ) for w ∈ W p . To treat the numerator, let w ∈ W 0 be an
element satisfying the condition w−1 (∆) r Φ p = 1. Accordingly, denote by βw
the unique element of ∆ ∩ w(Φ p ), and let ∆±w := ∆ ∩ w(Φ±p ). Then
∆ = ∆+w t ∆−w t { βw }. (12.36)

Lemma 12.6. Let w ∈ W 0 be an element satisfying w−1 (∆) r Φ p = 1. We have

∆+w = ∆ ∩ w(∆ p ).

Proof. By definition, we have ∆+w ⊇ ∆ ∩ w(∆ p ). To prove the opposite inclusion,

we note that w(∆ p ) ⊂ Φ− t ∆. Hence, we can write an arbitrary element α ∈ ∆+w
in a form
X X X
α= n j w(α j ) = n j w(α j ) + n j w(α j ) n j ≥ 0, w(α j ) ∈ Φ− t ∆.
j: j,p j: j,p j: j,p
w(α j )∈Φ− w(α j )∈∆

But the root α is simple, so only the second sum remains. That is, α ∈ ∆ ∩ w(∆ p ).
This then implies that ∆+w ⊆ ∆ ∩ w(∆ p ). 

To go further, we give a natural decomposition for Φ+ for later used.

Definition 12.7. We define, for integers k and h, the subsets Σ p (k) and Σ p (k, h) of
Φ by
Σ p (k) := α ∈ Φ : hλ p , α∨ i = k ,

(12.37)
Σ p (k, h) := α ∈ Σ p (k) : hρ, α∨ i = h .


Obviously, for all but finitely many k and (k, h), Σ p (k) and hence Σ p (k, h)’s are
empty. Moreover, we have
G∞ G∞
Φ+ r Φ+p = Σ p (k) and Σ p (k) = Σ p (k, h). (12.38)
k=1 h=1
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 311

Algebraic and Analytic Structures and Weak RH 311

We will use these decompositions in the proof for case B below.

Proposition 12.3. Let w ∈ W 0 be an element satisfying w−1 (∆) r Φ p = 1. Then,

either Φ+ r Φ+p ⊂ w−1 Φ+ , or Φ+ r Φ+p ⊂ w−1 Φ− .

Proof. We arrange this proof according to w(α p ) ∈ Φ± .

Case (A) Assume w(α p ) ∈ Φ+ . By definition, βw belongs to either Φ+ ∩w Φ+rΦ+p

or Φ+ ∩ w Φ− r Φ−p . In other words, either βw ∈ Φ+ ∩ w Φ+ r Φ+p or −βw ∈



Φ− ∩ w Φ+ rΦ+p .

Case (A.1) Assume βw ∈ Φ+ ∩ w Φ+ rΦ+p . We claim that, if Φ− ∩ w(Φ+ rΦ+p ) is

not an empty set, it contains −αk for a certain simple root αk . Indeed, if − ai αi ∈
P
Φ− ∩ w(Φ+ rΦ+p ), we have ai ≥ 0 for all i, and − ai w−1 (αi ) = w−1 − ai αi =
P P

b p α p + j: j,p b j α j (b p > 0, b j ≥ 0). Consequently, since, for all 1 ≤ i ≤ n,
P
w−1 (αi ) ∈ Φ, there exists at least one i0 such that w−1 (−αi0 ) = b0p α p + j: j,p b0j α j ∈
P
Φ+ rΦ+p (b0p > 0, b0j ≥ 0). Hence,

−αi0 ∈ Φ− ∩ w(Φ+ rΦ+p ). (12.39)

Case (A.1.i) Assume αi0 ∈ ∆+w . Then, αi0 ∈ ∆ ∩ w(∆ p ) ⊂ Φ+ ∩ w(Φ+ ). This
contradicts with (12.39).

Case (A.1.ii) Assume αi0 ∈ ∆−w . Then, αi0 ∈ Φ+ ∩ w(Φ−p ). This again contradicts
with (12.39).

Case (A.1.iii) Assume αi0 = βw . Then, αi0 = βw ∈ Φ+ ∩ w(Φ+ rΦ+p ), by definition.

This contradicts with (12.39) as well.

Since αi0 ∈ ∆ = ∆+w t∆−w t{ βw }, the above discussion implies Φ− ∩w(Φ+rΦ+p ) =

∅, namely, Φ+ ⊃ w(Φ+ rΦ+p ), or better, Φ+ rΦ+p ⊂ w−1 (Φ+ ).

Case (A.2) Assume −βw ∈ Φ− ∩ w Φ+ rΦ+p . As to be expected, in this sub-case,

parallel arguments in (A.1) can be translated and applied here. Consequently, as
above, we have Φ+ ∩ w(Φ+ r Φ+p ) = ∅, namely, Φ− ⊃ w(Φ+ r Φ+p ), or better,
Φ+ rΦ+p ⊂ w−1 (Φ− ).

Case (B) Assume w(α p ) ∈ Φ− . We claim that Φ+ r Φ+p ⊂ w−1 Φ− .

Lemma 12.7. Let w ∈ W 0 and α ∈ Φ+ . Assume that α ∈ (Φ+ rΦ+p ) ∩ w−1 (Φ− ) and
α + α j ∈ Φ+ rΦ+p for some α j ∈ ∆ p . Then α + α j ∈ (Φ+ rΦ+p ) ∩ w−1 (Φ− ).
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 312

312 Algebraic, Analytic Structures and Riemann Hypothesis

Proof. We have w(α) ∈ Φ− , and w(α j ) ∈ ∆ t Φ− since w ∈ W 0 . Assume first that

w(α j ) ∈ Φ− . Since α + α j ∈ Φ, we have w(α + α j ) ∈ Φ− , as required. Assume
next that w(α j ) ∈ ∆. Since w(α) ∈ Φ− , w(α + α j ) = w(α) + w(α j ) is a root, written
as a negative root minus a simple one. So, this root must be negative. That is,
w(α + α j ) ∈ Φ− . 

To complete our proof, we make an induction using decomposition for Φ+rΦ+p

above. By Lemma 12.7, since w(α p ) ∈ Φ− , we have Σ p (1) ⊂ (Φ+ rΦ+p ) ∩ w−1 (Φ− ).
Assume inductively that Σ p (k − 1) ⊂ (Φ+ rΦ+p ) ∩ w−1 (Φ− ) for a certain k ≥ 2, we
next show Σ p (k) ⊂ (Φ+ rΦ+p ) ∩ w−1 (Φ− ). For this, let α−p (k) be the lowest root in
Σ p (k). Then there exists a root β ∈ Σ p (k − 1) such that α−p (k)∨ = β∨ + α∨p since
(α−p (k)∨ − α∨j )∨ is not a root for j , p by the lowest property of α−p (k). By the
inductive assumption, we have w(β) ∈ Φ− . Recall that w(α p ) ∈ Φ− , a condition
we start with in case (B), we conclude that w(α−p (k)) = (w(β)∨ + w(α p )∨ )∨ ∈ Φ− .
That is, α−p (k) ∈ (Φ+ rΦ+p ) ∩ w−1 (Φ− ). Therefore, by Lemma 12.7 again, Σ p (k) ⊂
(Φ+rΦ+p ) ∩ w−1 (Φ− ). This, together with the decomposition Φ+rΦ+p = ∞ k=1 Σ p (k),
F
implies that Φ+ r Φ+p ⊂ w−1 Φ− . 

Corollary 12.3. Let w ∈ W p≥ r W p . Then w−1 (∆) r Φ p ≥ 2.

Proof. This is a direct consequence of the previous proposition. Indeed, if

w−1 (∆) r Φ p = 1, then Φ+ r Φ+p ⊂ w−1 Φ+ , or Φ+ r Φ+p ⊂ w−1 Φ− . When


Φ+ r Φ+p ⊂ w−1 Φ+ , by Lemma 12.5, w ∈ W p and hence w ∈ W p , a contradic-
tion. On the other hand, when Φ+ r Φ+p ⊂ w−1 Φ− , ` p,w = 0. Hence, w ∈ W p< , a
contradiction as well. 

12.3.4 Leading Polynomials

Proposition 12.4. Assume the discriminant of (G, P) over F is not zero. We have
   
deg a p (s) ≥ deg a p,v (s) + 1 ∀v ∈ W p≥ r W p . (12.40)
That is to say, a p (s) is a dominant polynomial, comparing with polynomials a p,v (s)
for all v ∈ W p≥ r W p .

Proof. Recall that, by (12.34), we have

Y 1
d p,v
hλ p , α∨ is + hρ, α∨ i − 1
a p,v (s) α∈v−1 (∆)rΦ p
= X " .
a p (s) r p,w
Y 1  #
cF c p,w d p,w 1 + o(1)
hλ p , α∨ is + hρ, α∨ i − 1
w∈W p α∈w (∆)rΦ p
−1
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 313

Algebraic and Analytic Structures and Weak RH 313

By Corollary 12.2, for w ∈ W p , we have w ∈ W p , and hence w−1 (∆) r Φ p =

{w−1 (α p )}. Since w−1 (α p ) is simply α p modulo Φ p , hλ p , w−1 (α p )i = 1. Conse-
quently, as |s| → ∞, for w ∈ W p , we have
Y
hλ p , α∨ is + hρ, α∨ i − 1 = s · (1 + o(1)).


α∈w−1 (∆)rΦ p
X r
Thus, when cFp,w c p,w d p,w , 0,
w∈W p

a p,v (s) d p,v Y 1

= X r ·s .
, α∨ is + hρ, α∨ i − 1

a p (s) cF c p,w d p,w 1 + o(1)
p,w hλ p
α∈v−1 (∆)rΦ p
w∈W p

Y 1
By Corollary 12.3, v−1 (∆) r Φ p ≥ 2. Hence, is
hλ p , α∨ is + hρ, α∨ i − 1
α∈v−1 (∆)rΦ p
at least of order s , since, for all a ∈ v (∆) r Φ p , hλ p , α∨ i , 0. Therefore,
−2 −1

a p,v (s) a p (s) = o(1)


X r
provided that ∆G/P
F = cFp,w c p,w d p,w , 0. 
w∈W p

We end this section by a remark that, unlike the rest of this chapter, the re-
sults in this and the previous subsection are mainly supplied by the author, as the
corresponding arguments in [56] do not work.

12.4 Zero Free Regions

12.4.1 Lie Theoretic Structures Involved

α p,v (s)
With studied for v ∈ W 0 rW p , naturally, we next consider the zeta quotient
a p (s)
z p,v (s)
. Recall that, by Lemma 12.2,
z p (s)
Y
z p (s) = z p,id (s) = ξF (hλ p , α∨ i + hρ, α∨ i + 1). (12.41)
α∈Φ+ rΦ+p

Hence, by (12.20),
z p,v (s) Y ξF (hλ p , α∨ i + hρ, α∨ i)
= . (12.42)
z p (s) ξF (hλ p , α∨ i + hρ, α∨ i + 1)
α∈(Φ+ rΦ+p )∩v−1 (Φ− )
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 314

314 Algebraic, Analytic Structures and Riemann Hypothesis

To estimate these quotients, as above, we need to understand what are the struc-
tures involved for (Φ+ r Φ+p ) ∩ v−1 (Φ− ). This naturally leads to the decomposition
∞
G ∞
G
Φ+ r Φ+p = Σ p (k) and Σ p (k) = Σ p (k, h), (12.43)
k=1 h=1

where Σ p (k) and its subsets Σ p (k, h) are defined in (12.37) by

Σ p (k) := α ∈ Φ : hλ p , α∨ i = k Σ p (k, h) := α ∈ Σ p (k) : hρ, α∨ i = h .
 
and

Obviously, for all but finitely many k, resp. (k, h), the set Σ p (k), resp. Σ p (k, h),
is empty. More precisely, we have the following

Lemma 12.8. (Lemma 1.4.5 of [28]) Let ni=1 ki αi (ki > 0) be the highest root of
P
Φ+ . Then, for k ∈ Z>0 , the set Σ p (k) is not empty if and only if 1 ≤ k ≤ k p .

Definition 12.8. For a positive integer k with Σ p (k) , ∅, we define the lowest
root α−p (k), resp. the highest root α+p (k), of Σ p (k) to be the root of Σ p (k) such
that β − α−p (k), resp. α+p (k) − β, is (possibly empty) sum of simple roots for every
β ∈ Σ p (k).

Lemma 12.9. (Proposition 1.4.2 of [28]) For k ∈ Z>0 , if Σ p (k) , ∅, there exists a
unique lowest root α−p (k), resp. unique highest root α+p (k), in Σ p (k).

Since, for α ∈ Σ p (k), hλ p , α∨ i = k. Accordingly, we may and hence will write

α±p (k)∨ =: kα∨p + β±p (k)∨ , (12.44)
X
where β±p (k) ∈ Z≥0 (∆∨p ) := Z≥0 α∨j . For example, β−p (1) = 0 as α−p (1) = α p .
j: j,p

Lemma 12.10. For 1 ≤ k ≤ k p ,

α−p (1) = α p , w p α−p (k) = α+p (k) β−p (k) − w p β+p (k) = k β+p (1).



and
In particular, w p β+p (1) = −β+p (1) and β+p (k) − w p β−p (k) = k β+p (1).

Proof. The first two are simply given in Lemma 1.4.6 of [28]. For the third, by
definition, w p α−p (k)∨ = α+p (k)∨ = kα∨p + β+p (k)∨ . Hence, by Lemma 12.9 (2),

α−p (k)∨ = w p α+p (k)∨ = kw p α∨p + w p β+p (k)∨

= kw p α−p (1)∨ + w p β+p (k)∨ = kα+p (1)∨ + w p β+p (k)∨

= kα∨p + kβ+p (1)∨ + w p β+p (k)∨ .

But α−p (k)∨ = kα∨p + β−p (k)∨ . Therefore, β−p (k) − w p β+p (k) = kβ+p (1).



December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 315

Algebraic and Analytic Structures and Weak RH 315

Definition 12.9. For 1 ≤ k ≤ k p ,

(1) the invariants h±k are defined by h±k := hρ, β±p (k)∨ i.
(2) the p-level of α ∈ Σ p (k) is defined to be hρ, (α − kα p )∨ i.

Obviously, h±k ≥ 0 and h+k , resp. h−k , is the highest level , resp. the lowest level,
of (the elements in) Σ p (k).

Lemma 12.11. With the same notation as above,

(1) h+k1 ≤ h+k2 for 1 ≤ k2 < k2 ≤ k(p).

(2) h+k + h−k = kh+1 for 1 ≤ k ≤ k(p). In particular, h+k ≤ h+k+1 ≤ (k + 1)h+1 − 1.
In addition, for α ∈ Σ p (k), if c p > 2,2
(2.1) 1 ≤ hρ, α∨ i ≤ c p − 1 when k = 1, and
(2.2) k + 1 ≤ hρ, α∨ i ≤ kc p − k − 1 when k ≥ 2.
(3) 2 + h+1 = 1 + hρ, α+p (1)∨ i = 1 + hρ, w p α∨p i = c p .

Proof. (1) By definition,

α+p (k2 )∨ − α+p (k1 )∨ = (k2 − k1 )α∨p + β+p (k2 )∨ − β+p (k1 )∨ .
 

Moreover, by Corollary of [53], we have α+p (k2 )∨ −α+p (k1 )∨ ∈ Φ∨ . Since k2 −k1 > 0
and β+p (k)∨ ∈ j: j,p Z≥0 α∨j , we conclude that β+p (k2 ) − β+p (k1 ) ∈ j: j,p Z≥0 α j .
P P
Therefore, by definition, h+k + h−k = kh+1 .
(2) It suffices to prove h+k + h−k = kh+1 , since others are direct consequences of
this relation. By the proof of Lemma 10.1, hρ p , α∨i i = 1 for i , p. Hence, for
β∨ ∈ j: j,p Z≥0 α∨j ,
P

hρ, β∨ i = hρ p , β∨ i = hw p ρ p , w p β∨ i = −hρ p , w p β∨ i = −hρ, w p β∨ i.

Therefore, by definition,
k + h−k = hρ, α−p (k)∨ i = hρ, w p α+p (k)∨ i = hρ, w p kα∨p + β+p (k)∨ i

= khρ, w p α∨p i + hρ, w p β+p (k)∨ i = khρ, w p α∨p i − hρ, β+p (k)∨ i

= k(1 + h+1 ) − h+k .

(3) First, by Lemma 12.10, w p (α p ) = α+p (1). Thus, by w p (ρ) = ρ − 2ρ p (see e.g.
the proof of Lemma 10.1) in Part 4, and hρ, α∨p i = h ni=1 λ j , α∨p i = 1, we have,
P

c p = 2hλ p − ρ p , α∨p i = 2 − h2ρ p , α∨p i = 2 − hρ − w p (ρ), α∨p i

= 1 + hw p (ρ), α∨p i = 1 + hρ, w p α∨p i = 1 + hρ, α∨p + β+p (1)∨ i

= 2 + hρ, β+p (1)∨ i = 2 + h+1 ,

as desired. 
2 When c p = 2, by (3) below, α+p (1) = α−p (1) = α p . Hence Φ is of type A1 and in this case k ≥ 2 does
not occur. So the assumption that c p > 2 is designed to exclude the trivial case.
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 316

316 Algebraic, Analytic Structures and Riemann Hypothesis

Now we are ready to decompose Σ p (k). We start with a property of Σ p (k, h).

Lemma 12.12. Let k and h be positive integers such that Σ p (k, h) is not empty.
Then, for Σ p (k, h) =: β1 , . . . , βn p (k,h) , there exists simple roots α j1 , . . . , α jn p (k,h) in

∆ p (not necessary distinct) satisfying the conditions

(1) β∨n + α∨jn ∈ (φ+ r Φ+p )∨ (1 ≤ n ≤ n p (k, h)); and

(2) β∨n + α∨jn , β∨m + α∨jm (1 ≤ m , n ≤ n p (k, h)).

With this, we can effectively construct Σ p (k) starting from α−p (k). For example,
when k = 1, we have the following

Lemma 12.13. There is a decomposition for Σ p (1) given by

M(1)
G
Σ p (1) = Lm (1) (12.45)
m=1

satisfying the conditions

(1) L1 (1) 3 α±p (1) and L1 (1) = h ρ, α+p (1) − α−p (1) i + 1;
(2) L1 (1) > L2 (1) , L3 (1) , . . . , L M(1) (1) ≥ 1;
(3) There exist some βm (1) ∈ Σ p (1) with β1 (1) = α−p (1) and some αm, j (1) ∈ ∆ p
(1 ≤ m ≤ M(1), 1 ≤ j ≤ Lm (1) − 1) satisfying
n o
(3.1) Lm (1)∨ := { βm (1)∨ } t βm (1)∨ + lj=1 αm, j (1)∨ : 1 ≤ l ≤ Lm (1) − 1 ,
P
(3.2) 1 + h−1 = hρ, β1 (1)∨ i, hρ, β M(1) (1)∨ i ≤ c p /2 and
hρ, β1 (1)∨ i < hρ, β2 (1)∨ i ≤ hρ, β3 (1)∨ i ≤ . . . ≤ hρ, β M(1) (1)∨ i,
D E
(3.3) For all 1 ≤ k ≤ M(1), ρ , 2βm (1)∨ + |Lj=1
P m (1)|−1
αm, j (1)∨ = c p .

We will delay a detailed proof of this lemma to the appendix of this chapter. In-
stead, we give a similar result for all the k’s.

Proposition 12.5. For 1 ≤ k ≤ k p , there is a decomposition for Σ p (k) given by

M(k)
G
Σ p (k) = Lm (k) (12.46)
m=1

satisfying the conditions

(1) L1 (k) 3 α±p (k) and L1 (k) = h ρ, α+p (k) − α−p (k) i + 1;
(2) L1 (k) > L2 (k) , L3 (k) , . . . , L M(k) (k) ≥ 1;
(3) There exist some βm (k) ∈ Σ p (k) with β1 (k) = α−p (k) and some αm, j (k) ∈ ∆ p
(1 ≤ m ≤ M(k), 1 ≤ j ≤ Lm (k) − 1) satisfying
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 317

Algebraic and Analytic Structures and Weak RH 317

n o
(3.1) Lm (k)∨ := { βm (k)∨ } t βm (k)∨ + lj=1 αm, j (k)∨ : 1 ≤ l ≤ Lm (k) − 1 ,
P
(3.2) k + h−k = hρ, β1 (k)∨ i, hρ, β M(k) (k)∨ i ≤ kc p /2 and
hρ, β1 (k)∨ i < hρ, β2 (k)∨ i ≤ hρ, β3 (k)∨ i ≤ . . . ≤ hρ, β M(k) (k)∨ i,
D E
(3.3) For all 1 ≤ k ≤ M(k), ρ , 2βm (k)∨ + |Lj=1
P m (k)|−1
αm, j (k)∨ = kc p .

Proof. By §1.4 of [28], we may reduce the problem to decompose Σ p (1) for irre-
ducible root systems. Hence it suffices to apply the previous lemma to complete
the proof. 

Accordingly, we set, for w ∈ W p0 ,

Lm (k)w := Lm (k) ∩ w−1 (Φ− ),
(12.47)
Λw (k) := m : Lm (k)w , ∅, 1 ≤ n ≤ Mk ,


and let
hm (k) := max hρ, α∨ i : α ∈ Lm (k) ,

(12.48)
hm,w (k) := max hρ, α∨ i : α ∈ Lm (k)w .


Here in the last definition for hm,w (k) we assume that m ∈ Λw (k).

Lemma 12.14. The value hm (k) is attained by one of α ∈ Lm (k)w . More strongly,

(1) For α ∈ Φ+ r Φ+p satisfying hρ p , α∨ i < 0, there exists α j ∈ ∆ p such that

α∨ + α∨j ∈ (Φ+ r Φ+p )∨ .
(2) For α ∈ (Φ+ r Φ+p ) ∩ w−1 (Φ− ) satisfying hρ p , α∨ i < 0, there exists α j ∈ ∆ p
such that α∨ + α∨j ∈ (Φ+ r Φ+p ) ∩ w−1 (Φ− ) ∨ .

Proof. It suffices to prove (1) and (2). For (1), from the condition hρ p , α∨ i < 0,
there exists α j ∈ ∆ p such that hα j , α∨ i < 0, since
X X
2ρ p = α= n jα j (n j ∈ Z>0 ).
α∈Φ+p 1≤ j,p≤n

Thus hα j , α i < 0, and hence

∨
hα∨j , α∨ i
< 0. By standard Lie theory, this implies
that α∨ + α∨j is in (Φ+ )∨ . Moreover, since α∨ ∈ (Φ+ r Φ+p )∨ and α∨p ∈ Φ+p , we
conclude that α∨ + α∨j is in fact contained in (Φ+ r Φ+p )∨ . (2) With (1), it follows
directly from Lemma 12.7. 

We end this discussion wth the following easy

Lemma 12.15. Let k ≥ 1 and α ∈ Σ p (k, h). We have

2h = kc p + 2hρ p , α∨ i. (12.49)
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 318

318 Algebraic, Analytic Structures and Riemann Hypothesis

In particular,

> 0 h > kc p /2,

hρ p , α i 
∨

(12.50)
< 0 h < kc p /2.


Proof. By the proof of Lemma 10.1, we have c p λ p = 2ρ − 2ρ p . Hence,

2hρ, α∨ i − hλ p , α∨ ic p = 2hρ, α∨ i − h2ρ − 2ρ p , α∨ ic p = 2hρ p , α∨ i.
This together with the definition of Σ p (k, h) then proves the lemma. 

12.4.2 Estimations on Right Half Plane

As the first application of the Lie theoretic structures we exposed, we now can
treat the quotient z p,w (s) z p (s) of (12.42) for w ∈ W 0 r W p . From now on till the

end of this chapter, we assume that F = Q.

Proposition 12.6. We have, for every w ∈ W 0 r W p ,

z p,w (s) cp z p,w (s) cp
<1 <(s) > − and ≤1 <(s) = − . (12.51)
z p (s) 2 z p (s) 2

Proof. By (12.42), we have

z p,w (s) Y ξF (hλ p , α∨ i + hρ, α∨ i)
= . (12.52)
z p (s) + +
ξF (hλ p , α∨ i + hρ, α∨ i + 1)
α∈(Φ rΦ p )∩w−1 (Φ− )

Since Φ+ r Φ+p = tk=1

k k
p
Σk (p) = tk=1
p M(k)
tm=1 Lm (k), we have, in terms of defini-
tion (12.47), (Φ+ r Φ+p ) ∩ w−1 (Φ− ) = tk=1
k k
p M(k)
tm=1 Lm (k)w = tk=1
p
tm∈Λw (k) Lm (k)w .
Thus, in terms of (12.48),
k(p) Y m (k)
hY k(p)
z p,w (s) Y ξ(ks + hm ) Y Y ξ(ks + hm,w (k))
= = . (12.53)
z p (s) k=1 m∈Λ (k) h =h (k) ξ(ks+hm +1) k=1 m∈Λ (k) ξ(ks+hm (k)+1)
w m mow w

Here in the last step, we have used (3.1) of Lemma 12.5 and Lemma 12.14. Con-
sequently, by (11.26), we have, for every m ∈ Λw (k),
ξ(ks + hm,w (k)) hm,w (k) + hm (k)
<1 <(s) > − .
ξ(ks + hm (k) + 1) 2k
Now, by Lemma 12.5(3.3), hm,w (k) ≥ hρ, βm (k)∨ i = kc p − hm (k). Hence, for m ∈
hm,w (k) + hm (k) c p
Λw (k), ≥ . This proves the first inequality of (12.51), unless
2k 2
hm,w (k) + hm (k) = kc p . Clearly, in the exceptional case when hm,w (k) + hm (k) = kc p ,
by (11.26) again, we have
ξ(ks + hm,w (k)) cp
=1 <(s) = − ,
ξ(ks + hm + 1) 2
as desired. 
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 319

Algebraic and Analytic Structures and Weak RH 319

12.4.3 Normalization Factor

As the second application of the Lie structure exposed in §12.4.1, we here explic-
itly determine M p (k, h) and ∆ p (s) used in the normalization process.

Definition 12.10. We define the constant D0p and the function D1p (s) by
∞
Y ∞ Y
Y ∞
D0p := ζ(h)n p (0,h−1)−M p (0,h) , D1p (s) :=
b ζF (ks+h)n p (k,h−1)−M p (k,h) , (12.54)
b
h=2 k=1 h=2

where, as in Definition 12.1(1),

n p (k, h) := # α ∈ Φ− : hλ p , α∨ i = k, hρ, α∨ i = h ,

n p,w (k, h) := # α ∈ w−1 (Φ− ) : hλ p , α∨ i = k, hρ, α∨ i = h ∀w ∈ W p0 , (12.55)

M p (k, h) := max n p,w (k, h − 1) − n p,w (k, h) .

w∈W p0

Obviously, by (12.54), or better, the proof of Proposition 10.3,

D p (s) := D0p · D1p (s). (12.56)

Lemma 12.16. Let k ≥ 1.

(1) For w ∈ W 0 , n p,w (k, h) ≤ n p,w (k, h + 1) provided that 2h + 1 ≤ k c p .


n p (k, h − 1) − n p (k, h) 2h − 1 > kc p

(2) M p (k, h) = 

0
 2h − 1 ≤ kc p .

Proof. (1) This is a direct consequence of Lemmas 12.7 and 12.12.

(2) By (1), if 2h − 1 ≤ kc p , we have M p (k, h) = 0. To prove the rest, we use
functional equation (see e.g. Proposition 10.3)
D p(1) (−c p − s) = D p(1) (s).
Following (12.54), write both sides explicitly (using M p (k, h) = 0 for 2h − 1 ≤
kc p ), we have
 
Y∞ 
 Y Y 
D p (s) =
(1) 
 ζ(ks
b + h)n p (k,h−1)
ζ(ks
b + h) n p (k,h−1)−M p (k,h)  ,

k=1 h≤(kc p +1)/2 h>(kc p +1)/2
 
Y∞   Y Y 
D p (−c p − s) =
(1) 

bζ(ks + h)n p (k,h)
ζ(ks
b + h)n p (k,h)−M p (k,kc p −h+1)  .

k=1 h≥(kc p +1)/2 h<(kc p +1)/2

Consequently, for h > (kc p + 1)/2, by comparing corresponding terms in these

two formulas, we conclude M p (k, h) = n p (k, h − 1) − n p (k, h). 
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 320

320 Algebraic, Analytic Structures and Riemann Hypothesis

As a direct consequence, by adding special zeta value factor D p(1) , we arrive at

Corollary 12.4. The function ∆ p (s) is given by

 
∞ 
Y  Y Y 
∆ p (s) = 
 ξ(ks + h)n p (k,h−1) ξ(ks + h) n p (k,h) 
  .
k=1 h≤(kc p +1)/2 h>(kc p +1)/2

Lemma 12.17. Let k ≥ 1.

(1) n p (k, h) = n p (k, kc p − h) if h ≥ 1.

(2) n p (k, h) ≤ n p (k, h + 1) if 2h + 1 ≤ kc p .

Proof. (1) is proved in Lemma 10.2. As for (2) (and in fact also (1)), please refer
to Proposition 1 of [73]. 

Corollary 12.5. Let w0 be the longest element in W. Then

M p (k, h) = n p,w0 (k, h − 1) − n p,w0 (k, h) ∀(k, h).

Proof. Since n p (k, h) = n p,w0 (k, h), this is a direct consequence of Lemmas 12.16
and 12.17. 

12.4.4 Zero Free Region on Right Half Plane

With the normalization factor treated, we are now ready to study zero free regions
of the function A p (s). We start with a simple case over a half plane on the right.

ζQG/P (s) is not zero.

Lemma 12.18. Assume that the discriminant of b

z p (s) cp
(1) There exists no zero for in the right half plane <(s) ≥ − .
∆ p (s) 2
z p (s) cp
(2) There exists no zero for in the region <(s) ≥ − − δ(t), where δ(t)
∆ p (s) 2
is a certain positive function satisfying δ(t) log |t| → ∞ (|t| → ∞).

Proof. Recall that by Lemma 12.3 and Corollary 12.4, we have, respectively,
∞ Y
Y ∞
z p (s) = ξ(ks + h)N p (k,h−1) ,
k=1 h=2

∞ 

 (12.57)
Y Y Y
∆ p (s) = ξ(ks + h) ξ(ks + h)  .
 n p (k,h−1) n p (k,h) 



k=1 h≤(kc p +1)/2 h>(kc p +1)/2
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 321

Algebraic and Analytic Structures and Weak RH 321

This implies
∞
z p (s) Y Y
= ξ(ks + h)n p (k,h−1)−n p (k,h) . (12.58)
∆ p (s) k=1 h>(kc +1)/2
p

In particular, the right hand contains no zeta factor in the denominator, since
n p (k, h − 1) − n p (k, h) ≥ 0 by Lemma, 12.17.
1−h
Recall that ξ(s) , 0 for <(s) ≥ 1. Hence ξ(ks + h) , 0 for <(s) ≥ .
k
kc p + 1 cp 1 − h
Therefore, it suffices to show that when k ≥ 1, h > implies ≥ .
2 2 k
kc p + 1 1−h c p
But this is rather trivial, since, if kc p is even, h > implies ≤− ,
2 k 2
1−h cp 1 cp
and if kc p is odd, ≤− − < − . This proves (1) and hence also
k 2 2k 2
(2) by a standard work on zero free region of ξ(s) in [116]. In fact, follow-
ing [116], with the Vinogradov method, δ(t) can be taken to satisfy the condition
1
δ when |t| → ∞. 
(log t)2/3 (log log t)1/3
B p (s)
Recall that, in (12.29), A p (s) = is an entire function defined using
R p (s) ∆ p (s)
 
 X 1 X  r
B p (s) :=   +  c p,w c a (s) z (s),
 F p,w p,w p,w
w∈W >
2 w∈W =
(12.59)
p p
n o
R p (s) := g.c.d a p,w (s) : w ∈ W 0 .
As a direct consequence of Propositions 12.4, 12.6 and Lemma 12.18, we have

ζQG/P (s) is not zero.

Corollary 12.6. Assume that the discriminant of b

(1) There are at most finitely many zeros of the function A p (s) lying in the right
cp
half-plane <(s) ≥ − .
2
(2) There are at most finitely many zeros of the function A p (s) lying in the re-
cp
gion <(s) ≥ − − δ(t), where δ(t) is a certain positive function satisfying
2
δ(t) log |t| → ∞ ( |t| → ∞).

Proof. Similarly as above, it suffices to prove (1). By an easy calculation,

B p (s) = a p (s)z p (s) 1 + r p (s)

where  
1 X  a p,w (s) z p,w (s)
 X !
r p (s) :=  +  · .
>
w∈W p rW p
2 w∈W =  a p (s) z p (s)
p
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 322

322 Algebraic, Analytic Structures and Riemann Hypothesis

Thus, if we set
   

  X 1 X  a p,w (s) 

+ ,
 
D p :=  s ∈ : ≥ 1
   
C  
2 a (s)
 


  >
w∈W rW p w∈W p=
 p 

p

by Proposition 12.4, D p ⊂ C is bounded. Hence, by Proposition 12.6,

cp
r p (s) < 1 for <(s) ≥ − with s < D p . (12.60)
2
Therefore, (1) is a direct consequence of Lemma 12.18. 

12.4.5 Zero Free Region on Left Half Plane

We next treat the zero free region on the left. This is more crucial and complicated.

Proposition 12.7. There exists no zero for the function B p (s) in the region on the
left defined by <(s) ≤ −κ p log |=(s)| + 10 , where κ p is a certain positive real


number.

Proof. Write C 3 s = σ + it with σ, t ∈ R. By Lemma 12.2, we have

 
B p (s) = z p (s) a p (s) + b p (s) ,
where
 
z p,w (s)
!
 X 1 X 
b p (s) :=  +  a (s) .
>
w∈W p rW p
2 w∈W =  p,w z p (s)
p

Since z p (s) is a finite product of zeta functions and hence can be treated rather
easily as above using the functional equation. Hence it suffices to find the zero
free region on the left for the function a p (s) + b p (s).
As in the proof of Proposition 12.6, we use
k(p) Y
z p,w (s) Y ξ(ks + hm,w (k))
= , (12.61)
z p (s) k=1 m∈Λw (k)
ξ(ks + hm (k) + 1)

to analyze its asymptotic structure as σ → −∞.

Lemma 12.19. For fixed integers k, a, b (k > 0), we have

(1) There exists real constant Cn (k; a, b) such that, as σ → ∞,

ζ (1 − ks) − a

X Cn (k; a, b)

=1+ .
ζ (1 − ks) − b − 1 n≥2
n−s
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 323

Algebraic and Analytic Structures and Weak RH 323

(2) As σ → −∞, for any fixed n ≥ 0, there exists real constants c j (a, b) such that
! a−b−1
ξ(ks + a)
!
2π 2
c1 (a, b) cn (a, b)
= − s 1+ + ... + + O |s| −n−1

ξ(ks + b + 1) k s sn
 
 X Cm (k; a, b) 
× 1 +  .
m≥2
m−s 
Proof. (1) To write down the left hand side as a Dirichlet series, it suffices to use
∞
1 X
the well-known formula = µ(n) n−s , where, as usual, µ(n) denotes the
ζ(s) n=1
n-th Mobius number.
(2) is a direct consequence (1). Indeed by the well-known Stirling formula,
we have, for | arg(s) | < π − ε,
√
!
1 1 1
Γ(s) = 2πss e 1 +
s −s
+ + ... + + O |s| −n−1

. (12.62)
12s 288s2 cn sn
Therefore, with | arg(s)| < π−ε, as |s| → ∞, there are real coefficients polynomials
a j (x) such that
Γ(s + λ)
!
a1 (λ) an (λ)
= sλ 1 + + . . . + n + O |s|−n−1


(12.63)
Γ(s) s s
where O |s|−n−1 depends on λ and ε. Applying this, resp. (1), to the Gamma


ξ(ks + a)
factors, resp. the zeta factors, of , we obtain (2) with an obvious
ξ(ks + b + 1)
change of variables. 
z p,w (s)
Directly applying this result to each term of in (12.61), we obtain
z p (s)
!  
z p,w (s) c1 (w) cn (w) X Cm (w) 
= (−2πs)a p,w eb p,w 1+ + . . . + n + O |s|−n−1
 1 +
 
z p (s) s s m−s  
m≥2
for some real constants ci (w) and C j (w). Here
k(p)
1X X
hm (k) − hm,w (k) + 1 ,


a p,w :=
2 k=1 m∈Λ (k)
w
(12.64)
k(p)
1 X X
hm (k) − hm,w (k) + 1 .


b p,w := log k
2 k=1 m∈Λw (k)
Since, for 1 ≤ k ≤ k(p) and m ∈ Λw (k), by definition, hm (k) − hm,w (k) ≥ 0, we
1
have, a p,w ∈ Z>0 . Hence, for sufficiently large n,
2
∞  n
  
X  Qµ (s1/2 )  X ck (µ)
a p (s) + b p (s) = 1 + + O |s|

−(n+1)/2

 (12.65)
µ

−s k/2

µ=1 k=1
s
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 324

324 Algebraic, Analytic Structures and Riemann Hypothesis

n o
where Qµ (s) are polynomials and M := max deg(Qµ (s)) : µ = 1, 2, . . . < ∞.
Let µ0 be the smallest positive integer X = M. Then, we can


XsuchX
that deg
X Qµ0 (s)
rewrite the summation in (12.65) as = + + so as to get
µ≥1 µ=µ0 µ≤µ0 −1 µ≥µ0 +1
n
1/2
 
Qµ0 (s )  X c1 (µ 0 )
a p (s) + b p (s) = 1 + + O |s|−(n+1)/2

µ0

−s k/2
k=1
s
µ −1 ∞ 
  n

 X0 X 1 Qµ (s1/2 )  X c1 (µ)
+  + 1 + + O |s|  .
 
 −(n+1)/2


µ=1 µ=µ +1
(µ/µ0 )−s Qµ0 (s1/2 )  k=1
sk/2
0

Now we take σ0 > 0 such that, for <(s) < −σ0 and |=(s)| > 1,
∞ n
 
1 Qµ (s1/2 )  (µ)
c  < 1 .
X X 1
1 + + O |s|−(n+1)/2

(µ/µ )−σ Q (s 1/2 ) s k/2 2
µ=µ +1 0 µ0 k=1
0

Hence, for <(s) < −σ0 , when |=(s)| → ∞, with our choice of µ0 , we have
µX0 −1
 n

1 Qµ (s1/2 )  X c1 (µ)  = O µσ0 |s|−1/2 .
 
1 + + O |s|

−(n+1)/2

(µ/µ0 )−σ 1/2
Qµ0 (s ) s k/2
µ=1 k=1

Therefore, for sufficiently large R > 0, when <(s) < σ0 and |=(s)| > R,
Qµ0 (s1/2 )   
a p (s) + b p (s) = 1 + ϕ(s; 1/2) + O µ−σ
0 |s|
−1/2
(12.66)
µ0 −s

for a certain function ϕ(s; 1/2) satisfying ϕ(s; 1/2) < 1/2. Consequently, when
|s| → ∞ with σ < κ p log(|t| + 10),
 
B p (s) = z p (s) a p (s) + b p (s) = z p (s) Qµ0 (s1/2 ) µ0s 1 + φ(s)


(12.67)
for a certain function φ(s) satisfying |φ(s)| < 1. If necessary, by enlarging κ p , we
deduce the assertion for zeros of B p (s) using (12.67). 

Corollary 12.7. All but finitely many zeros of the function A p (s) lie in the region
 cp 
s = σ + it ∈ C : −κ p log |t| + 10 < σ < − .


(12.68)
2
Proof. Indeed, form the proof above, particularly, (12.67), we have
z p (s) 1   z p (s) 1  
A p (s) = a p (s) + b p (s) = Qµ0 (s1/2 ) µ0s 1 + φ(s) .
∆ p (s) R p (s) ∆ p (s) R p (s)
If necessary, by enlarging κ p , this expression of A p (s) takes care of the lower
bounded of σ. As for the upper bound of σ, we apply Corollary 12.6 and Propo-
sition 12.7. 
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 325

Algebraic and Analytic Structures and Weak RH 325

12.5 Hadamard Product

12.5.1 Distributions for Zeros of A p(s)

We first recall a modification of Lemma 11.3, taken from [116], see e.g. page 213.

Lemma 12.20. Fix σ0 > 0, T > 10. Let 0 ≤ α < β < σ0 , where α and β may
depend on T . Let f (s) be an analytic function admitting no zero on the line t = T ,
real for real variables, and regular for σ ≥ α except for possibly finitely many
poles on the real line. Choose constants m, Mσ,t satisfying

< f (σ + it) ≥ m > 0 and f (σ0 + it0 ) ≤ Mσ,t (∀σ0 ≥ σ, 1 ≤ t0 ≤ t).

Then, for σ ≥ β,
π   3
arg f (σ + iT ) ≤
log Mα,T +2 + log m + π. (12.69)


log (σ0 − α)/(σ0 − β) 2

We use this lemma to analysis the distributions for the zeros of A p (s).
cp
Proposition 12.8. Let T > 1 and σ > . Denote by N(T ; σ) be the number of
2
zeros of A p (s) in the region
n o
s ∈ C : −σ < <(s) < −c p /2 − δ(t), 0 < =(s) < T . (12.70)

Assume that the discriminant of b ζQG/P (s) is not zero. There exist real constants
σL > 0, and c1 > 0, c2 , c3 such that
N(T ; σL ) = c1 T log T + c2 T + O(log T ),
(12.71)
N(T ; +∞) = c1 T log T + c3 T + O(log2 T ).
In particular,

N(T ; +∞) − N(T ; σL ) = O(T ). (12.72)

Proof. This is rather standard. Indeed, from the proof of Proposition 12.7, there
exist a positive integer µ0 and a suitable function φ(s) together with a positive real
number σL such that
 
(1) a p (s) + b p (s) = Qµ0 (s1/2 ) µ0s 1 + φ(s) .
(2) | φ(−σL + it) | < 1 for some fixed σL > 0 as |t| → ∞.
(3) | φ(s) | < 1 for <(s) < κ p log |=(s)| + 10 as |s| → ∞.


December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 326

326 Algebraic, Analytic Structures and Riemann Hypothesis

Moreover, using the function r p (s) introduced in the proof of Corollary 12.6, i.e.
 
 X 1 X  a p,w (s) z p,w (s) !
r p (s) :=  +  a p (s) · z p (s) ,
 (12.73)
>
w∈W rW
2 w∈W =
p p p

we obtain, by the definition of R p (s) in (12.59) and the proof of Corollary 12.6, in
particular, (12.60),

a p (s1/2 )
(4) is a polynomial,
R p (s)
z p (s) a p (s1/2 )  
(5) A p (s) = 1 + r p (s) , and
∆ p (s) R p (s)
(6) |r p (s)| < 1 for <(s) > −c p /2 if s < D p , where
   

  X 1 X  a p,w (s) 

+ .
 
D p :=  s ∈ : ≥ 1 (12.74)
   
C  
2 a (s)
 


  >
w∈W rW w∈W =
 p 

p p p

Assume that T is sufficiently large, and introduce a rectangle RT with vertices

−σL + ci, σR + ci, σR + iT, −σL + iT , where σR , c are sufficient large positive real
numbers satisfying the conditions that

(i) A p (s) has no zeros on the boundary of RT , and

(ii) |φ(−σL + ct)| < 1 and |r p (σR + it)| < 1 for |t| ≥ c.

By Cauchy’s residue formula, we have

Z
1 1  
N(T ; σL ) = d log A p (s) = ∆1 + ∆2 + ∆3 + ∆4 arg A p (s) ,


2πi RT 2πi
where ∆1 , ∆2 , ∆3 , ∆4 denote the variations from σR + ci to σR + iT , from σR + iT
to −σL + iT , from −σL + iT to −σL + ci, from −σL + ci to σR + ci, respectively.
By definition, obviously, ∆4 arg A p (s) = O(1). To calculate ∆1 arg A p (s) and



∆3 arg A p (s) , we use (4), (5) and (1), (3) to deduce the calculations to either


polynomials or the Riemann zeta functions. Standard calculations yields the cor-
responding T log T , T and O(log T ) terms, respectively. Thus what left is to treat
∆2 arg A p (s) . Set

∞
Y Y
Γ∗ (s) := γ(ks + h)n p (k,h−1)−n p (k,h) ,
k=1 h>(kc p +1)/2
∞
a p (s) Y Y
L(s) := ζ(ks + h)n p (k,h−1)−n p (k,h) ,
R p (s) k=1 h>(kc +1)/2
p
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 327

Algebraic and Analytic Structures and Weak RH 327

where γ(s) := s(s − 1)Γ(s). Then

A p (s) = Γ∗ (s) · L(s) 1 + φ p (s) ,

  

∆2 arg A p (s) = ∆2 arg Γ∗ (s) + ∆2 arg L(s) 1 + φ p (s) .



Obviously, it suffices to treat the term ∆2 arg L(s) 1+φ p (s) , since a standard cal-
 
culation using the Stirling formula easily evaluates ∆2 arg Γ∗ (s) . By definition,
 "
 X 1 X  a p,w (s)
L(s) φ p (s) =  + 
>
w∈W p rW p
2 w∈W =  R p (s)
p
# (12.75)
γ hλ p , α is + hρ, α∨ i
∨
Y

×
ζ hλ p , α is + hρ, α i .
∨ ∨

γ hλ p , α∨ is + hρ, α∨ i + 1
α∈(Φ+ rΦ+p )∩w−1 (Φ− )

a p,w (s)
Since is a polynomial, it is also quite standard to estimate this term. In-
R p (s)
deed, by Chapter V of [116], there exists a positive number M such that, when
−σL ≤ <(s) ≤ σR , we have L(s) 1 + φ p (s)  T M . Therefore, by Lemma 12.20,



∆2 arg L(s) 1 + φ p (s) = O log T .

This proves the asymptotic formula for N(T ; σL ).

To understand N(T ; +∞), it suffices to study N(T ; −κ p log T ) by Proposi-
tion 12.7. Accordingly, we form a rectangle R∗T with vertices −κ p log T + c0 i, σR +
c0 i, σR + iT, −κ p log T + iT , where c0 is a sufficient large positive real number. We
assume that A p (s) has no zeros on boundary of R∗T . Similarly, as above, we have
1  ∗ 
N(T ; σ + ∞) = ∆1 + ∆∗2 + ∆∗3 + ∆∗4 arg A p (s) ,


2πi
Almost all pave the same way as above, except that now, by (12.75), when σ =
−κ log T ≤ <(s) ≤ σR , we have, for a suitable constant κ∗p > κ p ,

L(s)φ p (s)  T κ p log T .

∗

Therefore, by applying Lemma 12.20 again with α = −2κ p log T, β = −κ p log T

and σ0 = σR ,

 

∆∗2 arg L(s) 1 + φ p (s) = O log2 T and ∆∗4 arg L(s) 1 + φ p (s) = O log T .

This proves the asymptotic formula for N(T ; ∞) and hence the proposition as
well. 
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 328

328 Algebraic, Analytic Structures and Riemann Hypothesis

12.5.2 Hadamard Product

We here show that A p (z) admits a refined structure of the Hadamard product, using
Proposition 12.8. Set for simplicity
 c 
ep (z) := A p − p + i z ,
A (12.76)
2

ζQG/P (s) is not zero. Then the

Proposition 12.9. Assume that the discriminant of b
ep (z) admits a refined Hadamard product
function A
ep (z) = κ eαz G(z) H1 (z) H2 (z),
A (12.77)
where

(1) both κ and α are real and κ , 0,

(2) G(z) is a polynomial, admits no real zeros in =(z) > 0,
∞  ! ∞  !
Y z  z Y z  z
(3) H1 (z) = 1− 1+ and H2 (z) = 1− 1+
n=1
ρn ρn n=1
ηn ηn
are uniformly convergent on every compact subset in C. Here, for all n,
<(ρn ) > 0, <(ηn ) > 0 and
0 < δ(t) < =(ρn ) < σL + 1 < =(ηn ) < κ p log <(ηn ) + 10 ,


(12.78)
where δ(t), resp. κ p , resp. σL , is the function, resp. the positive real number,
resp. the positive real number, introduced in Proposition 12.6, resp. Proposi-
tion 12.7, resp. Proposition 12.8.

The important parts are that α is a real number and the locations of the zeros
of G(z) H1 (z) H2 (z).

Proof. The function A p (z) is a linear combination of the products of finite zeta
function factors b ζ(ks + h), or better ξ(ks + h), with the coefficients of rational
function. Hence A ep (z) is an entire function of order at most one, since the ξ(ks +
h)’s are entire functions of order one (and maximal type). In addition, A p (z) takes
real values over real axis. Hence, by the functional equation, if ρ is a zero of
ep (z), so is −ρ. Therefore, by the theory of entire functions, using Corollary 12.6
A
and Proposition 12.7 on the locations of zeros of A p (z), we conclude that A ep (z)
admits an Hadamard product of the form
ep (z) = κ eβz G(z) h1 (z) h2 (z),
A (12.79)
where κ, resp. β, resp. G(z), is a non-zero real number, resp. a complex num-
ber, resp. a non-zero polynomial having no zero in =(z) > 0 except for purely
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 329

Algebraic and Analytic Structures and Weak RH 329

imaginary zero, and

∞ "  ! !#
Y z  z z z
h1 (z) = 1− 1+ exp − ,
n=1
ρn ρn ρn ρn
∞ "  ! !# (12.80)
Y z  z z z
h2 (z) = 1− 1+ exp − .
n=1
ηn ηn ηn ηn
Here, for all n, <(ρn ) > 0, <(ηn ) > 0 and =(ρn ) and =(ηn ) satisfy the condition
(12.78) in the proposition, and (12.80) converge uniformly on every compact sub-
set in C. In addition, if we set ρn = an + ibn and ηn = cn + idn with an , cn ∈ R and
bn , dn ∈ R>0 ,
∞ ∞ ∞
X 1 1 X bn X 1
− ≤2 ≤ 2(σ L + 1) ,
n=1
ρn ρn n=1
|ρn |2
n=1
|ρn |2
∞ ∞ ∞ ∞
(12.81)
X 1 1 X dn X log(cn + 10) X 1
− ≤2 ≤ 2κ p ε .
n=1
ηn ηn n=1
|ηn |2 n=1
|ηn |2 n=1
|ηn |2−ε
By Proposition 12.8, the sums on the right hand sides of (12.81) are !all fi-
z z
nite. Hence we can and hence will move the factors exp − and
! ρn ρn
z z
exp − out of the infinite product. This then implies the exis-
ηn ηn
tence of structural product in the statement of the proposition, if we set
∞ ! X ∞ !
X 1 1 1 1
α := β + − + − .
n=1
ρn ρn n=1
ηn ηn

Lemma 12.21. With the same notation as above, α ∈ R, or the same =(α) = 0.

Proof. This is quite structural as a consequence of the functional equation for

ep (z) = κ eαz G(z) H1 (z) H2 (z),
ep (z). Indeed, since A
A
∞ ∞
ep (−iy)
A G(−iy) X iy + ρn X iy + ηn
log = 2y =(α) + log + log +2 log .
ep (iy)
A G(iy) n=1
iy − ρn n=1
iy − ηn
G(−iy)
Easily log = o(1), and, by Proposition 12.8, as y → +∞, by 0 < bn <
G(iy)
σL + 1 and 0 < dn < κ p log(cn + 10), we have
∞ ∞
∞
a2 + (y + bn )2 X

X 4ybn X y log n 
log n2 ≤ = O  = O(log y)
an + (y − bn )2 a2 + (y − bn )2

n=1 n=1 n n=1
n2 + y2 
∞ ∞
∞
c2n + (y + dn )2 X

X 4ydn X y log n 
log 2 ≤ = O   = O(log y),
+ 2 + (y − d )2

c (y − d ) 2 c (n2 + y2 )1−ε 
n=1 n n n=1 n n n=1
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 330

330 Algebraic, Analytic Structures and Riemann Hypothesis

respectively. Therefore, to obtain =(α) = 0, it is sufficient to prove the following

Lemma 12.22. As y → +∞, there exists some constant A , 0 and m ≥ 0 such

that
A p − c p /2 + y


 

= A ym 1 + o(1) . (12.82)
A p − c p /2 − y

Proof. By definition and the functional equation,

A p − c p /2 + y B p − c p /2 + y R p − c p /2 − y ∆ p − c p /2 − y






=
A p − c p /2 − y B p − c p /2 − y R p − c p /2 + y ∆ p − c p /2 + y

B p − c p /2 + y


=
.
B p − c p /2 − y
Thus it suffices to verify the statement in the latest lemma for the function BP (s)
instead of A p (s). Now, by Propositions 12.4 and 12.6, we have, as y → +∞,

 
B p − c p /2 + y = a p − c p /2 + y · z p − c p /2 + y · 1 + o(1) .



(12.83)

Hence, what left is to verify the following

Claim 12.1. As y → +∞, there exists some constant A , 0 and m ≥ 0 such that
B p − c p /2 − y


 

= A−1 y−m 1 + o(1) . (12.84)
a p − c p /2 + y · z p − c p /2 + y

B p − c p /2 − y


Proof. By definition,
is given by
a p − c p /2 + y z p − c p /2 + y


 
1 X   a p,w − c p /2 − y z p,w − c p /2 − y 
 X


 +
 .
2 w∈W =  a p − c p /2 + y z p − c p /2 + y
  

w∈W > rW
p p p

By Proposition 12.4 on the leading polynomial a p (s), we have, as y → ∞, there

exists real numbers a1 (w), . . . , an (w) such that
a p,w − c p /2 − y


a1 (w) an (w)

= δ0w + + . . . + n + O y−n−1 ,


a p − c p /2 + y y y

where δ0w = 1 if w ∈ W p and w−1 (∆) r Φ p = 1, and 0 otherwise. As for the zeta
factor, we write
z p,w − c p /2 − y z p,w − c p /2 + y z p,w − c p /2 − y





=
·
.
z p − c p /2 + y z p − c p /2 + y z p,w − c p /2 + y
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 331

Algebraic and Analytic Structures and Weak RH 331

In addition, similar to the proof of Proposition 12.7, we have, as σ → +∞, there

exist real numbers b p,w , b1 (w), . . . , bn (w) such that
z p,w (s)
!
b1 (w) bn (w)
 
= −2πs a p,w eb p,w 1+ +. . .+ n +O |s|−n−1 1+O(2−σ ) (12.85)


z p (s) s s
where a p,w ∈ 12 Z>0 and b p,w are the constants in (12.86), defined by
k(p)
1X X
hm (k) − hm,w (k) + 1 ,


a p,w :=
2 k=1 m∈Λ (k)
w
(12.86)
k(p)
1X X
hm (k) − hm,w (k) + 1 .


b p,w := log k
2 k=1 m∈Λw (k)

z p,w − c p /2 − y


Hence, it suffices to analyze
. Recall that, by Lemma 12.19, as
z p,w − c p /2 + y
σ → +∞,
ξ k(−c p − s) + h + m

ξ ks + h + m


!− kc p −2h−2m+1 !
2π 2
c1 (k, h) cn (k, h) 
= 1+ + ... + n
+ O(|s|−n−1
) 1+O(2−σ ) .
ks s s
Thus, by §12.4.1, there exist real numbers b01 (w), . . . , b0n (w) such that, as σ → +∞,
z p,w (−c p − s) Y ξ hλ p , α∨ i(−c p − s) + hρ, α∨ i + δα,w

=
ξ hλ p , α∨ is + hρ, α∨ i + δα,w


z p,w (s) α∈Φ+ rΦ+ p
+
p +hk
k(p) kcY
ξ k(−c p − s)+h n p,w (k,h) ξ k(−c p − s)+h+1 n p (k,h)−n p,w (k,h) (12.87)
Y
!
!
=
ξ ks + h ξ ks + h + 1



k=1 h=1
 2π −a0p,w 0 !
b1 (w) bn (w)
 
= eb p,w 1 + + . . . + n + O |s|−n−1 1 + O(2−σ .
s s s
0 0
Here, a p,w and b p,w are real constants with
+
k(p) k+h
X Xk  
a0p,w := n p,w (k, h)(kc p − 2h + 1) + (n p (k, h) − n p,w (k, h))(kc p − 2h − 1) .
k=1 h=k+h−k

To see the sign of a0p,w , we rewrite it as

+
k(p) k+h
X Xk  
a0p,w = n p (k, h)(kc p − 2h) + 2n p,w (k, h) − n p (k, h)
k=1 h=k+h−k
+ +
k(p) k+h
X Xk  k(p) k+h
Xk 
 X 
= n p (k, h)(kc p − 2h) + 2n p,w (k, h) − n p (k, h) .
k=1 h=k+h−k k=1 h=k+h−k
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 332

332 Algebraic, Analytic Structures and Riemann Hypothesis

By Lemmas 12.17 and 12.11, we have, respectively,

n p (k, h) = n p (k, kc p − h) and h+k + h−k = k(c p − 2).

This implies
+
p +hk
k(p) kcX
X  
a0p,w = 2n p,w (k, h) − n p (k, h)
k=1 h=k+h−k
+ +
p +hk
k(p) kcX
X p +hk
k(p) kcX
  X
=2 2n p,w (k, h) − n p (k, h) + n p (k, h)
k=1 h=k+h−k k=1 h=k+h−k

= −2 φ+ r Φ+p − l p (w) + Φ+ r Φ+p = − Φ+ r Φ+p + 2l p (w).

 

>

> 0 w ∈ W p

Therefore, −a p,w = 
0

Consequently, by (12.85) and (12.87), we
0
 w ∈ W p= .
have, as y → +∞, there exist real numbers b∗1 (w), . . . , b∗n (w) such that
B p − c p /2 − y

a p − c p /2 − y z p − c p /2 − y



(12.88)
b∗1 (w) b∗n (w)
!
−(a p,w −a0p,w ) b∗p,w

=y e δw +
0
+ . . . + n + O |y| −n−1

1 + O(2 .
−y
y y
This completes the proof of claim and hence two lemmas and the proposition. 

12.6 Proof of Theorem

Proof. Now we are ready to complete a proof of Theorem 12.1 following [56].
By Proposition 12.2,

ξQG/P (s) = A p (s) ± A p (−c p − s). (12.89)

In addition, if the discriminant ∆G/P Q of the zeta function bζQG/P (s) is not zero, by
Theorem 12.9, A p (s) admits a refined Hadamard product (12.77). Hence, by the
conditions (1), (2) and (3) listed in that theorem, it is possible to apply Lemma 11.6
for W(z) = A ep (z) = A p (−c p /2 + iz) to conclude that all but finitely many zeros of
cp
ζQG/P (s) lie on the central line <(s) = − .
b
2
To prove the related zeros are simple, we use the algebraic structures of A p (s)
again. Indeed, from (12.89), we have


  
ξQG/P − c p /2 + it = A p − c p /2 + it eiθ(t) ± eiθ(t) .
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 333

Algebraic and Analytic Structures and Weak RH 333



Here θ(t) := arg A p − c p /2 + it . Thus, by a standard argument, it suffices to
show that θ(t) is strictly increasing as t → +∞. By Propositions 12.4 and 12.6, on
c
the central line <(s) = − 2p , we have, as |s| → ∞

a p (s) z p (s)
B p (s) = a p (s) z p (s) 1 + o(1) and hence A p (s) = 1 + o(1) .



R p (s) ∆ p (s)

Thus, by (12.58), we have, as t → +∞, there exists a suitable constant c such that
d
θ − c p /2 + it = c log t · 1 + o(1) .



dt
z (s)
Indeed, the factor log t comes from the Gamma factors of ∆pp (s) in (12.58), and the
following standard estimation (see e.g. Theorem 5.7 of [116])

ζ0
!
log t
(1 + it) = O = o(log t),
ζ log log t

in the case3 kc p is even and n p (k, h − 1) − n p (k, h) > 0 for some 1 ≤ k ≤ k(p) with
h = 1 + kc p /2. Therefore, we complete a proof of Theorem 12.1. 

12.7 Discriminant and Residue of Period

12.7.1 Residue of Period

We first rewrite (7.85) in the proof of Proposition 7.6 as follows.

Proposition 12.10. The residue of ωGF (λ) at the Weyl vector λ = ρ is given by
X r −1 −
Resλ=ρ ωGF (λ) = ζF (2)−|∆∩w (Φ )|
cFp,w b
w∈W o
Y 1 Y ζF (hρ, α∨ i) (12.90)
.
b
×
hρ, α∨ i − 1 ζF (hρ, α∨ i + 1)
α∈w−1 (∆)r∆ α∈(Φ+ r∆)∩w−1 (Φ− )
b

Proof. This result is first proved in [57] at page 1684 (see also (10.4) of [56]). By
the proof of Proposition 7.6 we have

Resλ=ρ ωGF (λ) = Res s=0 ωG/P

F (s).
3 This worst case does happen, say, for G = SL , the factor ζ(1 + it) occurs in Ã(s) with c = −2 and
2 p
A(s) = ξ(s + 2) on <(s) = −1.
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 334

334 Algebraic, Analytic Structures and Riemann Hypothesis

To continue, we use (9.13) to write

X r −1
(Φ− )|
ωG/P
F (s) = ζF (2)−|∆ p ∩w
cFp,w · b
w∈W 0
Y 1 Y ζF (hρ, α∨ i)
b
×
hρ, α∨ i −1 ζF (hρ, α∨ i + 1)
α∈w−1 (∆)∩(Φ p r∆ p ) α∈(Φ+p r∆ p )∩w−1 (Φ− )
b
Y 1 Y ζF (hλ p , α∨ is + hρ, α∨ i)
b
× .
hλ p , α∨ is + hρ, α∨ i − 1 ζF (hλ p , α∨ is + hρ, α∨ i + 1)
α∈w−1 (∆)∩Φ p α∈(Φ+ rΦ+p )∩w−1 (Φ− )
b

When taking residue of ωG/P

F (s) at s = 0, for each w ∈ W , we should concentrate
0

on corresponding terms for which either

(i) hλ p , α∨ is + hρ, α∨ i − 1 = 0 for a certain α ∈ w−1 (∆) ∩ Φ p , or

ζF (hλ p , α∨ is + hρ, α∨ i) admits a pole, i.e. hλ p , α∨ is + hρ, α∨ i = 1 for a certain
(ii) b
α ∈ (Φ+ r Φ+p ) ∩ w−1 (Φ− ).

On the other hand, by the Lie structures exposed in §12.4.1, we have either


{α p } α p ∈ w (∆),
 −1
(1) ∆ ∩ w (∆) r Φ p = {α p } ∩ w (∆) = 
−1
 −1
or
∅
 otherwise,

{α p } α p ∈ w (Φ ),
 −1 −
+ +
(2) ∆ ∩ (Φ r Φ p ) ∩ w (Φ ) = 
−1 −


∅
 otherwise,

and two cases cannot happen at the same time. Therefore, using hλ p , α∨p i = 1, we
conclude that the residue Res s=0 ωG/P
F (s) is equal to

X r −1
(Φ− )|
Res s=0 ωG/P
F (s) = ζF (2)−|∆ p ∩w
cFp,wb
w∈W 0
α p ∈w−1 (∆)
Y 1 Y ζF (hρ, α∨ i)
b
×
hρ, α i − 1
∨
ζF (hρ, α∨ i + 1)
α∈w−1 (∆)∩(Φ p r∆ p ) α∈(Φ+p r∆ p )∩w−1 (Φ− )
b
Y 1 Y ζF (hρ, α∨ i)
b
×
hρ, α i − 1
∨
ζF (hhρ, α∨ i + 1)
α∈w−1 (∆)r(Φ p ∪{α p }) α∈(Φ+ rΦ+p )∩w−1 (Φ− )
b
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 335

Algebraic and Analytic Structures and Weak RH 335

X r cF b −1 −
+ cFp,w ζF (2)−|∆ p ∩w (Φ )|
w∈W 0
ζF (2)
b
α p ∈w−1 (Φ− )
Y 1 Y ζ(hρ, α∨ i)
b
×
hρ, α∨ i −1 ζ(hρ, α∨ i + 1)
α∈w−1 (∆)∩(Φ p r∆ p ) α∈(Φ+p r∆ p )∩w−1 (Φ− )
b
Y 1 Y ζ(hhρ, α∨ i)
b
×
hρ, α∨ i −1 ζ(hhρ, α∨ i + 1)
α∈w−1 (∆)rΦ p α∈(Φ+p r(Φ+p ∪{α p }))∩w−1 (Φ− )
b
X r −1
(Φ− )|
Y 1 Y ζ(hρ, α∨ i)
= ζ(2)−|∆ p ∩w
b
cFp,wb
hρ, α∨ i −1 ζ(hρ, α∨ i + 1)
α∈w−1 (∆)r∆ α∈(Φ+p r∆)∩w−1 (Φ− )
b
w∈W 0
α p ∈w−1 (∆)
X r +1 −1 −
Y 1 Y ζ(hρ, α∨ i)
+ ζ(2)−|∆∩w (Φ )|
b
cFp,w b
hρ, α i − 1
∨
ζ(hρ, α∨ i + 1)
α∈w−1 (∆)r∆ α∈(Φ+ r∆)∩w−1 (Φ− )
b
w∈W 0
α p ∈w−1 (Φ− )
as desired. 

Remark 12.1. We will prove in Part 6 that this residue coincides with the total
mass of semi-stable principle G-lattices with certain fixed invariants, as predicted
in the conjecture on stability, parabolic induction and masses.

12.7.2 Period of Pseudo Root System

Recall that, for the definition of periods of G and (G, P) associated to a reduc-
tive group G and its a standard maximal group P, what really matters is their
associated root system (V, h·, ·i; Φ, Φ± , ∆; W; ρ) and the induced data (∆ p , W p ).
In this sense, we may also call these periods as the periods for the root system
(V, (·, ·); Φ, Φ± , ∆; W; ρ) and (∆ p , W p ), respectively. We claim that more general
type periods can be introduced for what we call pseudo root systems.

Definition 12.11. By a pseudo root system (Φ, ∆), we mean the data
(V, (·, ·); Φ, Φ± , ∆; W; ρ) consisting of a metrized R-vector space (V, h·, ·i), four fi-
nite sets Φ, Φ± , ∆, a finite Coxeter group W and a vector ρ ∈ V, satisfying
X 1X
(1) Φ = Φ+ t Φ− , ∆ is a base of V, Φ− = −Φ+ ⊂ Z≥0 α, ρ = α and
α∈∆
2 α>0
W := h σα : α ∈ Φ+ i = h σα : α ∈ ∆ i and W(Φ) = Φ
(2) there is a decomposition Φ = tm Φ
i=1 i such that (V i , (·, ·) i , Φi , ∆i , Wi ) satisfies
the condition (1) with Vi ⊥ V j (i , j), V = ⊕i=1 Vi , W(Φ) = m
m Q
i=1 W(Φi ), and
the data (Φ+i = Φ+ ∩ Φi , ∆i = ∆ ∩ Φi ) is reduced in an obvious sense.
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 336

336 Algebraic, Analytic Structures and Riemann Hypothesis

In particular, a pseudo root system is called restricted if Φ is a subset of a root

system with (V, (·, ·); Φ, Φ± , ∆; W; ρ) the natural induced structure from the root
system.

Similar to root systems, for a pseudo one of rank n, we set ∆∨ = α∨ : α ∈ ∆


∆ the basis of V which is dual to ∆∨ with λα the element corresponding to
and b
α ∈ ∆∨ . Quite often, we also write ∆ = α1 , α2 , . . . , αn , ∆∨ := {α∨1 , α∨2 , . . . , α∨n
∨ 
2
with α := α and ∆∨ := λ1 , λ2 , . . . , λn with λi (α j ) = δi j for all 1 ≤ i, j ≤ n.

(α, α)
The λi ’s are sometimes called the fundamental domination weights.

Example 12.1. (1) Recall that a finite set Φ is called a root system if

(a) Φ is a finite subset in a metrized R-vector space, and does not contain 0;
(b) For α ∈ Φ, Zα ∩ Φ = {±α};
(c) For α ∈ Φ, σα (Φ) = Φ; and
2
(d) h β, α i := (β, α∨ ) = (β, α) ∈ Z for all α, β ∈ Φ.
(α, α)

Hence a restricted pseudo root system is a root system. Moreover, for a Φ satis-
fying (a, c, d) but (b), then, by Exercise 9 of §9.4 in [44], for α ∈ Φ, among all
multiples of α, only ± 21 α, ±α, ±2α ∈ Φ. These Φ’s are clearly pseudo root sys-
tems. For a concrete example of pseudo root system which is not a root system,
please see the same exercise cited above.

(2) Let (V, (·, ·); Φ, ∆) be a root system and write ∆ = α1 , α2 , . . . , αn . For each

fixed 1 ≤ p ≤ n, set ∆ p = {α1 , . . . , αbp , . . . , αn and let Φ p be the subset of Φ
consisting of the roots which are perpendicular to the p-th fundamental dominate
weight λ p . Set Φ±p := Φ± ∩ Φ p . Then (Φ p , Φ±p , ∆ p ) is certainly a restricted pseudo
root system, hence a (sub) root system (of Φ).

(3) Let (V, (·, ·); Φ, ∆) be a root system and write ∆ = α1 , α2 , . . . , αn . For each

fixed 1 ≤ p ≤ n, set ∆ p = {α1 , . . . , αbp , . . . , αn and denote by H p := α⊥p ⊂ V be the
cxdimension one hyperplane in V which is perpendicular to λ p . Obviously, ∆ p is
a R-basis of H p . Denote by Φ fp , resp. Φ f±p , be the collection of the projections of
the elements of Φ, resp. Φ , on H p . Then (Φ
± fp , Φ
f±p , ∆ p ) is a pseudo root system.

pseudo-root system V, (·, ·); Φ, Φ± , ∆, ∆∨ , b

∆; W; ρ and a


Definition 12.12. For aX
coordinate system λ = sα λα + ρ of V, the associated period ωF (λ) over a
(Φ,∆)

σ∈∆
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 337

Algebraic and Analytic Structures and Weak RH 337

number field F is defined by

X 1 Y ζF (hλ, αi)
ω(Φ,∆) .
b
(λ) := (12.91)
F
ρ, αi
Q
α>0,wα<0 ζF (hλ, αi + 1)
w∈W α∈∆ hwλ − b

Proposition 12.11. Let (Φ, ∆) be a pseudo root system of rank n and fix 1 ≤ p ≤ n.

(1) For the period ω(Φ,∆)

Q (λ) of (Φ, ∆) over Q,
X −1 −
Resλ=ρ ω(Φ,∆)
Q (λ) = ζ(2)−|∆∩w (Φ )|
b
w∈W
w(∆)⊂∆∪Φ−
(12.92)
Y 1 Y ζ(hρ, α∨ i)
.
b
×
hρ, α∨ i − 1 ζ(hρ, α∨ i + 1)
α∈w−1 (∆)r∆ α∈(Φ+ r∆)∩w−1 (Φ− )
b
(Φ ,∆ p )
(2) For the period ωQ p (λ) of (Φ p , ∆ p ) over Q,
(Φ ,∆ )
X −1 −
Resλ=ρ p ωQ p p (λ) = ζ(2)−|∆ p ∩w (Φ p )|
b
w∈W p
w(∆ p )⊂∆ p ∪Φ−p
(12.93)
Y 1 Y ζ(hρ∨ , αi)
.
b
×
hρ p , α∨ i − 1 ζ(hρ, α∨ i + 1)
α∈w−1 (∆ p )r∆ p α∈(Φ+p r∆ p )∩w−1 (Φ−p )
b

Proof. (1) comes directly from Proposition 12.10. Moreover the proof there is
obviously valid for (2) as well. 

12.7.3 Residue and Leading Coefficient

The residue (2) in Proposition 12.11 above is closely related with the leading coef-
ficient of the polynomial a p (s) in (12.31) of §12.3.2. Recall that, for each w ∈ W 0 ,
the invariant d p,w is defined by
1 −1 +
Y
:= 2|∆ p ∩w (Φ )| hρ, α∨ i − 1


d p,w
α∈w−1 (∆)∩(Φ p r∆ p )
Y 
 (12.94)
hρ, α∨ i + δα,w hρ, α∨ i + δα,w − 1 .


×
α∈Φ+p r∆ p

Proposition 12.12. Let (Φ, ∆) be a rank n root system. Then for each 1 ≤ p ≤ n,
(Φ ,∆ ) 1 X 1
Resλ=ρ p ωQ p p (λ) = Q · c p,w d p,w ,
p αw i
, ∨
α∈Φ+ r∆ ζ(hρ, α i + 1)
b ∨
w∈W
hλ
p p p
|w−1 (∆)rΦ p |=1

where αw is the unique element of w−1 (∆) r Φ p .

December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 338

338 Algebraic, Analytic Structures and Riemann Hypothesis

Proof. By the definition and (12.93), it suffices to prove that

c p,w d p,w
−1 +
Y 1 Y
(12.95)
=bζ(2)−|∆ p ∩w (Φ )| ζ(hρ∨ , αi + δα,w ).
b
hρ p , α∨ i − 1 α∈Φ+ r∆
−1 α∈w (∆)∩(Φ p r∆ p ) p p
+
Since |∆ p ∩ w −1
(Φ )| = |∆ p ∩ w−1 (∆)| for w ∈ W , 0
−1
X 1 X b ζ(2)−|∆ p ∩w (∆)| Y 1
c d p,w =
hλ p , α∨w i p,w hλ p , α∨w i hρ p , α∨ i − 1
w∈W p w∈W p α∈w−1 (∆)∩(Φ p r∆ p )
|w−1 (∆)rΦ p |=1 |w−1 (∆)rΦ p |=1
Y Y
× ζ(hρ∨ , αi + 1)
b ζ(hρ∨ , αi).
b
α∈(Φ+p r∆ p )∩w−1 (Φ+ ) α∈(Φ+p r∆ p )∩w−1 (Φ− )
Consequently,
−1
X 1 X ζ(2)−|∆ p ∩w (∆)|
c d p,w =
b

w∈W p
hλ p , α∨w i p,w w∈W p
hλ p , α∨w i
|w−1 (∆)rΦ p |=1 ∆ p ⊂w−1 (∆ p ∪Φ−p )
Y 1 Y Y
× ζ(hρ∨ , αi+1)
b ζ(hρ∨ , αi).
b
hρ p , α∨ i−1
α∈w−1 (∆)∩(Φ p r∆ p ) α∈(Φ+p r∆ p )∩w−1 (Φ+ ) α∈(Φ+p r∆ p )∩w−1 (Φ− )
Furthermore, for w ∈ W p , easily,
w−1 (∆) ∩ (Φ p r ∆ p ) = w−1 (∆ p ) r ∆ p and ∆ p ∩ w−1 (∆) = ∆ p ∩ w−1 (∆ p ).
Also, for w ∈ W p , by the facts that,
(Φ+p r ∆ p ) ∩ w−1 (Φ± ) = (Φ+p r∆ p ) ∩ w−1 (Φ±p ) ∪ (Φ+p r∆ p )∩w−1 (Φ± rΦ±p )

and that (Φ+p r ∆ p ) ∩ w−1 (Φ± r Φ±p ) = ∅, we have

(Φ+p r ∆ p ) ∩ w−1 (Φ± ) = (Φ+p r ∆ p ) ∩ w−1 (Φ±p ) .

Therefore,
−1
X 1 X ζ(2)−|∆ p ∩w (∆ p )|
=
b
C p,w D p,w
w∈W
hλ p , α∨w i w∈W
hλ p , w−1 (α∨p )i
p p
|w−1 (∆)rΦ p |=1 ∆ p ⊂w−1 (∆ p ∪Φ−p )
Y 1 Y Y
× ζ(hρ∨ , α∨ i+1)
b ζ(hρ∨ , αi)
b
hρ p , α∨ i−1
α∈w−1 (∆ p )r∆ p α∈(Φ+p r∆ p )∩w−1 (Φ+p ) α∈(Φ+p r∆ p )∩w−1 (Φ−p )
X −1
Y 1
= ζ(2)−|∆ p ∩w (∆ p )|
b
hρ p , α∨ i − 1
w∈W p α∈w−1 (∆ p )r∆ p
∆ p ⊂w−1 (∆ p ∪Φ−p )
Y Y
× ζ(hρ∨ , α∨ i + 1)
b ζ(hρ∨ , αi).
b
α∈(Φ+p r∆ p )∩w−1 (Φ+p ) α∈(Φ+p r∆ p )∩w−1 (Φ−p )

Here, in the last step, we have used relations hλ p , w−1 (α∨p )i = hw(λ p ), α∨p i = 1.
This then proves the proposition by using (12.93). 
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 339

Algebraic and Analytic Structures and Weak RH 339

12.A Decomposition of Σ p(1)

We use the same numbering as in [49]. We write α∨ = ni=1 ai α∨i as the sequence
P
a1 , . . . , an with a p = 1. Then, same as in [56], we have:
Type An
By the symmetry of root system of An type, it suffices to consider the case
Pj
when 1 ≤ p ≤ [n/2] + 1. Let α∨i, j := k=i αk . Then for 1 ≤ j ≤ p, we have

n o
L j (1)∨ = α∨j, n , α∨j, n−1 , . . . , α∨j, p+ j−1 , α∨j+1, p+ j−1 , . . . , α∨p, p+ j−1 . (12.96)

Example 12.2. When n = 6, we have:

• p=1
L1 (1)∨ : 111111 111110 111100 111000 110000 100000
• p=2
L1 (1)∨ : 111111 111110 111100 111000 110000 010000
L2 (1)∨ : 011111 011110 011100 011000
• p=3
L1 (1)∨ : 111111 111110 111100 111000 011000 001000
L2 (1)∨ : 011111 011110 011100 001100
L3 (1)∨ : 001111 001110
• p=4
L1 (1)∨ : 111111 111110 111100 011100 001100 000100
L2 (1)∨ : 011111 011110 001110 000110
L3 (1)∨ : 001111 000111
• p=5
L1 (1)∨ : 111111 111110 011110 001110 000110 000010
L2 (1)∨ : 011111 001111 000111 000011
• p=6
L1 (1)∨ : 111111 011111 001111 000111 000011 000001

Type Bn
Let
j
X j
X k
X
α∨i, j := α∨i and β∨i, j, k := α∨i + 2 α∨i .
l=i l=i l= j+1
Then, for 1 ≤ j ≤ min{p, n − p + 1}, let
n
L j (1)∨ = β∨j, p, n , β∨j, p+1, n , . . . , β∨j, n−1, n ,
o
α∨j, n , α∨j, n−1 , . . . , α∨j, p+ j−1 , α∨j+1, p+ j−1 , . . . , α∨p, p+ j−1 .
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 340

340 Algebraic, Analytic Structures and Riemann Hypothesis

Moreover, if n − p + 1 < p ≤ n, then, for n − p + 1 < j ≤ min{p, 2n − 2p + 1},

n o
L j (1)∨ = β∨j, p, n , β∨j, p+1, n , . . . , β∨j, 2n−p− j+1, n , β∨j+1, 2n−p− j+1, n , . . . , β∨p, 2n−p− j+1, n .

Example 12.3. When n = 6, we have:

• p=1
L1 (1)∨ : 122222 112222 111222 111122 111112 111111 111110 111100
111000 110000 100000
• p=2
L1 (1)∨ : 112222 111222 111122 111112 111111 111110 111100 111000
110000 010000
L2 (1)∨ : 012222 011222 011122 011112 011111 011110 011100 011000
• p=3
L1 (1)∨ : 111222 111122 111112 111111 111110 111100 111000 011000
001000
L2 (1)∨ : 011222 011122 011112 011111 011110 011100 001100
L3 (1)∨ : 001222 001122 001112 001111 001110
• p=4
L1 (1)∨ : 111122 111112 111111 111110 111100 011100 001100 000100
L2 (1)∨ : 011122 011112 011111 011110 001110 000110
L3 (1)∨ : 001122 001112 001111 000111
L4 (1)∨ : 000122 000112
• p=5
L1 (1)∨ : 111112 111111 111110 011110 001110 000110 000010
L2 (1)∨ : 011112 011111 001111 000111 000011
L3 (1)∨ : 001112 000112 000012
• p=6
L1 (1)∨ : 111111 011111 001111 000111 000011 000001

Type Cn
j
X j
X n−1
X
Let α∨i, j := α∨i and β∨i, j := α∨i + 2 α∨i + α∨n .
l=i l=i l= j+1
Then, for 1 ≤ j ≤ min{p, n − p + 1},
n
L j (1)∨ = β∨j, p , β∨j, p+1 , . . . , β∨j, n−2 ,
o
α∨j, n , α∨j, n−1 , . . . , α∨j, p+ j−1 , α∨j+1, p+ j−1 , . . . , α∨p, p+ j−1 .
Moreover, if n − p + 1 < p ≤ n − 1, then, for n − p + 1 < j ≤ min{p, 2n − 2p},
n o
L j (1)∨ = β∨j, p , β∨j, p+1 , . . . , β∨j, 2n−p− j , β∨j+1, 2n−p− j , . . . , β∨p, 2n−p− j .
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 341

Algebraic and Analytic Structures and Weak RH 341

Finally, in the case p = n, for 1 ≤ j ≤ (n + 1)/2,

n o
L j (1)∨ = β∨j, j−1 , β∨j, j , β∨j, j+1 . . . , β∨j, n− j , β∨j+1, n− j , . . . , β∨n− j+1, n− j .

Example 12.4. When n = 6, we have:

• p=1
L1 (1)∨ : 122221 112221 111221 111121 111111 111110 111100 111000
110000 100000
• p=2
L1 (1)∨ : 112221 111221 111121 111111 111110 111100 111000 110000
010000
L2 (1)∨ : 012221 011221 011121 011111 011110 011100 011000
• p=3
L1 (1)∨ : 111221 111121 111111 111110 111100 111000 011000 001000
L2 (1)∨ : 011221 011121 011111 011110 011100 001100
L3 (1)∨ : 001221 001121 001111 001110
• p=4
L1 (1)∨ : 111121 111111 111110 111100 011100 001100 000100
L2 (1)∨ : 011121 011111 011110 001110 000110
L3 (1)∨ : 001121 001111 000111
L4 (1)∨ : 000121
• p=5
L1 (1)∨ : 111111 111110 011110 001110 000110 000010
L2 (1)∨ : 011111 001111 000111 000011
• p=6
L1 (1)∨ : 222221 122221 112221 111221 111121 111111 011111 001111
000111 000011 000001
L2 (1)∨ : 022221 012221 011221 011121 001121 000121 000021
L3 (1)∨ : 002221 001221 000221

Type Dn
Let
j
X j
X n−2
X n−2
X
α∨i, j := α∨i , β∨i, j := α∨i + 2 α∨i + α∨n−1 + α∨n , γi∨ := α∨i + α∨n .
l=i l=i l= j+1 l=i
Then, in the case 1 ≤ p ≤ n − 2, for 1 ≤ j ≤ min{p, n − p + 1},
n o
L j (1)∨ = β∨j, p , β∨j, p+1 , . . . , β∨j, n−2 , α∨j, n−1 , . . . , α∨j, p+ j−1 , α∨j+1, p+ j−1 , . . . , α∨p, p+ j−1 .
Moreover, if n − p + 1 < p ≤ n − 2, then, for n − p + 1 < j ≤ min{p, 2n − 2p − 1},
n o
L j (1)∨ = β∨j, p , β∨j, p+1 , . . . , β∨j, 2n−p− j−1 , β∨j+1, 2n−p− j−1 , . . . , β∨p, 2n−p− j−1 .
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 342

342 Algebraic, Analytic Structures and Riemann Hypothesis

In addition, for j = min{p + 1, 2n − 2p},

n o
L j (1)∨ = γ1∨ , γ2∨ , . . . , γ∨p .
In the case p = n − 1,
n o
L1 (1)∨ = β∨1, 1 , β∨1, 2 , . . . , β∨1, n−2 , α∨1, n−1 , α∨2, n−1 , . . . , α∨n−1, n−1
and for 2 ≤ j ≤ n/2,
n o
L j (1)∨ = β∨j, j , β∨j, j+1 , . . . , β∨j, n− j , β∨j, n− j , α∨j+1, n− j , . . . , β∨n− j, n− j .
Finally, in the case p = n, for 1 ≤ j ≤ n/2, L j (1)∨ is the same as that of
p = n − 1 but with αn−1 replaced by αn .

Example 12.5. When n = 6, we have:

• p=1
L1 (1)∨ : 122211 112211 111211 111111 111110 111100 111000 110000
100000
L2 (1)∨ : 111101
• p=2
L1 (1)∨ : 112211 111211 111111 111110 111100 111000 110000 010000
L2 (1)∨ : 012211 011211 011111 011110 011100 011000
L3 (1)∨ : 111101 011101
• p=3
L1 (1)∨ : 111211 111111 111110 111100 111000 011000 001000
L2 (1)∨ : 011211 011111 011110 011100 001100
L3 (1)∨ : 001211 001111 001110
L4 (1)∨ : 111101 011101 001101
• p=4
L1 (1)∨ : 111111 111110 111100 011100 001100 000100
L2 (1)∨ : 011111 011110 001110 000110
L3 (1)∨ : 001111 000111
L4 (1)∨ : 111101 011101 001101 000101
• p=5
L1 (1)∨ : 122211 112211 111211 111111 111110 011110 001110 000110
000010
L2 (1)∨ : 012211 011211 011111 001111 000111
L3 (1)∨ : 001211
• p=6
L1 (1)∨ : 122211 112211 111211 111111 111101 011101 001101 000101
000001
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 343

Algebraic and Analytic Structures and Weak RH 343

L2 (1)∨ : 012211 011211 011111 001111 000111

L3 (1)∨ : 001211

Type E6

• p=1
L1 (1)∨ : 123212 123211 122211 122111 122101 112101 111101 111001
111000 110000 100000
L2 (1)∨ : 112211 112111 111111 111110 111100
• p=2
L1 (1)∨ : 112211 112111 111111 111110 111100 111000 110000 010000
L2 (1)∨ : 012211 012111 012101 011101 011100 011000
L3 (1)∨ : 112101 111101 111001 011001
L4 (1)∨ : 011111 011110
• p=3
L1 (1)∨ : 111111 111110 111100 111000 011000 001000
L2 (1)∨ : 111101 111001 011001 001001
L3 (1)∨ : 011111 011110 011100 001100
L4 (1)∨ : 011101 001101
L5 (1)∨ : 001111 001110
• p=4
L1 (1)∨ : 122111 122101 112101 111101 111100 011100 001100 000100
L2 (1)∨ : 112111 111111 111110 011110 001110 000110
L3 (1)∨ : 012111 012101 011101 001101
L4 (1)∨ : 011111 001111
• p=5
L1 (1)∨ : 123212 123211 122211 122111 112111 111111 111110 011110
001110 000110 000010
L2 (1)∨ : 112211 012211 012111 011111 001111
• p=6
L1 (1)∨ : 123211 122211 122111 122101 112101 111101 111001 011001
001001 000001
L2 (1)∨ : 112211 112111 111111 011111 001111 001101
L3 (1)∨ : 012211 012111 012101 011101

Type E7

• p=1
L1 (1)∨ : 1343212 1243212 1233212 1233211 1232211 1222211 1122211
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 344

344 Algebraic, Analytic Structures and Riemann Hypothesis

1122111 1121111 1111111 1111110 1111100 1111000 1110000 1100000

1000000
L2 (1)∨ : 1232212 1232112 1232102 1232101 1222101 1122101 1121101
1111101 1111001 1110001
L2 (1)∨ : 1232111 1222111 1221111 1221101 1221001 1121001
• p=2
L1 (1)∨ : 1122211 1122111 1121111 1111111 1111110 1111100 1111000
1110000 1100000 0100000
L2 (1)∨ : 0122211 0122111 0121111 0111111 0111110 0111100 0111000
0110000
L3 (1)∨ : 1122101 0122101 0121101 0111101 0111001 0110001
L4 (1)∨ : 1121101 1111101 1111001 1110001
L5 (1)∨ : 1121101 1111101 1111001 1110001
• p=3
L1 (1)∨ : 1111111 1111110 1111100 1111000 1110000 0110000 0010000
L2 (1)∨ : 1111101 1111001 1110001 0110001 0010001
L3 (1)∨ : 0111111 0111110 0111100 0111000 0011000
L4 (1)∨ : 0111101 0111001 0011001
L5 (1)∨ : 0011111 0011110 0011100
L5 (1)∨ : 0011101
• p=4
L1 (1)∨ : 1221111 1221101 1221001 1121001 1111001 1111000 0111000
0011000 0001000
L2 (1)∨ : 1121111 1121101 1111101 0111101 0111100 0011100 0001100
L3 (1)∨ : 1111111 1111110 0111110 0011110 0001110
L4 (1)∨ : 0121111 0121101 0121001 0111001 0011001
L5 (1)∨ : 0111111 0011111 0011101
L6 (1)∨ : 1111100
• p=5
L1 (1)∨ : 1232112 1232102 1232101 1222101 1221101 1121101 1111101
1111100 0111100 0011100 0001100 0000100
L2 (1)∨ : 1232111 1222111 1221111 1121111 1111111 1111110 0111110
0011110 0001110 0000110
L3 (1)∨ : 1122111 1122101 0122101 0121101 0111101 0011101
L4 (1)∨ : 0122111 0121111 0111111 0011111
• p=6
L1 (1)∨ : 2343212 1343212 1243212 1233212 1232212 1232112 1232111
1222111 1221111 1121111 1111111 1111110 0111110 0011110 0001110
0000110 0000010
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 345

Algebraic and Analytic Structures and Weak RH 345

L2 (1)∨ : 1233211 1232211 1222211 1122211 1122111 0122111 0121111

0111111 0011111
L3 (1)∨ : 0122211
• p=7
L1 (1)∨ : 1233211 1232211 1232111 1232101 1222101 1122101 1121101
1111101 1111001 1110001 0110001 0010001 0000001
L2 (1)∨ : 1222211 1222111 1221111 1221101 1221001 1121001 0121001
0111001 0011001
L3 (1)∨ : 1122211 1122111 1121111 1111111 0111111 0011111 0011101
L4 (1)∨ : 0122211 0122111 0122101 0121101 0111101
L5 (1)∨ : 0121111

Type E8

• p=1
L1 (1)∨ : 13456423 12456423 12356423 12346423 12345423 12345422
12345322 12344322 12334322 12334312 12234312 12234212 12233212
12223212 12223211 12222211 12222111 12222101 11222101 11122101
11112101 11111101 11111001 11111000 11110000 11100000 11000000
10000000
L2 (1)∨ : 12345323 12345313 12345312 12344312 12344212 12334212
12333212 12333211 12233211 11233211 11223211 11222211 11222111
11122111 11112111 11111111 11111110 11111100
L3 (1)∨ : 12234322 11234322 11234312 11234212 11233212 11223212
11123212 11123211 11122211 11112211
• p=2
L1 (1)∨ : 11234322 11234312 11234212 11233212 11223212 11223211
11222211 11222111 11222101 11122101 11112101 11111101 11111001
11111000 11110000 11100000 11000000 01000000
L2 (1)∨ : 01234322 01234312 01234212 01233212 01223212 01123212
01123211 01122211 01112211 01112111 01111111 01111110 01111100
01111000 01110000 01100000
L3 (1)∨ : 11233211 01233211 01223211 01222211 01222111 01222101
01122101 01112101 01111101 01111001
L4 (1)∨ : 11123212 11123211 11122211 11112211 11112111 11111111
11111110 11111100
L5 (1)∨ : 11122111 01122111
• p=3
L1 (1)∨ : 11123212 11123211 11122211 11122111 11122101 11112101
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 346

346 Algebraic, Analytic Structures and Riemann Hypothesis

11111101 11111001 11111000 11110000 11100000 01100000 00100000

L2 (1)∨ : 01123212 00123212 00123211 00122211 00112211 00112111
00111111 00111101 00111001 00111000 00110000
L3 (1)∨ : 01123211 01122211 01122111 01122101 01112101 01111101
01111001 01111000 01110000
L4 (1)∨ : 11112211 11112111 11111111 11111110 11111100 01111100
00111100
L5 (1)∨ : 01112211 01112111 01111111 01111110 00111110
L6 (1)∨ : 00122111 00122101 00112101
• p=4
L1 (1)∨ : 11112211 11112111 11112101 11111101 11111001 11111000
11110000 01110000 00110000 00010000
L2 (1)∨ : 01112211 00112211 00012211 00012111 00012101 00011101
00011001 00011000
L3 (1)∨ : 11111111 11111110 11111100 01111100 01111000 00111000
L4 (1)∨ : 01112111 01111111 01111110 00111110 00111100 00011100
L5 (1)∨ : 01112101 00112101 00111101 00111001
L6 (1)∨ : 00112111 00111111 00011111 00011110
L7 (1)∨ : 01111101 01111001
• p=5
L1 (1)∨ : 11111111 11111110 11111100 11111000 01111000 00111000
00011000 00001000
L2 (1)∨ : 11111101 11111001 01111001 00111001 00011001 00001001
L3 (1)∨ : 01111111 01111110 01111100 00111100 00011100 00001100
L4 (1)∨ : 00111111 00111110 00011110 00001110
L5 (1)∨ : 01111101 00111101 00011101 00001101
L6 (1)∨ : 00011111 00001111
• p=6
L1 (1)∨ : 12222111 12222101 11222101 11122101 11112101 11111101
11111100 01111100 00111100 00011100 00001100 00000100
L2 (1)∨ : 11222111 11122111 11112111 11111111 11111110 01111110
00111110 00011110 00001110 00000110
L3 (1)∨ : 01222111 01222101 01122101 00122101 00112101 00012101
00011101 00001101
L4 (1)∨ : 01122111 00122111 00112111 00012111 00011111 00001111
L5 (1)∨ : 01112111 01112101 01111101 00111101
L6 (1)∨ : 01111111 00111111
• p=7
L1 (1)∨ : 12345313 12345312 12344312 12344212 12334212 12333212
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 347

Algebraic and Analytic Structures and Weak RH 347

12333211 12233211 12223211 12222211 12222111 11222111 11122111

11112111 11111111 11111110 01111110 00111110 00011110 00001110
00000110 00000010
L2 (1)∨ : 12334312 12234312 12234212 12233212 12223212 11223212
11123212 11123211 11122211 11112211 01112211 01112111 01111111
00111111 00011111 00001111
L3 (1)∨ : 11234312 11234212 11233212 11233211 11223211 11222211
01222211 01222111 01122111 00122111 00112111 00012111
L4 (1)∨ : 01234312 01234212 01233212 01233211 01223211 01123211
01122211 00122211 00112211 00012211
L5 (1)∨ : 01223212 01123212 00123212 00123211
• p=8
L1 (1)∨ : 12333211 12233211 12223211 12222211 12222111 12222101
11222101 11122101 11112101 11111101 11111001 01111001 00111001
00011001 00001001 00000001
L2 (1)∨ : 11233211 01233211 01223211 01123211 00123211 00122211
00112211 00012211 00012111 00011111 00001111 00001101
L3 (1)∨ : 11223211 11222211 11222111 11122111 11112111 11111111
01111111 00111111 00111101 00011101
L4 (1)∨ : 11123211 11122211 11112211 01112211 01112111 00112111
00112101 00012101
L5 (1)∨ : 01222211 01222111 01222101 01122101 01112101 01111101
L6 (1)∨ : 01122211 01122111 00122111 00122101

Type F4

• p=1
L1 (1)∨ : 1342 1242 1232 1231 1221 1220 1120 1110 1100 1000
L2 (1)∨ : 1222 1122 1121 1111
• p=2
L1 (1)∨ : 1122 1121 1120 1110 1100 0100
L2 (1)∨ : 0122 0121 0120 0110
L3 (1)∨ : 1111 0111
• p=3
L1 (1)∨ : 1111 1110 0110 0010
L2 (1)∨ : 0111 0011
L3 (1)∨ :
• p=4
L1 (1)∨ : 1231 1221 1121 1111 0111 0011 0001
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 348

348 Algebraic, Analytic Structures and Riemann Hypothesis

L2 (1)∨ : 0121

Type G2

• p=1
L1 (1)∨ : 13 12 11 10
• p=2
L1 (1)∨ : 11 01

12.B Geometric Interpretation of c p

For simplicity, we work over C. Let G be a reductive group over C with P its
maximal parabolic subgroup. Then the quotient G/P is proper. Denote by λ p the
fundamental dominant weight associated to P, TXP the tangent bundle of G/P, and
Pic(G/P) the Picard group of G/P. Then by [102],
Pic(G/P) ' Zλ p and det(TX/P ) = c p Lλ p , (12.97)
where LλP denotes the ample line bundle generating Pic(G/P).
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 349

Chapter 13

Weak Riemann Hypothesis for

Zeta Functions of Exceptional Groups

As we have seen in Theorem 8.6, the zeta functions for the exceptional group
G2 satisfy the Riemann hypothesis. The same method can also be applied to the
exceptional group F4 . In this chapter, we follow our work [133] to prove that all
the zeta functions for the exceptional groups of type E, namely E6 , E7 and E8 ,
satisfy a weak Riemann hypothesis.

13.1 Weyl Groups

13.1.1 Root Systems

Following the structures of the associated Weyl groups, to realize the set of simple
roots for the exceptional group of types En (n = 6, 7, 8), we choose a basis of R8
as follows.
α1 = (0, 0, 0, −1, 1, 0, 0, 0), α2 = (0, 0, −1, 1, 0, 0, 0, 0),
α3 = (0, −1, 1, 0, 0, 0, 0, 0), α4 = (1, 1, 0, 0, 0, 0, 0, 0),
1 (13.1)
α5 = (−1, 1, 0, 0, 0, 0, 0, 0), α6 = (1, −1, −1, −1, −1, −1, −1, 1),
2
α7 = (0, 0, 0, 0, −1, 1, 0, 0), α8 = (0, 0, 0, 0, 0, −1, 1, 0).
Accordingly, introduce the finite sets
∆6 = {α1 , α2 , α3 , α4 , α5 , α6 },
∆7 = {α1 , α2 , α3 , α4 , α5 , α6 , α7 }, (13.2)
∆8 = {α1 , α2 , α3 , α4 , α5 , α6 , α7 , α8 }.
It is not too difficult to calculate their related Cartan matrices and hence to confirm
that ∆n is indeed the set of the simple roots for the root system Φn associated to
the exceptional group En (n = 6, 7, 8), with the Dynkin diagrams given by

349
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 350

350 Algebraic, Analytic Structures and Riemann Hypothesis

α4

E6 (13.3)
α1 α2 α3 α5 α6

α4

E7 (13.4)
α7 α1 α2 α3 α5 α6

α4

E8 (13.5)
α8 α7 α1 α2 α3 α5 α6

Fig. 13.1: The Dynkin Diagrams for E6 , E7 , E8

Note that the numbering here for the simple roots are very different from some
standard numbering used in [16] and [44]. We will use this numbering in this
chapter, for the purpose of writing down the Weyl groups of En more effectively.

13.1.2 Weyl Group for D5

Let Wn be the Weyl groups of the exceptional group of type En (n = 6, 7, 8).

Induced from the natural inclusions Φ6 ⊂ Φ7 ⊂ Φ8 of the roos systems, we obtain
a filtration W6 ≤ W7 ≤ W8 of the associated Weyl groups. If, in addition, we set
∆5 := {α1 , α2 , α3 , α4 , α5 }, then ∆5 becomes the set of simple roots for the group of
type D5 in the D-series with the Dynkin diagram

α4

D5 (13.6)
α1 α2 α3 α5

Fig. 13.2: The Dynkin Diagrams for D5

Denote its associated roots system and the Weyl group by Φ5 and W5 , respec-
tively. We obtain two chains

Φ5 ⊂ Φ6 ⊂ Φ7 ⊂ Φ8 and W5 < W6 < W7 < W8 . (13.7)

December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 351

Weak RH for Zeta Functions of Exceptional Groups 351

Here, as usual, A < B means that A is a proper substructure of B. It is well known

that W5 admits a semi-product structure S5 o Signev , where S5 denotes the 5-th
symmetric group. Here, for any σ ∈ S5 ,

(x1 , x2 , x3 , x4 , x5 , x6 , x7 , x8 ) 7→ (xσ(1) , xσ(2) , xσ(3) , xσ(4) , xσ(4) , x6 , x7 , x8 ), (13.8)

and Signev consists of even sign changes for the first 5 coordinates. In particular,

W5 = S5 × 2(0)+(2)+(4) = 1920.
5 5 5

13.1.3 Weyl Group for E6

From now on, we will view all the groups Wn , n = 5, 6, 7, 8 as subgroups of

AutR (R8 ). Thus, to obtain the group W6 , we only need to find out what are the

representatives of the cosets in W6 W5 . For this, we then introduce the set of roots
Φ+6,5 := α ∈ Φ+6 : α < Φ+5 . It consists of 16 elements. Namely,


(− 2 , − 2 , − 2 , − 2 , 2 , − 2 , − 2 , 2 ), (− 12 , − 21 , − 12 , 12 , − 12 , − 12 , − 21 , 12 ),
 1 1 1 1 1 1 1 1 


 

 
1
, ,
1 1
, 1
, 1
, 1
, ,
1 1 1
, ,
1 1 1 1
, , , 1
, ,
1 1
 




 (− 2 − 2 2 − 2 − 2 − 2 − 2 2 ), (− 2 − 2 2 2 2 − 2 − 2 2 ), 




 
,
1 1
, 1
, 1
, 1
, 1
, ,
1 1
,
1 1
, ,
1 1 1
, , 1
, ,
1 1
 




 (− 2 2 − 2 − 2 − 2 − 2 − 2 2 ), (− 2 2 − 2 2 2 − 2 − 2 2 ), 




 
,
1 1 1
, , ,
1 1
, 1
, ,
1 1
,
1 1 1 1
, , , 1
, 1
, ,
1 1
 
+

 (− − − − ), (− − − − ), 

Φ6,5 =  2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2

 

( 2 , − 2 , − 2 , − 2 , − 2 , − 2 , − 2 , 2 ), ( 2 , − 2 , − 2 , 2 , 2 , − 2 , − 2 , 2 ), 
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

 



 


 
1
, ,
1 1
, ,
1 1
, 1
, ,
1 1 1
, ,
1 1 1
, , 1
, 1
, ,
1 1
 




 ( 2 − 2 2 − 2 2 − 2 − 2 2 ), ( 2 − 2 2 2 − 2 − 2 − 2 2 ), 




 
,
1 1
, 1
, ,
1 1
, 1
, ,
1 1
,
1 1
, ,
1 1
, 1
, 1
, ,
1 1
 




 ( 2 2 − 2 − 2 2 − 2 − 2 2 ), ( 2 2 − 2 2 − 2 − 2 − 2 2 ), 





 ( 1 , 1 , 1 , − 1 , − 1 , − 1 , − 1 , 1 ), 
, ,
1 1 1 1 1
, , , 1
, ,
1 1
 
2 2 2 2 2 2 2 2 ( 2 2 2 2 2 − 2 − 2 2 ) 
(13.9)
which we denote by {α6,1 , α6,2 , . . . , α6,16 }, in the order of 1st left-1st right-2nd
left-2nd right, .... Denote by σ6,i (1 ≤ i ≤ 16) the reflections associated to α6,i .
Our first result, jointly with Katayama ([51]), is obtained using Mathematica.

Lemma 13.1. A set of complete representatives for W6 W5 can be chosen to be
Id, σ6,1 , σ6,2 , σ6,3 , σ6,4 , σ6,5 , σ6,6 , σ6,7 , σ6,8 , σ6,9 , σ6,10 , 
 


 

 
σ , σ , σ , σ , σ , σ ,

 

 6,11 6,12 6,13 6,14 6,15 6,16
 
Σ6 =  .
 
(13.10)

σ σ , σ σ , σ σ , σ σ , σ σ ,
 

6,1 6,8 6,1 6,12 6,1 6,14 6,1 6,15 6,1 6,16


 



 

σ σ , σ σ , σ σ , σ σ , σ σ
 

6,2 6,15 6,2 6,16 6,3 6,16 6,5 6,16 6,9 6,16

As to be expected, it consists of 27 elements, since W6 = 27×1920 = 51, 840.

Based on this calculation, we have calculated the zeta functions associated to the
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 352

352 Algebraic, Analytic Structures and Riemann Hypothesis

pair (E6 , P1 ), where P1 denotes the maximal parabolic subgroup of E7 correspond-

ing to the subset {α1 } ⊂ ∆6 . For later use, we renumber the elements of Σ6 by
setting

Σ6 := {τ6,i 0 ≤ i =≤ 26}, (13.11)

where τ6,i stands for the (i + 1)-st element in the display (13.10) for the set Σ6 .
Our calculations, using Mathematica, is not very difficult. On the other hand,
the choices we made in §§ 13.1.1 and 13.1.2 for the orders of the simple roots
proves to be rather crucial.

13.1.4 Weyl Group for E7

To construct the Weyl group W7 for the exceptional group of type E7 , we use
the Weyl group W6 of type E6 as presented in the previous lemma. Similarly, by
(13.7), W6 is a subgroup of W7 , it suffices to find the 56 complete representatives

for the cosets in W7 W6 .
For this, we first introduce a subset Φ+7,6 := {α ∈ Φ+7 : α < Φ+6 } consisting of
27 positive roots in Φ7 which do not belong to Φ6 . Namely,

(− 21 , − 12 , − 12 , − 12 , − 21 , 12 , − 21 , 12 ),
 


 (−1, 0, 0, 0, 0, 1, 0, 0), 

 
1
, 1
, ,
1 1 1 1
, , , ,
1 1 1
, ,
1 1
, ,
1 1 1
, , ,
1 1
 




 (− 2 − 2 − 2 2 2 2 − 2 2 ), (− 2 − 2 2 − 2 2 2 − 2 2 ), 




 
1
, ,
1 1 1
, , ,
1 1
, ,
1 1
,
1 1
, 1
, ,
1 1 1
, , ,
1 1
 




 (− 2 − 2 2 2 − 2 2 − 2 2 ), (− 2 2 − 2 − 2 2 2 − 2 2 ), 




 
,
1 1
, ,
1 1
, ,
1 1
, ,
1 1
,
1 1 1
, , 1
, ,
1 1
, ,
1 1
 




 (− 2 2 − 2 2 − 2 2 − 2 2 ), (− 2 2 2 − 2 − 2 2 − 2 2 ), 




 
, ,
1 1 1 1 1 1
, , , , ,
1 1
 




 (− 2 2 2 2 2 2 − 2 2 ), (0, −1, 0, 0, 0, 1, 0, 0), 





 





 (0, 0, −1, 0, 0, 1, 0, 0), (0, 0, 0, −1, 0, 1, 0, 0), 





 

+

 (0, 0, 0, 0, −1, 1, 0, 0), (0, 0, 0, 0, 0, 0, −1, 1), 

Φ7,6 =  . (13.12)

 






 (0, 0, 0, 0, 1, 1, 0, 0), (0, 0, 0, 1, 0, 1, 0, 0), 





 





 (0, 0, 1, 0, 0, 1, 0, 0), (0, 1, 0, 0, 0, 1, 0, 0), 




 
1
, 1
, 1
, ,
1 1 1
, , ,
1 1 1
, 1
, ,
1 1
, ,
1 1
, ,
1 1
 




 ( 2 − 2 − 2 − 2 2 2 − 2 2 ), ( 2 − 2 − 2 2 − 2 2 − 2 2 ), 




 
1
, 1
, 1
, 1
, 1
, 1
, 1
, 1 1
, 1
, 1
, 1
, 1
, 1
, 1
, 1
 




 ( 2 − 2 2 − 2 − 2 2 − 2 2 ), ( 2 − 2 2 2 2 2 − 2 2 ), 




 
1
, 1
, 1
, 1
, 1
, 1
, 1
, 1 1
, 1
, 1
, 1
, 1
, 1
, 1
, 1
 




 ( 2 2 − 2 − 2 − 2 2 − 2 2 ), ( 2 2 − 2 2 2 2 − 2 2 ), 




 
1
, 1
, 1
, 1
, 1
, 1
, 1
, 1 1
, 1
, 1
, 1
, 1
, 1
, 1
, 1
 




 ( 2 2 2 − 2 2 2 − 2 2 ), ( 2 2 2 2 − 2 2 − 2 2 ), 





 

 (1, 0, 0, 0, 0, 1, 0, 0) 

We denote them, with above displayed order, by {α7,1 , α7,2 , . . . , α7,27 }, and denote
by σ7, j the reflection associated to α7, j (1 ≤ j ≤ 27).
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 353

Weak RH for Zeta Functions of Exceptional Groups 353


Proposition 13.1. A set of complete representatives for W7 W6 can be chosen to
be
Id, σ7,1 , σ7,2 , σ7,3 , σ7,4 , σ7,5 , σ7,6 , σ7,7 , σ7,8 , σ7,9 , σ7,10 ,
 


 



 

σ , σ , σ , σ , σ , σ , σ , σ , σ , σ ,

 

7,11 7,12 7,13 7,14 7,15 7,16 7,17 7,18 7,19 7,20


 



 

σ , σ , σ , σ , σ , σ , σ ,


 





 7,21 7,22 7,23 7,24 7,25 7,26 7,27 



 
σ7,1 σ7,14 , σ7,1 σ7,19 , σ7,1 σ7,20 , σ7,1 σ7,21 , σ7,1 σ7,22 ,


 




 



 

Σ7 :=  σ σ , σ σ , σ σ , σ σ , σ σ , .

 

 7,1 7,23 7,1 7,24 7,1 7,25 7,1 7,26 7,1 7,27 


 

σ7,2 σ7,22 , σ7,2 σ7,24 , σ7,2 σ7,25 , σ7,2 σ7,26 , σ7,2 σ7,27 ,


 




 



 

σ σ , σ σ , σ σ ,


 




 7,3 7,25 7,3 7,26 7,3 7,27 



 

σ σ , σ σ , σ σ , σ σ , σ σ ,

 

7,4 7,26 7,4 7,27 7,5 7,27 7,6 7,26 7,6 7,27


 




 


σ σ , σ σ , σ σ , σ σ , σ σ σ
 

7,7 7,27 7,8 7,27 7,9 7,27 7,14 7,27 7,1 7,14 7,27
(13.13)

With | W7 | = 56×|W6 | = 2, 903, 040, to verify this proposition, the calculations

for the data involved using Mathematica are still manageable. In fact, the whole
process of finding W7 from W6 can be completed in one night and all data can
even be stored as a single unit in our MacBook Power. Here, as above, the key to
our success is certainly the special orders for the simple roots made in §§ 13.1.1
and 13.1.2.
Similar to the case of Σ6 , for later use, we denote Σ7 by
n o
Σ7 := τ7, j : 0 ≤ j ≤ 55 , (13.14)
where τ7, j denotes the ( j + 1)-st elements in the previous display for Σ7 .

13.1.5 Weyl Group for E8

To construct W8 , as above, we use the inclusion W7 < W8 . However, this proves

to be extremely challenging, because W8 consists of 696, 729, 600 elements. Even
in terms of representatives τ8,k of W8 W7 , there are 240 elements. This time, our

Power Mac is not powerful enough due to the limited storage. To solve this, within
the decomposition W8 = t26,56,239
i, j,k=0 W5 τ6,i τ7, j τ8,k , we regroup it according to the in-
dex i and obtain hence 27 independent pieces. With our program in Mathematica,
on a Mac Power of 6 kernels, it took a couple of weeks to find the answer.
In more details, we first calculate the subset Φ+8,7 of positive root defined by
Φ+8,7 := {α ∈ Φ+8 : α < Φ+7 }. (13.15)
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 354

354 Algebraic, Analytic Structures and Riemann Hypothesis

It consists of 57 elements. Namely,

 



 (−1, 0, 0, 0, 0, 0, 0, 1), (−1, 0, 0, 0, 0, 0, 1, 0),  


1
, 1
, 1
, 1
, 1
, ,
1 1 1
, 1
, 1
, 1
, , ,
1 1 1 1 1 
, ,
 
(− − − − − − ), (− − − − ),


 



 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 


1
, 1
, ,
1 1
, ,
1 1 1 1
, , 1
, 1
, ,
1 1 1
, , ,
1 1 1 
,





 (− 2 − 2 − 2 2 − 2 2 2 2 ), (− 2 − 2 − 2 2 2 − 2 2 2 ),




1
, ,
1 1
, 1
, ,
1 1 1 1
, , 1
, ,
1 1
, ,
1 1
, ,
1 1 1 
,
 
(− − − − ), (− − − − ),


 



 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2  

1
, 1
, 1
, 1
, 1
, 1
, 1
, 1 1
, 1
, 1
, 1
, 1
, 1
, 1
, 1
 




 (− 2 − 2 2 2 − 2 − 2 2 2 ), (− 2 − 2 2 2 2 2 2 2 ), 




, , , , , , , , , , , , , ,
 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

(− − − − ), (− − − − ),


 




 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 


,
1 1
, ,
1 1
, 1
, ,
1 1 1
, 1 1
, , , ,
1 1 1 1 1 1
, , ,
 




 (− 2 2 − 2 2 − 2 − 2 2 2 ), (− 2 2 − 2 2 2 2 2 2 ), 




(− 2 , 2 , 2 , − 2 , − 2 , − 2 , 2 , 2 ), (− 2 , 2 , 2 , − 2 , 2 , 2 , 2 , 2 ), 


 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1




 


 
1
, 1
, 1
, 1
, 1
, 1
, 1
, 1 1
, 1
, 1
, 1
, 1
, 1
, 1
, 1
 
(− − ), (− − ),

 

2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2


 



 





 (0, −1, 0, 0, 0, 0, 0, 1), (0, −1, 0, 0, 0, 0, 1, 0), 




 
(0, 0, −1, 0, 0, 0, 0, 1), (0, 0, −1, 0, 0, 0, 1, 0),


 




 


 




 (0, 0, 0, −1, 0, 0, 0, 1), (0, 0, 0, −1, 0, 0, 1, 0), 




 
(0, 0, 0, 0, −1, 0, 0, 1), (0, 0, 0, 0, −1, 0, 1, 0),


 




 


 




 (0, 0, 0, 0, 0, −1, 0, 1), (0, 0, 0, 0, 0, −1, 1, 0), 




Φ+8,7 = . (13.16)
 
(0, 0, 0, 0, 0, 0, 1, 1), (0, 0, 0, 0, 0, 1, 0, 1),
 

 


 




 (0, 0, 0, 0, 0, 1, 1, 0), (0, 0, 0, 0, 1, 0, 0, 1), 




 
(0, 0, 0, 0, 1, 0, 1, 0), (0, 0, 0, 1, 0, 0, 0, 1),


 




 


 
(0, 0, 0, 1, 0, 0, 1, 0), (0, 0, 1, 0, 0, 0, 0, 1),


 




 


 



 (0, 0, 1, 0, 0, 0, 1, 0), (0, 1, 0, 0, 0, 0, 0, 1), 



 
( 2 , − 2 , − 2 , − 2 , − 2 , 2 , 2 , 2 ),
1 1 1 1 1 1 1 1 
 
(0, 1, 0, 0, 0, 0, 1, 0),


 

 

, , , , , , , , , , , , , ,

 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1






 ( 2 − 2 − 2 − 2 2 − 2 2 2 ), ( 2 − 2 − 2 2 − 2 − 2 2 2  ),



1
, 1
, 1
, 1
, 1
, 1
, 1
, 1 1
, 1
, 1
, 1
, 1
, 1
, 1
, 1
 
( − − ), ( − − − − ),


 




 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2  

1
, ,
1 1
, , ,
1 1 1 1 1
, , 1
, ,
1 1 1
, , ,
1 1 1 1
, ,
 




 ( 2 − 2 2 − 2 2 2 2 2 ), ( 2 − 2 2 2 − 2 2 2 2 ), 




1
, 1
, 1
, 1
, 1
, 1
, 1
, 1 1
, 1
, 1
, 1
, 1
, 1
, 1
, 1
 
( − − ), ( − − − − ),


 




 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 


1
, 1
, 1
, 1
, 1
, 1
, 1
, 1 1
, 1
, 1
, 1
, 1
, 1
, 1
, 1
 




 ( 2 2 − 2 − 2 2 2 2 2 ), ( 2 2 − 2 2 − 2 2 2 2 ), 




, , , , , , , , , , , , , ,
 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

( − − ), ( − − ),


 




 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 


,
1 1 1
, , ,
1 1
, ,
1 1 1
, ,
1 1 1 1
, , , 1
, ,
1 1 1
,
 




 ( 2 2 2 − 2 2 − 2 2 2 ), ( 2 2 2 2 − 2 − 2 2 2 ), 




, , , , , , ,
 1 1 1 1 1 1 1 1

( ), (1, 0, 0, 0, 0, 0, 0, 1),


 




 2 2 2 2 2 2 2 2 


 
(1, 0, 0, 0, 0, 0, 1, 0)

 


We denote them, with the above displayed order, by α8,1 , α8,2 , . . . , α8,57 , and

denote by σ8,k the reflection associated to α8,k (1 ≤ k ≤ 57). This then implies our
first main result on the construction of W8 .
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 355

Weak RH for Zeta Functions of Exceptional Groups 355

Theorem 13.1. A set Σ8 of complete representatives for W8 W7 is given by



Id, σ8,1 , σ8,2 , σ8,3 , σ8,4 , σ8,5 , σ8,6 , σ8,7 , σ8,8 , σ8,9 , σ8,10 ,
 


 


σ , σ , σ , σ , σ , σ , σ , σ , σ , σ ,


 




 8,11 8,12 13 8,14 8,15 8,16 8,17 8,18 8,19 8,20 

σ , σ , σ , σ , σ , σ , σ , σ , σ , σ ,

 


8,21 8,22 8,23 8,24 8,25 8,26 8,27 8,28 8,29 8,30


 


σ , σ , σ , σ , σ , σ , σ , σ , σ , σ ,


 




 8,31 8,32 8,33 8,34 8,35 8,36 8,37 8,38 8,39 8,40 


σ , σ , σ , σ , σ , σ , σ , σ , σ , σ ,


 




 8,41 8,42 8,43 8,44 8,45 8,46 8,47 8,48 8,49 8,50 


σ , σ , σ , σ , σ , σ , σ ,

 

8,51 8,52 8,53 8,54 8,55 8,56 57


 


σ σ , σ σ , σ σ , σ σ , σ σ , σ σ ,


 




 8,1 8,20 8,1 8,22 8,1 8,24 8,1 8,26 8,1 8,28 8,1 8,29 


σ σ , σ σ , σ σ , σ σ , σ σ , σ σ ,


 




 8,1 8,31 8,1 8,33 8,1 8,35 8,1 8,37 8,1 8,39 8,1 8,40 


σ σ , σ σ , σ σ , σ σ , σ σ , σ σ ,

 

8,1 8,41 8,1 8,42 8,1 8,43 8,1 8,44 8,1 8,45 8,1 8,46


 


σ σ , σ σ , σ σ , σ σ , σ σ , σ σ ,


 




 8,1 8,47 8,1 8,48 8,1 8,49 8,1 8,50 8,1 8,51 8,1 8,52 


σ σ , σ σ , σ σ , σ σ , σ σ ,


 




 8,1 8,53 8,1 8,54 8,1 8,55 8,1 8,56 8,1 8,57 


σ σ , σ σ , σ σ , σ σ , σ σ , σ σ ,

 

8,2 8,29 8,2 8,40 8,2 8,41 8,2 8,42 8,2 8,43 8,2 8,44


 


σ σ , σ σ , σ σ , σ σ , σ σ , σ σ ,


 




 8,2 8,45 8,2 8,46 8,2 8,47 8,2 8,48 8,2 8,49 8,2 8,50 


σ σ , σ σ , σ σ , σ σ , σ σ , σ σ , σ σ ,


 




 8,2 8,51 8,2 8,52 8,2 8,53 8,2 8,54 8,2 8,55 8,2 8,56 8,2 8,57 


σ σ , σ σ , σ σ , σ σ , σ σ , σ σ , σ σ ,

 

8,3 8,29 8,3 8,43 8,3 8,45 8,3 8,46 8,3 8,47 8,3 8,49 8,3 8,50


 


σ8,3 σ8,51 , σ8,3 σ8,52 , σ8,3 σ8,53 , σ8,3 σ8,54 , σ8,3 σ8,55 , σ8,3 σ8,56 , σ8,3 σ8,57 ,


 




 


σ σ , σ σ , σ σ , σ σ , σ σ , σ σ ,


 




 8,4 8,29 8,4 8,46 8,4 8,47 8,4 8,50 8,4 8,51 8,4 8,52 


σ σ , σ σ , σ σ , σ σ , σ σ ,

 

8,4 8,53 8,4 8,54 8,4 8,55 8,4 8,56 8,4 8,57


 


 σ8,5 σ8,29 , σ8,5 σ8,47 , σ8,5 σ8,51 , σ8,5 σ8,52 , σ8,5 σ8,53 ,


 


.
 

σ8,5 σ8,54 , σ8,5 σ8,55 , σ8,5 σ8,56 , σ8,5 σ8,57 ,


 



 

σ8,6 σ8,29 , σ8,6 σ8,52 , σ8,6 σ8,53 , σ8,6 σ8,54 , σ8,6 σ8,55 , σ8,6 σ8,56 , σ8,6 σ8,57 ,


 



 

σ σ , σ σ , σ σ , σ σ , σ σ , σ σ , σ σ ,


 




 8,7 8,29 8,7 8,47 8,7 8,51 8,7 8,53 8,7 8,54 8,7 8,55 8,7 8,56 


σ σ , σ σ , σ σ , σ σ , σ σ , σ σ , σ σ ,
 
%

 



 8,8 8,29 8,8 8,50 8,8 8,54 8,8 8,55 8,8 8,56 8,8 8,57 8,7 8,57  

σ σ , σ σ , σ σ , σ σ , σ σ ,

 

8,9 8,29 8,9 8,49 8,9 8,55 8,9 8,56 8,9 8,57


 


σ8,10 σ8,29 , σ8,10 σ8,48 , σ8,10 σ8,56 , σ8,10 σ8,57 , & σ8,11 σ8,57 , 


 



 


σ σ , σ σ , σ σ , σ σ , σ σ , σ σ , σ σ ,


 




 8,11 8,29 8,11 8,47 8,11 8,51 8,11 8,53 8,11 8,54 8,11 8,55 8,11 8,56 


σ σ , σ σ , σ σ , σ σ , σ σ , σ σ ,

 

8,12 8,29 8,12 8,46 8,12 8,54 8,12 8,55 8,12 8,56 8,12 8,57


 


σ σ , σ σ , σ σ , σ σ , σ σ ,


 




 8,13 8,29 8,13 8,45 8,13 8,55 8,13 8,56 8,13 8,57 


σ σ , σ σ , σ σ , σ σ ,


 




 8,14 8,29 8,14 8,44 8,14 8,56 8,14 8,57 


σ σ , σ σ , σ σ , σ σ , σ σ ,

 

8,15 8,29 8,15 8,43 8,15 8,55 8,15 8,56 8,15 8,57


 


σ σ , σ σ , σ σ , σ σ ,


 




 8,16 8,29 8,16 8,42 8,16 8,56 8,16 8,57 


σ σ , σ σ , σ σ , σ σ ,


 




 8,17 8,29 8,17 8,41 8,17 8,56 8,17 8,57 


σ σ , σ σ , σ σ , σ σ ,


 



 8,18 8,29 8,18 8,40 8,18 8,56 8,18 8,57 

σ σ , σ σ , σ σ , σ σ , σ σ ,


 




 8,19 8,29 8,19 8,39 8,19 8,57 8,20 8,29 8,20 8,38 


σ σ , σ σ , σ σ , σ σ , σ σ , σ σ ,
 
&

 



 8,21 8,29 8,21 8,37 8,21 8,57 8,22 8,29 8,22 8,36 8,25 8,57 


σ σ , σ σ , σ σ , σ σ , σ σ , σ σ , σ σ ,


 



 8,23 8,29 8,23 8,35 8,23 8,57 8,24 8,29 8,24 8,34 8,25 8,29 8,25 8,33  
σ σ , σ σ , σ σ , σ σ , σ σ ,


 


8,26 8,29 8,26 8,32 8,27 8,29 8,27 8,31 8,27 8,57


 


 σ σ ,σ σ ,σ σ ,σ σ ,σ σ ,σ σ ,σ σ

 


8,28 8,29 8,28 8,30 8,30 8,57 8,32 8,57 8,34 8,57 8,36 8,57 8,38 8,57
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 356

356 Algebraic, Analytic Structures and Riemann Hypothesis

For later use, we rewrite Σ8 as n o

Σ8 = τ8,k : 0 ≤ k ≤ 239 , (13.17)
where τ8,k denotes the (k + 1)-st elements in the above display.

Theorem 13.2. With the same notation as above, the Weyl groups for the excep-
tional groups of types E6 , E7 , E8 are given respectively by
G26
W6 = τ6,i W5 ,
i=0
55 G
G 26
W7 = τ7, j τ6,i W5 , (13.18)
j=0 i=0

239 G
G 55 G
26
W8 = τ8,k τ7, j τ6,i W5 .
k=0 j=0 i=0

As mentioned above, for this theorem to be hold, the special orders for the
simple roots made in §13.1.1 and §13.1.2 play an extremely crucial role.
Our calculations of W8 in the winter 2015 and the spring of 2016 were rather
long and complicated, with hundreds bugs searching and programs fixing. In
addition, there was a huge problem about how to store the W8 data used in our
Mathematica programs with a Mac Power. This forced our work stations stoping
calculations repeatedly. At some point, we realized the main reason causing all
these stops was merely the lack of storage. Soon after, we offered a not-perfect-
but-practical solution. To be more precisely, instead, we introduced
G239 G
55
W8,i := τ8,k τ7, j τ6,i W5 , (0 ≤ i ≤ 26) (13.19)
k=0 j=0
and hence divided our calculations into 27 parallel units mentioned. Consequently,
for W8 , we were able to start our second phrase of calculations on zeta functions
themselves and on the verifications of the weak Riemann Hypothesis for these
Weng zetas associated to the exceptional groups of types E6 , E7 , E8 . This lasted
for another 2 months. One of the up-shot is the above theorem.

13.2 Weak Riemann Hypothesis for Zeta Functions of E n

13.2.1 Conditions for Weak Riemann Hypothesis

In this subsection, as a notational preparation, we recall the numerical condition

stated in Theorem 12.1 on the weak Riemann Hypothesis for the Weng zeta func-
tions bζQG/P (s).
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 357

Weak RH for Zeta Functions of Exceptional Groups 357

For each w ∈ W, we set

Aw := {α ∈ ∆P : wα ∈ ∆} and Bw := {α ∈ ∆P : wα ∈ Φ− }. (13.20)
Clearly, Aw t Bw ⊆ ∆P . Recall that w is called special, if Aw t Bw = ∆P . Introduce
WP0 := w ∈ W : ∆P ⊂ w−1 (∆ ∪ Φ− ) .

(13.21)
For each w ∈ WP0 , set
+

1 α ∈ w (Φ )
 −1
δα,w = ,

(13.22)
0 α ∈ w−1 (Φ− )


and define X
kP (w) := (1 − δα.w ) = (Φ+ r Φ+P ) ∩ w−1 (Φ− ) ,
α∈Φ+ rΦ+P (13.23)
W p := w ∈ WP : kP (w) = 0 .


Write ξ(s) := s(s − 1)b ζ(s) as usual, and set

|∆P ∩w−1 Φ+ |
Y
cP,w := ξ(2) ξ(hρ, α∨ i + δα,w ),
α∈Φ+P r∆P

1 −1
Φ+ |
Y
:= 2|∆P ∩w (hρ, α∨ i − 1) (13.24)
dP,w
α∈(w−1 ∆)∩(Φ P r∆P )
Y
× (hρ, α∨ i + δα,w )(hρ, α∨ i + δα,w − 1).
α∈ΦP r∆P

ζQG,P (s) is given by

Then the discriminant of zeta function b
X 1
∆G,P
Q := · cP,w · dP,w , (13.25)
w∈W
hλP , α∨w i
P
|(w−1 ∆)rΦP |=1

where αw is the only element of w−1 (∆) r ΦP . By Theorem 12.1, to show that


ζQG,P (s), it suffices to verify that
the weak Riemann Hypothesis holds for b
∆G,P
Q > 0. (13.26)
Set, accordingly,
−1
Φ+ |
Y 1 Y
ζ(2)|∆P ∩w
BP,w := b ζ(hρ, α∨ i + δα,w ).
b
hρ, α∨ i − 1 α∈Φ+ r∆
α∈(w−1 ∆)∩(ΦP r∆P ) P P

(13.27)
By (12.95), we have
X BP,w
∆G,P
Q = . (13.28)
w∈WP
hλP , α∨w i
|(w−1 ∆)rΦP |=1
Here λ1 , λ2 , . . . , λn is the dual bases of {α∨1 , α∨2 , . . . , α∨n } for n = 6, 7, 8.

December 26, 2017 9:48 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 358

358 Algebraic, Analytic Structures and Riemann Hypothesis

13.2.2 Special and Very Special Weyl Elements

To use the notations in our Mathematica program, set

sp
WP := w ∈ W : w special ,


WP† := w ∈ W sp : |∆ r wΦP | = 1 ,

(13.29)
WP‡ := w ∈ W † : |w(Φ+ r Φ+P ) ∩ Φ− | = 0 .


Then W sp = WP0 since

Aw t Bw = ∆P ⇐⇒ ∆P ⊂ w−1 (∆ t Φ− ). (13.30)

Moreover, for the group G of type En (n = 6, 7, 8), we have

WP† = w ∈ W sp : |∆ ∩ wΦP | = n − 1 .

(13.31)

To prove ∆QEn ,P , 0 for every n = 6, 7, 8 and for all the associated maximal
parabolic subgroups, we first search on the selections of Weyl elements belonging
to W sp , W † and W ‡ respectively. Accordingly, we carry out our calculations in
Mathematica programs into three steps.
sp
Step 1. Calculate special Weyl elements in WP , denoted as ‘sp’ in the program.
This takes very very long time.

Step 2. Calculate elements of WP† , denoted as ‘sps’ in the program. This takes
rather long time. Finally,

Step 3. Calculate elements of WP‡ , called very special in the program. This is very
pleasant as we have the following very beautiful

Proposition 13.2. For the exceptional group of type En (n = 6, 7, 8), we have

‡ ‡ ‡
W6,Pi
⊂ W7,Pi
and W7,P j
⊂ W8†j ⊂ W8‡i (1 ≤ i ≤ 6, 1 ≤ j ≤ 7). (13.32)

More importantly, we have

†
Wn,P = 2n and ‡
Wn,P = 2n−1 . (13.33)

Example 13.1. When (G, P) = (E8 , P1 ), for simplicity, we denote the Weyl ele-
ment w5,s τ6,i τ7, j τ8,k simply by (s + 1, i + 1, j + 1, k + 1), where w5,s ∈ W5 is with
the index s predesigned by our program. Then, from our Mathematica program
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 359

Weak RH for Zeta Functions of Exceptional Groups 359

named ddag.nb, we have

 


 (1, 1, 1, 1), (1, 1, 1, 27), (1, 1, 14, 27), (1, 1, 14, 29), 


 
(1375, 3, 1, 1), (1375, 3, 1, 27), (1375, 3, 14, 27), (1375, 3, 14, 29),


 




 


(1, 9, 1, 1), (1, 9, 1, 27), (1, 9, 14, 27), (1, 9, 14, 29),


 




 


 



 (193, 9, 1, 1), (193, 9, 1, 27), (193, 9, 14, 27), (193, 9, 14, 29), 




 





 (301, 9, 1, 1), (301, 9, 1, 27), (301, 9, 14, 27), (301, 9, 14, 29), 




 




 (325, 9, 1, 1), (325, 9, 1, 27), (325, 9, 14, 27), (325, 9, 14, 29), 




 




 (393, 9, 1, 1), (393, 9, 1, 27), (393, 9, 14, 27), (393, 9, 14, 29), 




 




 (423, 9, 1, 1), (423, 9, 1, 27), (423, 9, 14, 27), (423, 9, 14, 29), 




 
(441, 9, 1, 1), (441, 9, 1, 27), (441, 9, 14, 27), (441, 9, 14, 29),


 




 


(447, 9, 1, 1), (447, 9, 1, 27), (447, 9, 14, 27), (447, 9, 14, 29),


 




 


(663, 13, 1, 1), (663, 13, 1, 27), (663, 13, 14, 27), (663, 13, 14, 29),


 




 


(687, 13, 1, 1), (687, 13, 1, 27), (687, 13, 14, 27), (687, 13, 14, 29),


 




 


 



 (781, 13, 1, 1), (781, 13, 1, 27), (781, 13, 14, 27), (781, 13, 14, 29), 





 





 (805, 13, 1, 1), (805, 13, 1, 27), (805, 13, 14, 27), (805, 13, 14, 29), 




 





 (963, 15, 1, 1), (963, 15, 1, 27), (963, 15, 14, 27), (963, 15, 14, 29), 




 
 (1375, 16, 1, 1), (1375, 16, 1, 27), (1375, 16, 14, 27), (1375, 16, 14, 29),

W8,1 = 
‡
.
 






 (1, 23, 1, 1), (1, 23, 1, 27), (1, 23, 14, 27), (1, 23, 14, 29), 




 




 (361, 23, 1, 1), (361, 23, 1, 27), (361, 23, 14, 27), (361, 23, 14, 29), 




 
(687, 23, 1, 1), (687, 23, 1, 27), (687, 23, 14, 27), (687, 23, 14, 29),


 




 


(775, 23, 1, 1), (775, 23, 1, 27), (775, 23, 14, 27), (775, 23, 14, 29),


 




 


(961, 23, 1, 1), (961, 23, 1, 27), (961, 23, 14, 27), (961, 23, 14, 29),


 




 


(963, 23, 1, 1), (963, 23, 1, 27), (963, 23, 14, 27), (963, 23, 14, 29),


 




 


 



 (985, 23, 1, 1), (985, 23, 1, 27), (985, 23, 14, 27), (985, 23, 14, 29), 




 





 (987, 23, 1, 1), (987, 23, 1, 27), (987, 23, 14, 27), (987, 23, 14, 29), 




 




 (1321, 23, 1, 1), (1321, 23, 1, 27), (1321, 23, 14, 27), (1321, 23, 14, 29),




 




 (1323, 23, 1, 1), (1323, 23, 1, 27), (1323, 23, 14, 27), (1323, 23, 14, 29),




 




 (1327, 23, 1, 1), (1327, 23, 1, 27), (1327, 23, 14, 27), (1327, 23, 14, 29),




 
(1335, 23, 1, 1), (1335, 23, 1, 27), (1335, 23, 14, 27), (1335, 23, 14, 29),


 




 


(1345, 23, 1, 1), (1345, 23, 1, 27), (1345, 23, 14, 27), (1345, 23, 14, 29),


 




 


(1347, 23, 1, 1), (1347, 23, 1, 27), (1347, 23, 14, 27), (1347, 23, 14, 29),


 




 


(1375, 23, 1, 1), (1375, 23, 1, 27), (1375, 23, 14, 27), (1375, 23, 14, 29),


 




 

 (1407, 23, 1, 1), (1407, 23, 1, 27), (1407, 23, 14, 27), (1407, 23, 14, 29) 
 


Totally, there are 128 = 27 elements.

For other types of (En , P), please refer to the appendix where W ‡ is hidden in
Appendix C of [133].
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 360

360 Algebraic, Analytic Structures and Riemann Hypothesis

13.2.3 Constants hλ P , α∨w i

Very special Weyl elements are these which appeared in WP‡ giving non-trivial
contributions to ∆G,PQ . To calculate ∆Q for G = E 6,7,8 , we here calculate the con-
G,P

stant hλP , α∨w i. Here, for each very special Weyl element w, αw is the only element
in w−1 (∆)rΦP . Our calculation using Mathematica confirms the following known
result.

Proposition 13.3. For all very special elements w ∈ WP‡ ,

hλP , α∨w i = 1.

Proof. Fix a very special Weyl element w ∈ WP‡ . Note that h·, ·i is w-invariant, it
is sufficient to calculate hw(λP ), β∨w i where βw is the unique element in ∆ r w(ΦP ).
Our calculation then shows that
‡
Lemma 13.2. For all w ∈ Wn,P , when applicable,
w(λ1 ) = (0, 0, 0, 0, 1, 1, 1, 3), w(λ2 ) = (0, 0, 0, 1, 1, 1, 1, 4),
1 1 1 1 1 1 1 5
w(λ3 ) = (0, 0, 1, 1, 1, 1, 1, 5), w(λ4 ) = ( , , , , , , , ),
2 2 2 2 2 2 2 2
(13.34)
1 1 1 1 1 1 1 7
w(λ5 ) = (− , , , , , , , ), w(λ6 ) = (0, 0, 0, 0, 0, 0, 0, 2),
2 2 2 2 2 2 2 2
w(λ7 ) = (0, 0, 0, 0, 0, 1, 1, 2), w(λ8 ) = (0, 0, 0, 0, 0, 0, 1, 1).

To go further, we calculate βw which is listed in the Mathematica program as the

set named ‘com’. Finally, we are able to verify the following
hwλP , β∨w i = 1 ‡
∀w ∈ Wn,P ,
as desired. 

13.2.4 Discriminants ∆ En,P

Q

BP,w
Recall that ∆QEn ,P =
X
where
w∈WP
hλP , α∨w i
|(w−1 ∆)rΦP |=1

−1
Φ+ |
Y 1 Y
ζ(2)|∆P ∩w
BP,w := b ζ(hρ, α∨ i + δα,w ).
b
hρ, α∨ i − 1 α∈Φ+ r∆
α∈(w−1 ∆)∩(ΦP r∆P ) P P

(13.35)
By Proposition 13.3, hλP , α∨w i = 1. Hence, only BP,w becomes essential now. Thus
to calculate the constants ∆QEn ,P (n = 6, 7, 8) for all associated maximal parabolic
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 361

Weak RH for Zeta Functions of Exceptional Groups 361

subgroups, it suffices to further calculate, for each very special w ∈ W ‡ ,

(1) The cardinal number of ∆P ∩ w−1 (Φ+ );

Y 1
(2) The product ;
hρ, α∨ i − 1
a∈(w−1 ∆)∩(ΦP r∆P )

(3) The constant δα,w for α ∈ Φ+ r ∆P ; and

Y
(4) The product ζ(hρ, α∨ i + δα,w ).
b
α∈P +P r∆P

We have calculated all these. The reader may find some details in our website.
The up-shot is the following explicit formulas for the discriminants ∆QEn ,P (n =
6, 7, 8), obtained using Mathematica.

Theorem 13.3. The discriminants ∆QEn ,P for exceptional group En (n = 6, 7, 8) are

given as follows.

(E8 , P1 )

1 5 4 3 2
− ζ(2) b
b ζ(3) (3 + 4b
ζ(2)(−2 + 3b
ζ(3)))b
ζ(4) bζ(5) bζ(6)b
ζ(7)
3628800
3 2
(9450 + 5040b
ζ(2) (−9 + 10b ζ(3)) + 60b
ζ(2) (1435 + 72b ζ(3)(−21 + 10b
ζ(4)))
+ 7b
ζ(2)(27b
ζ(3)(225 + 16b
ζ(4)(−19 + 25b
ζ(5)))
+ 200(−37 + 27b
ζ(4)b
ζ(6)(b
ζ(4) − 8b
ζ(5)b
ζ(8)))))

(E8 , P2 )

1 b 5b 3b 2
ζ(2) ζ(3) ζ(4)(9 + 18b
ζ(2) − 4bζ(2)(8 + 9b
ζ(3)(−1 + 2b
ζ(4))))
51840
(−45 + b
ζ(2)(200 + 8b
ζ(2)(−25 + 36bζ(3)) − 9b
ζ(3)(25 + 16b
ζ(4)(−2 + 5b
ζ(5)))))

(E8 , P3 )

1 b 4 2
− ζ(2) (−1 + 2b
ζ(2))b
ζ(3) (3 + 4b
ζ(2)(−2 + 3b
ζ(3)))b
ζ(4)
17280
(−45 + b
ζ(2)(200 + 8b
ζ(2)(−25 + 36bζ(3)) − 9b
ζ(3)(25 + 16b
ζ(4)(−2 + 5b
ζ(5)))))
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 362

362 Algebraic, Analytic Structures and Riemann Hypothesis

(E8 , P4 )

1 6 5 4 3 2
− ζ(2) b
b ζ(3) bζ(4) b ζ(5) bζ(6) bζ(7)
12700800
3
(99225 + 2b
ζ(2)(99225b ζ(2) + 44100(−8 + 9b ζ(3))
2
ζ(2) (425 + 18b
− 1568b ζ(3)(−46 + 24b
ζ(3) + 25b
ζ(4)))
2
+ 42b
ζ(2)(19600 + 45b
ζ(3) (385 + 12b
ζ(4)(−64 + 35b
ζ(4) + 70b
ζ(5)))
+ 24b
ζ(3)(−1540 + b
ζ(4)(1568 + 25b
ζ(5)(−64 + 63b
ζ(6)))))
ζ(3)b
− 144b ζ(4)(3528 + 25b
ζ(5)(−196 + 9b
ζ(6)(32 + 49b
ζ(7)(−1 + 4b
ζ(8)))))))

(E8 , P5 )

1 5 4 3 2
ζ(2) (−1 + 2b
b ζ(2))b
ζ(3) bζ(4) bζ(5) bζ(6)
1814400
2
(14175 + b
ζ(2)(11025(−8 + 9bζ(3)) + 4(882b ζ(2) (−25 + 36b
ζ(3))
+ 5b
ζ(2)(8330 − 9b
ζ(3)(15b
ζ(3)(−49 + 96b
ζ(4)) + 2(833 + 16b
ζ(4)(−49 + 45b
ζ(5)))))
+ 36b
ζ(3)b
ζ(4)(−882 + 25b
ζ(5)(49 + 36b
ζ(6)(−2 + 7b
ζ(7)))))))

(E8 , P6 )

1 6 6 5 5 4 3 2 2
ζ(2) b
b ζ(3) bζ(4) bζ(5) bζ(6) bζ(7) bζ(8) bζ(9) bζ(10)bζ(11)
4191264000
4 3
ζ(2) (−215+264b
(1164240b ζ(3))+41580b ζ(2) (14980 + 3b ζ(3)(−6139+3840b
ζ(4)))
2
− 10914750(3 + 16b
ζ(3) bζ(4)b
ζ(5))
2 2
+ 140b
ζ(2) (−14850b
ζ(3) (126 + b
ζ(4)(−387 + 224b
ζ(4) + 252b
ζ(5)))
ζ(3)(−58135 + 48b
− 99b ζ(4)(1487 − 225b
ζ(5) + 630b
ζ(4)b
ζ(6)))
+ 175(−24431 + 72b
ζ(4)b
ζ(6)(187b
ζ(4) − 216b
ζ(5)b
ζ(8))))
2
+ 99b
ζ(2)(6048000b
ζ(3) bζ(4)b
ζ(5)
+b
ζ(3)(−2149875 + 16b
ζ(4)(202419 − 25b
ζ(5)(2695 + 72b
ζ(6)(−72 + 245b
ζ(7)))))
− 9800(−245 + 9b
ζ(4)b
ζ(6)(15b
ζ(4)
ζ(6)b
− 8(−5b ζ(9)b
ζ(10) + 3b
ζ(8)(b
ζ(5) + 20b
ζ(7)b
ζ(10)b
ζ(12)))))))
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 363

Weak RH for Zeta Functions of Exceptional Groups 363

(E8 , P7 )
1 5 5 5 4 3 3 2
− bζ(2) (−1 + 2b
ζ(2))b
ζ(3) bζ(4) bζ(5) bζ(6) bζ(7) bζ(8) bζ(9)b
ζ(10)b
ζ(11)
6652800
3
(−51975 + 44bζ(2) (10255 + 48b
ζ(3)(−448 + 225b ζ(3)))
2
ζ(2) (1540 + 3b
− 440b ζ(3)(−819 + 210b
ζ(3) + 8b
ζ(4)(39 − 70b
ζ(5))))
+ 7b
ζ(2)(47300 + 9(b
ζ(3)(−4950 + 32b
ζ(4)(132 + 25b
ζ(5)(−11 + 6b
ζ(6))))
2
ζ(6)(3b
− 275b ζ(4) − 32b
ζ(4)b
ζ(5)b
ζ(8) + 192b
ζ(5)b
ζ(8)b
ζ(9)b
ζ(12)))))
(E8 , P8 )
1 6 6 6 6 5 5 4
ζ(2) b
b ζ(3) b ζ(4) bζ(5) bζ(6) bζ(7) bζ(8)
22054032000
4 3 3 2 2
ζ(9) b
b ζ(10) bζ(11) b ζ(12) b ζ(13) b ζ(14)bζ(15)bζ(16)bζ(17)
4
(−172297125 + 91891800b
ζ(2) (−9 + 10b
ζ(3))
3
ζ(2) (−122605 + 36b
−24310b ζ(3)(6153 − 3200b
ζ(4) + 150b
ζ(3)(−20 + 21b
ζ(4))))
2 2
+ 170b
ζ(2) (−1081080b
ζ(3) (10 + 3b
ζ(4)(−5 + 4b
ζ(5)))−

ζ(3)(−252175 + 16b
− 117b ζ(4)(13046 + 175b
ζ(5)(−55 + 36b
ζ(6))))
+ 3850(−4745 + 36b
ζ(4)b
ζ(6)(13b
ζ(4) − 72b
ζ(5)b
ζ(8))))
ζ(2)(51b
− 39b ζ(3)(606375 + 16b
ζ(4)(−40887 + 25b
ζ(5)(1925+
ζ(6)(−119 + 110b
18b ζ(7)))))
2
+ 1925(−16660 + 3b
ζ(6)(765b
ζ(4) + 288b
ζ(4)b
ζ(8)(−17b
ζ(5) + 20b
ζ(6)b
ζ(10))
+ 5440b
ζ(8)b
ζ(12)(b
ζ(5)b
ζ(9) − 18b
ζ(10)b
ζ(14)b
ζ(18))))))
(E7 , P1 )
1 b 4 4 3 2
− ζ(2) (−1 + 2b
ζ(2))b
ζ(3) bζ(4) bζ(5) bζ(6)bζ(7)
604800
3 2
ζ(2) (−9 + 10b
(5040b ζ(3)) + 378(25 − 32b ζ(3) b ζ(4))
2
+ 60b
ζ(2) (1435 + 72b
ζ(3)(−21 + 10b
ζ(4)))
+ 35b
ζ(2)(27b
ζ(3)(45 + 16b
ζ(4)(−3 + 5b
ζ(5))) + 40(−37
+ 27b
ζ(4)b
ζ(6)(b
ζ(4) − 8b
ζ(5)b
ζ(8)))))
(E7 , P2 )
1 b 4b 2
− ζ(2) ζ(3) (3 + 4b
ζ(2)(−2 + 3b
ζ(3)))b
ζ(4)
8640
(−45 + b
ζ(2)(200 + 8bζ(2)(−25 + 36b
ζ(3)) − 9b
ζ(3)(25 + 16b
ζ(4)(−2 + 5b
ζ(5)))))
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 364

364 Algebraic, Analytic Structures and Riemann Hypothesis

(E7 , P3 )
1 b 3
− ζ(2) (−1 + 2bζ(2))b
ζ(3)(3 + 4b
ζ(2)(−2 + 3b
ζ(3)))
1728
2
(9 + 18b
ζ(2) − 4bζ(2)(8 + 9b
ζ(3)(−1 + 2b
ζ(4))))
(E7 , P4 )
1 b 5b 4b 3b 2b
ζ(2) ζ(3) ζ(4) ζ(5) ζ(6)
907200
2
(14175 + b ζ(2)(11025(−8 + 9bζ(3)) + 4(882b
ζ(2) (−25 + 36b
ζ(3))
+ 5b
ζ(2)(8330−9b
ζ(3)(15b
ζ(3)(−49+96b
ζ(4))+2(833 + 16b
ζ(4)(−49+45b
ζ(5)))))
+ 36b
ζ(3)b
ζ(4)(−882 + 25b
ζ(5)(49 + 36b
ζ(6)(−2 + 7b
ζ(7)))))))
(E7 , P5 )
1 b 4 3 2
ζ(2) (−1 + 2b
ζ(2))b
ζ(3) bζ(4) bζ(5)
43200
2
(−675 + 2b
ζ(2)(675b
ζ(2) − 40b ζ(2)(65 + 9bζ(3)(−12 + 5b
ζ(3) + 10b
ζ(4)))
+ 9(200 + b
ζ(3)(−225 + 16b
ζ(4)(18 + 25b
ζ(5)(−1 + 3b
ζ(6)))))))
(E7 , P6 )
1 5 5 4 4 2 2
ζ(2) b
b ζ(3) bζ(4) b ζ(5) bζ(6) bζ(7) bζ(8)b
ζ(9)
9072000
4 3
(141750 + 453600b ζ(2) − 1800b ζ(2) (833 + 15b ζ(3)(−35 + 48b
ζ(4)))
2
ζ(2) (−3773 + 18b
− 500b ζ(3)(245 − 328b
ζ(4) + 42b
ζ(3)(−2 + 3b
ζ(4)))
2
+ 2016b
ζ(4) bζ(6))
ζ(2)(43000 + b
− 21b ζ(3)(−37125 + 48b
ζ(4)(1197 + 25b
ζ(5)(−25 + 72b
ζ(6))))
ζ(4)b
− 5400b ζ(6)(5b
ζ(4) − 8b
ζ(8)(b
ζ(5) − 10b
ζ(6)b
ζ(10)))))
(E7 , P7 )
1 5 5 5 4 3 3 2
ζ(2) b
b ζ(3) bζ(4) bζ(5) bζ(6) bζ(7) bζ(8) bζ(9)b
ζ(10)b
ζ(11)
3326400
2
(51975 + b
ζ(2)(3850(−86 + 81b ζ(3)) − 44b ζ(2) (10255
+ 48b
ζ(3)(−448 + 225b
ζ(3)))
+ 440b
ζ(2)(1540 + 3b
ζ(3)(−819 + 210b
ζ(3) + 8b
ζ(4)(39 − 70b
ζ(5))))
+ 63(−32b
ζ(3)b
ζ(4)(132 + 25b
ζ(5)(−11 + 6b
ζ(6)))
2
+ 275b
ζ(6)(3b
ζ(4) − 32b
ζ(4)b
ζ(5)b
ζ(8) + 192b
ζ(5)b
ζ(8)b
ζ(9)b
ζ(12)))))
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 365

Weak RH for Zeta Functions of Exceptional Groups 365

(E6 , P1 )
1 b 4b 4b 3b 2b b
− ζ(2) ζ(3) ζ(4) ζ(5) ζ(6)ζ(7)
302400
3 2
(9450 + 5040bζ(2) (−9 + 10b
ζ(3)) + 60b
ζ(2) (1435 + 72b
ζ(3)(−21 + 10b
ζ(4)))
+ 7b
ζ(2)(27b
ζ(3)(225 + 16b
ζ(4)(−19 + 25b
ζ(5)))
+ 200(−37 + 27b
ζ(4)b
ζ(6)(b
ζ(4) − 8b
ζ(5)b
ζ(8)))))

(E6 , P2 )
1 b 3 2
− ζ(2) (−1 + 2b
ζ(2))b
ζ(3) bζ(4)
1440
(−45 + b
ζ(2)(200 + 8b
ζ(2)(−25 + 36b ζ(3)) − 9b
ζ(3)(25 + 16b
ζ(4)(−2 + 5b
ζ(5)))))

(E6 , P3 )
1 b 2
ζ(2) (−1 + 2b
ζ(2))(3 + 4b
ζ(2)(−2 + 3b
ζ(3)))2
288
(E6 , P4 )
1 b 4b 3b 2b
ζ(2) ζ(3) ζ(4) ζ(5)
21600
2
(−675 + 2b
ζ(2)(675bζ(2) − 40b
ζ(2)(65 + 9b
ζ(3)(−12 + 5b
ζ(3) + 10b
ζ(4)))
+ 9(200 + b
ζ(3)(−225 + 16b
ζ(4)(18 + 25b
ζ(5)(−1 + 3b
ζ(6)))))))

(E6 , P5 )
1 b 3 2
− ζ(2) (−1 + 2b
ζ(2))b
ζ(3) bζ(4)
1440
(−45 + b
ζ(2)(200 + 8b
ζ(2)(−25 + 36b ζ(3)) − 9b
ζ(3)(25 + 16b
ζ(4)(−2 + 5b
ζ(5)))))

(E6 , P6 )
1 b 4b 4b 3b 2b b
− ζ(2) ζ(3) ζ(4) ζ(5) ζ(6)ζ(7)
302400
3 2
(9450 + 5040bζ(2) (−9 + 10b
ζ(3)) + 60b
ζ(2) (1435 + 72b
ζ(3)(−21 + 10b
ζ(4)))
+ 7b
ζ(2)(27b
ζ(3)(225 + 16b
ζ(4)(−19 + 25b
ζ(5)))
+ 200(−37 + 27b
ζ(4)b
ζ(6)(b
ζ(4) − 8b
ζ(5)b
ζ(8)))))

Corollary 13.1. For all n = 6, 7, 8 and for all their maximal parabolic subgroups
of En , the discriminants ∆QEn ,P are (strictly) positive.
December 26, 2017 9:55 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 366

366 Algebraic, Analytic Structures and Riemann Hypothesis

Proof. This is a direct consequence of the theorem above. For example, with
(E8 , P1 ), we have
(π3 ζ(3)4 · (4π − 3(3 + ζ(3))) · ζ(5)2 · ζ(7))
∆QE8 ,P1 = −
75243740696496046080000000
· (−567000π3 (−63 + 2ζ(3)) + 7938000π(185 + 27ζ(3))
− 79380π2 (5125 + 136ζ(3)) + 8π8 (−35 + 12ζ(5))
− 4465125(360 + ζ(3)(135 + 2ζ(5)))).
Approximately,
206392.
∆QE8 ,P1 ≈ > 0. (13.36)
75243740696496046080000000
Our verifications for other cases are the same. 

We mention in passing that, even positive, all the ∆QEn ,P ’s are indeed very very
small. This phenomenon is also observed for the case of S Ln .

13.2.5 Weak Riemann Hypothesis for Zeta Functions of Exceptional

Groups

As a direct consequence of Theorem 13.3, or better, of Corollary 13.1, by Theo-

rem 12.1, we have proved the following

ζQEn /P(s) (n = 6, 7, 8) are simple and

Theorem 13.4. All but finitely many zeros of b
1
lie on the line Re(s) = − cEn ,P .
2
We end this discussion by the following comments. Structures for W ‡ and
∆QEn ,P
as stated above are strikingly uniform. Besides |W ‡ | = 2n−1 , the number of
non-trivial factors of ∆QEn ,Pi in fact is the same as the number of connected com-
ponents of the Dynkin diagram by deleting the i-th vertex. All these are not just
a coincidence and can all be explained in terms of our conjecture on ‘Stability,
Parabolic Reduction and Masses’ to be introduced and proved in Part 6. For ex-
ample, |W ‡ | = 2n−1 implies that ∆QEn ,P consists of 2n−1 terms (as mentioned above).
On the other hand, 2n−1 is also exactly the cardinal number of a power set associ-
ated to standard parabolic subgroups appeared in the Arthur’s analytic truncation
theory. More importantly, we know from there that ∆QEn ,P is simply the mass of
principal En -lattices of a certain fixed slope which are P-semi-stable. So, the
positivity statement above comes from some theoretic structure.
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 367

PART 6

Geometric Structures and

Riemann Hypothesis

Stability plays a key role in various branches of mathematics and is used here to
complete a proof of the weak Riemann Hypothesis for our zeta functions. We
first show that, for special linear groups and more generally for reductive groups,
the geometric truncations of stability on the moduli spaces of arithmetic principal
torsors essentially coincide with Arthur’s analytic truncations (on the associated
adelic spaces). Then, we confirm the conjecture on Stability, Parabolic Reduction
and Masses, calculating the semi-stable masses (of arithmetic principal torsors)
in terms of parabolic reduction of stability, as a natural continuation of the works
of Atiyah-Bott, Harder-Narasimhan, Mumford in geometry and Arthur, Lafforgue
in arithmetic. Based on these, we establish the special uniformity of zeta func-
tions on the equivalence of the high rank non-abelian zeta functions and the Weng
zeta functions for special linear groups. Finally, we complete a proof of the weak
Riemann hypothesis for our zeta functions over the field of rationals, by showing
that their discriminants are positive, being the semi-stable masses of the arith-
metic principal torsors for which the canonical types coincide with the maximal
parabolic subgroups used. All the constructions and results of this part are new.

367
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 369

Chapter 14

Analytic and Geometric Truncations:

Special Linear Groups

14.1 Geometric Truncation and Parabolic Reduction

14.1.1 Canonical Polygon and Compactness

Let F be a number field with OF its ring of integers. Denote by S = S ∞ t S fin the
collection of inequivalent places of F with S ∞ , resp. S fin , the set of infinite, resp.
finite, places. For v ∈ S , denote by Fv the v-completion of F. For example, if
v ∈ S ∞ , Fv = R or C. Accordingly, we call v real or complex. Denote by r1 , resp.
r2 , the number of real, resp. complex, places in S ∞ . In addition, for v ∈ S fin , Fv is a
non-Archimedean local field. Denote by Ov its ring of integers. Let A be the adelic
ring of F, and let GLn (A), resp. SLn (A), be the adelic group associated to GLn ,
Q Q Q
resp. SLn . Let K := v∈S fin GLn (Ov ) × v∈S ∞ On (R) × v∈S ∞ U n (C), resp.
Q Q Q v:real v:complex
SK := v∈S fin SLn (Ov ) × v∈S ∞ SOn (R) × v∈S ∞ SUn (C), be the maximal
v:real v:complex
compact subgroup of GLn (A), resp. of SLn (A). Here On , resp. Un , denotes the
orthogonal group, resp. the unitary group, of size n, and set SOn := SLn ∩ On and
SUn := SLn ∩ Un . Accordingly, we obtain the quotient spaces GLn (F)\GLn (A)/K
and SLn (F)\SLn (A)/SK. By §§1.4.1-2 and §6.2.3, we have

Proposition 14.1. There are natural identifications among the following spaces.

(1) The quotient space GLn (F)\GLn (A)/K, resp. SLn (F)\SLn (A)/SK.
(2) The quotient space GLn (OF )\GLn (R)r1 × GLn (C)r2 On (R)r1 × Un (C)r2 , resp.

SLn (OF )\SLn (R)r1 × SLn (C)r2 SOn (R)r1 × SUn (C)r2 .

(3) The moduli space Mtot tot
F,n , resp. MF,n [1], of OF -lattices of rank n, resp. of rank
n and volume one. And
(4) The moduli space Mtot OF ,n
, resp. MtotOF ,n
(d), of rank n metrized vector bundles
of all degrees, resp. of degree d, on X = Spec OF .

369
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 370

370 Geometric Structures and Riemann Hypothesis

To go further, let g = (gfin ; g∞ ) ∈ GLn (A). For a parabolic subgroup Q

of GLn , we denote by Eg∗fin ;Q the filtration of the vector bundles Egfin induced
by the parabolic subgroup Q, and hence a filtration of hermitian vector bundles
(Eg∗fin ;Q , ρ∗g∞ ;Q ), where the hermitian metrics ρgi ∞ ;Q on the vector bundles Egi fin ;Q are
obtained as the restrictions of ρg∞ . This then further induces a filtration of OF -
lattices (Pg;Q g;Q
∗ , ρ∗ ). Consequently, by Definition 6.5, we obtain the convex poly-
gon pgQ : [0, r] → R of g with respect to Q. In addition, by Propositions 1.6, 6.2
and 6.3, we obtain the following:

Proposition 14.2. Let g ∈ GLn (A) and let P be a parabolic subgroup of GLn .

(1) Among all the pgQ , where Q runs over all parabolic subgroup of GLn , there is a
unique maximal convex polygon pg . Moreover there exists a unique parabolic
g
subgroup Q such that pg g = pg .
Q
(2) Let d ∈ R. Forn p : [0, n] → R be a convex polygon satisfying
o p(0) = p(n) =
0. The space g ∈ GLn (F)\GLn (A) : degar g = d, pg ≤ p is compact. More
generally,
(3) Among all the pgQ , where Q runs over all parabolic subgroups contained in P
of GLn , there is a unique maximal element pgP . Moreover there exists a unique
g
parabolic subgroup QP ⊆ P of GLn such that pg g = pgP .
QP
n o
(4) The space g ∈ GLn (F)\GLn (A) : degar g = d, p̄gP ≤ p, pgP ≥ −p is compact.

We call pg , resp. Q, resp. (Eg;Q g,Q

∗ , ρ∗ ), the canonical polygon, resp. the canon-
ical parabolic subgroup, resp. the canonical filtration of the metrized vector bun-
g
dle, associated to g. Similarly, we call pgP , resp. QP , the P-canonical polygon,
resp. the P-canonical parabolic subgroup, of g. Easily, G-canonical is simply
canonical.

14.1.2 Parabolic Reduction and Geometric Truncation

Let p, q : [0, n] → R be two normalized convex polygons. Following Lafforgue

( [63]), we say that q is bigger than p with respect to Q , denoted by q >Q p, if
q(ni ) − p(ni ) > 0 for all 1 ≤ i ≤ |Q| − 1.

Definition 14.1. Let p be a normalized convex polygon on [0, n]. The character-
istic function 1(p∗ ≤ p) and 1(p∗P ≤P p) associated to a parabolic subgroup P, on
GLn (F)\GLn (A)1 are defined by
 g
 g
1, if p ≤ p, 1, if pP ≤P p,
1(p ≤ p) =  and 1(pP ≤P p) = 
 
g
 g 
(14.1)
0, otherwise 0, otherwise

December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 371

Analytic and Geometric Truncations: Special Linear Groups 371

respectively. Here pg and pgP , denote the canonical polygon and the polygon asso-
ciated to P of g, respectively.

Theorem 14.1. Let g ∈ G(F)\G(A)1 and let p : [0, n] → R be a convex polygon

satisfying p(0) = p(n) = 0. Then

1(pg ≤ p) = (−1)|P|−1 1(pδg

X X
P >P p). (14.2)
P δ∈P(F)\G(F)

where P runs over all standard parabolic subgroups of GLn .

Remark 14.1. (1) This result and its proof below are arithmetic analogues of a
result and its proof obtained by Lafforgue ([63]) on vector bundles of curves over
finite fields.
(2) In algebraic geometry, or better in geometric invariant theory of Mumford
(see e.g. [80]), a fundamental principle claims that, for an action of a reductive
group over a space, if a point is not GIT stable, there should exist a parabolic
subgroup which destroys the corresponding stability. Our result above gives a
concrete realization of this principle in our arithmetic setting.

Proof. Since all parabolic subgroups of G which conjugate to P are exactly the
δ-conjugates of P, where δ ∈ P(F)\G(F), by the easy relation pδg g
P = pδPδ−1 , it
suffices to prove the following equivalence relation

1(pg ≤ p) = (−1)|P|−1 1(pgP >P p)

X
(14.3)
P

where P runs over all parabolic subgroups of GLn .

Assume first that pg ≤ p. Then the left hand side of (14.3) is simply 1. On the
other hand, for a proper parabolic subgroup Q , G, by definition, pQ ≤ pg ≤ p.
This implies 1(pgQ >Q p) = 0. Hence, the right hand side of (14.3) degenerates to
the one consisting of a single term involving G only, which, from the definition, it
is certainly 1.
Therefore, we may assume that pg  p, or the same 1(pg ≤ p) = 0. Hence, it
suffices to show that

1(pg ≤ p) = 0 =⇒ (−1)|P|−1 1(pgP >P p) = 0.

X
(14.4)
P

The proof below is rather involved. To start with, we recall that, for a parabolic
subgroup P, by Proposition 14.2, there are a P-canonical polygon pgP of g and a
g g,P
P-canonical parabolic subgroup QP ⊆ P of G. Denote by Λ∗ the P-canonical
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 372

372 Geometric Structures and Riemann Hypothesis

g
filtration of Λg induced by QP . Certainly, for different P and P0 , it may happen
g g
that QP = QP0 .
Lafforgue’s idea is to regroup the P’s in (14.4) according to their associated
P-canonical parabolic subgroups Q and then to show that for each fixed Q, the
summations on all involved P results exactly zero. Namely, after rewriting

(−1)|P|−1 1(pgP >P p) = (−1)|P|−1 1(pgP >P p),

X X X
(14.5)
Q
g
P P: pP =Q

it suffices to prove the following

Lemma 14.1. Let g ∈ G(F)\G(A)1 and let Q be a parabolic subgroup of GLn .

Assume that pg  p. Then

(−1)|P| 1(pPp >P p) = 0

X
(14.6)
g
P: QP =Q

g
where P runs over all parabolic subgroups of GLn satisfying QP = Q.

Proof. We begin with a definition.

Definition 14.2. Let µ ∈ R≥0 be a constant and denote by Λg the OF -lattices

associated to g.
g
(1) µ Q is defined to be the unique parabolic subgroup of G such that, for the
g g
induced filtration µ Λ∗ of Λg (by µ Q ),
nµ g o n g  g. g   g . g o
Λ∗ = Λ j : µ Λ j Λ j−1 > µ Λ j+1 Λ j + µ , (14.7)
g
where Λ∗ is the canonical filtration associated with Λg .
g
(2) µ QP is defined to be the unique parabolic subgroup of G such that, for the
g,P
induced filtration µ Λ∗ of Λg ,
nµ g,P o n o [ n g,P  g,P . g,P   g,P . g,P  o
Λ∗ = Λg,P j Λ j : µ Λ j Λ j−1 > µ Λ j+1 Λ j + µ , (14.8)

where Λg,P
j is the filtration of Λ induced from the parabolic subgroup P, and
g
g,P
Λ∗ is the P-canonical filtration associated with Λg .

Claim 14.1. With the same notation as above,

g g g g
Q ⊆ µQ and QP ⊆ µ QP ⊆ P. (14.9)
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 373

Analytic and Geometric Truncations: Special Linear Groups 373

Proof. The first inclusion is rather trivial. Indeed, for the canonical
 g .filtration
g g g . g g 
Λ∗ of Λ , for all 1 ≤ j ≤ |Q | − 1, Λ j+1 Λ j are semi-stable and µ Λ j Λ j−1 >
g
 g . g nµ g o g
µ Λ j+1 Λ j . On the other hand, the filtration Λ∗ of Λg associated to µ Q is a
partial filtration of the canonical one, obtained
 g . g by selecting
 gonly
. gthe
 special pieces
for which some stronger conditions µ Λ j Λ j−1 > µ Λ j+1 Λ j + µ should be
g g
satisfied. Therefore, Q ⊆ µ Q .
g
Next we show that µ QP ⊆ P. This is simple, since by our definition, the
µ g
filtration {µ Λg,P
∗ } associated to QP is a refinement of the filtration of Λ associated
g

to P.
g g
Finally, we prove QP ⊆ µ QP . This also quite direct. Indeed, by definition,
g,P
for the P-canonical filtration Λ∗ of Λg , the P-Harder-Narasimhan conditions
g,P . g,P  g,P . g,P   g,P . g,P 
hold. Namely, Λ j Λ j−1 are semi-stable and µ Λ j Λ j−1 > µ Λ j+1 Λ j for
all the j’s, except for possibly the indices j induced directly from P. But up to
g
the indices j induced from P, all other indices for µ QP is, by definition, character-
ized as the parts of the P-canonical filtration for which much stronger conditions
 g,P . g,P   g,P . g,P 
µ Λ j Λ j−1 > µ Λ j+1 Λ j + µ are assumed to be satisfied. 

Now, for any two parabolic subgroups Q1 ⊆ Q2 , we introduce the subsets of

non-negative integers by
n o
II(Q1 ) := rkΛg,Q
i
1
: i = 1, 2, . . . , |Q1 | − 1 ,
n o
I(Q2 ) := rkΛg,Qj
2
: j = 1, 2, . . . , |Q2 | − 1 ,
n  g,Q1 . g,Q1   g,Q1 . g,Q1  o
J0 (Q1 ) := rkΛg,Q
k
1
: µ Λk Λk−1 > µ Λk+1 Λk , (14.10)
n  g,Q1 . g,Q1   g,Q1 . g,Q1  o
Jµ (Q1 ) := rkΛg,Q
l
1
: µ Λl Λl−1 > µ Λl+1 Λl +µ ,
n    o
K p (Q2 ) := rkΛg,Q
m
2
: pgQ2 − p rkΛm g,Q2
>0 .

Lemma 14.2. With the same notation as above, the assignment

g g
P ∈ P : QP = Q1 , µ QP = Q2 , pgP >P p
n o





y
(14.11)
Σ ⊆ I(Q2 ), Jµ (Q1 ) ⊆ I(Q2 )
 
 
Σ 

 

  
I(Q1 ) − J0 (Q1 ) ∪ I(Q2 ) − Jµ (Q1 ) ⊆ Σ ⊆ K p (Q2 ) 


 


characterized by
n o
P 7−→ I(P) := rkΛg,P
i : i = 1, 2, . . . , |P| − 1 (14.12)
is a well-defined bijection. Here P denotes the set of parabolic subgroups in GLn .
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 374

374 Geometric Structures and Riemann Hypothesis

Proof. We first show that the assignment is well-defined, by examining what are
the conditions satisfied by P in the set
g g
P ∈ P : QP = Q1 , µ QP = Q2 , pgP >P p .
n o
(14.13)
g
Since the convex polygon pgQ1 of QP = Q1 associated to the filtration
{0} = Λg,Q
0
1
⊂ Λg,Q
1
1
⊂ Λ2g,Q1 ⊂ . . . ⊂ Λ|Q
g,Q1
1 |−1
⊂ Λg,Q
|Q1 |
1
= Λg
n g o
is the unique maximal element among all pQ : Q ⊆ P , Q1 ⊆ P, and except for
the indices associated to P,
 g,Q1 . g,Q1   g,Q1 . g,Q1 
µ Λk Λk−1 > µ Λk+1 Λk , (14.14)
g
since Q1 = QP is P-canonical. Therefore,
I(Q1 ) − J0 (Q1 ) ⊂ I(P). (14.15)
g
µ
In parallel, from QP = Q2 , by definition, the filtration
{0} = Λg,Q
0
2
⊂ Λg,Q
1
2
⊂ Λ2g,Q2 ⊂ . . . ⊂ Λ|Q
g,Q2
2 |−1
⊂ Λg,Q
|Q2 |
2
= Λg
coincides with the filtration
n g,P o n  g,Q1 . g,Q1   g,Q1 . g,Q1  o
Λ j : j = 1, 2, . . . , |P| − 1 ∪ Λkg,Q1 : µ Λk Λk−1 > µ Λk+1 Λk +µ
Hence, by examining the ranks involved in all the filtrations, we have
I(Q2 ) = I(P) ∪ Jµ (Q1 ).
Consequently,
I(P) ⊂ I(Q2 ) and I(Q2 ) − Jµ (Q1 ) ⊂ I(P). (14.16)
In addition, by definition, the relation pgP >P p is equivalent to the following
group of inequalities
 g  
pP − p rkΛg,Pj > 0. (14.17)
Hence I(P) ⊆ K p (Q2 ) since I(P) ⊆ I(Q2 ). This together with (14.15) and (14.16)
implies that P 7→ I(P) is well-defined.
The injectivity is rather clear since with a fixed partition I, there exists only
one unique standard parabolic subgroup PI such that I(PI ) = ∆ − I.
n o
Finally, we verify the surjectivity. Let Σ be a subset of 1, 2, . . . , r satisfying
   
Σ ⊆ I(Q2 ), Jµ (Q1 ) ⊆ I(Q2 ), I(Q1 ) − J0 (Q1 ) ∪ I(Q2 ) − Jµ (Q1 ) ⊆ Σ ⊆ K p (Q2 )
Since there obviously exists a parabolic subgroup P = PΣ , it suffices to check
whether there exists parabolic subgroups Q1 ⊆ Q2 of GLn contained in P satisfy-
g g
ing the conditions that QP = Q1 , µ QP = Q2 and pgP >P p.
g
It is clear that pgP >P p since Σ ⊂ K p (Q2 ). On the other hand, QP = Q1 and
µ g
QP = Q2 are direct  consequences of the definition
 of P-canonical
 filtration and
the conditions that I(Q1 ) − J0 (Q1 ) ⊆ Σ and I(Q2 ) − Jµ (Q1 ) ⊆ Σ as explained in
the proof above of the well-defined statement. 
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 375

Analytic and Geometric Truncations: Special Linear Groups 375

We are now ready to complete our proof of Lemma 14.1 and hence the theo-
rem. Indeed, from Lemma 14.2,

(−1)|P| 1(pP >P p) =

X X
(−1)#(σ)
g g
µ
P:QP =Q1 , QP =Q2
   
Σ: I(Q1 )−J0 (Q1 ) ∪ I(Q2 )−Jµ (Q1 ) ⊆Σ⊆K p (Q2 )
    
0, if I(Q1 ) − J0 (Q1 ) ∪ I(Q2 ) − Jµ (Q1 ) , K p (Q2 );

=

1, if I(Q1 ) − J0 (Q1 ) ∪ I(Q2 ) − Jµ (Q1 ) = K p (Q2 ).


To prove the theorem, assume otherwise that the value 1 does appear in the case
pg  p. In particular,
   
I(Q1 ) − J0 (Q1 ) ∪ I(Q2 ) − Jµ (Q1 ) = K p (Q2 ). (14.18)
Let then µ = 0. Obviously, this implies that Q1 = Q2 = Q. Consequently,
(−1)|P| 1(pP >P p) = (−1)|P| 1(pP >P p)
X X
g g g g
P:QP =Q1 , µ QP =Q2 P:QP =Q= 0 QP

(−1)|P| 1(pP >P p).

X
=
g
P:QP =Q

This is nothing but the summation appeared in the Proposition. Hence, by defi-
nition and the convexity property of p, we have, from (14.18), I(Q1 ) = J0 (Q1 ),
or better K pg (Q2 ) = ∅. Consequently, pg ≤ p, a contradiction. This completes the
proof of Lemma 14.1 and hence the theorem. 

14.2 Analytic and Geometric Truncations

14.2.1 Micro Bridge

For simplicity, we, in this subsection, work only with Q, the field of rationals, and
use mixed languages of adeles and lattices. Without loss of generality, we assume
that all the Z-lattices here are of volume one. Accordingly, set G = SLn .

Definition 14.3. Let p : [0, n] → R be a convex polygon.

(1) p is called normalized if p(0) = p(n) = 0.

(2) A character T of the minimal parabolic subgroup M0 = P1,1,...,1 is called
a character associated to a normalized p, denoted by T (p), if, for all i =
1, 2, . . . , n − 1,
αi (T ) = p(i) − p(i − 1) − p(i + 1) − p(i) ,



(14.19)
where αi = ei − ei+1 denotes the i-th simple root of SLn as in §6.1.3.
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 376

376 Geometric Structures and Riemann Hypothesis

Lemma 14.3. Let T (p) be the character of P1,...,1 associated to a normalized con-
vex polygon p : [0, n] → R. Then
T (p) = p(1) − p(0), p(2) − p(1), . . . , p(i) − p(i − 1),
(14.20)
. . . , p(n − 1) − p(n − 2), p(n) − p(n − 1) .

Proof. Write T = T (p) in more details as


T = T (p) = t1 (p), t2 (p), . . . , tn (p) = t1 , t2 , . . . , tn ∈ a0 .


(14.21)
Then, for all i = 1, 2, . . . , n − 1, we have αi (T ) = ti − ti+1 , and
ti − ti+1 = p(i) − p(i − 1) − p(i + 1) − p(i) = ∆p(i) − ∆p(i + 1)



(14.22)
where ∆p(i) := p(i) − p(i − 1). Hence
t1 −ti+1 = ∆p(1)−∆p(i+1) or equivalently ti+1 = t1 +∆p(i+1)− p(1). (14.23)
In particular,
t1 − tn = ∆p(1) − ∆p(n) = p(1) − p(n − 1), (14.24)
since ∆p(1) = p(1) − p(0) = p(1) and ∆(n) = p(n) − p(n − 1) = −p(n − 1). Now
n
X n−1
X
0= ti = t1 − p(1) + ∆p(i + 1)


i=1 i=0 (14.25)
= (n − 1) t1 − p(1) + p(0) − p(r) = (n − 1) t1 − p(1) .

Therefore t1 = p(1) and hence ti+1 = ∆p(i + 1) for all i = 1, 2, . . . , n − 1. 

Let g ∈ G(A) and let P be the parabolic subgroup of G associated to the

ordered partition n = n1 + n2 + . . . , nk . Denote by Λg the lattice associated to g,
and denote the filtration pf Λg induced from P by
0 = Λg,P g,P g,P
0 ⊂ Λ1 ⊂ . . . ⊂ Λ|P| = Λ .
g
(14.26)
Since the volume of Λg is one, the associated normalized convex polygon pgP =
pΛP : [0, n] → R is characterized by the conditions that, for all i = 1, 2, . . . , k − 1,
g

pgP is affine on the closed intervals [ni , ni+1 ] and pgP (ni ) = degar Λg,P .


i
g
We can write down these values of pP more precisely. Indeed, if P = NP MP
denotes the Levi decomposition of P, and g = n · m · a(g) · k denotes the Iwasawa
decomposition of g with n ∈ NP (A), m ∈ MP (A)1 , k ∈ K := p SL(OQ p ) × SOn (R)
Q
and a = a(g) = diagar a1 In1 , a2 In2 , . . . , ak Ink ∈ A+ , we have, for all i = 1, . . . , k,

i
Y i
X
nj
  
degar Λg,P
i = − log a j = − n j log a j . (14.27)
j=1 j=1
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 377

Analytic and Geometric Truncations: Special Linear Groups 377

Proposition 14.3 (Micro Bridge). Let p : [0, n] → R be a normalized convex

polygon, and let P = Pn1 ,n2 ,...,nk be a parabolic subgroup of SLn . Then, for all
g ∈ SLn (A),

τP − H0 (g) − T (p) = 1 pgP >P p .

   
b (14.28)

Proof. By definition, 1(pgP >P p) = 1 if and only if

pgP (n1 ) − p(n1 ) > 0, pgP (n2 ) − p(n2 ) > 0, . . . , pgP (nk−1 ) − p(nk−1 ) > 0.

This is equivalent to the conditions that

     
degar Λg,P
1 − p(n1 ) > 0, degar Λg,P
2 − p(n2 ) > 0, . . . , degar Λg,P
|P|−1
− p(nk−1 ) > 0.

That is to say, we have proved the following:

Lemma 14.4. The condition 1(pgP >P p) = 1 is equivalent to the following system
of inequalities

− n1 log a1 − p(r1 ) > 0,






 − n1 log a1 − n2 log a2 − p(n2 ) > 0,





(14.29)
... ... ... ...







 − n1 log a1 − n2 log a2 − . . . − nk−1 log ak−1
− p(nk−1 ) > 0.



τP − H0 (g) − T (p) . Recall that

To go further, we next analysis the function b
for P = PI with I ⊆ ∆ a subset of simple roots,
n τP is the characteristic function
b o of
the subset of a0 = a0I ⊕ aI defined by a0I ⊕ H ∈ aI : $i (H) > 0 ∀i ∈ ∆ − I .
|P|
X
(n )
 
By §6.1.3, H ∈ aI if and only if H = H1(n1 ) , H2(n2 ) , . . . , H|P||P| and ni Hi = 0.
i=1
Therefore, an element H ∈ a0 belongs to aI and satisfies $i (H) > 0 for all i =
n1 , n2 , . . . , n|P|−1 if and only if
(n )
  



 H = H1(n1 ) , H2(n2 ) , . . . , H|P||P| ,


 |P|
X
ni Hi = 0,







i=1



n1 H1 > 0,

(14.30)




n1 H1 + n2 H2 > 0,







... ... ... ...







n1 H1 + n2 H2 + . . . + n|P|−1 H|P|−1 > 0.


December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 378

378 Geometric Structures and Riemann Hypothesis

 
On the other hand, for H = H1 , H2 , . . . , Hr ∈ a0 , the projections of H to aI , resp.
a0I , is given by
 (n1 )  (n2 )  N|P|
(n|P| ) 
n1 N2
 1  1 X
 X    1 X  
H j  ,  H j  , . . . ,  H j   ∈ aI

n1 n2 n|P|

j=N1 +1 j=N|P|−1 +1

j=1

resp.
 N1 N1
H1 − 1
 X 1 X H1 + H2 + . . . + Hr1
H j , H2 − H j , . . . , Hn1 − , ...,
 n1 j=1 n1 j=1 d1
N|P| N|P| N|P|

1 X 1 X 1 X 
HN|P|−1+1 − H j , HN|P|−1+2 − H j , . . . , HN|P| − H j 
n|P| j=N +1 n|P| j=N +1 n|P| j=N +1
|P|−1 |P|−1 |P|−1
 
= H1 , H2 , . . . , Hr
 (n1 )  (n2 )  N|P|
(n|P| ) 
n1 N2
 1  1 X
 X    1 X  
−  H j  ,  H j  , . . . ,  H j   ∈ a0I .

 n1 n2 j=N +1 n|P| j=N|P|−1 +1

j=1 1

Here N1  = n1 , N2 =  n1 + n2 , . . . , N|P| = n1 + n2 + . . . + n|P| . There-

τP − H0 (g) − T (p) = 1 if and only if all the coordinates Hi ’s of
fore, b
−H0 (g) − T (p) =: H = H1 , H2 , . . . , Hn satisfy the conditions that

H + H2 + . . . + HN1 > 0,

1



   
H1 + H2 + . . . + HN1 + HN1 +1 + HN1 +2 + . . . + HN2 > 0,







........................ ............

(14.31)



   
H1 + H2 + . . . + HN1 + HN1 +1 + HN1 +2 + . . . + HN2






  
+ . . . + HN +1 + HN +2 + . . . + HN > 0.



|P|−1 |P|−1 |P|−1

Recall now that, by definition,

 
−H0 (g) − T (p) = − t1 − log b1 , −t2 − log b2 , . . . , −tn − log bn

(n|P| )
   ) (n ) 
where a(g) = diag b1 , b2 , . . . , bn = diag a(n
1 , a2 , . . . , a|P|
1 2
and

 
T (p) = t1 , t2 , . . . , tn = p(1), p(2) − p(1), . . . , p(n − 1) − p(n − 2), −p(n − 1) .

As such, the system of inequalities (14.31), after replacing the Hi ’s with the pre-
cise coordinates of −H0 (g) − T (p), are then changed to the following system of
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 379

Analytic and Geometric Truncations: Special Linear Groups 379

inequalities
     



 −log b1 − p(1) + −log b2 − p(2)+ p(1) + . . . + −log bN1 − p(N1 )+ p(N1−1) > 0,
      
−log b1 − p(1) + −log b2 − p(2)+ p(1) + . . . + −log bN1 − p(N1 )+ p(N1−1)








    
+ − log bN1 +1 − p(N1 + 1) + p(N1 ) + − log bN1 +2 − p(N1 + 2) + p(N1 + 1)








  
+ . . . + − log bN2 − p(N2 ) + p(N2 − 1) > 0,








..........................................







      
+ + . . . +





  −log b1 − p(1) −log b2 − p(2)+ p(1) −log b N1
− p(N1 )+ p(N 1−1)
    
+ + + + + + +





 − log b N 1 +1 − p(N1 1) p(N 1 ) − log b N 1 +2 − p(N 1 2) p(N 1 1)

  
+ . . . + +






 − log b N 2
− p(N2 ) p(N 2 − 1)


..........................................






  
+ − log bN|P|−2 +1 − p(N|P|−2 + 1) + p(N|P|−1 )







  
+ − log bN|P|−2 +2 − p(N|P|−2 + 2) + p(N|P|−2 + 1)







  
+ . . . + − log bN − p(N|P|−1 ) + p(N|P|−1 − 1) > 0.



 |P|−1

This, after the obvious simplification that

B1 := − log b1 , B2 = − log b2 , . . . , Bn = − log bn and N0 := 0,
gives the system of inequalities
B1 + B2 + . . . + BN1 − p(N1 ) + p(0) > 0,






B1 + B2 + . . . + BN1 + BN1 +1 + BN1 +2 + . . . + BN2 − p(N2 ) > 0,








.......................................

(14.32)




B1 + B2 + . . . + BN1 + BN1 +1 + BN1 +2 + . . . + BN2









+ . . . + BN|P|−2 +1 + BN|P|−2 +2 + . . . + BN|P|−1 − p(N|P|−1 ) > 0.




But
(N −N )

b1 , b2 , . . . , bn = a(g) = a1(N1 −N0 ) , a(N 2 −N1 )
, . . . , a|P||P| |P|−1 .


2
Consequently, we obtain the following:

τP − H0 (g) − T (p) = 1 is equivalent to the follow-

Lemma 14.5. The condition b
ing system of inequalities



 −n1 log a1 − p(N1 ) > 0,

 − n1 log a1 − n2 log a2 − p(N2 ) > 0





(14.33)
............ ............







 − n log a − n log a − . . . − n


1 1 2 2 log a
|P|−1


− p(N
|P|−1 |P|−1 ) > 0.

Therefore, to complete our proof, it suffices to use Lemmas 14.4 and 14.5. 
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 380

380 Geometric Structures and Riemann Hypothesis

14.2.2 Analytic and Geometric Truncations

We are finally ready to expose the following beautiful intrinsic relation between
geometric truncations using stability and analytic truncations using positivity.

Theorem 14.2 (Analytic and Geometric Truncations). Let p : [0, n] → R be a

normalized convex polygon, and let
 
T = T (p) = p(1), p(2) − p(1), . . . , p(n − 1) − p(n − 2), −p(n − 1)
be the corresponding element in a0 associated to SLn . Then, for all g ∈ SLn (A),
1(pg ≤ p) = ΛT (p) 1 (g).
 
(14.34)

Proof. This is a direct consequence of Proposition 14.1 and Proposition 14.3, on

the parabolic reduction for stability and a micro bridge, respectively. 

We end this discussion by mentioning that the above theorem makes it possible
to identify the geometric Eisenstein periods and the analytic Eisenstein periods,
defined as the integrations of Eisenstein series over the moduli spaces of vector
bundles and the integrations of truncated Eisenstein series over suitable funda-
mental domains for SLn , respectively. In later chapters, we will develop a parallel
theory for general reductive groups.
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 381

Chapter 15

Special Uniformity of Zeta Functions and

Weak Riemann Hypothesis

As an application of the relation between geometric and analytic truncations, we

first establish the special uniformity of zeta functions, claiming that the rank n zeta
functions coincides with the SLn -zeta functions. Then, we calculate the volume
of the moduli spaces of lattices of rank n and volume one. Finally, using all these
results, we prove a weak Riemann hypothesis for the above zeta functions.

15.1 Special Uniformity of Zeta Functions

15.1.1 Basic Properties of Truncations

We start with some generalizations of basic properties for analytic truncations ΛT ,

using Theorem 14.2 to remove the condition that T is sufficiently regular.

15.1.1.1 Idempotence

We first show that the operator ΛP,T is idempotent. As usual, we denote by C ⊂ a0

the positive Weyl chamber associated to SLn , and let T (p) ∈ a0 be the element
associated to a normalized convex polygon p : [0, n] → R.

Lemma 15.1. Let T ∈ C t {0} and let P be a parabolic subgroup of SLn . Then
ΛP,T ΛP,T 1 = ΛP,T 1,
 
(15.1)
ΛP,T (p) 1(p∗ ≤ p) = 1(p∗ ≤ p).
 

Proof. By Theorem 14.2, we may identify the geometric truncation 1(p∗ ≤ p) and
the analytic truncation ΛP,T 1, and hence it suffices to prove the second equality.
But for geometric truncations, this is simply a direct consequence of the proof of
Theorem 14.1. Indeed, for g ∈ SLn (A). when pgP ≤P p, by definition, the right

381
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 382

382 Geometric Structures and Riemann Hypothesis

hand side is one, and the left hand side becomes ΛP,T (p) 1 (g) = 1(pg ≤ p) = 1 by


Theorem 14.2 again. On the other hand, if pgP P p, by the proof of Theorem 14.1,
the right hand side becomes 0, while the right hand side gives ΛP,T (p) 0, and hence
becomes 0 as well. 

15.1.1.2 Self-Adjoint Property

Next, we show that ΛP,T is self-adjoint with respect to natural inner product.

Lemma 15.2. Let T be an element of C ∪ {0}.

(1) Let φ1 and φ2 be two continuous functions on SLn (F)\SLn (A). Assume that
φ1 is slowly increasing, and that φ2 is rapidly decreasing. Then
(ΛT φ1 , φ2 ) = (φ1 , ΛT φ2 ). (15.2)
(2) Let P be a parabolic subgroup of SLn with Levi decomposition P = N M. Let
φ1 and φ2 be two continuous functions on P(F)N(A)\SLn (A). Assume that φ1
is slowly increasing, and that φ2 is rapidly decreasing. Then
(ΛP,T φ1 , φ2 ) = (φ1 , ΛP,T φ2 ). (15.3)

Proof. Since a similar proof of (1) below works for (2) as well, we only give the
details of a proof for (1). For this, we use the proof of the self adjoint property
in [3, 4]. Indeed, since the inner product (ΛT φ1 , φ2 ) is defined by the absolutely
convergent integrals, we have
Z X
(ΛT φ1 , φ2 ) = (−1)n−1−|P|
G(F)\G(A)1 P
X Z
× φ1 (nδg)b
τP (H(δg) − T )φ2 (g) dn dg
δ∈P(F)\G(F) N(F)\N(A)
X Z Z
= (−1)n−1−|P| φ1 (ng)φ2 (g)b
τP (H(g) − T ) dg dn
P N(F)\N(A) P(F)\SLn (A)
X Z Z (15.4)
= (−1)n−1−|P|
φ1 (g)φ2 (ng)b
τP (H(g) − T ) dg dn
N(F)\N(A) P(F)\SLn (A)
ZP X
= (−1)n−1−|P|
SLn (F)\SLn (A) P
X Z
× φ1 (g)b
τP (H(δg) − T )φ2 (nδg) dn dg.
δ∈P(F)\SLn (F) N(F)\N(A)

This is simply (φ1 , ΛT φ2 ). 

December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 383

Special Uniformity of Zeta Functions and Weak RH 383

15.1.2 Analytic and Geometric Periods

We are now ready to prove the following fundamental

Lemma 15.3. Let T ∈ Ct{0} and denote by Σ(T ) ⊂ SLn (F)\SLn (A)/SK the com-
pact subset with ΛT 1 as its characteristic function. Then, for any Eisenstein series
E SLn /P (φ, λ, g) induced from an L2 -automorphic form φ on the Levi subgroup of a
parabolic subgroup P of SLn ,
Z Z
SLn /P
(φ, λ, g)dµ(g) = ΛT E SLn /P (φ, λ, g) dµ(g). (15.5)


E
Σ(T ) SLn (F)\SLn (A)/SK

Proof. We first need to justify that Σ(T ) makes sense. By Theorem 14.2, for all
g ∈ SLn (F)\SLn (A)/SK, 1(pg ≤ p) = ΛT (p) 1 (g). The right hand side, be-
 

ing taking values either zero or one, is easily a characteristic function of a com-
pact subset of DSLn ,F := SLn (F)\SLn (A)/SK. Hence Σ(T ) makes sense. Conse-
quently, by Lemmas 15.1 and 15.2, we have
Z Z
ΛT E SLn /P (φ, λ, g) dµ(g) = ΛT ΛT E SLn /P (φ, λ, g) dµ(g)



D DSLn ,F
Z SLn ,F  Z
Λ 1 (g) · Λ E SLn /P
E SLn /P (φ, λ, g)dµ(g)
 

= T
T
(φ, λ, g) dµ(g) =
DSLn ,F Σ(T )
as desired. 

Theorem 15.1 (Special Uniformity of Zeta Functions). Let F be a number

field. Then the rank n zeta function and the SLn zeta functions of F are (es-
sentially) the same. More precisely,
SL /P
ζF,n (s) = b
b ζF n n−1,1 (ns − n). (15.6)

Proof. This is a combination of several previous results. Indeed, by §§3.2.3,

8.1.1, 8.1.2 and 8.4.2, particularly, with λ = n−2 i=1 (si + 1)λi + (ns − n + 1)λn−1 ,
P
Z
ζF,n (s) =
b bSLn /Pn−1,1 (1, g, s)d(g)
E
MF,n [1]
Z !
= Norm(ns − n) · Res s1 =0,...,sn−2 =0 E SLn /P1,1,...,1 (1, g, s)d(g)
MF,n [1]
Z !
= Norm(ns − n) · Res s1 =0,...,sn−2 =0 Λ0 E SLn /P1,1,...,1 (1, g, s)d(g)
DSLn ,F
SL /P
  
= Norm(ns − n) · Res s1 =0,...,sn−2 = 0 ωSL n
F (λ) = b
ζF n n−1,1 (ns − n).
Here in the second, resp, the third, resp. the fourth, resp. the last, we have used
Theorem 15.1, resp. Theorem 8.2, resp. Lemma 15.3, resp. Definitions 8.12 and
8.13. 
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 384

384 Geometric Structures and Riemann Hypothesis

Similarly, if we introduce a parameter T ∈ Ct{0}, then the same constructions

lead to the following:

Definition 15.1. Let F be a number field.

(1) The T -version of a non-abelian zeta function of F is defined by

Z  0 
ζF,n
b T
(s) = eh (F,Λ) − 1 · vol(Λ) s dµ(Λ). (15.7)
Σ(T )

(2) Let (G, P) be a pair consisting of a reductive group of rank n and its maximal
parabolic subgroup. Denote by α p its associated simple root. The T-version
of the (G, P)-zeta function of F is defined by
  G,T 
ζF,T
b G/P
(ns − n) := Norm(s) · Res s1 =0,..., s[
p =0,...sn = 0
ωF (λ) (15.8)
where
X ehwλ−ρ,T i Y ζF (hλ, α∨ i)
ωG,T
F (λ) = .
b
(15.9)
α∈∆ hwλ − ρ, α i α>0,wα<0 b
∨
Q
w∈W ζF (hλ, α∨ i + 1)

In addition, a similar proof as above proves the following:

Corollary 15.1 (Special Uniformity of T -Zeta Functions). We have

SL /P
ζF,n
bT
(s) = b
ζF,Tn n−1,1 (ns − n). (15.10)

15.2 Weak Riemann Hypothesis for Non-Abelian Zeta Functions

15.2.1 Parabolic Reduction, Stability and Volumes

In the study of rank n zeta functions, the author formulates the following:

Conjecture 15.1 (Parabolic Reduction, Stability, and the Volumes). Let F be

a number field. Then the volume of the moduli space MF,n [1] of semi-stable OF -
lattices of rank n and volume one is given by

X X vF,n1 . . .b vF,nk
vol MF,n [1] = .
b
(−1)k−1 (15.11)
k≥1 n ,...,n >0
(n1 + n2 ) . . . (nk−1 + nk )
1 k
n1 +...+nk =n
m
Y
vF,m :=
Here b ζF (i) with b
b ζF (s).
vF,1 := Res s=1b
i=1

Theorem 15.2. The conjecture on parabolic reduction, stability, and the volumes
holds.
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 385

Special Uniformity of Zeta Functions and Weak RH 385

Proof. By (9.46) and (9.47), and more importantly, Theorem 15.1, we have
n
X X vF,k1 . . .b
b vF,k p 1
ζF,n (s) =
b (−1) p+r−1 ·
a=1 k1 ,...,k p >0
(k1 + k2 ) . . . (k p−1 + k p ) ns − n + a + k p
k1 +···+k p =n−a

ζF (ns − n + a)
×b
X 1 vF,l1 . . .b vF,lr
.
b
× ·
l1 ,...,lr >0
−ns + n − a + 1 + l1 (l1 + l2 ) . . . (lr−1 + lr ) (15.12)
l1 +···+lr =a−1

ζF,n (s)
Therefore, by Theorem 3.2(3), for the residue of non-abelian zeta function b
at s = 1, we have
 
vol MF,n [1] = Res s=1 bζF,n (s)

n
X X vF,k1 . . .b
b vF,k p 1
= (−1) p+r−1 · ζF (a)b
·b vF,a−1
a=2 k1 ,...,k p >0
(k1 + k2 ) . . . (k p−1 + k p ) a + k p
k1 +...+k p =n−a

X vF,k1 . . .b
b vF,k p 1
+ (−1) p+r−1 · ζF (1)
·b
k1 ,...,k p >0
(k1 + k2 ) . . . (k p−1 + k p ) 1 + k p
k1 +...+k p =n−1
n
X X vF,k1 . . .b
b vF,a
vF,k pb 1 (15.13)
= (−1) p+r−1 ·
a=2 k1 ,...,k p ,a>0
(k1 + k2 ) . . . (k p−1 + k p ) k p + a
k1 +...+k p +a=n

X vF,k1 . . .b
b vF,1
vF,k pb 1
+ (−1) p+r−1 ·
k1 ,...,k p ,1>0
(k1 + k2 ) . . . (k p−1 + k p ) k p + 1
k1 +...+k p +1=n
X X vF,n1 . . .b
vF,nk
=
b
(−1)k−1
k≥1 n1 ,...,nk >0,
(n1 + n2 ) . . . (nk−1 + nk )
n1 +...+nk =n

as desired. 

We end this with an inverse formula to (15.11) of Kontsevich-Soibelman.

Theorem 15.3. ( [61]) We have

n k
  X X 1 Y  
vol Mtot [1] = vol MQ,n j [1] .
Q,n
n (n +n2 ) · · · (n1 +. . .+nk ) · · · nk j=1
k=1 n1 +...+nk =n 1 1
n1 ,...,nk >0
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 386

386 Geometric Structures and Riemann Hypothesis

15.2.2 Existence of Stable O F -Lattices

Proposition 15.1 (Existence of Semi-Stable OF -Lattices). There exist stable

OF -lattices of rank n and volume one. In particular, the volume of the moduli
space MF,n [1] is strictly positive.

Proof. The second half is a direct of the first since the stability condition is open,
it suffices to prove the existence. As the proof below works for F as well, without
loss generality, we work only over Q. Then, it suffices to show that there is a stable
lattice in MQ,n [T ], for one and hence for all T . We use an induction on the rank
n. When n = 1, every rank one lattice is stable. Assume that all Λ ∈ MQ,n [1] are
not stable. Choose stable lattices Λi (i = 1, 2) of rank ni and degree di . Consider a
non-trivial extension of Λ2 by Λ1
0 → Λ1 → Λ → Λ2 → 0. (15.14)
Assume that the degree of Λ equals zero. Since Λ is non-stable, d1 > 0. Let Λ0 be
a sub lattice of Λ such that 0 ⊂ Λ0 ⊂ Λ is the canonical filtration of Λ. We obtain
an induced filtration
0 → Λ1 ∩ Λ0 → Λ0 → Λ0 /(Λ1 ∩ Λ0 ) → 0. (15.15)
Since Λ1 ∩ Λ0 ,→ Λ1 and Λ0 /(Λ1 ∩ Λ0 ) ,→ Λ2 , we have, using the stability of
the Λi ’s, the slope of Λ0 is bounded from below, resp. from above, by −d2 /n1 ,
resp. by d2 /n2 . Therefore, the composition of morphisms Λ0 ,→ Λ  Λ2 is an
isomorphism. This implies that (15.14) splits, a contradiction. 

For an alternative proof, see e.g. Theorem 1.1 of [99].

15.2.3 Weak Riemann Hypothesis

Theorem 15.4 (Weak Riemann Hypothesis). Let n ≥ 2.

(1) All but finite many zeros of the non-abelian zeta function bζQ,n (s) lie on the
1
central line <(s) = .
2
SL /P
(2) All but finite many zeros of the Weng zeta function bζQ n n−1,1 (s) lie on the
n
central line <(s) = − .
2
Proof. By Theorems 11.2 and 15.1, it suffices to verify the non-vanishing of the
volume of MQ,k [1] for 1 ≤ k ≤ n − 1. This comes from Theorem 15.2 and
Proposition 15.1. 

We will give a similar result for Chevalley groups over Q below.

December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 387

Chapter 16

Analytic and Geometric Truncations:

Reductive Groups

16.1 Lie Theoretic Preparation

16.1.1 Maximal Split Sub or Quotient Torus

Let F be an algebraic number field, a finite extension field of Q. Let G be a split

reductive group over F. Denote by g its associated Lie algebra. A subgroup P of
G is called parabolic if G/P is projective. Fix a minimal parabolic subgroup P0 of
G and denote its Levi subgroup by M0 . By definition, a parabolic subgroup P of G
is called standard if P0 ⊆ P. In this case, there exists one and only one maximal
reductive subgroup MP of P such that M0 ⊆ MP . Moreover, if NP denotes the
nilpotent radical of P, then
P = MP NP . (16.1)
We call MP , resp. (16.1), the Levi subgroup, or the Levi factor, resp. the Levi
decomposition, of P.
For a standard parabolic subgroup P of G, let AP be the maximal split torus in
the center ZP of MP , and let A0P be the maximal quotient split torus of MP , or the
same of MPab , where MPab denotes the maximal abelian quotient of MP . With the
splitness property, easily, we have
AP ' Hom(Gm , ZP ) × Gm A0P ' Hom Hom(MPab , Gm ), Gm (16.2)


and
where Hom denotes the group of morphisms for group schemes over F. Moreover,
the composition of the natural morphisms
AP ,→ ZP ,→ MP  MPab  A0P (16.3)
is an isogeny. Consequently, for the spaces
X∗ (AP ) := Hom(Gm , Z p ) X∗ (A0P ) := Hom Hom(MPab , Gm ), Z


and (16.4)

387
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 388

388 Geometric Structures and Riemann Hypothesis

(1) X∗ (AP ), resp. X∗ (A0P ), is the group of one-parameter subgroup in AP , resp. A0P .
(2) X∗ (AP ) and X∗ (A0P ) are free abelian groups of the same finite rank.
(3) Induced from (16.3), there is a natural injective morphism
X∗ (AP ) ,−→ X∗ (A0P ). (16.5)
Accordingly, set
X∗ (AP )R = X∗ (A0P )R =: aP . (16.6)

Let P ⊆ Q be two standard parabolic subgroups of G. Their associated AP , AQ

and A0P , A0Q are naturally related by the morphisms
AQ ,−→ AP −→ A0P − A0Q . (16.7)
In particular, induce from the injection AQ ,→ AP and surjection A0P  A0Q , there
exist naturally an embedding aQ ,→ aP and a surjection aP  aQ and hence the
canonical decompositions
aP = aQ
P ⊕ aQ and a∗P = aQ∗
P ⊕ aQ .
∗
(16.8)

More generally, let P ⊆ Q ⊆ R be three standard parabolic subgroups of G.

Applying above discussions, particularly (16.8), we obtain natural decompositions
aP = aQ R
P ⊕ aQ ⊕ aR and a∗P = aQ∗
P ⊕ aQ ⊕ aR .
R∗ ∗
(16.9)

P , [·]Q and [·]R the canonical projections of aP to aP , aQ ,

For later use, we denote [·]Q R Q R

Q
and aR , respectively. For simplicity, if P is fixed, we sometimes use [·] instead
of [·]Q
P.

16.1.2 Generalized Root System

For a standard parabolic subgroup P of G, denote the group of the characters of

AP by
X ∗ (AP ) := Hom(AP , Gm ). (16.10)
In addition,

(1) Let ΦP ⊂ aG∗ ∗ ∗

P ⊆ aP be the set of non-trivial elements of X (AP ), which occur
in the Lie algebra g of G.
(2) Let Φ+P ⊂ ΦP be the collection of non-trivial characters of AP which occur in
the subalgebra nP , the Lie algebra of the unipotent radical NP of P, and set
Φ−P := −Φ+P . (16.11)
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 389

Analytic and Geometric Truncations: Reductive Groups 389

If P is a general standard parabolic subgroup of G, ΦP may not be a root

system. In the special case when P = P0 is minimal, Φ0 := ΦP0 does form a root
system. Accordingly, denote by Φ+0 := Φ+P0 and Φ−0 := Φ−P0 the set of positive roots
and negative roots, respectively. Then

Φ0 = Φ+0 t Φ−0 . (16.12)

(3) Let ∆0 := ∆P0 ⊂ Φ+0 be the set of simple roots of (Φ0 , Φ+0 ).

It is well-known that ∆0 forms a basis of the R-linear space aG∗

P0 , that the name
+
of positive roots is justified by the inclusion Φ0 ⊂ α∈∆0 Z≥0 α, and that, for each
P
α ∈ Φ0 , there exists a corresponding coroot α∨ so that ∆∨0 := α∨ : α ∈ ∆0

becomes an R-basis of the real vector space

aG0 := aGP0 ⊂ aP0 = a0 . (16.13)

On the other hand, (even it may not be a root system,) many constructions for
a root system may be in parallel introduced for ΦP . To be more precise,

(4) Let ∆P ⊂ Φ+P be the set of non-trivial restrictions to AP , or equivalently, to aP

of simple roots in ∆0 . That is,
n o
∆P := α a ∈ a∗P : α ∈ ∆0 r{0}. (16.14)
P

(5) For each α ∈ ∆P , there exists one and only one β ∈ ∆0 such that α = β a .
P

Accordingly, we can define its co-version by α∨ := β∨ GP , the projection of

 
β∨ ∈ aG0 to aGP . In addition, write

∆∨P := α∨ : α ∈ ∆P .

(16.15)

∆∨P forms an R-basis of aGP , and denote its dual basis in aG∗
P by

∆P := $Gα : α ∈ ∆P .

b (16.16)

(6) There exists a canonical inner product (·, ·)P on aGP . Let WP be the Weyl group
generated by reflections σα on aGP associated to α ∈ ΦP defined by
(α, β)P
σα (β) := β − 2 α ∀β ∈ aGP . (16.17)
(α, α)P
X
∆P ⊂
One checks that b Z≥0 α, that WP is generated by σα with α ∈ ∆P , and that
α∈∆P
(·, ·)P is WP -invariant.
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 390

390 Geometric Structures and Riemann Hypothesis

16.1.3 Partial Order Induced by Positive Weyl Cone

Recall that, following Arthur, we have the following:

Definition 16.1.

(1) The (acute) Weyl chamber in aGP is defined by

n o
P := H ∈ aP : hα, Hi > 0 ∀α ∈ ∆P .
aG+ G
(16.18)

(2) The (obtuse) Weyl cone in aGP is defined by

+ G
n o
aP := H ∈ aGP : h$Gα , Hi > 0 ∀α ∈ ∆P . (16.19)
+ G
(3) The characteristic functions of subsets aG+
P and aP in aGP is denoted by τGP
τGP , respectively.
and b
+ G
For example, when P = P0 , the sets aG+
0 and a0 are the so-called positive
Weyl chamber and positive Weyl cone, respectively.

Lemma 16.1.

(1) The positive chamber is contained in the positive cone. Namely,

+ G
aG+
P ⊆ aP . (16.20)
(2) W acts transitively on all Weyl chambers. More precisely,
G
aG0 = WaG+
0 ∪α∈∆0 Pα . (16.21)

Here Pα denotes the hyperplane in aG0 which is perpendicular to α.

(3) + aG0 is the dual cone of aG+
0 with respect to (·, ·). That is to say,
+ G
x∈ a0 i f and only i f (x, y) ≥ 0 ∀y ∈ aG+
0 (16.22)
(4) We have
+ G
x ∈ aG+
0 i f and only i f x − w(x) ∈ a0 ∀w ∈ W. (16.23)

Proof. (1) comes from the fact that every $Gα ∈ b ∆P belongs to α∈∆P Z≥0 α. (2)
P
is a standard fact. For details, please refer to §10.1 of [44]. (3) comes from the
definition of fundamental weights $α . (4) is rather standard as well. For details,
please see Proposition 18 of Chapter VI of Bourbaki [16]. 

Definition 16.2. Let x, y ∈ aG0 . x is defined to be less than or equal to y, denoted

by x ≤ y, if y − x ∈ aG+
0 .
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 391

Analytic and Geometric Truncations: Reductive Groups 391

Obviously, ≤ defines a partial order on aG0 . By Lemma 16.1(4),

x ∈ aG+
0 t {0} if and only if x ≥ w(x) ∀w ∈ W. (16.24)

Example 16.1. In the case when G = SLn , the cone aG+

0 t {0} is given by

n n
X o
0 t {0} := (x1 , x2 , . . . , xn ) ∈ R : x1 ≥ x2 ≥ . . . ≥ xn ,
aG+ xj = 0
n

j=1

and the standard inner product (x, y) =

Pn
i=1 xi yi can be written as

(x, y) =(x1 − x2 )y1 + (x2 − x3 )(y1 + y2 ) + . . .

+ (xn−1 − xn )(y1 + . . . + yn−1 ) + xn (y1 + . . . + yn ).
This implies that
i
X n
X
+ G
n o
a0 t {0} = (y1 , . . . , yn ) ∈ a0 : y j ≥ 0 ∀1 ≤ i ≤ n − 1 & yj = 0 .
j=1 j=1

Proposition 16.1. ( [7]) For x, y ∈ aG0 , the conditions below are equivalent.

(1) x ≤ y.
dx ⊆ Wy.
(2) W c Here W x denotes the W-orbit of x, and b
S denotes the convex cone
of a subset S ⊂ aG0 .
(3) φ(x) ≤ φ(y) for all W-invariant convex function φ : aG0 → R.
c ⊆ Gy.
(4) Gx c Here Gx denote the G-orbit of x for the associated adjoint action.

(5) ϕ(x) ≤ ϕ(y) for all G-invariant convex function ϕ : aG0 → R.

Proof. (2) ⇒ (1) This is clear. Indeed, W dx ⊆ Wy c implies that x ∈ Wy.c Hence,
x = w∈W aw w(y) with aw ≥ 0 and w∈W aw = 1. Thus, by (16.24), we have
P P
x = w∈W aw w(y) ≤ ( w∈W aw )y = y.
P P

(1) ⇒ (2) By continuity, it suffices to assume that x is an interior point of aG0

and that x − y is an interior point of + aG0 . Let z be an intersection of the boundary
of aG0 and the line L xy connecting x and y, and let [y, z] be the part of line segment
within L xy starting at y and ending with z. We much show that this line segment
[y, z] is entirely contained in Wy.
c

Claim 16.1. There exists a constant c(y, z) such that, if x = tz + (1 − t)y (0 ≤ t ≤ 1)

is a point in the line segment [y, z], then c(x, z) ≥ c(y, z), and for t ≤ c(y, z), we
have x ∈ Wy.
c
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 392

392 Geometric Structures and Riemann Hypothesis

Proof. Let αi be the simple roots normalized to have length one. Then the αi ’s
are the unit normal vectors to the faces of aG0 and form a basis of + aG0 . Since y is
assumed interior to aG0 , we have (x, αi ) > 0 for all i. But y − z is assumed interior
to + aG0 , we have y − z = ri=1 ai αi with ai > 0. Thus, if denote by σi = σαi the
P
tai
reflection associated to αi , and, for 0 ≤ t ≤ 1, introduce the numbers bi :=
2(y, αi )
Xr
and set b := 1 − bi , we have, since σi (y) = y − 2(y, αi )αi ,
i=1
r
X r
X r
X
by + bi σi (y) = y + bi (σi (y) − y) = y − 2 bi (yi , αi )αi
i=1 i=1 i=1
r
(16.25)
X
= y−t ai αi = tz + (1 − t)y = x.
i=1

Hence x ∈ Wy c provided bi ≥ 0 and b ≥ 0, and this will hold if, for all i, we
(y, αi ) (y, αi )
have 0 ≤ t ≤ 2 . Set now ci (y, z) := 2 . We want to know the behavior
rai rai
of ci (y, z)’s when y runs over points in the line segment [y, z]. Obviously, when
replacing y by the variable point x = tz + (1 − t)y, ai changes to (1 − t)ai and
(x, αi ) = t(z, αi ) + (1 − t)(y, αi ) ≥ (1 − t)(y, αi ) (16.26)
since z ∈ aG0 and t ≥ 0. Therefore ci (x, z) ≥ ci (y, z) and it suffices for us to take
c(y, z) := mini ci (y, z). 

Following this claim, since the relation x ∈ Wy c implies that W x ⊂ Wy, c a

1
finite number N of repetitions, say N > , will then prove that the whole line
c(y, z)
segment [y, z] is contained in Wy.
c
To understand why (3), (4) and (5) should be related, let us first recall a gener-
alization of Scour-Horn’s result by Kostant and a result of Aniyah-Bott on convex
invariant functions, respectively.

Lemma 16.2.

(1) Let x ∈ aG0 . Denote by Gx the G-orbit of x under the adjoint action. Then,
with respect to the induced orthogonal projection π : g → aG0 ,

π(Gx) = W
dx. (16.27)
(2) Let φ be a W-invariant convex function on t and ϕ be the corresponding G-
invariant function on g. Then ϕ is also convex.
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 393

Analytic and Geometric Truncations: Reductive Groups 393

Proof. We only prove (2) while refer (1) to [59]. For a function on Rn , denote
by Γ( f ) be the region above its graph, i.e. the collection of all points (x, y) with
x ∈ Rn and y ∈ R such that y > f (x). Then the convexity of the function f is
equivalent to the convexity of the grape Γ( f ). Recall that Γ is convex if every
boundary point a has a supporting hyperplane Ga , i.e. Γ is contained in one of the
two half-spaces complementary to Ha . Apply this to the functions φ, ϕ and their
associated regions Γ(φ) ⊂ aG0 ⊕ R and Γ(ϕ) ⊂ g ⊕ R. Because ϕ is G-invariant, so is
Γ(ϕ). Hence it suffices to prove the existence of a supporting hyperplane to Γ(ϕ)
at boundary points (λ, y) of Γ(φ). Since γ(φ) is assumed to be convex, there exists
a supporting hyperplane H ⊂ aG0 ⊕ R. Let then H 0 = π−1 (H) ⊂ g ⊕ R where π is
the orthogonal projection. Take then (x, y) ∈ Γ(ϕ) satisfying y > ϕ(x). By (1), or
more precisely, by π(Gx) ⊆ W dx, we have ϕ(x) ≥ φ(π(x)) since φ is convex. Hence
y > φ(pi(x)) and (π, x) ∈ Γ(φ). Therefore (π, x) is in one side of H and hence (x, y)
is on the corresponding side of H 0 . This means that H 0 is the required supporting
hyperplane for Γ(ϕ). 

Obviously, (1) is equivalent to (5), and (2) implies (3) and (4), and (3) implies
(5) and (4) implies (5). By Lemma 16.2(1), resp. (2), we have (4) implies (2),
resp. (5) implies (3). It remains to show that (3) implies (2). For this, we take
φ(x) := w∈W exp(w(tx), ei ) where t > 0 and ei are a basis of aG0 , choosing from
P
the edges of this cone. Note that for x, y ∈ aG0 , by Lemma 16.24(3), (4), we have
(x, ei ) ≥ (w(x), ei ) and (y, ei ) ≥ (w(y), ei ). Thus, by the continuity, when x and
y are interior to aG0 , we have (x, ei ) < (w(x), ei ) and (y, ei ) < (w(y), ei ) if w , 1.
Consequently, for sufficiently large t, the first term with w = 1 in the sum defining
φ is dominant. Hence by φ(x) ≤ φ(y) we have (y, ei ) ≥ (x, ei ) for all i. Therefore,
y − x ∈ + aG0 and x ≤ y. 

Corollary 16.1. Let x, y ∈ aG0 .

(1) x ≤ y if and only if ρ(x) ≤ ρ(y) for all unitary representation ρ of G.

(2) If φ(x) = φ(y) for all W-invariant convex functions φ, then W x = Wy.

Proof. (1) Let ρ : G → U(n) be an irreducible representation. Let λ1 , . . . , λn ∈ t

be the associated weights. Then any x ∈ t gives rise to a hermitian matrix with
eigenvalues x j = (x, λ j ). If λ1 is the maximal weight, then λ ∈ aG+ 0 , and for all
j > 1, we have λ1 > λ j , or the same λ1 − λ j ∈ + aG0 . This means that, if x ∈ aG+ 0 ,
then x1 ≥ x j and hence x1 is the largest eigenvalue. Hence if x, y ∈ aG+ 0 and
ρ(x) ≤ ρ(y), then x1 ≤ y1 , or equivalently, (x, λ1 ) ≤ (y, λ1 ). If we let ρ run over all
irreducible representations, λ1 runs over all integral dominant weights and these
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 394

394 Geometric Structures and Riemann Hypothesis

0 , since there are r basic integral wights ei that lie in the edges of a0 and
span aG+ G+

generate it.
(2) If φ(x) = φ(y) for all W-invariant convex functions φ, then W dx = Wy.c
Hence, the extreme points of these two convex polyhedra must coincide. But the
extreme points of W dx are certainly among the finite set W x. Hence W x and Wy
intersect and so coincide. 

16.1.4 Classical Example

In the case when G = Un , the partial order above can be described explicitly in
concrete terms, say, following Atiyah-Bott [7].
Let µ = (µ1 , µ2 , . . . , µn ) and λ := (λ1 , λ2 , . . . , λn ) ∈ Rn . By definition, µ ≤ λ,
if, after arranging each sequence in decreasing order, we have
P
 j=1 µ j ≤ j=1 λ j
i Pi


 1≤i≤n−1
(0) 
 n µj = n λj

P P
j=1 j=1

Lemma 16.3. ([42]) For λ, µ ∈ Rn , the following statements are equivalent.

i
X i
X
(1) For every convex function f : R → R, f (µ j ) ≤ f (λ j ).
j=1 j=1
(2) For every doubly stochastic matrix of size n, µ = Pλ.

Recall that a real square matrix M = (mi j ) is called stochastic if


mi j ≥ 0 ∀ 1 ≤ i, j ≤ n



 (16.28)
 n mi j = 1

 P
∀1≤i≤n
j=1

and that M is called doubly stochastic, if both M and its transpose t M are
stochastic.

Lemma 16.4. ([11]) The subset of doubly stochastic matrices of size n is the
convex hull of the permutation matrices.

Consequently, we have

Lemma 16.5. (1) and hence (2) are equivalent to

(3) S[n (µ) ⊆ Sn (λ). Here, as usual, Sn is the symmetric group acting on the co-
[
ordinates of Rn via σ : (x1 , x2 , . . . , xn ) 7→ (xσ(1) , xσ(2) , . . . , xσ(n) ).
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 395

Analytic and Geometric Truncations: Reductive Groups 395

On the other hand, the geometric notion of convexity can be transformed into
the dual statement on convex functions using the following well-known:

Lemma 16.6. Let C ⊂ Rn . Then the following conditions are equivalent.

(a) A point x belongs to C.

b

(b) For all convex function φ : Rn → R, we have φ(x) ≤ sup φ(y).

y∈C

Consequently, (1) and hence (2) and (3) are equivalent to

(4) For all convex symmetric functions φ on Rn , φ(µ) ≤ φ(λ).

Indeed, by the equivalence of (a) and (b), we see easily that (3) implies (4).
On the other hand, if we take φ(x1 , x2 , . . . , xn ) := ni=1 f (xi ), then (4) implies (1).
P

Lemma 16.7. Let λ, µ ∈ Rn .

(c) (Schur) If µ j (1 ≤ j ≤ n) are the diagonal entries of a hermitian matrix whose

eigenvalues are λ j (1 ≤ j ≤ n), then µ ≤ λ.
(d) (Horn) If µ ≤ λ, then there exists a hermitian matrix with diagonal entries
µ j (1 ≤ j ≤ n) and eigenvalues λ j (1 ≤ j ≤ n).

Consequently, (0) and hence (1), (2), (3), (4) are equivalent to

(5) The λ j ’s are eigen values of a hermitian matrix with diagonal entries the µ j ’s.

Now we are ready to state the following:

Theorem 16.1. For any two elements λ, µ ∈ Rn , the condition µ ≤ λ and hence
(1), (2), (3), (4) (5) are all equivalent to either of the following two conditions.
c ⊆ [λ],
(6) [µ] c where [λ] denotes the conjugacy class of hermitian matrices whose
eigenvalues are the λ j ’s.
(7) For all A ∈ [µ] and B ∈ [λ], and for all convex Sn -invariant function ϕ on the
space of hermitian matrices, ϕ(A) ≤ ϕ(B).

Proof. (0) ⇔ (6). A hermitian matrix B is determined, up to conjugacy by U(n),

by the unordered set of its eigenvalues, or equivalently, by the orbit Sn (λ) in Rn .
And if Sn (µ) corresponds to the conjugacy class [A] of a hermitian matrix A, then
(3) clearly implies that [A] lies in the convex hall [B].
c Conversely, if a diagonal
matrix B, with diagonal entries, or better, with eigenvalues µ j ’s, lies in the convex
hall [B],
c it must lie in the convex hull of the diagonal parts of the matrices in [B].
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 396

396 Geometric Structures and Riemann Hypothesis

But by (c) above, this means that µ ∈ Rn is in the convex hull of Sn (λ) for the
eigenvalues λ j of A.
(6) ⇒ (7) is trivial. Furthermore, by setting ϕ(B) = φ(λ), then (4) also implies
(7). So what is left is that (7) implies (4).! This is classical. Indeed, for every con-
α0
!
A B
vex invariant function ϕ, we have ϕ ≥ϕ for a blocked skew-hermitian
CD 0δ
matrix with α and δ the central components of A and ! D respectively. This follows
α0
from Horn’s theorem and the observation that is the convex hull of the Sn -
0δ
!
A B
orbit of the diagonal part of . As the same argument works for all types of
CD
blocked matrices, we are done. 

16.1.5 Relative ‘Root System’

The constructions above admit some natural generalizations to the pairs (P, Q)
consisting of standard parabolic subgroups P, Q of G satisfying P ⊆ Q. To start
with, recall that there are canonical morphisms
A0 ←- AP ,→ MP ,→ P = MP NP ,→ Q = MQ NQ ←- MQ , (16.29)
and that the elements of ΦP , resp. Φ+P , resp. ∆P are the restrictions of the roots in
Φ, resp. the positive roots in Φ+ , resp. the simple roots in ∆, to AP . Accordingly,
we let ΦQ P = ΦP∩MQ , resp. ΦP
Q+
= Φ+P∩MQ , resp. ∆Q P = ∆P∩MQ , be the set of
α ∈ ΦP , resp. α ∈ Φ+P , resp. α ∈ ∆P , which appear in the Lie algebra mQ of the
Levi factor MQ of Q. Indeed, the elements of ∆P are non-trivial characters of AP
which appear in nP , and there are natural morphisms nP ,→ p ,→ q ←- mQ , hence
nP ∩ mQ makes sense in q, and so does ∆Q P . Here, as the notation stands, p is the
Lie algebra of P, etc....
In terms of Lie spaces, there is a natural decomposition aP = aQ P ⊕ aQ , and the
set ∆Q
P forms an R-basis of a Q∗
P . In addition, being elements of ∆ P , each α ∈ ∆P
Q

admits a natural companion α∨ , the projection of the coroot β∨ for a uniquely

determined simple root β ∈ ∆0 satisfying β aG = α. By an abuse of notation, we
P

use the same notation α∨ to denote the projection of β∨ to aQ

P , and hence obtain a
natural companion
∨
n o
∆QP := α : α ∈ ∆P .
∨ Q
(16.30)
∨
Obviously, ∆Q Q
P forms an R-basis of aP . Accordingly, we set
∗
n o
∆Q∨
P := $αQ : α ∈ ∆Q P (16.31)
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 397

Analytic and Geometric Truncations: Reductive Groups 397

Q∨
P with respect to ∆P .
be the dual R-basis of aQ∗
Induced from the W-invariant inner product (·, ·) on a, we obtain a natural inner
product (·, ·)Q Q Q
P on aP by taking the restriction. Set WP be the group generated by
the reflections σa on the space aP defined by
Q

(β, α)
σα : β 7−→ β − 2 α ∀α ∈ ΦQ
P , ∀β ∈ aP
Q
(16.32)
(α, α)
since (β, α) = (β, α)QP and (α, α) = (α, α)P . Obviously, (·, ·)P is WP -invariant and
Q Q Q

one checks that WP is generated by σα for all α ∈ ∆P .

Q Q

Example 16.2. Let (P, Q, R) be a triple of standard parabolic subgroups of G

satisfying P ⊆ Q ⊆ R. We have
aP = aQ R
P ⊕ aQ ⊕ aR and aRP = aQ
P ⊕ aQ .
R
(16.33)
Naturally, MQ ⊆ MR and ∆Q P ⊆ ∆P . In addition, for a H ∈ aP with the decomposi-
R R

tion H = [H]P + [H]Q , we have

Q R

(1) If α ∈ ∆Q
P , by definition,

(α, H) = (α, [H]Q

P ) + (α, [H]Q ) = (α, [H]P ).
R Q
(16.34)
(2) If α ∈ ∆RQ , or better, for $Rα ∈ ∆R∨ ∗
Q , by definition,

($Rα , H) = ($Rα , [H]Q

P ) + ($α , [H]Q ) = ($α , [H]Q ).
R R R R
(16.35)

16.1.6 Relative Positive Chamber and Cone

Let (P, Q) be a pair of standard parabolic subgroups of G satisfying P ⊆ Q. Even
ΣQ
P := (ΦP , ΦP , ∆P , WP ) in general does not form a root system, as we see in
Q Q+ Q Q

§7.1, many structures for the root systems work for the relative setting as well.

Definition 16.3 (Definition 7.2). (1) The relative (acute) Weyl chamber aQ+
P is
defined by

P := H ∈ aP : hα, Hi > 0 ∀α ∈ ∆P .
aQ+
 Q Q
(16.36)
(2) The relative (obtuse) Weyl cone + aQ
P is defined by
+ Q
aP := H ∈ aP : h$αQ , Hi > 0 ∀α ∈ ∆Q
Q
P .

(16.37)
+ Q
(3) The characteristic functions of the subsets aQ+ Q
P and aP in aP are denoted by
τP and b
Q
τP , respectively.
Q

Lemma 16.8. ([3, 4]) Let P ⊆ Q be standard parabolic subgroups of G.

December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 398

398 Geometric Structures and Riemann Hypothesis

(1) The positive Weyl chamber is contained in the positive Weyl cone. That is,
+ Q
P ⊆ aP .
aQ+ (16.38)
(2) The positive Weyl cone is the dual cone of positive Weyl chamber. That is, + aQ
P
is the dual cone of aQ+ Q
P with respect to the inner product (·, ·) on aP
+ Q
n o
aP = x ∈ aQ P : (x, aP ) ⊂ R>0 .
Q+
(16.39)

P , for every α ∈ ∆P ,
Proof. (1) First we note that, for a fixed H ∈ aQ Q

(a) if either (α, H) > 0 or ($αQ , H) > 0, then, ($αQ , H) > 0 for all α ∈ ∆QP.
(b) if either (α, H) ≤ 0 or ($αQ , H) ≤ 0, then, $αQ , H ≤ 0 for all α ∈ ∆QP .

Since for H ∈ aQ+ P , hα, Hi > 0 for every α ∈ ∆P . Hence, from (a) above,
Q

($α , H) > 0 for all α ∈ ∆P . That is to say, H ∈ + aP .

Q Q Q

(2) Recall that, if V be a finite dimensional R-vector space with an inner

product (·, ·), we may identify V with V ∗ by the map v 7→ (v, ·). Thus if we let
e1 , . . . , en be an R-basis of V, the dual basis e1 , . . . , en of V ∗ satisfies the property
(ei , e j ) = δi j for all i, j = 1, . . . , n. Indeed, if we let v1 , . . . , vn be elements of
V such that (vi , ·) = ei for i = 1, . . . , n. Clearly, δi j = (vi , e j ). Consequently,
under the identification of V ∗ with V using (·, ·), we have vi = ei for i = 1, . . . , n.
Therefore, it suffices to verify the following:

Claim 16.2. For two elements α, β of ∆Q P , there exists a strictly positive constant
cαβ such that
α, $βQ QP = cαβ · δαβ .


(16.40)
Here, as usual δαβ denotes the Weierstrass delta symbol.
n o
Proof. By definition, $αQ : α ∈ ∆Q P is the dual basis of aQ P for the basis
n o
α : α ∈ ∆P of aP . Furthermore, β = [γ ]P ∈ aP ⊆ a0 ⊕ aP ⊕ aG∗
∨ Q Q∗ ∨ ∨ Q Q∗ P∗ Q∗
Q = a0 for
G∗

a unique simple root γ ∈ ∆0 . Therefore,

 Q 2  Q Q 2  Q Q
δαβ = $αQ , [γ∨ ]Q
P P = $α , [γ]Q P P = $α , [γ]QP P . (16.41)
(γ, γ) (γ, γ)
Recall that, by definition, [γ]Q
P is obtained first as [γ]P , the restriction of simple
root γ ∈ ∆0 to a0 ←- aP = aQ P ⊕ aQ , or equivalently, a character of AP appears in
nP , then viewed as an element of ∆Qp by checking whether it is belong to mQ ∩
nP . Since every elements in ∆Q P are obtained in this way and, conversely, this
2  Q Q
process does produce an element of ∆Q P ∪ {0}. Therefore, δαβ = $ ,β .
(γ, γ) α P
(γ, γ)
Consequently, it suffices to set cαβ := . This proves the claim and hence the
2
proposition as well. 
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 399

Analytic and Geometric Truncations: Reductive Groups 399

Moreover, we have the following standard facts for the Coxeter groups WPQ .

Proposition 16.2.
 
(1) Let σ ∈ WPQ , then σ aQ+ aQ+
S
T
σ∈WPQ P P is a union of finite number of hy-
perplanes of aQ
P , all passing through the origin. We call them the walls.
[
P =
(2) aQ Q σ aP
Q+

.
σ∈WP

In particular, aQ+
P is a ‘fundamental domain’ for the action of WPQ on the metrized

R linear space aP , (·, ·)P .
Q Q

16.1.7 Partial Order in Relative Theory

To facilitate ensuing discussion, we now introduce a partial order in aQ

P.

Definition 16.4. Let P ⊆ Q be standard parabolic subgroups of G. For two ele-

ments x, y in aG0 , we say that x is small or equal to y with respect to (P, Q), denoted
as x ≤Q
P y, if
+ Q
P ∈ aP .
[x − y]Q (16.42)

Similar to Proposition 16.1 for classical theory recalled in §16.1.3, we have

Proposition 16.3.

(1) Let x be an element of aQ

P . Then,

x ∈ aQ+
P if and only if x − w(x) ∈ + aQ
P ∀ w ∈ WPQ . (16.43)

In particular, if x ∈ aQ+ Q
P , then x ≥ w(x) for all w ∈ WP .
Q
(2) Let x, y be two elements in aP . Then
Q
x≤y if and only if x∈W
d
P y. (16.44)

Here WPQ y
denotes the WPQ -orbit
of the element y, and for any subset S ∈ aQ
P,
Sb denotes the convex cone of S in aQ
P .

Proof. (1) When (P, Q) = (P0 , G), this is recalled in Lemma 16.1, a standard fact
given in Bourbaki [16], Chapter VI, Proposition 18. However, the proof there
works for general Coxeter groups.
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 400

400 Geometric Structures and Riemann Hypothesis

(2) This may be proved following that of Proposition 16.1. Indeed, for a fixed
y ∈ aQ+
P , we may consider the set
+

P;≤y := x ∈ aP : x ≤ y .
aQ
 Q+
(16.45)

Then (2) is equivalent to the claim that

+

P;≤y = aP ∩ Wy.
aQ Q+ c (16.46)

One direction is easy. Indeed, if x ∈ Wy,

c then
X X
x= aw w(x) aw ≥ 0, aw = 1. (16.47)
w∈WPQ w∈WPQ

Hence, by (1),
X X
y= aw y ≥ aw w(y) = y. (16.48)
w∈WPQ w∈WPQ

To prove the converse, by the continuity, it is sufficient to assume that y is an

+ Q
P and that y − x is an interior point of aP . The straight line xy
interior point of aQ+
Q+
then meets a certain wall of aP . We must show that the whole interval [y, z] in line
c Since x ∈ Wy
xy is included in Wy. c is transitive (with respect to the W Q action), it
P
suffices to show that there exists a constant c(y, z) so that, if x = tz + (1 − t)y with
t ∈ [0, 1], is a point of [z, y], then

(i) c(x, z) ≥ c(y, z), and

(ii) x ∈ Wy
c if t ≤ c(y, z).

If so, a finite number N of repetitions, where N −1 < c(y, z) then proves that the
whole interval [z, y] belongs to Wy.
c

P , we have (y, α) > 0 for all α ∈P

Since y is in the interior of aQ+ ∆QP . Moreover,
+ Q
since y − z is assumed to be in the interior of aP , we have y − z = α∈∆Q aα α with
P

aα > 0 for all α ∈ ∆QP . Now let wα ∈ WP be the reflection in the wall (x, α) = 0 so
Q

(y, α)Q (α, α)Q

that wα (y) = y − 2 P
α, P
and t = 1 − α∈∆Q ba . Then
P
we set bα := taα
(α, α)P
Q
(y, α)P
Q P

X X X (y, α)Q
by + bα wα (y) = y + bα (wα (y) − y) = y − 2 ba P

α α α (α, α)Q
P
X
= y−t aα α = tz + (1 − t)y = x.
α
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 401

Analytic and Geometric Truncations: Reductive Groups 401

Hence x ∈ Wy c if bα ≥ 0 and b ≥ 0, and this will hold if, for all α ∈ ∆Q ,

P
(y, α)Q
P 1
0≤t≤ · . Accordingly, we are led to the problem on how the
(α, α)Q
P aα ∆QP
following system of equations depend on y when we very y in the interval [y, z]
(y, α)Q 1
cα (y, z) = P
· ∀α ∈ ∆Q
P.
(α, α)Q
P aα ∆Q
P

Replacing y by the variable point x0 = tz + (1 − t)y, aα is changed to (1 − t)aα and

(y, α)Q (y, α)Q
(x 0
, α)Q
P = t(z, α)Q
P + (1 − t) P
≥ (1 − t) P
(α, α)Q
P (α, α)Q
P

P and t ≥ 0. Therefore, cα (y , z) ≥ cα (y, z) and the proof completes if

since z ∈ aQ+ 0

we set c(y, z) = minα∈∆Q cα (y, z). 

P

16.2 Arithmetic Principal Torsors and Their Stability

16.2.1 Compatible Metrics

In this section, we first recall some basic facts on maximal compact subgroups
and the Carton involution associated to algebraic groups, mainly following [15]
and [31]. Then we introduce a notion of canonical metrics associated to the Carton
involution. When the groups involved are semi-simple, all these are classical.

16.2.1.1 Maximal Compact Subgroup

Let G be a real Lie group with finitely many connected components. Then any
compact subgroup of G is contained in a maximal one. If K is a maximal one, G
is diffeomorphic to the direct product of K with a euclidean space, G/G0 = K/K 0 ,
and any two maximal compact subgroups are conjugate in G.
If G1 is a closed normal subgroup of G, with finitely many connected compo-
nents, then the maximal compact subgroups of G1 are the intersections of G1 with
the maximal compact subgroups of G. Similarly, if G1 is a closed subgroup of
G with finitely many connected components such that all maximal compact sub-
groups of G are conjugate by elements of G1 , the maximal compact subgroups of
G1 are the intersections of G1 with the maximal compact subgroups of G. Conse-
quently, in both cases, by taking a maximal compact subgroup K of G containing
a maximal one of G1 , we see that G1 ∩ K is maximal compact in G1 for at least
one and hence for all maximal compact subgroups K of G by conjugacy.
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 402

402 Geometric Structures and Riemann Hypothesis

More generally, if G → G0 is a surjective morphism (of Lie groups) whose

kernel has finitely many connected components, the maximal compact subgroups
of G0 are the images of the maximal compact subgroups of G.

16.2.1.2 The Cartan Involution

Let G be an algebraic group defined over a base field k ⊆ R. Denote by RG the

radical of G and RuG the unipotent radical of G and Rd G the so-called split radical
of G, namely, the greatest connected k-split subgroup of RG. By definition, a Levi
subgroup of G is a maximal reductive k-subgroup of G. Let G1 = χ∈X(G)k kerχ2 ,
T
with X(G)k the group of k-morphism of G into Gm . The group G1 is normal in G,
defined over k. Note that for a character χ in X(G)k , its restriction to G1 is of order

0
≤ 2, hence is trivial on (G1 )0 . Consequently, (G1 )0 = χ∈X(G)k kerχ . Since any
T
character in X(G)k is trivial on RuG, we have G1 = L1 nRuG for any Levi subgroup
L of G. Hence, if A is a maximal k-split torus of RG, then G(R) = A(R)0 n G(R)1
and G(R)1 contains every compact subgroup of G(R). More generally, if A1 and
A2 are two k-tori in RG such that A1 is k-split, A1 A2 is a torus and A1 ∩ A2 is
finite, then here exists a normal k-subgroup G1 of G containing A2 and G1 such
that G(R) = A1 (R)0 n G1 (R).
Therefore, if P is a parabolic k-subgroup of G and A is a maximal k-split torus
of the split radical Rd G of G, then, for a maximal compact subgroup K of G(R),
we have K ∩ P is a maximal compact subgroup of P(R) and G(R) = K P(R) =
K A(R)0 P(R). Moreover, if K a P(R)0 = K a0 P(R)0 for some a, a0 ∈ A(R)0 ,
then a = a0 and the map G(R) → A(R)0 sending g to a = a(g) characterized by
g ∈ K a P(R)0 is real analytic.
As a direct consequence, when G is a reductive group, there exists one and
only one involutive automorphism θK of G(R) satisfying the following conditions.

(1) θK is ”algebraic,” i.e. the restriction to G(R) of an involutive automorphism

of algebraic groups of the Zariski-closure of G(R) in G.
(2) The fixed point set of θK is K.
(3) If G1 is a normal R-subgroup of G, then θK (G1 (R)) = G1 (R).
(4) θK leaves [g, g] and c = center(g) stable. Here, we use the same θK to denote
the induced involution on g := Lie(G).
(5) If o is the (-1)-eigenspace of θK in c, then V = exp o is a split component of G.
(6) If p is the (-1)-eigenspace of θK in Lie(G(R)) and set k = Lie(K), then there is
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 403

Analytic and Geometric Truncations: Reductive Groups 403

a decomposition Lie(G(R)) = k ⊕ p and

[k, k] ⊆ k, [k, p] ⊆ p, [p, p] ⊆ k, pk ⊂ p (∀k ∈ K).
(7) The map (k, X) 7→ k · exp(X) is an isomorphism of analytic manifolds of K × p
onto G(R).

Note that in the case when G is semi-simple, θK is the usual Cartan involution.
Motivate by this, we call θK the Cartan involution of G(R) with respect to K.
Moreover, the existence of the Cartan involution θK implies an existence of a non-
degenerate symmetric bilinear form B on gC × gC satisfying the follows.

(a) B is invariant under G and θK , and is real on g × g.

(b) The quadratic form of B is positive definite on p and negative definite on k. In
particular, if we set
(X, Y) := −B(X, θK y) and kXk2 := −B(X, θK X) ∀X, Y ∈ g (16.49)
then (·, ·) is a positive definite K-invariant and θK -invariant scalar product for
g × g with k · k its associated norm.
(c) If we set g1 := Lie(G1 ) and let B1 and θ1K be the restriction of B and θK to
g1 × g1 and G1 , respectively, then (G1 , K, θ1K , B1 ) inherit all the properties of
(G, K, θK , B) above.

In addition, since B is G-invariant, the following infinitesimal invariance holds.

(d) B is characteristic. That is,

B(φ(X), φ(Y)) = B(X, Y) ∀φ ∈ AutLie (g) ∀X, Y ∈ g
(16.50)
B([X, Y], Z) = B(X, [Y, Z]) ∀X, Y, Z ∈ g.

Therefore, [g, g] and c are mutually orthogonal with respect to B. In addition,

since B is θK -invariant, k and p is mutually orthogonal. Conversely, we may recon-
struct the bilinear form B using these conditions. To be more precisely, starting
with the Cartan-Killing form on [g, g], we may extend it to obtain B as the direct
sum of the Cartan-Killing form with a symmetric non-degenerate bilinear form on
c, which is negative definite on c ∩ k and positive definite on c ∩ p. Finally, we may
extend this latest B to the total space gC × gC . For later use, we call B =: BK the
canonical form on g associated to K.
Finally, let us point out that the Cartan involution can be applied in many
ways. For example, if G1 is a R-subgroup of G containing RuG such that all
maximal compact subgroups K of G(R) are conjugate under G1 (R), then, for a
Levi subgroup M of G(R), it makes sense to take about the Cartan involution θK
of M with respect to K. Moreover, in this case, the subgroup (G1 ∩ M)∩θK (G1 ∩K)
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 404

404 Geometric Structures and Riemann Hypothesis

is the unique θK -stable Levi subgroup of G1 (R) contained in M. Consequently, if

P is a parabolic R-subgroup of G, and K is a maximal compact subgroup of G(R)
and M is a Levi subgroup of G(R) containing K, then M ∩ P contains one and only
one Levi subgroup of P(R) stable under θK .

16.2.1.3 Cartan Involutions and Compatible Metrics

Motivated by the Cartan involution associated to a maximal compact group K of

G, we introduce the following:

Definition 16.5. Let H be a positive definite real symmetric bilinear form on g.

H is said to be K-compatible with the Lie structure of g if −H −1 B is a Lie algebra
automorphism of g.1

Lemma 16.9. Let H be a positive definite real symmetric bilinear form on g. Let
θ := H −1 B. Assume that H is K-compatible. Then

(1) θ is a Cartan involution of g. Namely,

(1.a) θ2 = 1;
(1.b) Let k, resp. p, be the (+1)-eigenspace, resp. the (−1)-eigenspace, of θ on g.
Then Bθ is negative definite on k and positive definite on p. In particular,
g = k ⊕ p.
(2) H is compatible with B. That is, B : g → g∗ is an isometry with respect to the
metric H on g and the metric H −1 on g∗ .

When G is semi-simple, this definition and lemma are due to Grayson ([34]).

Proof. By the infinitesimal invariance of B, namely, both relations in (16.50), we

have t θBθ = B since θ is compatible by definition. On the other hand, t θB =
−BH −1 B = Bθ. Therefore θ2 = 1. This proves (1.a). Moreover, on k, we have
Bθ = −h < 0; and, on p, we have Bθ = h > 0. So we obtain (1.b). Finally, since
θ2 = 1, we have t BH −1 B = h. That is to say, B : g → g∗ is an isometry with
respect to the metric H on g and the metric H −1 on g∗ . This proves (2). 

Definition 16.6. The morphisms θ in the previous lemma are called Cartan invo-
lutions associated to K.
1 Here as usual, we view the bilinear H as a linear map from g to g. Hence H −1 B is indeed a linear

endomorphism of g.
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 405

Analytic and Geometric Truncations: Reductive Groups 405

Denote by Mtotg;K the moduli space of K-compatible inner products on g. Ob-

viously, Mtot
g;K may be viewed as a closed subset of the space of all inner products
on g. Since the inner product (·, ·) in (16.49) is obviously K-compatible, Mtot
g;K is
not empty.

Proposition 16.4. Let G be a connected reductive group and let g = Lie(G).

(1) There is a natural action of G on Mtot

g;K .
(2) The action of G on Mtotg;K above induces a natural diffeomorphism
g;K .
G(R)/K ' Mtot (16.51)

Proof. Recall that, for φ ∈ AutLie (g), we have t φBφ = B since B is characteristic.
g;K , we have (− φHφ)B = φ θφ. This implies that φHφ
−1
Hence, for any H ∈ Mtot t t

belongs to Mtot
g;K as well. Consequently, the assignment
(H, g) 7−→ t (Ad g)H(Ad g) g;K , ∀g ∈ G
∀H ∈ Mtot (16.52)
defines a natural action of G on Mtot
Here, as usual, for g ∈ G, Ad g denotes its
g;K .
adjoint. This proves (1).
In terms of Cartan involutions θ, the action above is equivalent to
(θ, g) 7−→ t (Ad g)θ(Ad g) ∀g ∈ G. (16.53)
Recall that, for a general G, if we set oθ := { X ∈ c : θX = −X }. Then Vθ := exp oθ
is a split component of G. Moreover, by Proposition 2.1.10 of [31], the assignment
θ 7→ oθ gives a bijection from the set of Cartan involutions associated to K to the
set of split components of G. In particular, when G = G1 , this map gives a bijec-
tion from the set of Cartan involutions to the set of maximal compact subgroups
of G. Hence, in our case, since G is connected and reductive, all maximal compact
subgroups of G are conjugate to each other. This implies that G acts transitively
on Mtotg;K . Thus, to complete the proof, it suffices to show that the stabilizer group
of θK in G is exactly K itself. For this we choose ΘK : G(R) → G(R) to be a Car-
tan involution satisfying dΘK = θK . By definition, K = { g ∈ G(R) : Θ(g) = g }.
On the other hand, for g ∈ G, θK = (Adg)−1 θK (Adg) if and only if g−1 ΘK g is in
the center of G(R)0 . But this center is trivial by our assumption, hence g belongs
to the stabilizer group of θK if and only if g ∈ K. 

16.2.2 Principal Lattices

16.2.2.1 Canonical Infinitesimal Integral Structures

Let F be an algebraic number field with OF the ring of integers. Let G be a

connected split reductive group defined over F. Then g admits a rational structure
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 406

406 Geometric Structures and Riemann Hypothesis

gQ ⊂ g which can also be viewed as an F-structure. Our aim here is to introduce

a G(OF )-invariant projective OF -module gOF in g ⊗F R which is compatible with
the Lie structure.
We start with the case when F = Q. Since G is defined over Q, its Lie algebra g
admits a natural rational structure gQ and the adjoint representation G → Aut(gQ )
is a regular map defined over Q. Consequently, there always exist G(Z)-invariant
integral structures in gQ , since, for any integral structure in gQ , the image under
the action of G(Z) is again an integral structure in gQ . Obviously, the summation
of two G(Z)-invariant integral structures in gQ is again a G(Z)-invariant integral
structure. Moreover, if gZ is a G(Z)-invariant integral structure in gQ , we have
[gZ , gZ ] ⊂ gQ . Hence, by clearing up denominators, we can instead assume that
[gZ , gZ ] ⊆ gZ from the beginning. In this way, we obtain a unique maximal G(Z)-
invariant integral structure gZ in gQ satisfying the condition that [gZ , gZ ] ⊆ gZ .
For a general F, since G is defined over F, its associated Lie algebra gF admits
a natural F-linear space structure. Accordingly, introduce the space g∞ := gF ⊗Q
Q
R := σ∈S ∞ g ⊗ Fσ , where S ∞ denotes the connection of normalized inequivalent
Archimedean places of F. Similarly, since G → Aut(gF ) is a regular map defined
over F, there exist G(OF )-invariant OK -modules in gF . Using a similar argument
as above, we then obtain a unique maximal G(OF )-invariant integral structure gOF
in gF satisfying the condition that [gOF , gOF ] ⊆ gOF . For convenience, this final
gOF will be called the canonical (infinitesimal) OF -structure of G/F in the sequel.

16.2.2.2 Principal Lattices

When G is defined over a number field F. For each σ ∈ S ∞ , we fix a maximal

compact subgroup Kσ of G(Fσ ). This way, we obtain the spaces MtotgFσ;Kσ of the
compatible metrics with respect to Kσ on gFσ and an isomorphism
'
Y Y
G(Fσ )/Kσ −→ g ;Kσ .
Mtot
Fσ
(16.54)
σ∈S ∞ σ∈S ∞

For later use, set

Y Y
G(F∞ )/K(F∞ ) := G(Fσ )/Kσ and Mtot
g∞ ;K∞ := g ;Kσ .
Mtot
Fσ
(16.55)
σ∈S ∞ σ∈S ∞

Definition 16.7. Let F be a number field with OF its ring of integers. Let G be
a connected (split) reductive group over a number F. By a metrized G-torsor, or
equivalently, a principal G-lattice (over OF ), we mean a pair gOF , (Hσ ) consist-


ing of the canonical infinitesimal integral structure gOF of G/F and an element
(Hσ ) of Mtot
g∞ ;K∞ .
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 407

Analytic and Geometric Truncations: Reductive Groups 407

Q
Set now B := σ∈S ∞ Bσ , where Bσ denotes the canonical form on gFσ with
respect to Kσ . By our construction,

B(gOF , gOF ) ⊆ Z. (16.56)

Q 
Indeed, since Bσ is integral, σ∈S ∞ Bσ (gOF , gOF ) is a rank one Z-module in Q.
Therefore, it must be contained in Z since gOF is maximal.

16.2.3 Parabolic Reduction

16.2.3.1 Parabolic Sub-algebras

We start with a simple characterization of parabolic sub-algebra of a Lie algebra.

Lemma 16.10. Let G be a reductive group over R with g its associated Lie al-
gebra. Then a sub-algebra a ⊂ g is a parabolic sub-algebra if and only if it is
adg -nilpotent and the orthogonal component a⊥ with respect to a canonical form
on g is contained in a.

When G is semi-stable, all the lemmas in this subsection are due to [33, 34].
The proofs there work in our general setting with some modifications when
necessary.

Proof. With an extension of the coefficients to C, we can and hence will work
over C. Let b be a Borel sub-algebra of g. We must show that b ⊆ a.
Being a canonical form, B is associated to a maximal compact subgroup K
of G. Consequently, using the (±1) eigenspaces within b of the Cartan involution
θK , we obtain an
Xorthogonal decomposition b = k0 ⊕ p0 with respect to B. De-
note by p0 = gα where gα denotes the associated root space, and fix a base
α∈k∗, α>0
of b obtaining from gα ’s compatible with the order determined by positive roots
involved.
Let x ∈ a⊥ , the perpendicular subspace of a. Since it is solvable, a⊥ is con-
tained in b. Consequently, if we write x = k + p with k ∈ k0 and p ∈ p0 , then
adb (p) is an upper-triangle matrix and adb (k) is a diagonal matrix. Thus the nilpo-
tent property of adg a implies that adb (x) is nilpotent. Hence adb (k) = 0. This
implies that α(k) = adgα (k) = 0 for all α ∈ Φ+ . Thus k = 0 since simple roots form
a basis of the dual space of b. Therefore a⊥ ⊆ p0 .
To go further, note that B(k0 , gα ) = 0, hence k0 ⊆ p⊥0 . Moreover, since
B(gα , gb ) = 0 when α + β , 0. Hence p0 ⊂ p⊥0 . Therefore, p⊥0 = b since
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 408

408 Geometric Structures and Riemann Hypothesis

dim b + dim[b, b] = g and p0 = [b, b]. This then completes the proof since
b = p⊥0 ⊆ (a⊥ )⊥ = a.
Conversely, if a is a parabolic sub-algebra, then we may choose a Borel sub-
algebra b such that b ⊆ a. With this, it is rather obvious to track back to show the
only if direction left. 

As pointed in [33, 34], this lemma characterizes the parabolic subgroups of

g in terms of trilinear and bilinear data of Lie bracket [·, ·] : g ⊗ g → g and the
canonical form B : g ⊗ g → R.

16.2.3.2 Canonical Parabolic Sub-Groups

Let Λ = (gOF , (Hσ )) be a principal G-lattice over OF . By Proposition 1.6, there

exists a unique canonical filtration of Λ
Λ∗ : 0 = Λ0 ⊂ Λ1 ⊂ . . . ⊂ Λr = Λ. (16.57)
Denote by Gri Λ := Λi /Λi−1 and let νi be its slope and set
inf Λ = µr and sup Λ = µ1 . (16.58)
From the definition, the OF -lattices Gr1 Λ, Gr2 Λ, . . . , Grr Λ are all semi-stable
and µ1 > µ2 > . . . > µr . Accordingly, we define a subspace W j ⊆ g∞ by
n o
W j := X ∈ gOF : adg X(Λi ) ⊆ Λi+ j ∀i . (16.59)
Then, with respect to B = σ∈S Bσ , it is not difficulty to prove that
Q

(1) W j = (W − j+1 )⊥ within g∞ for all i.

(2) The canonical filtration (16.57) for Λ induces a filtration of subspaces of g
0 = W −r ⊆ . . . ⊆ W −1 ⊆ W 0 ⊆ W 1 ⊆ . . . ⊆ W r−1 = gOF (16.60)
(3) [W i , W j ] ⊆ W i+ j for all −r ≤ i, j ≤ r − 1 (with the obvious convention when
i + j is not in the range [−r, r − 1]).

 φ : W  ⊗ W → Λ/W . Since
0 0 0
To go further, consider next the Lie bracket
inf W 0 ⊗ W 0 = inf W 0 + inf W 0 = 0 and sup Λ/W 0 < 0, we have φ = 0. This

implies that W 0 is a Lie sub-algebra lattice of gOF . For the similar reason, since
[W − j , W −1 ] ⊆ W − j−1 for all j ≥ 0, W −1 is a nilpotent ideal. Consequently, W −1 is
the nilpotent radical of W 0 and W 0 /W −1 is a reductive quotient of W 0 . Therefore,
by Lemma 16.10, W 0 is a parabolic sub-algebra lattice over OF , since we have:

Lemma 16.11. Let W ⊆ g be a parabolic sub-algebra. There is a filtration

0 = W −r ⊆ . . . ⊆ W −1 ⊆ W 0 ⊆ W 1 ⊆ . . . ⊆ W r−1 = g (16.61)
such that
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 409

Analytic and Geometric Truncations: Reductive Groups 409

(1) W 0 = W,
(2) W j = (W − j+1 )⊥ within g∞ for all i,
(3) [W i , W j ] ⊆ W i+ j for all −r ≤ i, j ≤ r − 1.

Moreover, if instead W is an F-subspace, so are the W j ’s.

Proof. Set W 0 = W and let W 1 := W ⊥ be the orthogonal component with respect

to a canonical form B. Moreover, when i > 1, we define W i inductively by W i :=
[W i−1 , W 1 ], and when i < 0, we define W i := (W −i+1 )⊥ . Obviously, this does form
a filtration, and W s+1 = 0 for some s since W 1 is nilpotent by Lemma 16.10. To
prove that [W i , W j ] ⊆ W i+ j , we may first tensor with C and make use of the roots
relative to (−1)-eigenspace k1 within W 0 of the Cartan involution θK . Let φ be a set
of simple roots such that, if α ∈ φ the associated root space gα ⊂ W. Then there
exists a subset φ1 ⊆ φ such that W is generated by the root spaces gα for α ∈ φ
and g−α for α ∈ φ1 and k1 . Recall that the root spaces gα are one dimensional, and
if α, β and α + β are all roots, then [gα , gβ ] = gα+β , [gα , g−α ] ⊆ k1 and [k1 , gα ] ⊆ gα .
Furthermore, B(k1 , gα ) = 0 and B(gα , gβ ) = 0 if α + β , 0. Thus, if we write, for a
subset R ⊂ Z∆ with ∆ the set of simple roots, g(R) = ⊕α∈R∩∆ gα ,

W 0 = g Z≥0 φ + Zφ1 W 1 = g Z∆ r −(Z≥0 φ + Zφ1 ) .

and
nP o
α∈φ nα α ∈ Z∆ :
P
Thus, if we set T i := α∈φrφ1 nα ≥ i , then

W i := g(T i ) and W −i = g(Z∆ r −T i+1 ) = g(T −i ) ∀i ≥ 1,

as desired. 

Finally, from the parabolic sub-algebra lattice W 0 over OF , we obtain the cor-
responding parabolic subgroup of G and this is its own normalizer. Therefore, the
sub-algebra lattice W 0 over OF determines a reduction of the structural group G
of Λ to this parabolic subgroup Q. We denote this new principal P-lattice over OF
by ΛQ and call it the canonical parabolic reduction of Λ. All these then prove the
following:

Proposition 16.5. Let G be a connected (split) reductive group over F. Set gOF
be the canonical infinitesimal structure of G over the ring of integers OF of F,
Q Q
G∞ := σ∈S G(Fσ ) and g∞ := Lie(G∞ ). Assume that Kσ := σ∈S Kσ is a
maximal compact subgroup of G∞ . Then, for each (Hσ ) ∈ Mtot g∞ ;K∞ , the induced
principle G-lattice Λ = (gOF , (Hσ )) admits a canonical parabolic reduction ΛQ
for a unique F-parabolic subgroup Q of G.
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 410

410 Geometric Structures and Riemann Hypothesis

16.2.4 Torsors in Arithmetic

16.2.4.1 Torsors over Integer Ring

Let F be an algebraic number field with OF its ring of integers. By a group scheme
G on X = Spec OF , we mean G is a (smooth affine) algebraic group scheme over
X with connected fibers. Similarly, by a G-torsor E on X with a structural map
π : E → X, we means that π is faithfully flat and locally finite type (surjective)
projection, and there is a group action G × E → E satisfying the properties that it
is compatible with π and that the induced morphism G ×X E → E ×X E, charac-
terized by (g, x) 7→ (g x, x), is an isomorphism. Dente by G, resp. E, the generic
fiber GF and EF of G and E, respectively. We always assume that G is split over
F. Naturally, we obtain an induced F-action G × E → E. In the sequel, we will
be interested in the reductive group schemes G defined over F, that is, the alge-
braic group schemes such that all of its geometric fibers are (connected) reductive
algebraic groups.
Let now B be a dense open subscheme of X. Then for pairs (Gi , Ei ), i = 1, 2,
consisting of a reductive group scheme Gi and a Gi -torsor Ei on X, we say that
(G1 , E1 ) is B-isomorphic to (G2 , E2 ), denoted by (G1 , E1 ) 'B (G2 , E2 ), if G1 ' G2
as group schemes over B and there exists a B-isomorphism E1 ' E2 which is
compatible with the actions of G1 and G2 . We have the following recent result of
Javanpeykar-Loughran.

Lemma 16.12. ([48]) For a fixed n ∈ Z>0 and a dense open subscheme B of X,
the set of B-isomorphism classes of pairs (G, E) is finite, provided that G is an
algebraic reductive group scheme of rank n.

16.2.4.2 Arithmetic Torsors

Let G be an algebraic reductive group scheme over X and let E be a G-torsor

over X. Then E admits an induced G-torsor E. As usual, denote by S ∞ the col-
Q
lection of inequivalent Archimedean places of F, set G∞ := σ∈S ∞ G(Fσ ) and
denote by g∞ = Lie(G∞ ). Note that, for each real, resp. complex, embedding
of F, we obtain a G(R)-torsor E(R), resp. G(C)-torsor E(C). In this way, natu-
rally, E∞ = σ∈S ∞ E(Fσ ) becomes a G∞ -torsor, and its associated tangent bundles
`
T E∞ admits a natural flat g∞ structure. Consequently, it suffices to work over the
Q
associated adjoint space gE,∞ := σ∈S ∞ gE ⊗ Fσ . Thus, with respect to a fixed
maximal compact subgroup K∞ = σ∈S Kσ of G∞ , we may transform all com-
Q
patible metrics on g∞ to these on one and hence on all fibers of the tangent bundle
T E∞ . We denote by Mtot E∞ ,K∞ the collection of all such induced metrics, and call
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 411

Analytic and Geometric Truncations: Reductive Groups 411

elements of Mtot E∞ ,K∞ as the fiberwise admissible metrics on E∞ . Similarly, by the

flatness, the fiberwise admissible metrics induce natural metrics on E∞ .

Definition 16.8.

(1) By an arithmetic G torsor on X = Spec OF associated to K∞ , we mean a

pair E = (E, (Hσ )), consisting of a G-torsor E on Spec OF and a fiberwise
admissible metric (Hσ ) in Mtot
E∞ ,K∞ .
(2) Let E = (E, (Hσ )) be an arithmetic G-torsor on Spec OF . The slope of E,
denote by µ(E), is defined to be the element µ ∈ X∗ (AG0 ) satisfying
hχ, µi = c1,ar Eχ ∀χ ∈ X ∗ (AG0 )


(16.62)
where Eχ denotes the arithmetic line bundle, or the same, the metrized line
bundle, induced from E by an reduction of structure group scheme associated
χ
to the character G  AG0 −→ Gm .

Lemma 16.13. There exists a bijection between the moduli space of arithmetic G-
torsors over OF and the quotient space G(F)\G(A)/K, where K = v∈S fin G(Ov ) ×
Q
K∞ denotes the maximal compact subgroup of G(A) induced by the OF structure
on G and K∞ .

Proof. This is a direct consequence of Proposition 16.4 with a similar arguments

as in the proof of Propositions 14.1. 

16.2.4.3 Arithmetic Torsors for Parabolic Subgroups

Recall that, for a fixed maximal compact subgroup K of a connected reductive

group G(R), the constructions of the Cartan involution and more generally Cartan
involutions and hence the compatible metrics work well for the Levi subgroups
of (standard) parabolic subgroups, by §16.2.1, particularly, §16.2.1.2. Based on
this, we may, in parallel, introduce arithmetic P-torsors associated to (standard)
parabolic subgroups P of G.
To motivate our discussion below, we start with a local theory over R. Let K be
a maximal compact subgroup of a connected reductive group G over R, and P be
a (standard) parabolic subgroup of G. Denote its associated Levi decomposition
by P = N M with M the Levi subgroup and N the unipotent radical. By §16.2.1.1,
there exist compatible metrics on M with respect to K (and its associated Cartan
involution). To obtain the metrics on p = n ⊕ m, we still need to introduce metrics
on n. But, this is very simple since n is simply an affine R-space, or better, admits a
natural filtration structure with graded quotients one dimensional R-affine spaces.
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 412

412 Geometric Structures and Riemann Hypothesis

With the infinite places discussed, with Lemma 16.13, we are ready to give
an adelic version of compatible metrics, the compatible measures on G(A) with
respect to K = v∈S fin G(Ov ) × K∞ . Our basic principle is that all the compati-
Q
ble measures with respect to K on G(A) should be compatible with the Iwasawa
decomposition
G(A) = N0 (A) · A M0 (A) · M0 (A)1 · K = N(A) · A M(A) · M(A)1 · K. (16.63)
Here P0 = N0 M0 is the Levi decomposition of a minimal parabolic subgroup
of G and P = N M is the Levi decompositions as above for standard parabolic
subgroups P of G.
To be more precisely, for the compact group K∞ of G∞ , its associated compat-
ible measures on G∞ are discussed above. This induces naturally the compatible
measures dm on M0 (A) with respect to K ∩ M0 (A) with an induction argument,
since the works obviously for all split tori. Furthermore, since, for the unipotent
radical N of P, N0 (F) is discrete and N0 (F)\N0 (A) is compact, we can define the
compatible measures dn on N0 (A) to be the Haar measures on N0 (A) such that the
volume of N0 (F)\N0 (A) is one. With this, for all compactly supported continuous
functions f on G(A),
Z
f (uamk) dk a−2ρ0 da dm du


f 7→ (16.64)
N0 (A)·A M0 (A) ·M0 (A)1 ·K

defines a compatible Haar measure dg on G(A) with respect to K. Here u ∈

N0 (A), a ∈ A M0 (A) , m ∈ M0 (A) and k ∈ K. Furthermore, for any parabolic sub-
group P = MN, by restricting dg to M(A), resp, to N(A), we obtain a compatible
Haar measure dm on M(A) with respect to K ∩ M(A) and a Haar measure dn
on N(A). It is well-known that all these compatible Haar measures satisfy the
following compatibility condition (see e.g. §6.1.5).

For any compactly supported continuous function f on G(A),

Z Z
f (g) dg = f (uamk) dk a−2ρP da dm du (16.65)
G(A) N(A)·A M(A) ·M(A)1 ·K

where u ∈ N(A), a ∈ A M(A) , m ∈ M(A)1 and k ∈ K.

That is to say, once compatible measures with respect to N0 (A) and K =

Kfin × K∞ are defined, there are naturally compatible measures on P(A) for any
(standard) parabolic subgroup of G. Denote by Mtot
n∞ ;K∞ the moduli space of com-
patible metrics with respect to both n∞ and K∞ .

Definition 16.9. Let P be a parabolic subgroup scheme of a connected reductive

group scheme G over Spec OF . Denote by P ⊆ G the associated generic fiber,
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 413

Analytic and Geometric Truncations: Reductive Groups 413

and P = N M its Levi decomposition. Set p := Lie(P), n = Lie (N) and form
p∞ := σ∈S ∞ p ⊗ Fσ and n∞ := σ∈S ∞ n ⊗ Fσ . Denote by Mtot
Q Q
n∞ ;K∞ the moduli
space of the n∞ and K∞ -admissible metric on p∞ .

(1) By an admissible arithmetic P torsor on Spec OF (associated to K∞ ), we mean

a pair E = (E, (Hσ )), consisting of a P torsor E on Spec OF and a fiberwise
n∞ and K∞ -admissible metric (Hσ ) in Mtot n∞ ;K∞ .
(2) In (1), when E = P is a simply parabolic subgroup scheme, we call E an
admissible arithmetic parabolic group scheme over Spec OF (based on P).
(3) For an admissible arithmetic P-torsor E = (E, (Hσ )) on Spec OF , we define
the slope of E, denote by µ(E), to be the element µ ∈ X∗ (A0P ) such that
hχ, µi = c1,ar Eχ ∀χ ∈ X ∗ (A0P )


(16.66)
where c1,ar denotes the first arithmetic Chern class, Eχ denotes the arithmetic
line bundle, or the same, the metrized line bundle, obtained from E obtained
by an reduction of the structure group scheme associated to the character
χ
P  A0P −→ Gm .

All these, together with §16.2.1.2, then prove the following

Lemma 16.14. There exists a bijection between the moduli space of admissible
arithmetic parabolic group schemes over Spec OF associated to K∞ and the quo-
tient space P(F)\P(A)/K, where K = Kfin × K∞ with Kfin the maximal compact
subgroup of P(A) defined by the associated OF -structure.

16.2.5 Canonical Parabolic Subgroup Scheme

Let G be a connected reductive group scheme over Spec OF with G its generic
fiber. We assume that G is split over F. Fix a maximal compact subgroup scheme
K of G and a minimal parabolic subgroup scheme P0 of G over Spec OF . Denote
by N0 the unipotent radical of P0 . Let P be a standard parabolic subgroup scheme
over Spec OF with N the unipotent radical. As above, denotes the associated
generic fibers by P0 and P respectively.

Definition 16.10.

(1) An admissible arithmetic P-torsor E = (E, (Hσ )) on Spec OF with respect

to K and N is said to be compatible with respect to N0 if there exists an
admissible arithmetic P0 -torsor E0 = (E0 , (Hσ,0 )) on Spec OF with respect to
K and N0 such that
E = E0 ×P0 P. (16.67)
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 414

414 Geometric Structures and Riemann Hypothesis

(2) An admissible arithmetic P-torsor E = (E, (Hσ )) on Spec OF with respect to

K and N0 is said to be semi-stable if, for each parabolic subgroup scheme Q
of G, which is contained in P with relative rank one generic fibers, and, for
each admissible arithmetic Q-torsor EQ on Spec OF with respect to K and N0
such that EQ is the reduction of E associated to Q ,→ P, namely,
E = EQ ×Q P, (16.68)
the slope µ EQ ∈ X∗ (A0Q )


⊂ aQ satisfies the condition
hαQ , µ EQ i ≤ 0


(16.69)
where αQ is the unique element of ∆PQ .

Proposition 16.6. Let E = (E, (Hσ )) be an admissible arithmetic P-torsor on

Spec OF with respect to K and N0 of slope νP .

(1) There exist a unique standard parabolic subgroup scheme Q ⊆ P, and, up

to isomorphism, a unique semi-stable admissible arithmetic Q-torsor EQ with
respect to K and N0 over OF such that
E ' EQ ×Q P, ν(EQ )]PQ ∈ aP+ ν(EQ )]P = νP . (16.70)
 
Q and
Here Q ⊆ P denotes the generic fiber of Q.
(2) Let Q be as in (1). For each fixed ν ∈ a0P+ ∪ {0}, there exist a unique stan-
dard parabolic subgroup ν Q and an admissible arithmetic ν Q-torsor Ebν with
Q
respect to K and N0 over OF satisfying the following properties:
(a) Q ⊆ ν Q ⊆ P,
ν
(b) EP ' Eν Q × Q P,
(c) Denote by ν Q the generic fiber of ν Q and define a (standard) parabolic
subgroup ν Q0 to be the one corresponding to the subset Jν Q obtained from
the decomposition ∆ν Q = IP ∪ Jν Q where [IP ]Q
P = ∆P . Then

ν(Eν Q ) ν Q0 ∈ [ν]ν Q0 + aν Q0 ν(Eν Q )]P = νP .

 P P P+ 
and (16.71)

Proof. As the proof for the second one is a direct generalization of the first, we
here only prove (1) to illustrate. The existence is a direct consequence of the
constructions in §16.2.3, in particular, that of §16.2.3.2. As for the uniqueness,
if there exists two, say P and P0 . Let T be the maximal torus of G satisfying
T ⊂ P ∩ P0 . If T is not split, by passing to an extension of F, we may assume
that T is split over the extension field. On the other hand, it is clear that this does
not affect the positivity of the slopes involved. Consequently, there exists an even
larger parabolic subgroup scheme satisfying all the properties. This contradicts to
the maximality we start with. 
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 415

Analytic and Geometric Truncations: Reductive Groups 415

Remark 16.1. (1) As noticed above, the conditions EP ' EQ ×Q P and EP '
ν
EνQ × Q P imply automatically that ν(EQ )]P = νP and ν(EνQ )]P = νP , respectively.
 

(2) In the case when G = GLn , we have

(a) The filtration on EP induced by Q is an refined filtration of EP such that every

graded OF -lattice is semi-stable and the slopes of these graded OF -lattices are
strictly decreasing.
(b) The filtration on EP induced by ν Q is a refined filtration of EP such that, in
addition to that for EP , it also consists of those for which differences of slopes
for successive graded OF -lattices, say corresponding to simple root αk ∈ ∆PQ ,
are strictly bigger than hαk , νi.
 
Definition 16.11. In the situation of Proposition 16.6, the pair Q, ν(EQ ) , resp.
ν
 
Q, ν(EνQ ) , is called the canonical type of EP , resp. ν-canonical type of EP , and
the induced arithmetic torsors EQ and EνQ are called the canonical reduction and
the ν-canonical reduction of EP , respectively, or simply, the parabolic reduction
and the ν-parabolic reduction of EP , respectively.

N0 ,K∞
Accordingly, we denote by MG,F (νG ), or simply MG,F (νG ), to be the moduli
space of semi-stable admissible arithmetic G-torsors E over OF of slope νG with
respect to N0 and K∞ . Similarly, for each parabolic subgroup scheme P of G,
0 ,K∞
we denote by MN F;G
(P, νG ), or simply MF;G (P, νG ), the (sub) moduli space of
G-torsors EG over OF of slope νG which admit (P, νP ) as their canonical types
and EP as their parabolic reductions. We denote such P by P and write E := EP
and ν := νP . More generally, for a parabolic subgroup scheme P of G, we denote
0 ,K∞
by MN P,F
(νP ), or simply MP,F (νP ) the moduli space of semi-stable admissible
arithmetic P-torsors EP over OF of slope νP . And, for each parabolic subgroup
0 ,K∞
scheme Q contained in P, denote by MN P,F
(Q, νQ ), or simply MP,F (QνQ ), the
sub moduli space corresponding to the admissible arithmetic P-torsors EP over
OF of slope νP with respect to N0 and K∞ which admit (Q, νQ ) as their canonical
types, and EQ as their parabolic reductions. We denote such Q by QP , and write
EP := EQ and νP := νQ .
P P

16.3 Analytic and Geometric Truncations

Recall that, in the theory of automorphic forms and trace formula, when working
with Arthur’s analytic truncation ΛP,T , we always assume that T is sufficiently
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 416

416 Geometric Structures and Riemann Hypothesis

regular. This is mainly because, in the proof of Theorem 7.1 on the basic prop-
erties for Arthur’s analytic truncation ΛP,T , used are the classical reductions and
the Langland partition for G(A) via Langlands combinatorial lemma, for them the
sufficiently regular condition is crucial. As we will see below, for these funda-
mental properties of Arthur’s analytic truncation, by our result on the equivalence
between the analytic truncation and geometric truncation using stability (to be es-
tablished below), this sufficiently regular condition becomes artificial and hence
will be removed. What matters is the positive condition
T ∈ a0P+ ∪ {0}.

16.3.1 Stability and Parabolic Reduction

Let F be a number field. Denote by OF its ring of integers and denote by A its
adelic ring. Let G be a reductive group over F and K be a maximal compact
subgroup of G(A). Fixed once and for all an integral model G of G, a group
of reductive group scheme over X = Spec OF with G its generic fiber. (Recall
that by Lemma 16.12, there are only finitely many of them with a fixed B-type.)
For a element g ∈ G(F)\G(A)/K, let Eg be the associated G-torsor. Denote by
(P, νg ) the canonical type of Eg and by Eg,P the canonical reduction of Eg . More
generally, if P is a parabolic scheme with P its generic fiber, we let Eg,P be the P-
torsor obtained from Eg via the parabolic reduction associated to P ,→ G. Denote
by νg,P the slope of Eg,P . From the definitions, we have
ν p := νP,g ∈ aGP ⊆ a0P ⊕ aGP = aG0 and νP,p ∈ aGP ⊆ a0P ⊕ aGP = aG0 . (16.72)

Fix an parameter T > 0, i.e. T ∈ aG,+

0 . In the sequel, we will use the following
partial orders on the spaces aGP defined in §16.1.7.

(a) νg ≤ T if and only if T − νg ∈ + aG0 .

(b) νP,g >P T if and only if νP,g − T ∈ + aGP .

Theorem 16.2. Let T ∈ aG+

0 ∪ {0}. For g ∈ G(F)\G(A)/K, we have

(1) Denote by νg be the canonical type of g. Then

1(νg ≤ T ) = (−1)dim aP 1(νg,P >P T ).
X G
(16.73)
P: parabolic

(2) For all standard parabolic subgroups P of G,

1(νP,g ≤P T ) = (−1)dim aQ 1(νQ,g
X P
P
>PQ T ). (16.74)
Q:Q⊆P
parabolic
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 417

Analytic and Geometric Truncations: Reductive Groups 417

When G = GLn , this is essentially Theorem 14.2, for which the canonical
convex polygons and a natural partial order on a0 of type An−1 is used. However,
for general reductive groups, this language of polygons cannot be applied. Instead,
we use naturally defined partial orders recalled above.

Proof. Since a similar proof below for (1) works for (2) as well, we here only
work put the details for (1).
First assume that νg ≤ T . Then the left hand side is equal to 1. Hence it
suffices to show that the right hand side is equal to 1 as well. For this purpose,
we analysis the terms on the right hand side. For each fixed parabolic subgroup
P of G, its associated contribution (to the right hand side) is given by, up to sign,
1 νg,P >P νP,g . Easily, from the definition,


1 νg,P >P νP,g = 1 ⇐⇒ νg,P − νP,g ∈ + aGP .

 

Claim 16.3. Assume that νg ≤ T . Let P be a parabolic subgroup of G. Then

1 νg,P >P νP,g = 1 if and only if P = G.
 

Proof. Since νP,g is the canonical type of g, we have, in aG0 ,

νP,g ≥ νP,g .

This implies that

νP,g − νP,g ∈ + aG0 .

Lemma 16.15. ν ∈ + aG0 implies that [ν]GP ∈ + aGP . Or better,

[+ aG0 ]GP ⊆ + aGP .

n o
Proof. Indeed, by definition, ∆P = [α]GP : α ∈ ∆0 r{0}. Hence, by Example 16.2,
using the fact that [·]GP is a linear projection, we have
  G  
    
+ G G
h i  X    X 
=   α = G
   
a0 a : a ≥ 0 a [α] : a ≥ 0
   
α α α α
 

P  
 

 
 P 

α∈∆0
 
 α∈∆0
 

P
 

 

X  + G
= G
= aP ,
 
 a α [α] P : a α ≥ 0 


α∈∆G 

P

as desired. 
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 418

418 Geometric Structures and Riemann Hypothesis

Consequently, we have
νP,g − νP,g ∈ + aGP or the same νP,g ≥P νP,g ∀P. (16.75)
This is a contradiction unless P = G. That is to say,
1 νg,P >P νP,g = 0
 
unless P = G. (16.76)
Therefore, the right hand side is simply 1, since 1(νg,G >G νP,g ) ≡ 1. This verifies
the claim and hence also the theorem under the assumption that νg ≤ T . 

Next we assume that νg  T . Then the left hand side becomes zero from the
definition. Hence it suffices to show that the right hand side equal zero as well,
namely,
(−1)dim aP 1(νP,g >P T ) ≡ 0.
X G
(16.77)
P

For this purpose, recall that, for each parabolic subgroup P and the arithmetic
P-torsor associated to g, there exists a unique P-canonical subgroup QP,g . Ac-
cordingly, following an idea of Lafforgue used in the proof of Theorem 14.2, we
regroup the summation in (16.77) according to Q = QP,g . In this language, since,
for a fixed parabolic subgroup P of G, the P-canonical parabolic subgroup of g is
unique, to prove the theorem, it suffices to establish the following:

Lemma 16.16. Let Q be a parabolic subgroup contained in a parabolic subgroup

P of G. For g ∈ G(F)\G(A)/K, when νg  T ,
(−1)dim aP 1(νP,g >P T ) ≡ 0.
X G
(16.78)
P:QP,g =Q

Proof. Motivated by Lafforgue’s work [63] used in Theorem 14.2, we introduce

a new parameter ν ∈ aG+ 0 . By Proposition 16.6(2), there exists a unique ν-
canonical parabolic subgroup scheme ν QP,g of G such that, P ⊇ ν QP,g ⊇ QPg ,
and νg ≥P ν + νP,g . Consequently, we rewrite a ν-modified summation of (16.78)
as an alternate summation of the characteristic functions of all the subset in the
power set of a certain set described below.
ν
Claim 16.4. Let ν ∈ aG+ G+
0 t {0} and T ∈ a0 , and let Q ⊆ Q be two parabolic
subgroups of G. The function on Q(F)\G(A)/K defined by

(−1)dim a0 1(νP,g >P T )

X P
g 7→ (16.79)
P
QP,g =Q,ν QP,g = ν Q
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 419

Analytic and Geometric Truncations: Reductive Groups 419

is the characteristic function of a compact subset which contains

g ∈ Q(F)\G(A)/K : Qg = Q, ν Qg = ν Q, νg ≤ T .
n o
(16.80)
In addition, the compact subset coincides (16.80) if T dominates ν, i.e. if T −ν ∈
aG+
0 .

Proof. We first point out that, on the right hand side of (16.79), the sum is taken
over parabolic subgroups P such that, with respect to fixed parabolic subgroups Q
and ν Q, the canonical parabolic reduction within P of the admissible arithmetic
G-torsor Eg is given by Q and the ν-parabolic reduction within P of the admissible
G-torsor Eg is given by ν Q.
For a standard parabolic subgroup P of G, denote by P = MP NP its Levi
decomposition with MP , resp. NP , the standard Levi factor, resp. the nilpotent
radical, of P. It is well known, see e.g. [36] or, more recently, [21], that there exist
the simple factors of reductive groups MP,1 , . . . , MP,|P| , ordered according to Φ+0 ,
or the same ∆0 , such that

(i) the root system of MP,i are all reduced, and,

(ii) up to an isogeny, MP ∼ MP,1 × · · · × MP,|P| .

More generally, for a parabolic subgroup P of G, there exists δ ∈ P(F)\G(F) such

that δ−1 Pδ is standard. Hence, it makes sense to talk about the reductive factors of
all parabolic subgroups P of G. For simplicity, we write them by MP,1 , . . . , MP,|P|
as well.
ν
Applying this to the parabolic
n subgroups Q o and n Q, there are two collections
o
of simple reductive groups MQ,1 , . . . , MQ,|Q| and Mν Q,1 , . . . , Mν Q,|ν Q| . In this
language, our discussion is not within each MQ,k ’s, ie., not within the space a0Q ,
but between the MQ,i ’s, i.e. but within the space aG . Similar statements can be
Q
said for ν Q as well.
Accordingly, we introduce the sets
ν
∆ :=∆ν Q = αi : 1 ≤ i ≤ ν Q − 1 ,
n o
n o (16.81)
∆ :=∆ Q = β j : 1 ≤ j ≤ Q − 1 ,
and consider the assignment
n o
P 7−→ ∆P := γ j : 1 ≤ j ≤ |P| − 1 (16.82)
on the collection of parabolic subgroups P of G, which we denote by P. We want
to understand its affects on the collection
P(g) := P ∈ P : QP,g = Q, ν QP,g = ν Q, νP,g >P T .
n o
(16.83)
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 420

420 Geometric Structures and Riemann Hypothesis

For this, associated to g ∈ G(A), or better, to the principal P-bundle EP,g with
slope νP,g , we introduce the following sets:
g n o
∆0 := β j ∈ ∆ : hβ j , νQ,g i > 0 ,
g n o
∆ν := β j ∈ ∆ : hβ j , νQ,g i > hβ j , νi ≥ 0 , (16.84)
ν g ν
n o
∆T := αi ∈ ∆ : hαi , νν Q,g − T i > 0 .

Then, for P ∈ P(g), we have

(a) ∆P ⊆ ν ∆ since ν QP,g = ν Q ⊆ P. Similarly, ν ∆ ⊆ ∆ since Q ⊆ ν Q.

g
(b) ∆ν ⊆ ν ∆ since ν Q contains Q.
(c) ∆P ⊆ ν∆gT since νP,g >P T .
g
(d) ∆ = ∆P ∪ ∆0 since QP,g = Q.
ν g
(e) ∆ = ∆P ∪ ∆ν since ν QP,g = ν Q.

Consequently,
g g g
∆ν ⊆ ν ∆ ⊆ ∆ and (∆ r ∆0 ) ∪ ( ν ∆ r ∆ν ) ⊆ ∆P ⊆ ν∆gT . (16.85)

Therefore, the assignment P 7→ ∆P induces a well-defined map

g g g
ϕ(g) : P(g) −→ J ⊆ ν ∆ : ∆ν ⊆ ν ∆ ⊆ ∆, (∆ r ∆0 ) ∪ ( ν ∆ r ∆ν ) ⊆ J ⊆ ν∆gT .


Claim 16.5. The map ϕ is a bijection.

Proof. Both injectivity and surjectivity come directly from the fact that there
exists an order reversed bijection between the subsets of ∆0 and the standard
parabolic subgroups of G. Indeed, the infectivity is rather obvious. As for the sub-
jectivity, the existence of a parabolic subgroup comes from J ⊆ ν ∆. Moreover, by
reversing (a,. . . ,e) above, we see that QP,g = Q, ν QP,g = ν Q, and νP,g >P T. 

To go further, we recall the following well-known

Lemma 16.17. Let Y be a finite set and denote by Σ(Y) its power set, i.e. the set
consists of all the subsets of Y. Then

X  0 Y , ∅,

(−1) = 
|S |

(16.86)
 1 Y = ∅.

S ∈Σ(Y)
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 421

Analytic and Geometric Truncations: Reductive Groups 421

Consequently, for the fixed Q and ν Q as above, the condition that the right
hand of (16.79) is not zero implies that
g g g
∆ν ⊆ ν ∆ ⊆ ∆, ∆ r ∆0 ∪ ν ∆ r ∆ν = ν∆gT .
 
(16.87)

Therefore, the assignment (16.79) is a characteristic function of the subset

g g g
g ∈ Q(F)\G(A)/K : ∆ν ⊆ ν ∆ ⊆ ∆, ∆ r ∆0 ∪ ν ∆ r ∆ν = ν∆gT . (16.88)
n   o

Obviously, by definition, this set contains

g ∈ Q(F)\G(A)/K : Qg = Q, ν Qg = ν Q, νg ≤ T .
n o
(16.89)

Claim 16.6. Assume that T dominates ν. Then the set ν∆gT is empty provided that
g g g
∆ν ⊆ ν ∆ ⊆ ∆, ∆ r ∆0 ∪ ν ∆ r ∆ν = ν∆gT .
 
(16.90)

Proof. Assume otherwise that ν∆gT is not empty. Let k be the smallest number such
that βk of ∆ attains maximal among the hβ j , νg −T i’s for β j ∈ ∆. Since T dominates
g
ν, we certainly have βk ∈ ∆ν . On the other hand, βk ∈ ν ∆ and hence βk ∈ ν∆gT since,
by our assumption, ν∆gT , ∅. For the same reason, β j ∈ ∆0 . This contradicts with
g
g

the condition that ∆ r ∆0 ∪ ν ∆ r ∆ν = ν∆gT . Therefore, ν∆gT = ∅. 

g g g g
Since ∅ = ν∆gT = (∆ r ∆0 ) ∪ ( ν ∆ r ∆ν ), we have ∆ = ∆0 and ν ∆ = ∆ν . This to
say, Qg = Q and ν Qg = ν Q. Moreover, under the assumption that T dominates ν,
the relations νν Q,g ≤ T and ν Qg = ν Q implies that νg = νQ,g ≤ T . This verifies the
first claim, i.e, Claim 16.5. 

To prove Lemma 16.16, it suffices to take ν = 0 in the previous lemma. Indeed,

since T ∈ aG+
0 , hence it is T dominates 0. Moreover, since, by our assumption,
νg  T , we conclude that
n o
∅ = g ∈ Q(F)\G(A)/K : Qg = Q, νg ≤ T
(16.91)
= g ∈ Q(F)\G(A)/K : Qg = Q, 0 Qg = ν Q, νg ≤ T .
n o

Hence the map (16.79) is the constant map g 7→ 0. This proves Lemma 16.16 and
hence also Theorem 16.2. 
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 422

422 Geometric Structures and Riemann Hypothesis

16.3.2 Equivalence of Analytic and Geometric Truncations

Now we are ready to expose the following fundamental result on an intrinsic rela-
tion between Arthur’s analytic truncation and the geometric truncation of stability.

Theorem 16.3 (Arithmetic and Geometric Truncations). Let T ∈ aG+

0 ∪ {0}.
For every (standard) parabolic subgroup P of G,
ΛP,T 1(g) = 1(νP,g ≤P T ) ∀g ∈ P(F)\G(A)/K. (16.92)

In other words, the geometric truncation of stability is simply the analytic

truncation introduced by Arthur.

Proof. Recall that, for T ∈ a+0 and a (standard) parabolic subgroup P of G,

Arthur’s (relative) analytic truncation is defined by
ΛP,T 1(g) =
X P
X
(−1)dim aQ τPQ ([H(δg)] − T )
b
Q∈P:Q⊆P δ∈Q(F)\P(F)
standard (16.93)
X
dim aPQ
= (−1) τPQ ([H(g)] − T ).
b
Q∈P:Q⊆P

Moreover, by Definition 16.4, we have

τPQ (H(g) − T ) = 1(νQ,g
b P
>PQ T ), (16.94)
as desired. 

16.3.3 Properties of Analytic and Geometric Truncations

In this subsection, we will establish some fundamental properties for analytic and
hence geometric truncations. These properties are first proved by Arthur for ana-
lytic truncations ΛT under the condition that the parameter T is sufficiently regular
(see e.g. Theorem 7.1). Based on Theorem 16.3, we here show that all these prop-
erties can be generalized under a mild positivity condition
T ∈ a0P+ ∪ {0}. (16.95)
In the sequel, we call such a T positive.

16.3.3.1 Compactness

Theorem 16.3 is fundamental to many arithmetic applications. For our limited

purpose, we start with the compactness of moduli spaces.

Proposition 16.7. Let P be a standard parabolic subgroup of G.

December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 423

Analytic and Geometric Truncations: Reductive Groups 423

(1) ΛP,0 1 is the characteristic function of the compact subset

n o
g ∈ P(F)N(A)\G(A)1 /K : Eg,P semistable . (16.96)

(2) Let T ∈ a0P+ ∪ {0}. For any T 0 ∈ a0P+ , ΛP,T 1 is the characteristic function of
the compact subset
n o
g ∈ P(F)N(A)\G(A)1 /K : −T 0 ≤ νP,g ≤ T . (16.97)

When P = G and T is sufficiently regular, Arthur proves that ΛP,T 1 is a char-

acteristic function of a certain compact subset as precisely described in § 7.2.1.
Our result removes the condition that T is sufficiently regular.

Proof. Since a similar proof below for (2) works for (1), we only prove (2). For
this, by Theorem 16.3, ΛP,T 1 is the characteristic function of the moduli space
n o
g ∈ P(F)N(A)\G(A)/K : −T 0 ≤ νP,g ≤ T . (16.98)

So it suffices to prove the compactness. This consists of two steps. First, we

choose a sufficiently regular T 0 and prove that the compactness holds when T =
T 0 . Indeed, by Theorem 16.3, it suffices to verify the compactness for Arthur’s
analytic truncation ΛT 1. This is simply Theorem 7.3. To go further, we have to
0

show that (ΛP,T − ΛP,T )1 is the characteristic function of a compact subset. But
0

this is a direct consequence of Proposition 7.4 since, as a function of g, (ΛP,T −

ΛP,T )1 is both bounded and compactly supported modulo P(F)N(A) uniformly
0

for T varying in a compact set. 

This then leads to the problem whether there exist semi-stable compatible
arithmetic G-torsors of slope ν. For our limited purpose, we give the following:

Proposition 16.8. Assume that G is semi-simple. Then the moduli space MG,F (0)
is non-empty and hence with a non-zero volume.

Proof. This is a direct consequence of the works of Grayson in [33, 34], particu-
larly, §4 and §5 of [34]. Indeed, in the space G(F∞ )/K∞ , there exists a non-empty
manifold with boundary Xss such that its quotient under G(OF ) gives the moduli
space MG,F (0). Therefore, it suffices to prove that Xss is not empty. This is a
direct consequence of Properties (A1∼ A9) of [34]. Indeed, Xss is not only non-
empty, but a topological space whose boundary is homotopy equivalent to the Tits
building by a homotopy equivalence which respects the action of G(OF ). 
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 424

424 Geometric Structures and Riemann Hypothesis

16.3.3.2 Idempotence

We next show that the operator ΛP,T is idempotent for all positive T .

Proposition 16.9. Let T ∈ aG+

0 t {0} and let P be a parabolic subgroup of G. Then

ΛP,T ΛP,T 1 = ΛP,T 1,

 
(16.99)
ΛP,T (p) 1(νP,∗ ≤P T ) = 1(νP,∗ ≤P T ).
 

Proof. The proof is similar to that of Lemma 15.1. Indeed, by Theorem 16.3, we
may identify the geometric truncation 1(νP,g ≤P T ) and the analytic truncation
ΛP,T 1, and hence it suffices to prove the second equality. But for geometric trun-
cations, this is a direct consequence of the proof of Theorem 16.2. Indeed, for
g ∈ G(A), when νP,g ≤P T , from the definitions, the right hand side is one, and the
left hand side becomes ΛP,T 1 (g) = 1(νP,g ≤P T ) = 1 by Theorem 16.3 again.
On the other hand, if νP,g P T , by the proof of Theorem 16.2, the right hand
side becomes 0, while the right hand side gives ΛP,T 0, and hence becomes 0 as
well. 

16.3.3.3 Self-Adjoint Property

Next, we show that ΛP,T is self-adjoint with respect to the natural inner product.

Proposition 16.10. Let T be an element of aG+

0 ∪ {0}.

(1) Let φ1 and φ2 be two continuous functions on G(F)\G(A)1 /K. Assume that
φ1 is slowly increasing, and that φ2 is rapidly decreasing. Then
   
ΛT φ1 , φ2 = φ1 , ΛT φ2 . (16.100)

(2) Let P be a parabolic subgroup of G with Levi decomposition P = N M. Let φ1

and φ2 be two continuous functions on P(F)N(A)\G(A)1 /K. Assume that φ1
is slowly increasing, and that φ2 is rapidly decreasing. Then
   
ΛP,T φ1 , φ2 = φ1 , ΛP,T φ2 . (16.101)

Proof. When G = SLn , this is simply Lemma 15.2. Even the group G now is
more general, the proof for Lemma 15.2 works here as well, after changing C and
SLn in (15.4) to aG+
0 and G, respectively. 
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 425

Analytic and Geometric Truncations: Reductive Groups 425

16.3.3.4 Identifications among Analytic and Geometric Periods

As a direct consequence to the above basic properties for analytic and geometric
truncations, similar to Lemma 15.3, we have the following:

Theorem 16.4 (Analytic Periods are Geometric Periods). Let T ∈ aG+ 0 t {0}
and denote by Σ(T ) ⊂ G(F)\G(A)1 /K be the compact subset with ΛT 1 as its
characteristic function. Then, for an Eisenstein series EG/P (φ, λ, g) induced from
an L2 -automorphic form on the Levi subgroup of a parabolic subgroup P of G,
Z Z
E G/P (φ, λ, g)dµ(g) = ΛT E G/P (φ, λ, g) dµ(g).


(16.102)
Σ(T ) G(F)\G(A)1 /K

Proof. This is a direct consequence of Theorem 16.3 and Propositions 16.7, 16.9
and 16.10, since if we denote by DG,F a fundamental domain of G(F)\G(A)1 /K,
we have
Z Z 

ΛT E G/P (φ, λ, g) dµ(g) = ΛT ΛT E G/P (φ, λ, g) dµ(g)


D DG,F
Z G,F  Z
ΛT 1 (g) · ΛT E G/P (φ, λ, g) dµ(g) =
 

= E G/P (φ, λ, g)dµ(g)


DG,F Σ(T )

as desired. 

In the next chapter, we will use the results above to prove a weak Riemann
hypothesis for the Weng zeta functions of reductive groups over number fields.
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 427

Chapter 17

Weak Riemann Hypothesis for

Zeta Functions of Reductive Groups

17.1 Volumes of Semi-Stable Moduli Spaces

17.1.1 Parabolic Reduction, Stability and the Volumes

Let F be a number field and let G be a (split) reductive group of rank r over F.
Fix a minimal parabolic subgroup P0 of G with a Levi decomposition P0 = M0 N0 ,
where M0 , resp. N0 , denotes a Levi subgroup, resp. the nilpotent radical, of P0 .
For the constant function 1 on M0 (A), denote by E G/P0 (1, g, λ) the induced Eisen-
stein series on G(F)\G(A). In addition, for every non-negative T ∈ aG0 t{0}, denote
by Σ(T ) the compact subset in G(F)\G(A)1 /K defined by the characteristic func-
tion ΛT 1. By Theorems 16.3 and 16.4, Σ(T ) admits also a geometric interpretation
≤T
as the moduli spaces MG,F of K-admissible arithmetic G-torsors Eg satisfying the
condition that their associated canonical types νEg are ≤ T . In particular, Σ(0)
is the moduli space MG,F of semi-stable K-admissible arithmetic G-torsors. Our
≤T
main purpose is to calculate the volume of the moduli spaces MG,F .
Let P be a standard maximal parabolic subgroup and denote by α p ∈ ∆
be the corresponding simple root and set ∆ p = ∆ r {α p }. And, for the Weyl
group W associated to (G, P), introduce the set Wspa , resp. W 0 , of standard, resp.
(p-)special, elements by
n o
Wspa := w ∈ W : ∆ = α ∈ ∆ : wα ∈ ∆ ∨ wα < 0 ,

n o (17.1)
W 0 := w ∈ W : ∆ p = α ∈ ∆ p : wα ∈ ∆ ∨ wα < 0 .


Obviously, w ∈ Wspa , resp. w ∈ W 0 , if and only if Aw t Bw = ∆, resp. Iw t Jw = ∆ p ,

where
n o n o
Aw := α ∈ ∆ : wα ∈ ∆ and Bw := α ∈ ∆ : wα < 0 ,
n o n o (17.2)
Iw := α ∈ ∆ p : wα ∈ ∆ and Jw := α ∈ ∆ p : wα < 0 .

427
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 428

428 Geometric Structures and Riemann Hypothesis

In addition, set, for w ∈ Wspa and i ≥ 0,

n o n o
mi := α > 0 : hρ, α∨ i = i and mw,i := α > 0, wα < 0 : hρ, α∨ i = i ,
ni+1 := mi+1 − mi and nw,i+1 := mw,i+1 − mw,i .
ζF (1) := Res s=1b
Finally, for our own convenience, set b ζF (s). Then the volume of
Σ(T ) is evaluated in the following:

Proposition 17.1. Let T ∈ aG+

0 t {0}.

(1) The integration of E G/P0 (1; g, λ) over the truncated domain Σ(T ) is given by
ehwρ−ρ,T i ζ(hλ, α∨ i)
Z
E G/P0 (1; g, λ)dµ(g) =
X Y
.
b
α∈∆ hwλ − ρ, α i α>0,wα<0 b
∨
Q
≤T
MG,F w∈W ζ(hλ, α∨ i + 1)
In particular,
ζ(hλ, α∨ i)
Z
1
E G/P0 (1; g, λ)dµ(g) =
X Y
.
b
α∈∆ hwλ − ρ, α i α>0,wα<0 b
∨
Q
MG,F w∈W ζ(hλ, α∨ i + 1)
≤T
(2) The volume of MG,F is given by
X Q bni,w −ni (i)
i≥1 ζF
vol ∆ ∨
r2 r
· (2π) (−1) r−|Aw |
ehwρ−ρ,T i .
− hwρ, α∨ i)
Q
w∈Wspa α∈∆rwAw (1

In particular, the volume of moduli space MG,F is given by

Q bni,w −ni
i≥1 ζF (i)
X
vol ∆ · (2π)
∨
r2 r
(−1)r−|Aw |
.
α∨ i)
Q
w∈W α∈∆rwAw (1 − hwρ,
spa

When T is sufficiently regular, this result is proved in §7.3.4.

Proof. All becomes now very simple. (1) By Theorem 16.4,

Z Z
E G/P0 (1; g, λ)dµ(g) = ΛT E G/P0 (1; g, λ)dµ(g)
Σ(T ) G(F)\G(A)1 /K
X ehwρ−ρ,T i Y ζ(hλ, α∨ i)
= .
b
α∈∆ hwλ − ρ, α i α>0,wα<0 b
∨
Q
w∈W ζ(hλ, α∨ i + 1)
Here in the last step, we have used Corollary 7.5 and Lemma 7.4.
(2) The method in [57] works here, thanks for the basic properties established
for arithmetic geometric truncation in the previous section. Indeed, by a result of
Langland, the limit of E G/P0 (1; g, λ) when λ → ρ is a constant function indepen-
dent of g. More precisely,
Resλ→ρ E G/P0 (1; g, λ) = b
Y
ζF (1)r ζF (k)nk .
b (17.3)
k≥2
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 429

Weak Riemann Hypothesis for Zeta Functions of Reductive Groups 429

Z
Therefore, lim E G/P0 (1; g, λ)dµ(g) is equal to the product of the volume of
λ→ρ Σ(T )
Σ(T ) and b
ζF (1)r k≥2 b
ζF (k)nk . This leads to lim ωG,F
T
Q
(λ) where
λ→ρ
X e hwρ−ρ,T i Y ζF (hλ, α∨ i)
ωG,F .
T
b
(λ) := (17.4)
α∈∆ hwλ − ρ, α i α>0,wα<0 b
∨
Q
w∈W ζF (hλ, α∨ i + 1)
So the method of [57] can be used to give the formula above. 

Motivated by Theorem 7.6, we can write the above result in terms of parabolic
reduction. To explain this, we make some preparations first.
Recall that, if E is an arithmetic G-torsor of slope ν with (P, ν) its canonical
type of E, by Proposition 16.6, there exists a semi-stable arithmetic p-torsor EP
such that
E = EP ×P G and P , [ν]G = ν.
[νP ]GP ∈ aG+ (17.5)
Denote by MG,F (P, ν) the moduli space of arithmetic G-torsors with canonical
type (P, ν). In the sequel till the end of this subsection, for simplicity, we write P
for P.
Let P = MP NP be the Levi decomposition of P with MP , resp. NP , a Levi
subgroup, resp. the unipotent radical, of P. Denote by M MP ,F (ν) the moduli
spaces of semi-stable arithmetic MP -torsors of slope ν. There is a natural fibration
structure
π : MG,F (P, ν) −→ M MP ,F (ν)
. (17.6)
E 7−→ N\EP
Since NP (A)/NP (F), or better NP (R)/NP (Z) is a compact torus and NP admits
a filtration of one-dimensional affine spaces, by the conditions (a,b, c) for the
associated Haar measures in §6.1.5, the volume of the moduli space MG,F (P, ν) is
equal to the volume of the moduli space N\EP .
Moreover, from the theory of reductive groups explained in [36], if MP is a
connected semi-simple group over a field F, then the set {MP,i }i∈I of the minimal
non-trivial normal smooth connected F-subgroups of MP is finite, where, each
MP,i is F-simple,
Y the MP,i ’s are pairwise commutative, and the multiplication ho-
momorphism MP,i → MP is a central isogeny. More generally, for a connected
i
reductive group MP over a field F with Z its maximal central F- torus, if we de-
note by MP0 := D(MP ) its semi-simple derived group and let {MP,i } be the F-simple
factors of MP0 , then for the associated MP,i ’s, the multiplication map
Y
MP,i → MP (17.7)
i∈I
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 430

430 Geometric Structures and Riemann Hypothesis

is an central isogeny. (See [36], or more recently, [21].) Therefore, up to a central

isogeny, the reductive group MP may be viewed as a direct product of its central
torus and the associated semi-simple factors. Since for moduli spaces of arith-
metic torsos of fixed slopes, the central isogeny (17.7) can be ignored, to under-
stand MG,F (P, ν), it suffices to treat the product of moduli spaces i∈I M MP,i ,F (νi ).
Q
This argument works in particular for semi-simple groups. Therefore, with the
fact that all the MP,i ’s are semi-simple, by Proposition 16.8, we proves not only
the second part of the following theorem, but the first part on an explicit formula
for the volumes of MG,F (P, ν) in terms of i∈I Mtot
Q
MP,i ,F (νi ). Indeed, by applying
Theorem 2.1 and Theorem 2.4 in [71], we can obtain a close formula for the semi-
stable arithmetic principal G-torsors as what we have done in [130] for G-bundles
over curves. We leave this to the reader. Instead, we use an alternative analytic
method to achieve this same goal. For this, recall that, by the later half of §7.3.4,
there is a natural one-to-one correspondence between the elements w in Wspa and
the subsets Jw of ∆, and hence to the standard parabolic subgroups P of G. In
other words,
n o
Wspa = wP : P standard parabolic subgroup . (17.8)

Consequently, for a standard parabolic subgroup P, with the same notations as in

§7.3.4, we write wP ∈ Wspa the associated Weyl element in Wspa and set JP ⊆ ∆
be the corresponding subset. With all this, using Theorem 16.4 on the equivalence
between analytic and geometric periods, Theorem 7.6, is translate to the first part
of the theorem below. This, together with the discussions above, proves the fol-
lowing one of the main result of this book.

Theorem 17.1 (Parabolic Reduction, Stability & the Volumes). Let G be a

Chevalley groups over F. Then

≤
(1) The volume of MG,F T can be calculated in terms of parabolic reduction.
Namely,
  X 1
≤
vol MG,F T (ν) = (−1)|∆P | · Q
P:P⊆Q α∈∆\wP JP (1 − hwP ρ, α∨ i)
standard parabolic
  (17.9)
Y tot  hw ρ−ρ,T i
× vol  M MP.i (0) e J .
i
Q
(2) The volume of i∈I M Mi ,F (0), and hence the volumes of M M,F (0) and
MG,F (P, 0) are not zero.
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 431

Weak Riemann Hypothesis for Zeta Functions of Reductive Groups 431

17.2 Riemann Hypothesis for Zeta Functions of Reductive Groups

17.2.1 Geometric Interpretation of Discriminant

As a preparation for our proof of the weak Riemann hypothesis for zeta functions
of reductive groups, we expose the following ket connection.

Proposition 17.2. Let G be a reductive group over a number field F and let P be
a maximal subgroup of G. Then the discriminant of b ζFG/P (s) is (up to a non-zero
factor,) the volume of the moduli space MG,F (P, 0) of admissible arithmetic G-
torsors of slope zero and canonical type (P, 0).

Proof. By the proof of Proposition 17.1, the volume of the moduli space
MG,F (P, 0) is the same as the product of the volumes of the moduli spaces
M Mi ,F (0) of semi-stable admissible arithmetic Mi -torsors, where the Mi ’s are the
simple factors of the Levi subgroup of P. Hence it suffices to show that the dis-
criminant of b ζQG/P (s) coincides with this product. For this, we make a decomposi-
tion of the discriminant of b ζ G/P (s). We start with the following:
Q

Lemma 17.1. Let (Φ, Φ± , ∆, W, ρ) be the complete data of a reduced and irre-
ducible root system. Denote by λα : α ∈ ∆ be the set of fundamental dominate

weights. For a simple root α ∈ ∆, denote by Φα the subset of Φ which is orthog-
onal to λα . Then Φα , together with its intersections with Φ± and ∆, forms a root
system.

Proof. Write ∆ = α1 , . . . , αr and, if necessary by rearranging the order of the


αi ’s, we may and hence will assume that α = αr . Write Φr be the subset of
Φ which is orthogonal to λr and denote by Vr the subspace of the root space V
generated by Φr . Since hλr , α j i = 0 for all 1 ≤ j ≤ r − 1, we have α1 , . . . , αr−1

forms a basis of Vr . Since Φr ⊂ Φ and ∆ r {αr } ⊂ ∆, all the axiom for root system
satisfies by Φr , Φr ∩ Φ± , ∆r{αr }, Wr , ρr forms a root system as well, even it may


not be irreducible. Here Wr denotes the Weyl group associated to Φr , or the same
1 X
to ∆r{αr }, and ρr := α. 
2 α∈Φ+
r

Since Φr forms a root system, By Proposition in §11.3 of [44], there exists a

unique decomposition Φr = t`k=1 Φr (k) of Φr into the reduced and irreducible
root systems. For each 1 ≤ k ≤ `, denote by (Φr (k), Φr (k)± , ∆r (k), Wr (k), ρr (k))
the root system data of Φr (k). Assume, in addition, that the order of
α1 , . . . αr is compatible with this decomposition. That is, if we write ∆r (k) =
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 432

432 Geometric Structures and Riemann Hypothesis

α1 (k), . . . , αrk (k) and b ∆r (k) = λ1 (k), . . . , λrk (k) , then α1 , α2 , . . . , αr−1 =
 

α1 (1), . . . , αr1 (1), α1 (2), . . . , αr2 (2), . . . , α1 (`), . . . , αr` (`) , and


`
Y 1 X
∆r = t`k=1 ∆r (k), Wr = Wr (k) and ρr (k) = α.
k=1
2 α∈Φ (k)+
r

Consequently, if we rewrite the period ωGF (λ) as ωΦ ∆,F (λ) based on the fact that
ωΦ∆,F (λ) is essentially determined by the root system (Φ, ∆), then it makes sense for
Φr (k)
us to introduce the period ωΦ ω∆r (k);F λ(k) for all i = 1, . . . , `,
r


(λ
∆r r ) and the periods
where λ = i=1 (si +1)λi and λ(k) = j=1 (s j (k)+1)λi (k) with λ1 (k), . . . , λrk (k) =
Pr−1 Prk 

∆r (k) for all k = 1, , . . . , `. Assume that αr is the simple root corresponding to the
b
maximal parabolic subgroup P. Then, by Proposition,12.12 and the definition
of the discriminant ∆G/P F of the zeta function b ζFG/P (s), up to the multiple factor
α∈Φ+r r∆r ζF (hρ, α i + 1),
Q b ∨

∆G/P
F = lim ωΦ r
∆r ,F (λr ).
λr →ρr
In addition, from the orthogonality of the Φr (k)’s, we have easily
` `
Φr (k)
ωΦ Φr
lim ωΦ
Y Y
= ω λ(k) ω = ∆r (k),F λ(k) .
r (k)
r



∆r ;F (λ) ∆r (k);F and lim (λ
∆r ,F r )
λr →ρr λ(k)→ρr (k)
k=1 k=1
Moreover, since for each 1 ≤ k ≤ `, (Φr (k), ∆r (k)) is naturally associated to the
semi-simple algebraic group Mk , by Theorem 17.1, lim ωΦ ∆r (k),Q λ(k) coin-
r (k)


λ(k)→ρr (k)
cides with the volume of the moduli space M Mk ,Q (0) of the semi-stable admissible
arithmetic Mk -torsors of slope zero. This, together with Proposition 17.1, then im-
ζQG/P (s) is the volume of the moduli space MG,F (P, 0)
plies that the discriminant of b
of admissible arithmetic G-torsors of slope zero and canonical type (P, 0). 

17.2.2 Riemann Hypothesis for Weng Zeta Functions

Theorem 17.2 (Wean Riemann Hypothesis). Let G be a product of Chevalley

groups over Q, each of which simple factors is of rank r ≥ 1, and let P be a
maximal parabolic subgroup of G. Then, the Weng zeta function b ζQG/P (s) satisfies
ζQG/P (s) lie on the
a weak Riemann hypothesis, i.e. all but finitely many zeros of b
cP
central line <(s) = − .
2
Proof. This now becomes extremely simple. Indeed, by Theorem 12.1, we have
to show that the discriminant ∆G/P
F ζFG/P (s) is not zero. This then is a direct
of b
consequence of Propositions 17.2 and 16.8. 

We end this chapter by pointing out that when G is a group of type A, E, F, G,

this theorem has been proved earlier more concretely.
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 433

Chapter 18

Distributions of Zeros for Zeta Functions of

Reductive Groups

Following our work [131], we expose two levels of the distribution structures for
ζQ,n (s) and b
the zeros of b ζQG/P(s).

18.1 Distributions of Zeta Zeros

18.1.1 Distributions in Classical Style

A well-known conjecture on distributions of Riemann zeros claims that they re-

semble that of the Gaussian Unitary Ensembles. We here study the distributions
of zeros for the non-abelian zeta functions over Q defined by
Z  0 
ζQ,n (s) := eh (Q,Λ) − 1 · e−s degar (Λ) dµ.
b

MQ ,n
Here MQ,n denotes moduli space of semi-stable lattices of rank n. It is proved
in Theorem 3.2 that b ζQ,1 (s) = b ζ(s) coincides with the complete Riemann zeta
function and the bζQ,n (s)’s satisfy standard zeta properties. In particular, in Theo-
rem 15.4, we prove that, when n ≥ 2, all but finitely many zeros of b ζQ,n (s) lie on
the line <(s) = 21 . So it is natural to investigate the distributions of these zeta
zeros.
The initial works were independently done by Suzuki and myself on n = 2
with a disappointing outcome of the Dirac distribution, instead of that of GUE.
It took quite some times for the author to understand this. The turning point
was a joint work with Zagier ([134]) on the Riemann hypothesis for the high
rank zeta zeros of elliptic curves. From this, the author concludes that there are
two levels of the structures for the distributions of our zeta zeros. That is, the
classical one on the thetas measuring the arguments of the zeta zeros, and another
secondary one on a new big Theta, obtained from the original thetas by blowing-up
the infinitesimal structures around their accumulating points ([132]). In addition,

433
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 434

434 Geometric Structures and Riemann Hypothesis

while the classical theta leads to the Dirac distributions, the secondary big Theta
recovers the Sato-Tate type distributions for high rank zeta zeros (associated to
non-CM elliptic curves defined over Q). All these in turn motivate the current
works on a parallel structure for the zeros of bζQ,n (s).
√
ζQ,n (s) as ρ = 12 + −1 γ. Arrange the γ’s
To explain this, write the zeros of b
in an increasing order
0 ≤ γn,1 ≤ γn,2 ≤ · · · ≤ γn,3 ≤ . . . , (18.1)
and, as usual, denote by
Nn (T ) := # k : 0 < γn,k < T

(18.2)
ζQ,n (s) whose imaginary parts lie between 0 and T .
the number of zeros of b

ζQ,n (s), as T → ∞,
Theorem 18.1. Let n ≥ 21 . For the zeros of b
n n 2πe
(1) Nn (T ) = T log T − T log + O(1).
2π 2π n
2π k  log log k

(2) γn,k = 1+O .
n log k log k
2π 1  log log k 
(3) γn,k+1 − γn,k = +O .
n log k k log k
The structures presented in this theorem are pretty much similar to the one for
the Riemann zeros ( [116]). To go further, motivated by [76], based on (2) and (3)
above, we introduce

ζQ,n (s) is defined

Definition 18.1. The pair correlation function for the zeros of b
by
 n
  n 
δn,k := γn,k+1 − γn,k · log γn,k . (18.3)
2π 2π

Theorem 18.2. Let n ≥ 2. For the zeros of bζQ,n (s), as k → ∞,

 log log k 
δn,k = 1 + O . (18.4)
log2 k

18.1.2 New Secondary Distributions

Being of Dirac type, in particular, the distributions for the zeros of the non-abelian
zeta functions are very different from that of Riemann zeros, which conjecturally
coincide with that of GUE in the theory of random matrix. However, it turns out
1 Here and in the sequel, when n = 2, stronger results hold. For details, please see Proposition 18.1.
December 21, 2017 13:46 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 435

Distributions of Zeros for Zeta Functions 435

that there is yet another refined level of structure for the zeros of non-abelian zeta
functions. To explain this, motivated by our studies for non-abelian zeta functions
of function fields ([132]), we introduce the following:

Definition 18.2. The big Delta pair correlation functions for the zeros of high
rank non-abelian zeta function 
ζQ,n (s) are defined by
   n 
Δn,k := δn,k − 1 · log γn,k . (18.5)
2π
The distributions of Δn,k and δn for the high rank zeta zeros and the Riemann
zeros, respectively, are expected to be closely related ([83]). We, motivated by the
conjectural connection between Riemann zeros and GUE, make the following:

Conjecture 18.1. Denote by μ(Δn ) and μ(GUE) the probability measures associ-
ated to Δn,k and the Gaussian unitary ensemble, respectively. Then
 
limn→∞ Discrep μ(Δn,k ), μ(GUE) = 0. (18.6)

Here Discrep (μ, ν) denotes the Kolomogorov-Smirnov distance of μ and ν, up to

a normalization depending only on n.

This is supported by some very impressive numerical calculations on zeros of

low rank non-abelian zeta functions. For an example, based on numerical calcu-
lations for the first 138,068 zeros of the zeros rank two zeta function 
ζQ,2 (σ + it)
(with 0 ≤ t ≤ 50, 000), we obtain the following very impressive figures.

Fig. 18.1: Conjectural distributions Fig. 18.2: Numerical distributions

for the classical δk of the Riemann for the secondary Δ2,k of the first
zeros (oversimplified smooth curve) 138,068 rank two zeros
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 436

436 Geometric Structures and Riemann Hypothesis

Our method to prove the above results works for the zeta functions b ζQG/P (s)
as well. To explain this, let (G, P) be a pair consisting of a Chevalley group and
its maximal parabolic subgroup over Q. By Theorem 17.2, if the rank of G is
ζQG/P (s) satisfies a weak Riemann hypothesis. Consider the zeros
at least one, b
cP cP
ρ = − + γ i of b ζQG/P (s) on the central line <(s) = − . Arranging the γ’s in an
2 2
increasing order
0 ≤ γG/P
1 ≤ γG/P
2 ≤ · · · ≤ γG/P
k ≤ ..., (18.7)
and denote by
n o
N G/P (T ) := # k : 0 < γG/P
k <T (18.8)
the number of the zeros with imaginary parts lying between 0 and T . Set
∞
1X
dP := k · NP (k, [(kcP − 1)/2]),
2 k=1
∞
(18.9)
1X
eP := NP (k, [(kcP − 1)/2]) k log k
2 k=1

where NP (k, h) := # α ∈ Φ : hλ p , α∨ i = k, hρ, α∨ i = h .



Theorem 18.3. As T → ∞,
dP eP − dP log(2πe)
(1) N G/P (T ) = T log T + T + O(log T ).
π !! π
π n 1
(2) γG/P
n = 1+O .
dP log n log n
π 1
!
1
(3) γn+1 − γn =
G/P G/P
+O .
dP log n log2 n

Accordingly, introduce the pair correlation function, the small delta, for these
zeta zeros by
dP  G/P ! dP G/P
!
δG/P := γ − γG/P
· log γ . (18.10)
k π k+1 k π k

Theorem 18.4. As k → ∞,
!
1
δG/P
k =1+O . (18.11)
log k

ζ G/P
In other words, the pair correlation distributions for the zeros of b Q (s) are
simply of Dirac type. Motivated by ∆n,k , we introduce the following:
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 437

Distributions of Zeros for Zeta Functions 437

ζ G/P
Definition 18.3. The big Delta pair correlation functions for the zeros of b Q (s)
are defined by
!
 G/P  dP G/P
∆G/P := δ − 1 · log γ . (18.12)
k k π k

Similarly, we expect that the distributions of ∆G/P

n and δn are closely related.

18.2 Proof of Theorems

18.2.1 Rank Two Zeta Zeros

To start with, we prove a stronger result for rank two zeta zeros, obtained inde-
pendently by Suzuki and the author.

Proposition 18.1. For the γ2,k ’s of the zeros of b

ζQ,2 (s), as T → ∞,
1 1 log T

(1) N2 (T ) = T log T − T log(πe) + O .
π π log log T
k  1

(2) γ2,k = π 1+O .
log k log k
1 1
(3) γ2,k+1 − γ2,k = π +O


.
log k log k log log k
1
(4) δ2,k = 1 + O .


log log k

Proof. It suffices to prove the first three since (4) comes directly from (2) and (3).
1
We start with (1). Since bζ(2s − 1) = b ζ(2s) when s = + i t, by Theorem 5.9,
2
ζ(1 + 2it)
!
1 1 b
ζQ,2
b + it = · cos θ2 (t), (18.13)
2 2 − 12 + i t

where θ2 (t) denotes the argument of b ζ(1 + 2it)/(− 12 + it). Recall that in the proof
of the simplicity of the zeros, θ2 (t) is an increasing function in t. Hence,
θ2 γ2,n+1 − θ2 γ2,n = π.



(18.14)
Furthermore, it is not difficult to understand the asymptotic of
θ2 (t) = arg Γ 1/2 + it + arg ζ 1/2 + it − arg π 2 +it − arg − 1/2 + it . (18.15)


1

Indeed, by Stirling’s formula,

arg Γ(σ + it) = t log t − t + (σ − 1/2) π/2 + O(1/t). (18.16)
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 438

438 Geometric Structures and Riemann Hypothesis

In addition, from Theorem 5.16 of [116],

arg ζ(1 + 2it) = O log t/ log log t .

(18.17)

Consequently,
!
log t
θ2 (t) = t log t − t (1 + log π) + O . (18.18)
log log t
This, together with (18.14), implies that N2 (T ) = θ2 (T )/π, asymptotically, and
hence also (1). To prove (2), we use the relation

N2 (γ2,n − 1) ≤ n ≤ N2 (γ2,n + 1). (18.19)

This implies that γ2,n log γ2,n ∼ πn, and hence γn,2 ∼ π n/ log n. Consequently,
!
log(γ2,n ± 1)
π N2 (γ2,n ±1) = (γ2,n ±1) log(γ2,n ±1)−log π−1 +O . (18.20)


log log(γ2,n ± 1)
From this, with a simple manipulation, we obtain (2). To prove (3), we use (18.18),
or equivalently, (18.16) and (18.17) to conclude that, for sufficiently small h,
!! !
1 1
θ2 (t + h) − θ2 (t) = h log t 1 + O +O . (18.21)
log log t t
Applying this to t = γ2,n and h = γ2,n+1 − γ2,n , then (3) comes from (18.14). 

Following Proposition 18.1, we introduce the following modified

ζQ,2 (s)
Definition 18.4. The big Delta pair correlation functions for the zeros of b
are defined by
γ2,k

∆2,k = δ2,k − 1 · log log .


(18.22)
π

18.2.2 Proof of Theorems 18.3 and 18.4

ζQG/P (s).
Step 1. Fine symmetric structures of b
Without loss of generality, let P be a standard parabolic subgroup of G. De-
note by P = MP NP the Levi decomposition of P, nP the Lie algebra of NP , AP the
maximal central subgroup of MP , and aP := X ∗ (AP )R . Let ∆P be the set of roots for
(P, AP ), i.e. the finite subset of non-zero elements in X(AP )Q parametrizing the de-
composition nP = ⊕α∈ΦP nα of the eigenspace under the adjoint action Ad : AP →
GL(nP ) of AP , where nα := {Xα ∈ nP : Ad(a)Xα = aα Xα ∀a ∈ AP }. Note that
1X
ΦP ⊂ X(AP )Q ⊂ X(AP )Q ⊗ R ' a∗P . Accordingly, set ρP := (dim nα ) α.
2 α∈ΦP
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 439

Distributions of Zeros for Zeta Functions 439

By the theory of algebraic groups (see e.g. [14, 103]), there is an order revers-
ing bijection

P : standard parabolic subgroup of G ←→ ∆P ⊂ ∆

 

such that aP = H ∈ a : α(H) = 0 ∀α ∈ ∆P . This ∆P forms a basis of aP . Let ∆P be


the set of linear forms on aP obtained by the restrictions of the elements of ∆0 \∆0P .
X ∆P := α|aP ∈ aP : ∃α ∈ ∆0 \∆0 . It is well-known that for any α ∈ ΦP ,
 ∗ P
Namely,
α= nβ β with nβ ∈ Z≥0 . Even ∆P is not really a root system in the usual sense,
β∈∆P
with this property, it is still possible to introduce Φ±P such that ΦP = Φ+P t Φ−P ,
Φ−P = −Φ+P . Indeed, we can and hence will identify ΦP as a subset of Φ from
the above construction, so that, simply, Φ+P := Φ+ ∩ ΦP . In this language, then
1 X
ρP = α. In addition, set cP := 2hλP − ρP , α∨p i.
2 α∈Φ
P

From now on, assume that P is maximal. Then ∆P = {αP } = {α p } consisting of

a single element (1 ≤ p ≤ r). By definition, the period of (G, P)/Q is defined by
X
Q (s) =
ωG/P T w (s) (18.23)
w∈W

where

α∈∆ hλ − ρ, α i ζ (hλ, α∨ i)
Q ∨ Y
.
b
T w (s) := lim Q P
α∈∆ hwλ − ρ, α i
∨
λ→λP
α>0, wα<0 ζ (hλ, α∨ i + 1)
b

Since, for α ∈ ∆P , limλ→λ p hλ − ρ, α∨ i ≡ 0, to obtain a non-trivial T w (s) within the

period ωG/P
Q (s), there should be a total cancellation for all the factors hλ − ρ, α i,
∨

α ∈ ∆P . In particular, T w (s) . 0 if and only if ∆P ⊂ w−1 (∆∪Φ− ), since b ζ(s) admits

only two simple poles at s = 0, 1. Accordingly, by Proposition 9.1, we conclude
X n o
ωG/P
Q (s) = T
0 w
with W 0 := w ∈ W : ∆P ⊂ w−1 (∆ ∪ Φ− ) . (18.24)
w∈W

Even after taking the residues, there are still some zeta factors left in the de-
nominators of T w . To get ride of them, following (12.7), we first introduce an
‘overdone’ normalization for ωG/P
F (s) and set

Z p (s) := F p (s) ωG/P

F (s), (18.25)

ζF hλ p , α∨ i s+hρ, α∨ i . In addition, to make the discussion

Q

where F p (s) := α∈Φ− b
with in the space of entire functions, following (12.21), set
Y
Q p (s) := qP,w (s) and X p (s) := Q p (s) Z p (s), (18.26)
w∈W 0
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 440

440 Geometric Structures and Riemann Hypothesis

where
−1 +
Y Y
2|∆P ∩w Φ |
 
qP,w (s) := hλ s + ρ, α∨ i − 1
w∈W 0 α∈(w−1 ∆)∩∆P
Y   
× hλ s + ρ, α∨ i + δα,w hλ s + ρ, α∨ i + δα,w − 1 ,
α∈Φ+ \∆P

with δα,w = 1 if α ∈ w−1 Φ+ , 0 otherwise. Obviously, Q p (s) is a polynomial, and

by §12.2, X p (s) is entire,
Then, by (12.20) we may write down XP (s) as
X r
X p (s) = cFp,w c p,w a p,w (s) z p,w (s), (18.27)
w∈W 0

where
(Φ+ )|
−1
Y
c p,w := ξ(2)|∆ p ∩w ξF hρ, α∨ i + δα,w ,


α∈Φ+ r∆ p

(Φ+ )|
Y −1
Y  
a p,w (s) = 2|∆ p ∩v hλ p , α∨ is + hρ, α∨ i − 1
v∈W p0 r{w} α∈(v−1 (∆)r∆ p
Y   !
× hλ p , α is + hρ, α i + δα,v hλ p , α is + hρ, α i + δα,v − 1 ,
∨ ∨ ∨ ∨

α∈Φ+ r∆ p
Y
ξF hλ p , α∨ is + hρ, α∨ i + δα,w .


z p,w (s) :=
α∈Φ+ rΦ+p

Therefore, as in Definition 12.2, for w ∈ W 0 , set `±p,w := (Φ+ r Φ+p ) ∩ w−1 (Φ± ) ,
and
W p< := w ∈ W p0 : `+p,w < |Φ+ r Φ+p |/2 ,


W p> := w ∈ W p0 : `+p,w > |Φ+ r Φ+p |/2 ,


(18.28)
W p= := w ∈ W p0 : `+p,w = |Φ+ r Φ+p |/2 ,


W p≤ := W p< t W p= and W p≥ := W p> t W p= ,

then, by Proposition 12.1, we have,
X p (s) = B p (s) ± B p (−c p − s), (18.29)
 
 X 1 X  r
where B p (s) :=   +  c p,w c a (s) z (s).
 F p,w p,w p,w
w∈W >
2 w∈W =
p p

Finally, to remove the extra factors in the overdone normalization above, we

introduce the function
XP (s)
ξQG/P (s) := (18.30)
RP (s)DP (s)
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 441

Distributions of Zeros for Zeta Functions 441

where
∞ Y
Y ∞
DP (s) := ξ(ks + h)NP (k,h−1)−MP (k,h) ,
k=1 h=2

RP (s) := g.c.d. QP,w : w ∈ WP .



BP (s)
Then, by Proposition 12.2, ξQG/P (s) is entire, and, for AP (s) := ,
RP (s)DP (s)

ξQG/P (s) = AP (s) ± AP (−cP − s). (18.31)

Since, the quotient of ξG/P (s) and b

ζ G/P
Q (s) is a well-determined rational function,
ζ G/P (s), it suffices to study ξG/P (s).
to understand the distributions of zeros of b Q Q

− cP /2 + i t .


Step 2. Asymptotic behaviors for the arguments of AP
Let θP (t) be the argument of AP − cP /2 + it . By (18.31),


 c   c  
ξQG/P − + i t = AP − + i t · ei θP (t) ± e−i θP (t) ,
P P

(18.32)
2 2

since AP − cP /2 + it = AP − cP /2 − it . Hence, the zeros of ξQG/P (−cP /2 + i t)

correspond in one-to-one to the zeros of cos θP (t) or sin θP (t), or equivalently,
to the solutions of either θP (t) ∈ π/2 + π Z or θP (t) ∈ π Z. Therefore, it suffices
to obtain the asymptotic behaviors of the function θP (t) as |t| → +∞. For this
purpose, let W p := w ∈ W 0 : `−p,w = 0 and

X r
z p (s) := z p,id (s) and a p (s) := cFp,w c p,w a p,w (s). (18.33)
w∈W p

Then, by the proof of Corollary 12.6,

a p (s) z p (s)  
A p (s) = · · 1 + rP (s) and rP (s) < 1. (18.34)
RP (s) DP (s)
Since | rP (s) | < 1, we have arg 1 + rP (s) ≤ π2 . Consequently,


! !
a p (s) z p (s)
θP (t) = arg + arg + O(1).
RP (s) s=− c2P +it D p (s) s=− c2P +it

The first term is simply O(1), since a p (s), RP (s) are polynomials. To treat the
second term, we use the formula (12.58) that
∞
z p (s) Y Y
= ξ(ks + h)NP (k,h−1)−NP (k,h) .
D p (s) k=1 h>(kc +1)/2
P
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 442

442 Geometric Structures and Riemann Hypothesis

Note that 1 ≥ < (ks + h) = − c2P k + h > 12 for s = − c2P + it. We can assume
 cP 
that arg ζ − k + h + ikt = O(log |t|) obtained under the Riemann Hypothesis.
2
Thus, from the Stirling formula,
∞
z p (s) X X  
arg cP = NP (k, h − 1) − NP (k, h)
DP (s) s=− 2 +it k=1 h>(kc +1)/2
P

× arg(ks + h)(ks + h − 1) s=− cP +it
2
! !
cP h ikt cP h ikt  c
P
+ arg π 4 k− 2 − 2
+ arg Γ − k + + + arg ζ − k + h + ikt
4 2 2 2
X∞ X  
= NP (k, h − 1) − NP (k, h)
k=1 h>(kcP +1)/2
! ! ! !
kt kt kt 1
× O(1) − log π + log −1 +O + O(log t)
2 2 2 t
∞ " #! !
X kcP − 1 kt
= log t − log(2πe) + log k + O(log t) .


NP k, ·
k=1
2 2
All these then prove the following proposition and hence also Theorem
18.3(1).

Proposition 18.2. The asymptotic behaviors of θP (T ) is given by

θP (T ) = T log T · dP + T · eP − dP log(2πe) + O(log T ).

ζQG/P (s).
Step 3. Distributions of zeros for b

To complete our proof of Theorems 18.3 and 18.4, we use

Lemma 18.1. Assume that there are constants C, C 0 , C1 , C2 satisfying

   
θP γG/P
n+1 = θP γn
G/P
+ C,
1
N G/P (γG/P
n ) ∼ θP (γG/P
n ) + O(1),
(18.35)
C0
N (T ) = C1 T log T + C2 T + O(log T ).
G/P

Then !! !
1 n 1 1 1 1
γG/P
n = 1+O and γG/P
n+1 − γG/P
n = +O .
C1 log n log n C1 log n log2 n

Proof. We start with the dominant term for γn = γG/P

n . From our assumption on
N = N G/P , obviously,
N(γn ± 1) ∼ C1 (γn ± 1) log(γn ± 1) + C2 (γn ± 1) ∼ C1 γn log γn . (18.36)
December 21, 2017 12:32 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 443

Distributions of Zeros for Zeta Functions 443

In addition, by definition, N(γn − 1) ≤ n ≤ N(γn + 1). Hence

n ∼ C1 γn log γn and log n ∼ log γn . (18.37)
1 n
Consequently, γn ∼ . To get the precise asymptotic behaviors, we use
C1 log n
N(γn ± 1) = C1 (γn ± 1) log(γn ± 1) + C2 (γn ± 1) + O log(γn ± 1) .

1 n
As above, then n = C1 γn log γn + O(γn ), or better, since γn ∼ ,
C1 log n
 1

C1 γn log γn = n · 1 + O . (18.38)
log n
This verifies the first assertion.
For the second, we shift our attention to θ = θP . Then, for T  0, ∆T > 0,
θ(T +∆T ) − θ(T ) = C 0 N G/P (T + ∆T ) − C 0 N G/P (T )
T + ∆T T + ∆T
!
= C C1 T log
0
+ C C1 ∆T log(T + ∆T ) + C C2 ∆T + O log
0 0
T T
!
1
= C 0C1 ∆T log T + 1 + O ,


T
!T
T + ∆T ∆T
since T log = log 1 + = O(1). In particular, by taking T = γn and
T T
∆T = γn+1 − γn , we get
C = θ(γn+1 ) − θ(γn ) = C 0C1 (γn+1 − γn ) log γn + 1 + O 1/γn .

This implies
!
C 1 1
γn+1 −γn ∼ 0 and C = C C1 (γn+1 −γn ) log γn +O
0
. (18.39)
C C1 log γn log γn
Therefore,
!
C 1 1
γn+1 − γn = + O .
C 0C1 log n log2 n
This then completes the proof of the lemma and hence Theorems 18.3 and 18.4,
since we can take
dP eP − dP log(2πe)
C = C 0 = π, C1 = and C2 = (18.40)
π π
ζ G/P (s). Indeed, by the proof of the simplicity of zeros of
for the zeta function b Q
ζQG/P (s), θP (t) is monotone and hence θP (γG/P
b
k+1 ) − θP (γk ) = π. This proves Theo-
G/P

rem 18.3(1) under Proposition 18.2 and hence gives (18.40). 

January 2, 2018 9:13 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 444

444 Geometric Structures and Riemann Hypothesis

18.2.3 Proof of Theorems 18.1 and 18.2

Proof. Theorem 18.2 is a direct consequence of Theorem 18.1, so it suffices to

prove Theorem 18.1. For this, note that, when P = Pn−1,1 , Φ+ = ei − e j 1 ≤ i <

j ≤ n , ρ = 12 ni=1 (n+1−2i) ei , and λP = 1n e1 +· · ·+en−1 −(n−1) en . So, for i < j,
P

hλP , ei − e j i = δ jn . Consequently, NP (k, h) = 0 unless k ≤ 1. This implies that
eP = 0 and 2dP = NP 1, n − 1/2 when n ≥ 3. By a direct calculation, we know
 

that then dP = 1/2. Consequently, we have, for the k, h involved, −cP /2 k + h ≥ 1.
So, for zeros of non-abelian zeta functions, we do not really need to assume the
Riemann Hypothesis as in Step 2 above. This, together with Theorems 18.3 and
18.4, completes the proof of Theorem 18.1 and hence also Theorem 18.2. 

We end this discussion with the comment that Theorems 18.1 and 18.2 can
also be proved directly using Theorem 9.1.

Added during proofreading:

Motivated by our consideration of secondary structures for pair correlations of
non-abelian zeta zeros, Suzuki proves a refined result for the correlations distribu-
tions of rank 2 zeta zeros. This in turn motivates the following result of ours for all
rank n (n ≥ 2) zeta zeros. To explain this, introduce the well-known M-function
Z X ∞

1 λz (n)λz̄ (n)   
M(z, σ) :=

· exp − i<(z̄w) dw. (18.41)
n2σ
 
2π C n=1 

∞  ∞ !k
X i k Λk (n) k X Λk (n) ζ0
Here λz (n) := − z with Λk (n) defined by := (s) .
k=0
2 k! n=1
ns ζ
The M-function M(z, σ) is real valued, decays rapidly as |z| → ∞. Since
λz (m)λz (n) = λz (mn) for (m, n) = 1, M(z, σ), being a Fourier transform, admits a
natural convolution Euler product.

Theorem 18.5. Let ϕ(x) be a function in C 1 (R)! such that ϕ0 (x)  1 for |x| < 1,
d ζ0  n 
ϕ0 (x)  x−2 for |x| ≥ 1, and ϕ < + int is bounded on R. Then for n ≥ 2,
dt ζ 2
Z ∞ 
1 X 1 n
ϕ ∆n,k = ϕ(u) du.


lim m u, (18.42)
T →∞ Nn (T ) 2π −∞ 2
0<γk ≤T
Z ∞
Here m(u, σ) := M(u + iv, σ) dv.
−∞
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 445

APPENDICES

Five Essays on Arithmetic Cohomology

(Joint with K. Sugahara)

This part presents a series of joint works with K. Sugahara [106–108] on adelic
arithmetic cohomology groups over arithmetic varieties and applications. To be
more precise, in Appendix A, we start with a general construction for these co-
homology groups, based on the pioneer works of Parshin and Beilinson on adelic
cohomology groups for quasi-coherent sheaves over Noetherian schemes, a gener-
alization of the Osipov-Parshin tensor products in dimension one ind-pro topology
to all dimensions and a general uniformity principal. Then in Appendix B, we shift
our attention to arithmetic surfaces. Here we construct the arithmetic cohomol-
ogy groups for Weil divisors concretely and establish some of the basic algebraic
properties for them. In Appendix C, we give an intensive study of the ind-pro
topology in arithmetic dimension two. In Appendix D, we prove the topological
duality for the associated cohomology groups and show that the middle arithmetic
cohomology groups are always finite. Finally, in Appendix E, as an application of
these arithmetic cohomology constructions, we construct some arithmetic central
extensions and hence prove several reciprocity laws for arithmetic surfaces, using
some ideas of Osipov and Parshin for algebraic surfaces and the Arakelov theory.

445
 b2530   International Strategic Relations and China’s National Security: World at the Crossroads

This page intentionally left blank

b2530_FM.indd 6 01-Sep-16 11:03:06 AM

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 447

Appendix A

Arithmetic Adelic Complexes

A.1 Parshin-Beilinson’s Theory

We first recall some basic constructions of adelic cohomology theory for algebraic
varieties following Parshin and Beilinson ([10, 91, 92], see also [43]).

A.1.1 Local Fields for Reduced Flags

Let K be a number field with OK the ring of integers, and π : X → Spec OK be

an integral arithmetic variety. By a flag δ = (p0 , p1 , . . . , pn ) on X, we mean a
chain of integral subschemes pi satisfying pi+1 ∈ {pi } =: Xi , and we call δ reduced
if dim pi = n − i for each i. For a reduced δ, with respect to each affine open
neighborhood U = Spec B of the closed point pn , we obtain a chain, denoted also
by δ with an abuse of notation, of prime divisors on U. Consequently, through
processes of localizations and completions, we can associate to δ a ring
p0 . . . C pn S pn B.
Kδ := C p0 S −1 −1
(A.1)
Here, as usual, for a ring R, an R-module M and a prime ideal p of R, we write
p M for the localization of M at S p = R\p, and C p M = lim M/p M its p-adic
n
S −1
←n∈N
completion. The ring Kδ is independent of the choices of B. Indeed, following
0
[92], we can introduce inductively schemes Xi,αi
as in the following diagram
X0 ⊃ X1 ⊃ X2 ⊃ · · ·
↑ ↑
X00 ⊃ X1,α1 ⊃ ↑
↑ (A.2)
0
X1,α1
⊃ X2,α2
↑
..
.

447
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 448

448 Five Essays on Arithmetic Cohomology

where X 0 denotes the normalization of a scheme X, and Xi,αi denotes an integral

0
subscheme in Xi−1,αi−1
which dominates Xi . In particular,

(i) X1,α1 , being an integral subvariety of the normal scheme X00 , defines a discrete
valuation of the field of rational functions on X0 , whose residue field coincides
0
with the field of rational functions on the normal scheme X1,α 1
.
(ii) More generally, for a fixed i, 1 ≤ i ≤ n, Xi,αi , being an integral subvariety of
0
the normal scheme Xi−1,α i−1
, defines a discrete valuation of the field of rational
0
functions on Xi−1,αi−1 , whose residue field coincides with the field of rational
0
functions on the normal scheme Xi,α i
.

Accordingly, for each collection (α1 , α2 , . . . , αn ) of indices, the chain of field of ra-
tional functions K0 , K1,α1 , . . . , Kn,αn defines an n-dimensional local field K(α1 ,...,αn )
and hence an Artin ring

Kδ := ⊕(α1 ,...,αn )∈Λδ K(α1 ,...,αn ) .

Theorem A.1. ([91, 92, 136]) Let δ = (p0 , p1 , . . . , pn ) be a reduced flag on X, and
K(α1 ,...,αn ) be the n-dimensional local field associated to the collection of indices
(α1 , α2 , . . . , αn ) above. Then

(1) The n-dimensional local field K(α1 ,...,αn ) is isomorphic to

either Fv0 ((tn−1 )) · · · ((t1 )), or Fv0 {{tn }} · · · {tm+2 }}((tm )) . . . ((t1 ))

where Fv0 denotes a certain finite extension of some v-adic non-Archimedean

local field Fv .
(2) The ring Kδ is isomorphic to Kδ . In particular, it is independent of the choices
of U.

For example, if X is an arithmetic surface, and p1 is a vertical curve, then,

Kδ = Fπ(p2 ) {{u}}2 , where u denotes a local parameter of the curve p1 at the point
p2 , and Fπ(p2 ) denotes the finite residue field associated to the closed point π(p2 )
on Spec OK . On the other hand, if p1 is a horizontal curve, then Kδ = L((t)), where
t is a local parameter of p1 at p2 , and L/F is a finite field extension. Indeed, p1
corresponds to an algebraic point on the generic fiber XF of π, and L is simply the
corresponding defining field.
2 Definition of Fπ(p2 ) {{u}} will be recalled in §B.1.1.3, see e.g. (B.4).
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 449

Arithmetic Adelic Complexes 449

A.1.2 Adelic Cohomology Theory

Let X be a Noetherian scheme, and let P(X) be the set of (integral) points of X (in
the sense of Grothendieck). For p, q ∈ P(X), if q ∈ {p}, we write p ≥ q. Let S (X)
be the simplicial set induced by (P(X), ≥), i.e. the set of m-simplices of S (X) is
defined by
S (X)m := (p0 , . . . , pm ) | pi ∈ P(X), pi ≥ pi+1 ,

(A.3)
the natural boundary maps δni are defined by deleting the i-th point, and the de-
generacy maps σni are defined by duplicating the i-th point:
δm
i : S (X)m → S (X)m−1 , (p0 , . . . , pi , . . . , pm ) 7→ (p0 , . . . , p̌i , . . . , pm ),
(A.4)
σm
i : S (X)m → S (X)m+1 , (p0 , . . . , pi , . . . , pm ) 7→ (p0 , . . . , pi , pi , . . . , pm ).

Denote also by S (X)red

m the subset of S (X)m consisting of all non- degenerate sim-
plifies, i.e.

m = (p0 , . . . , pm ) ∈ S (X)m pi , p j ∀i , j .
S (X)red

(A.5)
For p ∈ P(X) and M an O p -module, set [M] p := (i p )∗ M, where i p : Spec(O p ) ,→
X denotes the natural induced morphism. Moreover, for K ⊆ S (X)m and a point
p ∈ P(X), introduce p K ⊆ S (X)m−1 by
p K := (p1 , . . . , pm ) ∈ S (X)m−1 | (p, p1 , . . . , pm ) ∈ K ∩ S (X)m .
 red red
(A.6)
Then, we have the following

Proposition A.1. ([10, 91, 92], see also Proposition 2.1.1 of [43]) There exists
a unique system of functors {A(K, ∗)}K⊆S (X) from the category of quasi-coherent
sheaves on X to the category of abelian groups, such that

(1) A(K, ·) commutes with direct limits.

(2) For a coherent sheaf F on X,
Y
lim F p mlp F p , m = 0,
 



 ←− l

 p∈K
A(K, F ) = 

(A.7)

 Y
lim A p K, [F p mlp F p ] p , m > 0.


 


 p∈P(V) ←− l



Here m p denotes the prime ideal associated to p.

Consequently, for any quasi-coherent sheaf F on X, there exist well-defined

adelic spaces

m ,F .
AmX (F ) := A S (X)red


(A.8)
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 450

450 Five Essays on Arithmetic Cohomology

Indeed, for coherent sheaves F , we can uniquely define {A(K, F )}K⊆S (X) by (2);
and, for quasi-coherent sheaves F , in addition, we use (1), since any such F can
be obtained as direct limits of coherent sheaves on X. Clearly, if we introduce
Ki0 ,...,im = (p0 , . . . , pm ) ∈ S (X)m | codim {pr } = ir ∀ 0 ≤ r ≤ m , and define


AX;i0 ,...,im (F ) := AX Ki0 ,...,im , F , then


M
AmX (F ) = AX;i0 ,...,im (F ). (A.9)
0≤i0 <···<im ≤dim X

Moreover, since A(K, F ) ⊂ (p0 ,...,pm )∈K A (p0 , . . . , pm ), F , we sometimes write

Q

an element K of A(K, F ) as f = ( f p0 ,...,pm ) or f = ( fX0 ,...,Xm ), where Xi = {pi } and

f p0 ,...,pm = fX0 ,...,Xm ∈ A (p0 , . . . , pm ), F .

To get an adelic complex associated to X, we next introduce boundary maps

dm : Am−1 m
X (F ) → AX (F ) as in Definition 2.2.2 of [43]. For K ⊆ S (X)m and L ⊆
S (X)m−1 such that δm i K ⊆ L for a certain i, we define a boundary map

dim (K, L, F ) : A(L, F ) −→ A(K, F ) (A.10)

as follows.

(a) For coherent sheaves F ,

(i) When i = 0, for p ∈ P(X), induced from the morphism F →
[F p /mlp F p ] p and the inclusion p K ⊆ L, we obtain natural induced
morphisms A(L, F ) → A L, [F p /mlp F p ] p and A L, [F p /mlp F p ] p →



A p K, [F p /mlp F p ] p . Their compositions form a projective system ϕlp :


A(L, F ) → A p K, [F p /mlp F p ] p . Accordingly, we set d0m (K, L, F ) :=

p∈P(X) lim←−l ϕ p ;
Q l

(ii) When i = m = 1, we obtain a projective system induced from standard

morphisms πlp : Γ X, [F p /mlp F p ] p → A p K, [F p /mlp F p ] p . Accordingly,



we set d11 (K, L, F ) := p∈P(X) lim←−l πlp ;
Q
(iii) When i > 0, m > 0, we use an induction on (i, m). That is to say, we set
dim (K, L, F ) := p∈P(X) lim←−l di−1 p K, p L, [F p /m p F p ] p .
Q m−1 l

(b) For quasi-coherent sheaves F , first we write F as an inductive limit of coher-

ent sheaves, then we use (a) to get boundary maps for the later, finally we use
the fact that in the definition of (a), all constructions commute with inductive
limits. One checks easily that the resulting boundary map is well-defined.

With this, we set

m
X
m , S (X)m−1 ; F .
(−1)i dim S (X)red red
dm :=


(A.11)
i=0
Then we have the following
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 451

Arithmetic Adelic Complexes 451

Theorem A.2. ([10,91,92], see also Theorem 4.2.3 of [43]) Let X be a Noetherian
scheme. Then, for any quasi-coherent sheaf F over X,

(1) A∗X (F ), d∗ forms a cohomological complex of abelian groups;

(2) The cohomology groups of the complex A∗X (F ), d∗ coincide with the


Grothendieck’s sheaf theoretic cohomology groups H i X, F . That is to say,

H i A∗X (F ), d∗ = H i X, F



∀i. (A.12)

A.1.3 Example

Let X be an integral regular projective curve defined over a field k. Denote its
generic point by η and its field of rational functions by k(X). For a divisor D on X,
let OX (D) be the associated invertible sheaf. Then, from definition, the associated
adelic spaces can be calculated as follows:
AX;0 (OX (D)) = A {η}, OX (D)


(A.13)
= lim OX (D)η mlη OX (D)η = lim k(X)/{0} = k(X),

←−l ←−l

AX;1 (OX (D)) = AX {p} | p ∈ X : closed point}, OX (D)

Y Y −ord (D)  −ord (D)+l
= lim OX (D) p mlp OX (D) p = lim m p p m p p

←−l ←−l (A.14)
p∈X p∈X
Y −ord (D)
n Y o
= mp p = (a p ) ∈ k(X) p ord p (a p ) + ord p (D) ≥ 0 ,
p∈X p∈X

and
AX;01 (OX (D)) = AX {η, p | p ∈ X : closed point}, OX (D)

= lim A {p} | p ∈ X : closed point}, [OX (D)η mlη OX (D)η ]η



←−l

= A {p} | p ∈ X : closed point}, [k(X)]η

= A {p} | p ∈ X : closed point}, lim OX (E)

−→E
(A.15)
[
= lim AX;1 (OX (E)) = AX;1 (OX (E))
−→E
E
n Y o
= (a p ) ∈ k(X) p a p ∈ O p ∀0 p .
p∈X

Remark A.1. To calculate AX;01 (OX (D)), when dealing with the constant sheaf
[k(X)]η , we cannot use Proposition A.1(2) directly, since [k(X)]η is not coherent.
Instead, above, we first expressed it as an inductive limit of coherent sheaves
OX (E) associated to divisors E, then get the result from the inductive limit of
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 452

452 Five Essays on Arithmetic Cohomology

adelic spaces for OX (E)’s. Indeed, if we had used Proposition A.1(2) directly,
then we would have obtained simply A {p} | p ∈ X : closed point}, [k(X)]η = {0},


a wrong claim.

Clearly, AX;01 (OX (D)) is independent of D. We will write it as AX;01 , or simply

AX . Consequently, the associated adelic complex
d1
0 −→ AX;0 (OX (D)) ⊕ AX;1 (OX (D)) −→ AX;01 (OX (D)) −→ 0 (A.16)
1
d
is given by 0 −→ k(X) ⊕ AX;1 (OX (D)) −→ AX −→ 0 where d1 : (a0 , a1 ) 7→ a1 − a0 .
Therefore,
H 0 AX (OX (D)) = k(X) ∩ AX:1 (OX (D)),


(A.17)
H 1 AX (OX (D)) = AX k(X) + AX;1 (OX (D)) .



Note that AX (OX (D)) is simply AX (D) of [98]. We have proved the following:

Proposition A.2. For a divisor D over an integral regular projective curve de-
fined over a field k, we have

H 0 AX (OX (D)) = H 0 X, OX (D) , H 1 AX (OX (D)) = H 1 X, OX (D) . (A.18)

A.2 Arithmetic Cohomology Groups

Let K be a number field with OK the ring of integers. Denote by S fin , resp. S ∞ ,
the collection of finite, resp. infinite, places of K. Write S = S fin ∪ S ∞ . Let
π : X → Spec OK be an integral arithmetic variety of pure dimension n + 1. That
is, an integral Noetherian scheme X, a proper morphism π with generic fiber XF
a projective variety of dimension n over K. For each v ∈ S , we write Fv the v-
completion of K, and for each σ ∈ S ∞ , we write Xv := X ×OK SpecFv and write
ϕσ : Xσ → XF for the map induced from the natural embedding F ,→ Fσ . In
particular, an arithmetic variety X consists of two parts, the finite one, which we
also denote by X, and an infinite one, which we denote by X∞ . These two parts
are closely interconnected.

A.2.1 Adelic Rings for Arithmetic Surfaces

The part of our theory on arithmetic adelic complexes for finite places now be-
comes very simple. Indeed, our arithmetic variety X is assumed to be Noetherian,
so we can apply the theory recalled in §A.1 directly. In particular, for a quasi-
coherent sheaf F on X, we have well-defined adelic spaces
X; i0 ,...,im (F ) := AX (Ki0 ,...,im , F ).
Afin (A.19)
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 453

Arithmetic Adelic Complexes 453

So to define AarX; i0 ,...,im (F ), we need to understand what happens on X∞ . For

this purpose, we next recall Osipov-Parshin’s construction of arithmetic adelic
ring Aar
X for an arithmetic surface X.

Definition A.1. ([87]) (Arithmetic adelic ring of an arithmetic surface) Let π :

X → Spec OK be an arithmetic surface, i.e. a 2-dimensional arithmetic variety,
with generic fiber XF .

(1) (Finite adelic ring) From the Parshin-Beilinson theory for the Noetherian
scheme X, we define
X := AX;012 (OX ) = lim
Afin

−→
lim
←−
AX;12 (D1 ) AX;12 (D1 ). (A.20)
D1 D2 : D2 ≤D1
Here D∗ ’s are divisors on X and AX;12 (D) := AX;12 (OX (D∗ )) for ∗ = 1, 2;
(2) (∞-adelic ring) Associated to the regular integral curve XF over K, we obtain
the adelic ring
AXF := AXF ;01 (OXF ) = lim

−→
lim
←−
AXF ;1 (D1 ) AXF ;1 (D1 ). (A.21)
D1 D2 : D2 ≤D1
Here D∗ ’s are divisors on XF and AXF ;1 (D) := AXF ;1 (OXF (D∗ )) for ∗ = 1, 2.
By definition,
 
A∞ ⊗Q R . (A.22)


X := AXF ⊗ Q R := lim lim AXF ;1 (D1 ) AXF ;1 (D1 ) b
b
−→ ←−
D1 D2 :D2 ≤D1
(3) (Arithmetic adelic ring) The arithmetic adelic ring of an arithmetic surface X
is defined by
M
Aar A∞X.
ar fin
X := AX;012 := AX (A.23)

The essential point here is, for divisors Di , i = 1, 2, over the curve XF , when

D2 ≤ D1 , the quotient AX;1 (D1 ) AX;1 (D1 ) is a finite dimensional K- and hence
Q-vector space.
To help the reader understand the above formal definition in concrete terms,
we add the following examples.

Example A.1. On X = P1Z , we have XQ = P1Q and Q P1Q = Q((t)). Easily,

Q((t)) ⊗Q R , R((t)). (A.24)

However, since Q((t)) = lim −n  −m −n  −m
lim t Q[[t]] t Q[[t]] and t Q[[t]] t Q[[t]]’s
−→ ←−
n m: m≤n
are finite dimensional Q-vector spaces, we have
⊗ Q R = lim lim t−n Q[[t]] t−m Q[[t]] ⊗ Q R


Q((t)) b −→ ←−
n m: m≤n
(A.25)
= lim lim t−n R[[t]] t−m R[[t]] = R((t)).


−→ ←−
n m: m≤n
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 454

454 Five Essays on Arithmetic Cohomology

Example A.2. Over an arithmetic surface X, for a complete flag (X, C, x) on X

(with C an irreducible curve on X and x a close point on C), let k(X)C,x its associ-
ated local ring. By Theorem A.1, k(X)C,x is a direct sum of two-dimensional local
fields. Denote by πC the local parameter defined by C in X. Then
Y0 Y0  Y 
X = AX,012 =
Afin k(X)C,x := k(X)C,x
x∈C C x:x∈C
 


 ∞
aiC ∈ AC,01 

 (A.26)
 X Y Y 
πiC

= .


:=  a ∈ k(X) : a 0 (i  0)
 
 iC C C C,x iC C 

 iC =−∞
 
C x:x∈C
min{iC : aiC , 0} = 0 (∀0C)

 


A.2.2 Adelic Spaces at Infinity

Now we are ready treat adelic spaces at infinite places for general arithmetic vari-
eties. Motivated by the discussion above, we make the following

Definition A.2. Let π : X → Spec OK be an arithmetic variety. Let S (XF ) be the

simplicial set associated to its generic fiber XF and K ⊆ S (XF )m , m ≥ 0, a subset.

(1) Let G be a coherent sheaf on XF . We define the associated adelic spaces by

Y
lim G p /mlp G p ⊗ R , m=0





 ←− l

 p∈K
A∞ (K, G) := 

(A.27)

 Y
lim A K, G p /m p G p , m > 0.
l




 p∈P(V) ←− l p




(2) Let {Gi }i be an inductive system of coherent sheaves on XF and F = lim Gi .

−→i
Then we define

A∞ (K, F ) := lim A∞ (K, Gi ). (A.28)

−→i

Naturally, the key to this definition is the one for m = 0.

Moreover, if F = lim G0i is another inductive limit of coherent sheaves,
−→i
lim A∞ (K, G0i ) ' lim A∞ (K, Gi ), by the universal property of inductive limits since
−→i −→i
G p /mlp G p is finite Q-dimensional. Therefore, A∞ (K, F ) is well-defined. More-
over, as a functor from the category of coherent sheaves on XF to that of Q-vector
spaces, A∞ (K, ∗) is additive and exact. Hence, by §1.2 of [43], A∞ (K, ∗) com-
mutes with the direct limits, even in general, for an inductive system {Fi }i of
quasi-coherent sheaves lim A∞ (K, Fi ) , A∞ (K, lim Fi ).
−→i −→i
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 455

Arithmetic Adelic Complexes 455

A.2.3 Arithmetic Adelic Complexes

As mentioned at the beginning, for arithmetic varieties, the finite and infinite parts
are closely interconnected. Therefore, when developing an arithmetic cohomol-
ogy theory, we should treat them as an unify one. For arithmetic curves, this is
done in §2.1.2 by an uniformity condition. To treat general arithmetic varieties,
we go as follows.
Let F be a quasi-coherent sheaf on X, denote its induced sheaf on the generic
fiber XF by FF . It is well-known that F is quasi-coherent as well. Moreover,
for a point p of XF , we denote its associated Zariski closure in X by Epi . Then,
motivated by §2.1.2, we introduce the following

Definition A.3. Let X be an arithmetic variety of dimension n + 1 and F a quasi-

coherent sheaf on X. Fix an index tuple (i0 , . . . , im ).

(1) The finite, resp. infinite, adelic space of type (i0 , . . . , im ) associated to F is
defined by
X; i0 ,...,im (F ) := AX KX; i0 ,...,im , F ,
Afin


(A.29)
A∞ , .


resp. X; i0 ,...,im (F ) := A X K X F ; i 0 ,...,im
F F

Here, for Z ⊆ X or XF , we set

KZ;i0 ,...,im := (p0 , . . . , pm ) ∈ S (Z)m codimZ {pr } = ir ∀ 0 ≤ t ≤ m ;

(A.30)
(ii) The arithmetic adelic space of type (i0 , . . . , im ) associated to F is defined by
AX; 0,1,...,n+1 (F ) =: Afin
 ar L ∞


 X; 0,1,...,n+1 (F ) AX; 0,1...,n (FF );



 L ∞
Aar Afin

X; 0,1,...,n (F ) ⊂ X; 0,1,...,n (F ) AX; 0,1...,n (FF )





=

consisting of f fp0 ,p1 ,...,pn ; (A.31)


 Ep0 ,Ep1 ,...,Epn



(F ) =: Afin
 L ∞
Aar (F ) AX; 0,...,î,...,n (FF )



X; 0,...,î,...,n+1 X; 0,...,î,...,n+1



∀i = 0, 1, . . . , n.




Remark A.2. (1) For any p ∈ P(XF ), OX,Ep = OXF ,p and k(X)Ep = k(XF )p . Conse-
quently, for any (p0 , . . . , pn ) ∈ S (XF )n , we have a natural morphism
A (Ep0 , . . . , Epn ), F = A (p0 , . . . , pn ), FF .



(A.32)
It is in this sense we use the relation fEp0 ,Ep1 ,...,Epn = fp0 ,p1 ,...,pn above. (In particu-
lar, if pi ’s are vertical, there are no conditions on the corresponding components.)
Clearly, this uniformity condition is an essential one, since it characters the nat-
ural interconnection between finite and infinite components of arithmetic adelic
elements.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 456

456 Five Essays on Arithmetic Cohomology

(2) When n = 0, the index set {0, . . . , î, . . . , n} is empty, so the space A∞ X; ∅ (FF )
for arithmetic curves has yet been defined. More generally, as to be seen below,
we need the spaces A∞ X; ∅ (FF ) for all arithmetic varieties. Here, to complete our
definition, for an arithmetic variety X, we view A∞ X; ∅ (FF ) as the (−1)-level of the
adelic complex for its generic fiber XF . That is to say, we define it as follows. By
p. 63 of citeY, we have the (-1)-simplex 1U for open U ⊆ X. Set then S (XF )−1 =
{1U | U ⊆ X : open}, and, for K ⊆ S (XF )−1 , let
A∞
X; ∅ (K, FF ) := { sF ∈ FF (U K,F ) ⊗ R | s ∈ F (U K ) } (A.33)
where U K := ∪1U ∈K U and sF denotes the section induced by s.

With all this, we are ready to introduce the reduced arithmetic adelic spaces.

Definition A.4. Let X be an arithmetic variety of dimension n + 1 and F a quasi-

coherent sheaf on X.

(1) For I := iα1 <···<iαk iα1 , . . . , iαk , define the type I reduced arithmetic adelic
T 
space Aar
X; I (F ) of F by
\
Aar
X; I (F ) := Aar
X; iα ,...,iα (F ); (A.34)
1 k
iα1 <···<iαk

(2) For m ≥ 0, define the m-th reduced arithmetic adelic space Aar
X; m (F ) of F by
M
ar, red ar
AX; m (F ) := AX; I (F ). (A.35)
I: |I|=m+1

Lemma A.1. For any subset I ⊂ {0, 1, . . . , n}, the type I reduced arithmetic adelic
space Aar
X; I (F ) for F above is well defined.

Proof. Recall that all subsets I ⊂ {0, 1, . . . , n + 1} can be expressed I as I =

iα1 <···<iαk {iα1 , . . . , iαk } for suitable choices of subsets {iα1 , . . . , iαk }. Hence, at least,
T
for each index set I, we do have an arithmetic adelic space Aar X; I (F ).
To prove AX;I (F ) is well defined, we write I = {i1 , . . . , i s } with i1 < · · · < i s ,
ar

and let { j1 , . . . , jt } := {0, 1, . . . , n + 1} − {i1 , . . . , i s } with j1 < · · · < jt .

(a) i s = n + 1. Then, by a direct calculation,
t
\
Aar
X; I (F ) = Aar j ,...,n+1
X; 0,..., b
(F ).
r
r=1
(b) i s , n + 1. Then, by a direct calculation,
t
\\ 
Aar
X; I (F ) = Aar
X; 0,...,n (F ) Aar j ,...,n+1
X; 0,..., b
(F ) ,
r
r=1
as desired. 
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 457

Arithmetic Adelic Complexes 457

To define arithmetic adelic complex for quasi-coherent sheaf F , we also need

non-reduced indexed arithmetic adelic spaces for the reduced spaces. Following
the standard procedure of the usual simplicial constructions, (see e.g. [43],) we
make the following

Definition A.5. Let X be an arithmetic variety of dimension n + 1 and let F be

a quasi-coherent sheaf on X. The k-th arithmetic adelic space associated to F is
defined by
k−1
Y
(kl)
Akar (X, F ) := Aar, X; l (F ) ,
red
X; k (F ) × Aar, red
(A.36)
l=0

the associated boundary morphisms by

L ar L ar
dim : AX; l0 ,...,lm−1 (F ) −→ AX; k0 ,...,km (F )
(A.37)
(al0 ,...,lm−1 ) 7→ (ak0 ,...,k̂i ,...,km ),

and dm = m
L L ar
i m
Aar
P
i=0 (−1) di : X; k0 ,...,lm−1 (F ) −→ AX; k0 ,...,km (F ).

Then, from the standard homotopy theory, we have

 
Proposition A.3. A∗ar (X, F ), d∗ defines a complex of abelian groups.

All in all, we are now ready to introduce the following main definition.

Definition A.6. Let π : X → Spec OK be an arithmetic variety with XF its generic

fiber. Let F be a coherent sheaf on X. The i-th adelic arithmetic cohomology
groups of F is defined by

Hari (X, F ) := H i (A∗ar (X, F ), d∗ ), (A.38)

the i-th cohomology group of the complex (A∗ar (X, F ), d∗ ).

By the construction, we have the following

Corollary A.1. If X is an arithmetic variety of dimension n + 1, then

Hari (X, F ) = 0 (A.39)

unless i = 0, 1, . . . , n + 1.

Proof. Indeed, outside the range 0 ≤ i ≤ n + 1, the complex consists of zero by

definition. 
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 458

458 Five Essays on Arithmetic Cohomology

A.2.4 Cohomology Theory for Arithmetic Curves

We here give an example of the above theory for arithmetic curves, which was
previously developed in §2.1.2 based on Tate’s thesis ([113]).
Let D = ri=1 ni pi be a divisor on X = Spec OK . Write ni = ordpi (D). For
P
simplicity, we use A∗• (D) instead of A∗• (OX (D)). Then, same as in §A.1.3,
 Y 
X;01 (D) = (ap ) ∈
Afin ap ∈ Op ∀p ∈ S fin .
p∈S fin
(A.40)
And, since Dη = 0 is trivial,
Y
A∞ X;0 (D) = lim OXF ,η mη OXF ,η ⊗Q R = F/{0} ⊗Q R = F ⊗Q R = Fσ ,
 l

←l
σ∈S ∞
Y
Aar OX (D) = Afin 01 (D) ⊕ A0 (D) = (ap ) ∈
∞
Fp ap ∈ Op ∀p ∈ S fin .


X;01 p∈S
In particular, Aar


X;01 OX (D) ' AF is independent of D.
To understand Aar fin
0 (D), we first calculate A0 (D). Same as in §A.1.3 again,
AX;0 (D) = F. Note that, from above, AX;0 (D) = F ⊗Q R. Thus, by definition,
fin ∞

X;0 (D) = (av ; aσ ) ∈ AX;0 (D) ⊕ AX;0 (D) (av ) = ifin ( f ), (aσ ) = i∞ ( f ) ∃ f ∈ F .
∞
Aar fin


Hence Aar X;0 (D) ' K is also independent of D.

X;1 (D) = AX;1 (OX (D)) ⊕ AX;∅ (OX (D)). To understand

∞
From our definition, Aar fin
fin
it, we first calculate AX;1 (D). With the same calculation as in §A.1.3, we have
X;1 (D) = (ap ) ∈ AX;01 ordp (ap ) + ordp (D) ≥ 0 . Next, we calculate AX;∅ (D).
∞
Afin fin


By definition, we have A∞ X;∅ (D) = {s ∈ F ordp (s) + ordp (D) ≥ 0 ⊗Q R, since

D = p ordp (D)p, and hence if U = X − {p1 , . . . , pr }, OX (U) is trivial.
P

In this way, we conclude that the following two arithmetic adelic complexes
coincide
d1
0 −→ Aar ar ar
X;0 (OX (D)) ⊕ AX;1 (OX (D)) −→ AX;01 (OX (D)) −→ 0
d1 (A.41)
0 −→ F ⊕ Aar
X;1 (OX (D)) −→ AF −→ 0
d1 : (a0 , a1 ) 7→ a1 − a0 .
Therefore,
Har0 F, OX (D) = F ∩ Aar


X;1 (OX (D)),
(A.42)
Har1 F, OX (D) = AF F + Aar .



X;1 (O X (D))
This coincides with the definition of Har0 F, g in §2.1.2, where g = (gp ; gσ ) ∈


GL1 (AF ) such that gσ = 1 ∀σ ∈ S ∞ and ordp (gp ) = ordp (D) ∀p ∈ S fin , i.e. the
P
associated divisor D(g) := p∈S fin ordp (gp ) p coincides with D.
As it stands, all above is compatible with cohomology theory for arithmetic
curves developed in §2.1.2, even a much more refined theory is developed there.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 459

Appendix B

Cohomology for Arithmetic Surfaces

In the sequel, by an arithmetic surface, we mean a 2-dimensional regular integral

Noetherian scheme X together with a proper (flat) morphism π : X → Spec OK .
Here OK denotes the ring of integers of a number field K. In particular, the generic
fiber XF is a geometrically connected, regular, integral projective curve defined
over K.

B.1 Local Residue Pairings

Theory of residues for arithmetic surfaces, as a special case of Grothendieck’s

residue theory, can be realized using Kähler differentials as done in Chapter III, §4
of [67]. However, here we follow ([78, 79]) to give a rather precise realization in
terms of structures of two-dimensional local fields.

B.1.1 Residue Maps for Local Fields

B.1.1.1 Continuous differentials

Let (A, − A ) be a local Noetherian ring and N an A-module N. Denote by N sep

the maximal Hausdorff quotient of N for the − A -adic topology, i.e. N sep =
N ∞ n
T
n=1 − A N. In particular, if A is an R-algebra for a certain ring R, then we
sep
have the differential module ΩA/R and hence ΩA/R . Thus, if K is a complete dis-
crete valuation field and K a subfield such that Frac(K ∩ OK ) = K, then we have
the space of the continuous differentials
sep
Ωcts
F/K := ΩOK /OK ∩K ⊗OK F. (B.1)

Consequently, if F 0 /F is a finite, separable field extension, then Ωcts

F 0 /K = ΩF/K ⊗F
cts

F and hence there is a natural trace map TrF 0 /F : ΩF 0 /K −→ ΩF/K .

0 cts cts

459
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 460

460 Five Essays on Arithmetic Cohomology

B.1.1.2 Equal characteristic zero

Let K be a two-dimensional local field of equal characteristic zero. Then K con-

tains a unique subfield kF of coefficients such that F ' kF ((t)) for a suitable uni-
sep
formizer t. In particular, ΩOK /Ok ' OK · dt is a free OK - module of rank one. We
F
define the residue map for K by

resF : Ωcts
F/kF −→ kF , ω = f dt 7→ coeftt−1 ( f ). (B.2)

By [78], this is well defined, i.e. independent of the choice of t. Moreover, for a
finite field extension F 0 /F, we have the following commutative diagram
resF 0
Ωcts
F 0 /kF −
−−−−→ kF 0
TrF 0 /F ↓ ↓ TrkF0 /kF (B.3)
resF
Ωcts −−−→ kF .
F/kF −

B.1.1.3 Mixed characteristic

Let L be a two-dimensional local field of mixed characteristics. Then the constant

field kL of L coincides with the algebraic closure of Q p within L for a certain
prime number p, and L itself is a finite field extension over kL {{t}} for a certain
uniformizer t. Here, by definition,
∞
nX o
ai ti : ai ∈ kL , inf νkL (ai ) > −∞ (∀i), ai → 0 (i → −∞) . (B.4)

kL {{t}} :=
i
i=−∞
sep sep
Moreover, by [78], ΩOk = OkL {{t}} dt Tors ΩOk . We define the
L

L {{t}}/Ok L L {{t}}/OkL
residue map, first, for kL {{t}}, by

reskL {{t}} : Ωcts

Ok {{t}}/O −→ kL , ω = f dt 7→ −coeftt−1 ( f ); (B.5)
L kL

then, for L, by the composition

TrL/kL {{t}} reskL {{t}}
resL : Ωcts −−−−−−→ Ωcts
L/kL − −−−−−→ kL .
kL {{t}}/kL − (B.6)

This is well defined by [78]. Consequently, if L0 /L is a finite field extension, we

have the commutative diagram
resL0
Ωcts
L0 /kL −−−−→ kL
0

TrL0 /L ↓ ↓ TrkL0 /kL (B.7)

resL
Ωcts −−−→ kL .
L/kL −
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 461

Cohomology for Arithmetic Surfaces 461

B.1.2 Local Residue Maps

As above, let K be a number field and π : X → Spec OK be an arithmetic surface

with XK its generic fiber.
For each closed point x ∈ X, and a prime divisor C on X with x ∈ C, by
Theorem A.1, the local ring k(X)C,x is a finite direct sum of two-dimensional local
fields, i.e.
M
k(X)C,x = k(X)Ci ,x , (B.8)

where Ci ’s are normalized branches of the curve C in a formal neighborhood U of

x. Set then
X
resC,x = resk(X)Ci ,x , (B.9)
i

which takes the values in Fπ(x) , the local field of K at the place π(x). Recall also
that, following [113], we have the canonical character
TrFπ(x) /Q p
λπ(x) : Fπ(x) −−−−−−−→ Q p −→ Q p /Z p −→ Q/Z ,→ R/Z ' S1 . (B.10)

Introduce accordingly

ResC,x := λπ(x) ◦ resC,x . (B.11)

On the other hand, for each closed point P ∈ XF ,

O
d Y  M M M
k(XF )P Fσ = R((t)) C((t)) (B.12)
F
σ∈S ∞ σ: real σ: complex

is a finite direct sum of local fields R((t)) and C((t)). Hence, similarly, for each
σ ∈ S ∞ , we have the associated residue maps resP,σ . Define

ResP,σ = λσ ◦ resP,σ . (B.13)

Here, as in [113], to make all compatible, we set λσ (x) = −TrFσ /R , i.e. with a
minus sign added.

B.2 Global Residue Pairing

The purpose here is to introduce a non-degenerate global residue pairing on the

arithmetic adelic ring of an arithmetic surface.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 462

462 Five Essays on Arithmetic Cohomology

B.2.1 Global Residue Pairing

Let π : X → Spec OK be an arithmetic surface with XF its generic fiber. Then, by

§A.2.1, we have the associated arithmetic adelic ring
M
Aar ar
X := AX,012 := AX
fin
A∞X (B.14)

X = AX,012 the adelic ring

with Afin for the 2-dimensional Noetherian scheme X and

O Y 
∞
AX := AXF ⊗Q R := lim lim AXF (D1 )/AXF (D2 )
b Fσ . By an abuse of
−→D1 ←−D2
D2 ≤D1 F σ∈S ∞
notation, we will write elements of Afin
X as ( fC,x )C,x , or even ( fC,x ), and elements of
A∞X as ( f P ) P , or even ( f P ).
Fix a rational differential ω = f (t) dt . 0 on X. Then, we define a global
pairing with respect to ω by:
h·, ·iω : Aar ar
X × AX
−→ S1
( fC,x , fP,σ ), (gC,x , gP,σ ) 7→ C⊆X,x∈C: π(x)∈S fin ResC,x ( fC,x gC,x ω)
P
(B.15)
+ P∈XF σ∈S ∞ ResP,σ ( fP,σ gP,σ ω).
P P

Lemma B.1. Let X be an arithmetic surface and ω is a non-zero rational differ-

ential on X. Then the global pairing with respect to ω above is well defined.

Proof. Write the summation C⊆X,x∈C: π(x)∈S fin ResC,x ( fC,x gC,x ω) as double sum-
P
mations C⊆X x∈C: π(x)∈S fin ResC,x ( fC,x gC,x ω). Then, by Example A.2, for all but
P P
finitely many curves C, ResC,x ( fC,x gC,x ω) = 0. So it suffices to show that for a
fixed curve C, x∈C: π(x)∈S fin ResC,x ( fC,x gC,x ω) is finite. This is well known.
P


B.2.2 Non-Degeneracy

Proposition B.1. The residue pairing h·, ·iω on Aar

X is non-degenerate.

Proof. Let g ∈ Aar X be an adelic element such that, for all f ∈ AX , h f, giω = 0. We
ar

show that g = 0.
Rewrite the summation in the definition of h·, ·iω according to prime horizontal
curves E Par associated to closed points P ∈ XF and prime vertical curves V ⊆ X
appeared in the fibers of π. Namely, P∈XF x∈EPar + V∈X x∈V . Then, note that,
P P P P
for a fixed adelic element g = (gC,x , gP,σ ) ∈ Aar
X,

h( fC,x , fP,σ ), (gC,x , gP,σ )i = 0 ∀( fC,x , fP,σ ) ∈ Aar

X. (B.16)
This implies that
X X XX
ResEP ,x (gEP ,x f ω)+ ResEP ,x (gEP ,x f ω) = 0 ∀ f ∈ k(X). (B.17)
P∈XF x∈E Par V∈X x∈V
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 463

Cohomology for Arithmetic Surfaces 463

Now assume, otherwise, that g , 0. There exists either some vertical curve
C such that 0 , gC,x ∈ k(X)C,x , or, a certain algebraic point P of the generic
ˆ
fiber XF such that 0 , LgP,σ ∈ k(XF )P ⊗F Fσ . In case gC,x , 0, by defini-
tion, gC,x = (gCi ,x ) ∈ i k(X) Ci ,x = k(X) C,x , where C i runs over all branches
of C, and k(X)Ci ,x is a two-dimensional local field. Fix a branch Ci0 such that
gCi0 ,x , 0. By definition, ResCi ,x := λπ(x) ◦ resCi ,x . So we can choose an el-
ement h ∈ k(X)Ci0 ,x such that ResCi0 ,x (hω) , 0. For such h, we then take
fCi0 ,x ∈ k(X)Ci0 ,x such that fCi0 ,x gCi0 ,x = h. Accordingly, if we construct an adelic
element K by taking all other components fCi ,x to be zero but the fCi0 ,x , then we
have h f, giω = ResCi0 ,x ( fCi0 ,x gCi0 ,x ω) = ResCi0 ,x (hω) , 0. This contradicts to
our original assumption that h f, giω = 0. Hence, all the components of g corre-
sponding to vertical curves are zero. Since we can use the same argument for the
components of g corresponding to algebraic points of XF to conclude that all the
related components are zero as well, so g = 0. This completes the proof. 

B.3 Adelic Subspaces

B.3.1 Level Two Subspaces

Let K be a number field with OK the ring of integers, and let π : X → Spec OK
be an arithmetic surface with XF its generic fiber. Then X is a regular integral
2-dimensional Noetherian scheme, π is proper, and XF is a geometrically con-
nected, regular, integral projective curve defined over K. Our purpose here is
to introduce certain level two intrinsic subspaces of the arithmetic adelic space
∞
Aar ar
X := AX,012 := AX ⊕ AX .
fin

To start with, we analyze the structures of Afin X,01 , one of the level two subspaces
of Afin
X = A fin
X,012 . By definition, see e.g. [91], an element ( fC,x )C,x ∈ Afin
X belongs
to Afin
X,01 , if, for all curves C, the partial components ( f )
C,x x∈C are independent of
fin
x. So we may simply write elements of AX,01 as ( fC )C . On the other hand, with
respect to π, curves on X may be classified as being either vertical or horizontal.
Therefore, we may and will write ( fC )C = ( fC )C: ver × ( fC )C: hor . Accordingly, set

X,01 = AX,01 ⊕ AX,01 ,

fin,v fin,h
Afin (B.18)

X,01 , resp. AX,01 denotes the collections of ( fC )C: ver , resp. ( fC )C: hor . Fur-
where Afin,v fin,h

thermore, if C is horizontal, there exists an algebraic point P of XF such that

X
C = {P} , the Zariski closure of P in X. For simplicity, write C = E P . Then
fEP ∈ F(X)XP . But F(X)EP = F(XF )P . So it makes sense for us to talk about
whether fEP = fP for a certain element fP ∈ F(XF )P .
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 464

464 Five Essays on Arithmetic Cohomology

To go further, fix a Weil divisor D = Dv + Dh on X, where Dv = F nV V

P
with V irreducible vertical curves and Dh = P nP E P with E P the horizontal
P
curves (corresponding to algebraic points P of XF ). In particular, D induces a
divisor DF = P nP P on XF . Motivated by the above discussion on Afin
P
X,01 and the
arithmetic cohomology theory for arithmetic curves (see §B.25), we introduce the
following level two intrinsic arithmetic adelic subspaces Aar X,01 , AX,02 , AX,12 (D) by
ar ar

X,01 = ( fC,x ) × ( fP ) ∈ AX ( fC,x )C,x = ( fC )C,x ∈ AX,01 , fE P = fP ∀P ∈ XF ,

Aar
 ar fin

M M
Aar =A fin ar fin
⊗K R ,


X,02 X,02 k(X F ), and A X,12 (D) := A X,12 (D) AXF (DF )b
where

X,12 (D) := ( fC,x ) ∈ AX ordC ( fC,x ) + ordC (D) ≥ 0 ∀C ⊆ X ,

Afin
 fin

AXF (DF ) := ( fP ) ∈ AXF ordP ( fP ) + ordP (DF ) ≥ 0 ∀P ∈ XF ,


 O 
⊗F R := lim0 AXF (DF )/AXF (D0F )
AXF (DF ) b R.
−→DF
F
D0F ≤DF

Here we have used the natural imbedding k(X) = k(XF ) ,→ AXF ,→ Aar
X.
Accordingly, we then also obtain three level one subspaces

X,0 := AX,01 ∩ AX,02 ,

Aar ar ar
and
(B.19)
Aar ar ar
X,1 (D) := AX,01 ∩ AX,12 (D), Aar ar ar
X,2 (D) := AX,02 ∩ AX,12 (D).

Lemma B.2. Let X be an arithmetic surface, D be a Weil divisor on X with DF

its induced divisor on XF . We have

X,0 = k(X);
(1) Aar
X,1 (D) = {( fC )C,x × ( fP ) ∈ AX : fP ∈ AX,1 (D), fE P = fP ∀P ∈ XF };
(2) Aar ar

(3) AX,2 (D) = {( f x )C,x × ( f ) ∈ AX :

ar ar

( f x )C,x ∈ AX,2 (D), f ∈ AXF (DF ) ∩ k(XF ) = H 0 (XF , DF )}

(4) Under the natural boundary map, (A∗ar (X, D), d∗ ) :

0 → Aar ar ar ar ar ar ar
X,0 ⊕ AX,1 (D) ⊕ AX,2 (D) → AX,01 ⊕ AX,02 ⊕ AX,12 (D) → AX,012 → 0

forms a complex, the adelic complex for D.

Proof. The first three are direct consequences of the construction, while (4) is
standard from homotopy theory. Indeed, we only need to write down the boundary
maps: the first is the diagonal embedding, the second is given by (x0 , x1 , x2 ) 7→
(x0 − x1 , x1 − x2 , x2 − x0 ) and the final one is given by (x01 , x02 , x12 ) 7→ x01 + x02 −
x12 . 
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 465

Cohomology for Arithmetic Surfaces 465

As a direct consequence of (2) and (3), we have the following

Corollary B.1. Let X be an arithmetic surface, D be a Weil divisor on X. Then,

we have the following induced ind-pro structures on Aar ar
X,01 and AX,02 :

X,01 = lim0 lim0 AX,1 (D) AX,1 (D ),

0
Aar ar  ar
←−D −→D
D0 ≤D
(B.20)
X,02 = lim0 lim0 AX,2 (D) AX,2 (D ).
0
Aar ar  ar
←−D −→D
D0 ≤D

One checks easily that our definitions here are specializations of the definitions
in §A.2.3 for the line bundle F = OX (D) over arithmetic surface X.

B.3.2 Perpendicular Subspaces

For late use, we here establish a fundamental property for the level two arithmetic
adelic subspaces introduced above. In fact, as we will see below, this property, in
turn, characterizes these subspaces.
Fix a non-zero rational differential ω on the arithmetic surface X. Then, by
Proposition B.1, we have a natural non-degenerate pairing h·, ·iω on Aar X . For a
ar
subspace V of AX , set

X hw, viω = 0 ∀v ∈ V
V ⊥ := w ∈ Aar

(B.21)

be its perpendicular subspace of V in Aar

X with respect to h·, ·iω . Then, we have the
following technical

Proposition B.2. Let X be an arithmetic surface, let D be a Weil divisor and let ω
be a non-zero rational differential on X. Denote by (ω) the divisor on X associated
to ω. Then
 ⊥
(1) Aar
X,01 = Aar
X,01 .
 ⊥
(2) A ar
=A .
ar
 X,02 ⊥ X,02
(3) AX,12 (D) = Aar
ar
X,12 ((ω) − D).

Proof. By an abuse of notation, we will write elements of Afin X as ( fC,x )C,x or even
( fC,x ), and elements of A∞ X as ( f P ) P or even ( f P ).
⊥
We begin with a proof of Aar ar ar
X,01 ⊆ (AX,01 ) . Let f, g ∈ AX,01 . By definition,
f = ( fC )C,x × ( fP )P , g = (gC )C,x × (gP )P , and, for all algebraic points P of XF ,
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 466

466 Five Essays on Arithmetic Cohomology

fEP = fP and gEP = gP . Consequently,

X X
hf, giω = ResC,x ( fC gC ω) + ResP ( fP gP ω)
x,C P
XX X
= ResC,x ( fC gC ω) + ResP ( fP gP ω).
C x:x∈C P

If C is a vertical curve on X, then, by the standard residue theorem for algebraic

curves, see e.g. [91], x:x∈C ResC,x ( fC gC ω) = 0. Hence
P
X X X
hf, giω = ResP ( fEP gEP ω) + ResP ( fP gP ω)
P∈XF x∈E P P
X X
= ResQ ( fP gP ω).
P∈XF Q∈E P

Here E P denote the complete arithmetic curve associated to the algebraic point
P ∈ XF and in the last step, we have used our definition condition fEP = fP and
gEP = gP for elements f, g in Aar X,01 . Now, by the residue theorem for horizontal
curves E P (Theorem 5.4 of [79]), Q∈E P ResQ ( fP gP ω) = 0. Therefore, hf, giω = 0,
P
⊥
and Aar ar
X,01 ⊆ (AX,01 ) .
⊥
Next, we show that Aar ar
X,01 ⊃ (AX,01 ) , based on the following ind-pro structure
ar
on AX,01 in Corollary B.1

X,01 = lim0 lim0 AX,1 (D) AX,1 (D ).

0
Aar ar  ar
←−D −→D
D0 ≤D

Let C be an irreducible curve C. Then, induced by the perfect pairing h·, ·iω :
Aar ar
X × AX → R/Z, for any divisor D on X, we obtain a pairing

X,12 (D) AX,12 (D − C) × AX,12 ((ω) + C − D) AX,12 ((ω) − D) −→ R/Z. (B.22)

Aar
 ar ar  ar

Moreover, directly from the definition, we have that Aar

 ar ar
X,12 (D) AX,12 (D−C) ' AC,01
for any divisor D. Hence, AX,12 ((ω) + C − D) AX,12 ((ω) − D) ' AC,01 as well. So
ar ar ar

ar
we can and will view (B.22) as a pairing on AC,01 .
If C is vertical, then, there exists ωC ∈ Ωk(C)/F p and an a = (av ) ∈ AC,01 ar
such
that (B.22) coincides with the pairing
X
ar
h·, ·iωC ,a : AC,01 ar
× AC,01 −→ Fq ; (f, g) 7→ Resv ( fv gv av ωC ). (B.23)
v
⊥
Since Aar⊆
X,01 (Aar
X,01 ) ,
and, directly from the definition, we have that

X,1 (D) AX,1 (D − C) ' AX,1 ((ω) + C − D) AX,1 ((ω) − D) ' AC,0 = k(C), (B.24)
Aar
 ar ar  ar ar

ar ar
we conclude that h·, ·iωC ,a : AC,01 ×AC,01 → F p annihilates k(C)×k(C). But h·, ·iωC ,a
ar ar
can be identified with the canonical residue pairing h·, ·i : AC,01 × AC,01 → Fp
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 467

Cohomology for Arithmetic Surfaces 467

associated to C. Consequently, K(C)⊥ = K(C).3 In particular, with respect to the

pairing hak(C), k(C)i = 0. So a ∈ k(C). Hence, if necessary, with a possible
modification on ω, without loss of generality, we may and will assume a = 1 and
write h·, ·iωC ,a simply as h·, ·iωC . Therefore, by (B.24) and the fact that k(C)⊥ =
k(C), we have
 ⊥
Aar X,1 ((ω) + C − D) AX,1 ((ω) − D).
 ar
' Aar
 ar
X,1 (D) AX,1 (D − C) (B.25)
Moreover, with a verbatim change, the same discussion is valid for horizontal
curves as well. Consequently, by applying this repeatedly, we have, for any irre-
ducible curves C1 , C2 , the following commutating diagram with exact columns
,→ A(D) A(D − C1 − C2 ) 
  
A(D − C1 ) A(D − C1 − C2 ) A(D) A(D − C1 )
k ↓ k
G ∩ B(D − C1 ) G ∩ B(D − C1 − C2 ) ,→ G ∩ B(D) G ∩ B(D − C1 − C2 )  G ∩ B(D) G ∩ B(D − C1 )
  

ar
⊥
where, to save space, we set A := Aar X,1 , B := AX,12 and G := AX,01 . Con-
ar

sequently, the vertical map in the middle is surjective. On the other hand, since
ar
⊥ 0
Aar
X,01 ⊆ AX,01 , this same map is also injective. Therefore, for any D ≤ D,

0 ⊥
 
0  ar
Aar
 ar
X,1 (D) AX,1 (D ) ' Aar
X,1 ((ω) − D ) AX,1 ((ω) − D)

with respect to our pairing

0 0  ar
Aar
 ar ar
X,1 (D) AX,1 (D ) × AX,1 ((ω) − D ) AX,1 ((ω) − D) −→ R/Z.

Consequently, we have

⊥
Aar
X,01 = Aar
X,01 ,

since, by (3), whose proof is independent of (1) and (2),

X,01 = lim0 lim0 AX,1 (D) AX,1 (D ) = lim lim0 AX,1 ((ω) − D ) AX,1 ((ω) − D).
0 0  ar
Aar ar  ar ar
←−D −→D −→D ←−D
D0 ≤D D0 ≤D

This proves (1).

⊥
To prove (2), we start with the inclusion Aar ar
X,02 ⊂ (AX,02 ) . By definition, every
element f ∈ AX,02 can be written as f = ( f x )C,x ×( fP )P with fC,x = f x and ( fP ) = ( f )
ar

for some f ∈ k(XF ) (since, by definition, the 02 type adeles are independent of
one dimensional curves). Thus, for f, g ∈ Aar X,02 , we have
X X
hf, gi = ResC,x ( fC,x gC,x ω) + ResP ( fP gP ω)
C,x P
X X X
= ResC,x ( f x g x ω) + ResP ( f gω).
x C:C3x P
3 Itis well-known that, see e.g. [46], if χ is a non-zero character on AC,01 ar such that χ(k(C)) = {0},
then the induced pairing h·, ·iχ : AC,01 × AC,01 → Fq ; (f, g) 7→ χ(f · g) is perfect and k(C)⊥ = k(C).
ar ar
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 468

468 Five Essays on Arithmetic Cohomology

Note that for a fixed x, f x and g x are fixed. Thus, by the residue theorem for the
point x (Theorem 4.1 of [78]), we have C:C3x ResC,x ( f x g x ω) = 0. So
P
X X X
hf, gi = 0+ ResP ( f gω) = ResP ( f gω).
x P P

On the other hand, since f, g ∈ k(XF ) and ω is a rational differential on XF ,

the standard residue formula for the curve XF /F (see e.g. [98]) implies that
P ResP ( f gω) = 0. Hence
P

hf, gi = 0, X,02 .
∀ f, g ∈ Aar
⊥
Therefore, Aar ar
X,02 ⊂ (AX,02 ) .
⊥
For the opposite direction Aar ar
X,02 ⊃ (AX,02 ) , similarly as in (1), we, in theory,
ar
can use the following ind-pro structure on AX,02 :

X,02 = lim0 lim0 AX,2 (D) AX,2 (D ).

0
Aar ar  ar
←−D −→D
D0 ≤D

However, due to the lack of details for horizontal differential theory in literature,
we decide to first use this ind-pro structure to merely prove the part that if f =
X,02 ) , for vertical C’s, fC,x = f x ; and then to take a more
⊥
( fC,x )C,x × ( fP )P ∈ (Aar
classical approach for the rest.
Choose f = ( fC,x )C,x × ( fP )P ∈ (Aar ⊥
X,02 ) . Then by our assumption, for any
element g = (g x )C,x × (g)P ∈ Aar X,02 , we have
X X X
0 = hf, gi = ResC,x ( fC,x g x ω) + ResP ( fP gω). (B.26)
x C:C3x P

Now note that the element (g x )C,x ∈ Aar X,02 and the element g ∈ k(XF ) can
be changed totally independently, we conclude that both of the summations,
x C:C3x ResC,x ( fC,x g x ω) and
P P P
P ResP ( fP gω), are constants independent of g.
P
This then implies that both of them are 0. Indeed, since P ResP ( fP gω) is a con-
stant independent of g, by choosing g to be constant function, we conclude that
P
P ResP ( fP gω) ≡ 0, ∀g ∈ k(XK ). Consequently, by (B.26),
X X
ResC,x ( fC,x g x ω) ≡ 0, ∀(g x ) ∈ Aar
X,02 . (B.27)
x C:C3x

To end the proof, we need to show that fC,x ’s are independent of C and fP are
independent of P. First, we treat the case when C is vertical. As said above, we
will use the associated ind-pro structures. So, assume for now that C is vertical.
Then, for any divisor D on X, we have
Aar
 ar
X,2 (D) AX,2 (D − C) 'AC,1 (D|C ),

X,2 ((ω) + C − D) AX,2 ((ω) + C − (D − C)) 'AC,1 ((ωC ) − D|C ),

0
Aar
 ar
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 469

Cohomology for Arithmetic Surfaces 469

for a certain ωC0 ∈ Ωk(C)/Fq satisfying (ωC0 ) = (ω) + C |C . By the adjunction

formula, (ωC0 ) may be viewed as a canonical divisor of the curve C. Thus,, with
respect to the canonical residue pairing on C, we have AC,1 (D|C )⊥ = AC,1 ((ωC0 ) −
D|C ). Consequently, as in (1), from

X,02 = lim0 lim0 AX,2 (D) AX,2 (D ) = lim lim0 AX,2 ((ω) − D ) AX,2 ((ω) − D),
0 0  ar
Aar ar  ar ar
←−D −→D −→D ←−D
D0 ≤D D0 ≤D

we conclude that fC,x = f x ∈ k(X) x and hence independent of C.

To prove the rest, we take a classical approach with a use of Chinese remainder
theorem, using an idea in the proof of Proposition 1 of [91]. To be more precise,
fix x0 ∈ X and a prime divisor H on X satisfying that x0 ∈ H and that V :=
Spec OX,x0 − H is affine. We claim that for any family of prime divisors D j on V
such that Di ∩ D j = ∅ if i , j, and any rational functions f0 , f1 , . . . fn on V, and
any fixed divisor D supported on Di ’s, there exists a rational function g such that

ordDi ( fi − g) ≥ ordDi (D), i = 0, 1, . . . , n,


i < {0, 1, . . . , n}.

ordD (g) ≥ ordD (D),

i i

Indeed, by clearing the common denominators for fi ’s, (with a modification of

fi ’s if necessary,) we may assume that fi ’s are all regular. Then by applying the
ordD (D)
Chinese remainder theorem to the fractional ideals pi i , i = 0, 1, . . . , n, and
ordD (D)
∩i<{0,1,...,n} pi i , where pi are the prime ideas associated to the prime divisors
Di , we see the existence of such a g.

⊥ 0
Associated to f ∈ Aar ar
X,02 , form a new adele f ∈ AX,012 by setting

 fC,x − fH,x , x = x0 , C horizontal,

fC,x = 
0

 fC,x ,
 otherwise.
Then
X
0
ResC,x0 ( fC,x g ω)
0 x0
C:C3x0
X X
= ResC,x0 ( fC,x0 g x0 ω) − ResC,x0 ( fH,x0 g x0 ω).
C:C3x0 C:C3x0

The first sum is zero, since we have (B.27) by our choice of f. The second sum,
being taken over all prime curves passing through x0 , is zero as well, since we can
apply the residue theorem for the point x0 as above (with fH,x0 being independent
of C). That is to say,
X
0
ResC,x0 ( fC,x g ω) = 0.
0 x0
C:C3x0
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 470

470 Five Essays on Arithmetic Cohomology

Now, applying the above existence to obtain a g satisfying that for any fixed
rational function f0 and any fixed curve C0 3 x0 , we have

ordC ( fC,x0 ) + ordC ( f0 − g) + ordC (ω) ≥ 0, C = C0

 0

 ordC ( f 0 ) + ordC (g) + ordC (ω) ≥ 0,


 C , C0 , H.
C,x0
0
Consequently, by the definition of the residue map, with fH,x 0
≡ 0 in mind, we get,
for any f0 ∈ k(X) and the corresponding g just chosen,
X X
0= 0
ResC,x0 ( fC,x 0
gω) = 0
ResC,x0 ( fC,x 0
gω)
C:C3x0 C:C3x0 ,{C:C,C0 ,H}∪{C:C=C0 ,H}
X
= 0
ResC,x0 ( fC,x0
gω) = ResC0 ,x0 ( fC0 0 ,x0 gω) = ResC0 ,x0 ( fC0 0 ,x0 f0 ω).
C:C3x0 ,C=C0 ,H

Since the last quantity is always zero for all f0 , this then implies that fC,x 0
0
= 0,
namely, fC0 ,x0 = fH,x0 .
To end the proof of (2), we still need to show that fP = fP0 for a fixed P0 ∈ XK
and all P ∈ XK . But this is amount to a use of a similar argument just said again,
based on Chinese remainder theorem. We leave the details to the reader. Thus, if
f = ( fC,x )C,x × ( fP )P ∈ (Aar
X,02 ) , then fC,x = fC0 ,x and fP = fP0 for fixed C 0 and P0 .
⊥

Therefore, f ∈ AX,02 . That is to say, (Aar

ar
X,02 ) = AX,02 . This proves (2).
⊥ ar

⊥
Finally, we prove (3). The inclusion Aar ar
X,12 ((ω) − D) ⊆ (AX,12 (D)) is easy.
Indeed, for elements f = ( fC,x )×( fP ) ∈ AX,12 ((ω)−D), g = (gC,x ) × (gP ) ∈ Aar
ar
X,12 (D),
we have hf, gi = C,x ResC,x ( fC,x gC,x ω) + P ResP ( fP gP ω) = 0, since every term
P P
in each of these two summations is zero by definition.
⊥
To prove the other direction Aar ar
X,12 ((ω) − D) ⊃ (AX,12 (D)) , we make the fol-
lowing preparations. Set e π : X → Spec OF → Spec Z. Then, by Theorem 5.7
of [78], the dualizing sheaf ωeπ of eπ is given by, for an open subset U ⊆ X,
ωeπ (U) = ω ∈ Ωk(X)/Q | ResC,x ( f ω) = 0 ∀x ∈ C(⊆ U), ∀ f ∈ OX,C .


By a similar argument as in [78] (used to prove the above result), we have, for a
fixed curve C0 ,
ωeπ,C0 = ω ∈ Ωk(X)/Q | ResC0 ,x ( f ω) = 0 ∀ x ∈ C0 , f ∈ OX,C0 .


This is nothing but the collection of differentials ω satisfying ordC0 ((ω)) ≥ 0.

Moreover, we have, for a fixed pair x0 ∈ C0 ,

k(X)C ,x /Qeπ(x ) | ResC0 ,x0 ( f ω) = 0 ∀ f ∈ OC0 ,x0 .

ωeπ,C0 ⊗OC0 OC0 ,x0 = ω ∈ Ωcts

0 0 0

This is simply the collection of differentials ω satisfying ordC0 ((ω)) ≥ 0. From

these, we conclude that the following conditions are equivalent for a non-zero
differential ω ∈ Ωk(X)/Q and a fixed pair x0 ∈ C0 :
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 471

Cohomology for Arithmetic Surfaces 471

(a) ∀ f ∈ OX,C0 ResC0 ,x0 ( f ω) = 0.

(b) ordC0 ((ω)) ≥ 0.
Furthermore, a similar argument for points instead of curves also gives us that the
following conditions are equivalent for a fixed point P:

(i) ∀ f ∈ OXF ,P0 ResP0 ( f ω) = 0.

(ii) ordP0 ((ω)) ≥ 0.

Now we are ready to continue our proof. Let f = ( fC,x ) × ( fP ) ∈ Aar ⊥

X,12 (D) and
g = (gC,x ) × (gP ) ∈ AX,12 (D). Then
ar

X X
0 = hf, gi = ResC,x ( fC,x gC,x ω) + ResP ( fP gP ω).
C,x P

Thus, by the independence of components of adeles, we see that, for fixed (C, x)
and P, ResC,x ( fC,x gC,x ω) = 0 and ResP ( fP gP ω) = 0. This, together with the equiv-
alence between (a) and (b) and the equivalence between (i) and (ii), we conclude
that f = ( fC,x ) × ( fP ) ∈ Aar
X,12 ((ω) − D), as desired. 

B.4 Arithmetic Cohomology Groups for Arithmetic Surfaces

B.4.1 Definitions

Let X be an arithmetic surface and D be a Weil divisor on X. By Lemma B.2,

particularly, (4), we obtain an arithmetic adelic complex. Accordingly, we define
the arithmetic cohomology groups Hari (X, D) by H i (A∗ar (X, D), d∗ ), the i-th coho-
mology groups of the complex (A∗ar (X, D), d∗ ), i = 0, 1, 2.

Proposition B.3.

(1) Arithmetic cohomology groups Hari (X, D) of D on X, i = 0, 1, 2, are given by

Har0 (X, D) = Aar ar ar

X,01 ∩ AX,02 ∩ AX,12 (D),
 . 
Har1 (X, D) = Aar X,01+AX,02 ∩AX,12 (D)
ar
ar
X,01 ∩AX,12 (D)+AX,02 ∩AX,12 (D) ,
Aar ar ar ar

. 
Har2 (X, D) = Aar
X,012 AX,01 + AX,02 + AX,12 (D) .
ar ar ar

(2) There exist natural isomorphisms


. ar 
X,01 ∩ AX,02 +AX,12 (D)
Har1 (X, D) ' Aar X,02 +AX,01 ∩AX,12 (D)
ar ar
AX,01 ∩Aar ar ar


. ar 
X,02 ∩ AX,01 +AX,12 (D)
' Aar X,02 +AX,02 ∩AX,12 (D) .
ar ar
AX,01 ∩Aar ar ar
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 472

472 Five Essays on Arithmetic Cohomology

Proof. This is an arithmetic analogue of [91]. Indeed, (1) comes directly from the
definition, and via the first and second isomorphism theorems in elementary group
theory, (2) is obtained using a direct group theoretic calculation for our arithmetic
adelic complex. 

Cohomology groups for the divisor D defined here coincide with the adelic
global cohomology defined in the previous section associated to the invertible
sheaf OX (D). That is, Har0 (X, D) = Har0 (X, OX (D)). In fact, our general construction
in the previous section is a higher dimensional generalization of our constructions
for arithmetic surfaces here in this section, which itself is motivated by that in
§B.27 for arithmetic curves.

B.4.2 Inductive Long Exact Sequences

B.4.2.1 Vertical Curves

Just like the classical cohomology theory, our arithmetic cohomology groups also
admit an inductive structure related to vertical geometric curves on arithmetic
surfaces. More precisely, we have the following

Proposition B.4. Let C be an irreducible vertical curve of X, then, for any D,

there exists a long exact sequence of cohomology groups
0 →Har0 (X, C) → Har0 (X, D + C) → Har0 (C, (D + C)|C )
→ Har1 (X, D) → Har1 (X, D + C) → Har1 (C, (D + C)|C )
→ Har2 (X, D) → Har2 (X, D + C) → 0.
Here Hari (C, (D + C)|C ), i = 0, 1 are the usual cohomology groups for vertical
geometric curves.

Proof. For C be an irreducible vertical curve in X. Then from definition, A direct

calculation gives that
(a) AarX,12 (D + C)/AX,12 (D) = AX,12 (D + C)/AX,12 (D) ⊕ {0}AC,01 ⊕ {0}.
ar fin fin

(b) AX,1 (D + C)/AX,1 (D) = k(C) as there are neither changes along horizontal
ar ar

curves nor along XF . And (c)

 
X,2 (D + C)/AX,2 (D) = AX,2 (Dfin + C)/AX,2 (Dfin ) ⊕ {0} = AC (Dfin + C)|C ⊕ {0}.
Aar ar fin fin

Consequently, for the morphism

φ
X,1 (D + C)/AX,1 (D) ⊕ AX,2 (D + C)/AX,2 (D) −→ AX,12 (D + C)/AX,12 (D),
Aar ar ar ar ar ar

(x, y) 7→ x − y,
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 473

Cohomology for Arithmetic Surfaces 473

we conclude that
(d) Ker φ is given by
   
X,1 (D + C)/AX,1 (D) ∩ AX,2 (D + C)/AX,2 (D) = Har (C, (D + C)|C ),
Aar ar ar ar 0

(e) Coker φ is given by

 . 
X,12 (D + C)/AX,12 (D)
Aar X,12 (D + C)/AX,12 (D) + AX,12 (D + C)/AX,12 (D)
ar
Aar ar ar ar

= Har1 (C, (D + C)|C ).

Therefore, by definition, we have the long exact sequence
0 → Har0 (X, C) → Har0 (X, D + C) → Har0 (C, (D + C)|C )
→ Har1 (X, D) → Har1 (X, D + C) → Har1 (C, (D + C)|C )
→ Har2 (X, D) → Har2 (X, D + C) → 0,
as desired. 

B.4.2.2 Horizontal Curves

For horizontal curve E P corresponding to an algebraic point P of XF , we have

X,12 (D + E P )/AX,12 (D)
(a) Aar ar

= Afin
X,12 (D + E P )/AX,12 (D) ⊕ AXF (DF + P)/AXF (DF ) ⊗Q R
fin

= AEP ,01 ⊕ k(E P ) ⊗Q R = Aar

EP ;

X,1 (D + E P )/AX,1 (D) = k(E P ), diagonally embedded in

(b) Aar ar

X,1 (Dfin + E P )/AX,1 (Dfin ) ⊕ AXF (DF + P)/AXF (DF ) ⊗Q R;

Afin fin

X,2 (D + E P )/AX,2 (D)

(c) Aar ar

 
= AEP (Dfin + E P )|EP ⊕ H 0 (XF , DF + E P ) ⊗ R/H 0 (XF , DF ) ⊗ R.
Similarly, we have the corresponding morphism
ϕ
X,1 (D + E P )/AX,1 (D) ⊕ AX,2 (D + E P )/AX,2 (D) −→ AX,12 (D + E P )/AX,12 (D)
Aar ar ar ar ar ar

(x, y) 7→ x − y,
and hence obtain the following proposition in parallel.

Proposition B.5. Let E P be the horizontal curve E P corresponding to an alge-

braic point P of XF , then, for any D, there exists a long exact sequence
0 → Har0 (X, E P ) → Har0 (X, D + E P ) → Ker ϕ
→ Har1 (X, D) → Har1 (X, D + E P ) → Coker ϕ
→ Har2 (X, D) → Har2 (X, D + E P ) → 0.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 474

474 Five Essays on Arithmetic Cohomology

However, unlike for vertical curves, we do not have the group isomorphisms
between Ker ϕ, reps. Coker ϕ, and Har0 (E P , (D+E P )|EP ), reps. Har1 (E P , (D+E P )|EP ).
This is in fact not surprising: different from vertical curves, for the arithmetic
cohomology, there is no simple additive law with respect to horizontal curves
when count these arithmetic groups: In Arakelov theory, we only have
1
χar (X, D + E P ) = χar (X, D) + χar (E P , (D + E P )|EP ) − dλ (E) (B.28)
2
with discrepancy − 21 dλ (E) resulting from the Green functions. (See e.g. p.114
of [67].)
On the other hand, recall that, on generic fiber XF , we have the long exact
sequence of cohomology groups

0 → H 0 (XF , DF ) → H 0 (XF , DF + P) → OXF (D + P)|P → Q → 0

(B.29)
(with Q defined by 0 → Q → H 1 (XF , DF ) → H 1 (XF , DF + P) → 0),

and that, in Arakelov theory, see e.g. Chapter VI, particularly, p.140 of [67],
what really used is the much rough version λ(DF + P) ' λ(DF ) ⊗ OXF (D + P)|P
where λ denotes the Grothendieck-Mumford determinant. One checks that the
exact sequence (B.29) does appear in our calculation (B.28) above. Indeed, for
the curve XF /F,

H 0 (XF , DF ) = k(XF ) ∩ AXF (DF ) & H 1 (XF , DF ) = AXF /k(XF ) + AXF (DF ).

Consequently, (B.29) is equivalent to the exact sequence

0 → H 0 (XF , DF + P)/H 0 (XF , DF ) → AXF (DF + P)/AXF (DF ) → Q → 0

0 → Q → H 1 (XF , DF ) → H 1 (XF , DF + P) → 0.

(Note that AXF (DF + P)/AXF (DF ) is supported only on P.) Clearly, all this can be
read from the calculations in (a,b,c) and the morphism ϕ above. So our construc-
tion offers a much more refined structure topologically.

B.4.3 Duality of Cohomology Groups

Let π : X → Spec OK be an arithmetic surface defined over the ring of integers

of a number field K. Then we have the adelic space Aar X , its level two subspaces
Aar
01 , Aar
02 and Aar
12 (D), and hence the cohomology groups Hari (X, OX (D) associated
to a Weil divisor D on X. Moreover, there is a natural ind-pro structure on Aar X

X = lim
Aar

lim AX,12 (D) AX,12 (E). (B.30)
−→D ←−E:E≤D
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 475

Cohomology for Arithmetic Surfaces 475


Note that AX,12 (D) AX,12 (E)’s are locally compact topological spaces. Conse-
quently, as explained in the appendix, induced from the projective limit, we get a
natural final topology on
AX,12 (D) = lim AX,12 (D) AX,12 (E).

(B.31)
←−E:E≤D

Similarly, induced from the projective limit, we get a natural initial topology on

X = lim AX,12 (D).

Aar
−→D
(B.32)

Moreover, by Theorem II, we know that all three level two subspaces Aar 01 , A02
ar
ar ar
and A12 (D) are closed in AX . Consequently, we obtain natural topological struc-
tures on arithmetic cohomology groups Hari (X, OX (D)) induced from the canonical
topology of AarX . Recall that, for a topology space, its topological dual is defined
by bT := {φ : T → S1 continuous} together with compact-open topology.

Theorem B.1. Let X be an arithmetic surface and D be a Weil divisor on X. Then

as topological groups, there are natural isomorphisms
(X, D) ' Har2−i ((ω) − D)
Hari[ i = 0, 1, 2. (B.33)

This theorem is proved in the following chapter, after we expose some basic
structural results for the ind-pro topology on Aar
X.
 b2530   International Strategic Relations and China’s National Security: World at the Crossroads

This page intentionally left blank

b2530_FM.indd 6 01-Sep-16 11:03:06 AM

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 477

Appendix C

Ind-Pro Topologies in Arithmetic

Dimension Two

In this chapter, based on ind-pro topology in arithmetic dimension two, we estab-

lish some basic properties for ind-pro topologies on various adelic spaces associ-
ated to arithmetic surfaces. The main result is the following:

Theorem C.1. Let X be an arithmetic surface. Then with respect to the canonical
ind-pro topology on Aar
X , we have

(1) Level two subspaces Aar X,01 , AX,02 and AX,12 (D) are closed in AX ;
ar ar ar
ar
(2) AX is a Hausdorff, complete, and compact oriented topological group;
(3) Aar
X is self-dual. That is, as topological groups,

X ' AX .
ar
Aar
d

C.1 Ind-Pro Topology on Adelic Spaces

C.1.1 Ind-Pro Topology Spaces and Their Duals

To begin with, let us recall some basic topological constructions for inductive
limits and projective limits of topological spaces.

(1) Let {Gm }m be an inductive system of topological spaces, G := lim Gm with

−→m
structure maps ιm : Gm → G. Then, the inductive topology on G is defined
by assigning subsets U of G to be open, if ι−1 m (U) is open in G m for each
m. Inductive topology is also called the final topology since it is the finest
topology on G such that ιm : Gm → G are continuous.
(2) Let {Gn }n be a projective system of topological spaces, G := lim Gn with struc-
←−n
ture maps πn : G → Gn . Then, the projective topology on G is defined as the
one generated by open subsets π−1 n (U n ), where U n are open subsets of G n .

477
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 478

478 Five Essays on Arithmetic Cohomology

Projective topology is also called the initial topology since it is the coarsest
topology on G such that πn : G → Gn are continuous.

For a topological space T , denote by bT := { f : T → S1 continuous}. There

is a natural compact-open topology on b T , generated by open subsets of the form
W(K, U) := { f ∈ T : f (K) ⊆ U}, where K ⊆ T are compact, U ⊂ S1 are open.
b
We call bT together with its canonical topology the (topological) dual of T .
For inductive and projective topologies, we have the following general results
concerning their duals.

Proposition C.1. Let {Pn }n be a projective system of topological groups with

structural maps πn,m : Pn → Pm and πn : lim Pn → Pn . Assume that all πn and
←−n
πn,m are surjective and open. Then, as topological groups,
[
lim Pn .
Pn ' lim d
←−n −→n

Proof. Denote by b πn,m : P

cm → Pbn , fm 7→ fm ◦ πn,m , the dual of πn,m : Pn → Pm .
Then, for an element lim fn ∈ lim d Pn , we have bπn,m ( fm ) = fn , or equivalently, fm ◦
−→n −→n
πn,m = fn , for sufficiently large m, n. Hence, for an element x = lim xn ∈ lim Pn ,
←−n ←−n
we have fm (xm ) = fm (πn,m (xn )) = fn (xn ). Based on this, we define a natural map
ϕ : lim d
Pn −→ lim
[ Pn , lim fn 7→ f
−→n ←−n −→n

where f : lim Pn → S1 , x = lim xn 7→ fn (xn ). From the above discussion, K is

←−n ←−n
well defined. Moreover, we have the following

Lemma C.1.

(1) K is continuous. In particular, ϕ is well defined;

(2) ϕ is a bijection;
(3) ϕ is continuous; and
(4) ϕ is open.

Proof. (1) Let U be an open subset of S1 . If x = lim xn ∈ f −1 (U), fn (xn ) ∈ U for

←−n
sufficiently large n. In particular, xn ∈ fn−1 (U). On the other hand, since fn is con-
tinuous, fn−1 (U) is open. So, lim fn−1 (U) is an open neighborhood of x = lim xn .
←−n ←−n
Note that f (lim fn−1 (U)) = fn ( fn−1 (U)) = U. Hence lim fn−1 (U) ⊆ f −1 (U). Conse-
←−n ←−n
quently, K is continuous, and hence ϕ is well defined.
(2) To prove that ϕ is injective, we assume that ϕ(lim gn ) =: g = f . Thus
−→n
fn (xn ) = gn (xn ) for sufficiently large n and for all x = lim xn ∈ lim Pn . Note that
←−n ←−n
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 479

Ind-Pro Topologies in Arithmetic Dimension Two 479

πn,m are surjective. So fn (xn ) = gn (xn ) for all xn ∈ Pn . This means that fn = gn for
sufficiently large n. Consequently, lim fn ≡ lim gn , and hence ϕ is injective.
−→n −→n
To show that ϕ is surjective, let f : lim Pn → S1 be a continuous map. Then,
←−n
for any open subset U ⊂ S1 containing 1, f −1 (U) is an open neighborhood of
0 in lim Pn . Hence, there exists an open subgroup V of lim Pn such that V ⊆
←−n ←−nY
f −1 (U). Being a subgroup, we may write V as V = lim Pn ∩ Kn where Kn ⊆
←−n
α
Pn are open subgroups and Kn = Pn for almost all n. However, by the continuity,
f (V) = 1. So, there exists a sufficiently large N such that f ( lim Pn ) = 1 for
←−n:n≤N
all n ≥ N. Built on this, we define, for n ≥ N, the maps fn : Pn → S1 , xn 7→
f (x) if πn (x) = xn . Note that f (x) always make sense, since πn : lim Pn → Pn
←−n
is surjective. Moreover, fn ’s are well defined. Indeed, if y ∈ lim Pn such that
←−n
πn (y) = xn , then πn (y) = xn = πn (x) for n ≥ N. Hence x − y ∈ V. This implies that
f (y) = f (x). Clearly, by definition, ϕ(lim fn ) = f . So ϕ is surjective.
−→n
(3) and (4) are direct consequences of the bijectivity of ϕ. Indeed, to prove
that ϕ is continuous, it suffices to show that for open subsets of lim
[ Pn in the
←−n
form U = W(K, 0), ϕ−1 (U) is open in lim d Pn , where K is a compact subset of
−→n
lim Pn . Since πn,m are continuous, Kn = πn (K) are compact. In this way, we get an
←−n
inductive system of open subsets {W(Kn , 0)}n . Set U = lim W(Kn , 0). Note that,
−→n
by the bijectivity of ϕ, f (U) = U. This shows that K is continuous.
To prove ϕ is open, let U be a subset of lim cPn . By definition, there exists n
−→n
such that U = ιn W(Kn , 0) for a compact subset Kn of Pn . But πn : lim Pn → Pn
−1

←−n
is open, K := π−1
n (Kn ) is compact in lim Pn . Consequently, W(K, 0) is open in
←−n
[
lim Pn . Note that, from the bijectivity of ϕ, we have ϕ(lim W(Kn , 0)) = W(K, 0).
←−n −→n
So ϕ is open. This proves the lemma and hence also the proposition. 

Next we treat inductive systems. By definition, an inductive system {Dn }n of

Hausdorff topological groups is called compact oriented, if for any compact subset
K ⊆ lim Dn , there exists an index n0 such that K ⊆ Gn0 .
−→n

Proposition C.2. Let {Dn }n be a compact oriented inductive system of Hausdorff

topological groups with structural maps ιn : Dn → lim Dn and ιn,n0 : Dn → Dn0 .
−→n
Assume that ιn,n0 and ιn are injective, and Dn = ιn (Dn ) ⊆ lim Dn are complete.
−→n
Then, as topological groups,
[
lim Dn .
Dn ' lim d
−→n ←−n
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 480

480 Five Essays on Arithmetic Cohomology

Proof. To start with, we define ψ : lim−→n d Dn −→ lim[ ←−n Dn , lim←−n fn 7→ f

where f : lim Dn → S1 , lim xn 7→ fn (xn ). Hence, it suffices to prove the following:
←−n −→n

Lemma C.2.

(1) K is well defined and continuous. In particular, ψ is well defined.

(2) ψ is a bijection;
(3) ψ is continuous; and
(4) ψ is open.

Proof. (1) Note that for n < n0 , fn = fn0 ◦ ιn,n0 and xn0 = ιn,n0 (xn ). Consequently,
fn (xn ) = fn0 ◦ ιn,n0 (xn ) = fn0 (ιn,n0 (xn )) = fn0 (xn0 ). So K is well defined.
To prove that K is continuous, let U ⊂ S1 be an open subset and
lim xn ∈ f −1 (U). By definition, xn ∈ fn−1 (U) =: Un and Un are open. Hence
−→n
U := lim Un is open and x = lim xn ∈ U. Moreover, f (U) ⊆ U. So K s continuous.
−→n −→n
(2) Assume that ψ(lim fn ) = f ≡ 0. Then, for any x = lim xn , f (x) = 0. This
←−n −→n
means that for all n, and xn ∈ Dn , fn (xn ) = f (x) = 0 where x is determined by the
condition that for all n0 ≥ n, xn0 = ιn,n0 xn . (Since ιn are injective, this is possible.)
Thus fn ≡ 0 and hence lim fn = 0.
←−n
Dn , let fn = f ◦ ιn : Dn → S1 . Clearly, fn is continuous.
(3) For any f ∈ lim d
←−n
So fn ∈ D cn . Moreover, for all n0 ≥ n, fn = f ◦ ιn = f ◦ ιn0 ◦ ιn,n0 = fn0 ◦ ιn,n0 . That
is, { fn }n forms a projective limit. Obviously, ψ(lim fn ) = f .
←−n
(4) This is rather involved: not only the just proved bijectivity of ψ, but
all assumptions for our injective system are used here. Let U = W(K, 0) be an
[
open subset of lim Dn , where K ⊆ lim Dn is compact. Since lim Dn is Haus-
−→n −→n −→n
dorff, and Dn are complete. Dn ⊆ lim Dn are closed. So Kn := K ∩ Dn
−→n
are compact. If lim fn is an element in ψ−1 (U) ⊆ lim Dn , we have fn ∈
←−n −→n
W(Kn , 0), and lim fn ∈ lim W(Kn , 0) ⊆ W(K, 0). But, by the bijectivity of ψ,
←−n ←−n
lim W(Kn , 0) = W(lim Kn , 0). So it suffices to show that W(lim Kn , 0) is open,
←−n −→n −→n
or the same, lim Kn is compact. This is a direct consequence of our assump-
−→n
tions. Indeed, since our inductive system if compact oriented, there exists a cer-
tain n0 such that lim Kn ⊆ Dn0 . Hence, Kn = Kn0 for all n ≥ n0 . Consequently,
−→n
lim Kn = Kn0 ⊆ Dn0 . So lim Kn is compact, since Kn0 is a compact subset of the
−→n −→n
closed subspace Dα0 .
(5) This is a direct consequence of the bijectivity of ψ. Indeed, let U ⊆ lim D
cn
←−n
be an open subset. By definition, without loss of generality, we may assume that
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 481

Ind-Pro Topologies in Arithmetic Dimension Two 481

there exists an n and an open subset W(Kn , 0) of D

cn such that U = π−1
n (W(Kn , 0)).
Since Kn is a compact subset of Dn and Dn is closed in lim Dn , Kn is compact in
−→n
lim Dn . On the other hand, ψ(U) = ψ(π−1
n (W(Kn , 0))) = W(Kn , 0). So ψ(U) is open
−→n
[
in lim Dn . 
−→n

C.1.2 Adelic Spaces and Their Ind-Pro Topologies

Let X be an arithmetic surface. For a complete flag (X, C, x) on X (with C an

irreducible curve on X and x a close point on C), let k(X)C,x its associated local
ring. By Theorem A.1, k(X)C,x is a direct sum of two-dimensional local fields.
Denote by (πC , tX,C ) the local parameter defined by the flag (C, x) of X, and fix a
Madunts-Zhukov lifting ([72])
Y0
bC,x πC 0
Y0
bC,x −lifting
. Y
hC = (hπC ,tX,C ) : AC,01 ' O O −−−→ ObC,x . (C.1)
x:x∈C x:x∈C x:x∈C
Then, following Parshin, see e.g. Example A.2,
Y0 Y0  Y 
X = AX,012 =
Afin k(X)C,x := k(X)C,x
x∈C C x:x∈C

aiC ∈ AC,01 ,
 
 


 ∞
X YY 


hC (aiC )πCiC aiC = 0 (iC  0); .


:=  ∈ k(X)C,x
 
 C 

iC =−∞
 
C x:x∈C
min{iC : aiC , 0} = 0 (∀ C)
0


 


This gives the finite adelic space for X. Moreover, from Parshin-Osipov, see e.g.
Definition A.1, we have the infinite adelic space A∞ X := AXF ⊗Q R, and hence the
b
total arithmetic adelic space Aar
X for the arithmetic surface X:

X := AX ⊕ AX .
∞
Aar fin
(C.2)
Moreover, there are natural ind-pro structures on and hence on Afin
X , A∞
X Aar
X, since
AX = lim lim AX,12 (D) AX,12 (E),
fin 
−→D ←−E:E≤D
(C.3)
AX = lim lim AXF ,1 (D) AXF ,1 (E)b
∞ 
⊗Q R.
−→D ←−E:E≤D

Consequently, induced from the locally compact topologies on AX,12 (D) AX,12 (E)
∞
⊗Q R, we get canonical ind-pro topologies on Afin

and AXF ,1 (D) AXF ,1 (E)b X , AX , and
ar
hence on AX . For example, by [72], a fundamental system of open neighborhood
of 0 in Afin
X is given by
 


 ∞
aiC ∈ UiC ⊆ AC,01 open subgroup 

 X 
−iC
fin
= .
 
h (a )π ∈ : U ∀ i  0 (C.4)
 

 C iC C C A X iC
AC,01 C 

 iC =−∞
 
min{iC : UiC = AC,01 } ≤ 0 (∀ C)
0

 


November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 482

482 Five Essays on Arithmetic Cohomology

We have a similar descriptions for A∞ X . However, while, with respect to these

fin ∞ ar
canonical topologies, AX , AX , and AX are additive topological groups, they are
not topological rings. That is, the multiplication operations are not continuous for
these spaces. Still, in §C.2.1 below, we will prove the following very useful

Proposition C.3. For a fixed a ∈ ACar , the scalar product by a, namely, the map
a×
X −→ AX , x 7→ ax, is continuous.
Aar ar

Remark C.1. This result can be used to establish a similar result for two-dimen-
sional local fields. Indeed, since k(X)C,x = lim lim AX,12 (nC) AX,12 (mC), as a

 −→n ←−m:m≤n
X = lim
subspace of Afin lim AX,12 (D) AX,12 (E), there is a natural ind-pro topol-
−→D ←−E:E≤D
ogy on k(X)C,x , induced from the ind-pro topology on AX,012 . On the other hand,
for a two-dimensional local field K, we have F = lim lim m−n
 −m
F mF , where mF
−→n ←−m:m≤n
denotes the maximal ideal of K. So from the natural locally compact topologies on
the quotient spaces m−n
 −m
F mF , we obtain yet another ind-pro topology on K, and
hence on k(X)C,x , since k(X)C,x is also a direct sum of two-dimensional local fields.
Induced from the same roots of locally compact topology on one-dimensional lo-
cal fields, these two topologies on k(X)C,x are equivalent. This, with the above
proposition, then proves the following

Corollary C.1. For a fixed aC,x ∈ k(X)C,x , the scalar product by aC,x , namely,
aC,x ·
the map k(X)C,x −→ k(X)C,x , α 7→ aC,x α is continuous. In particular, the scalar
product of a fixed element on a two-dimensional local field is continuous.

This result can be used to prove that, with respect to the canonical ind-pro
topology, two-dimensional local fields are self-dual as topological groups.

C.1.3 Adelic Spaces are Complete

∞
We will show that adelic spaces Afin
X and AX are complete. For basic facts of
completeness condition on topological groups, please refer to [17] and [30]. We
begin with the following:

∞
Proposition C.4. The subspaces AX,12 (D) ⊂ Afin X and AXF ,1 (D) ⊂ AX , and the
level two subspace AX,01 , AX,02 and AX,12 (D) of AX are complete and hence closed.
ar ar ar ar

Proof. As our proof below works for all other types as well, we only
treat AX,12 (D) ⊂ Afin X to demonstrate how our arguments work. Since

AX,12 (D) AX,12 (E)’s are finite dimensional vector spaces over one-dimensional
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 483

Ind-Pro Topologies in Arithmetic Dimension Two 483

local field, which is locally compact, so they are complete. Consequently, as

a projective limit of complete spaces, AX,12 (D) = lim AX,12 (D) AX,12 (E) is

←−E:E≤D
complete. It is also closed since AX,012 is Hausdorff. 

Proposition C.5. For an arithmetic surface X, its associated adelic spaces Afin
X
and A∞
X are complete.

Proof. We will give a uniform proof for both finite and infinite cases. For this
∞
reason, we use simply A to denote both Afin X and AX , and A(D) for both AX,12 (D)
and AXF (D)⊗Q R. Clearly, it suffices to prove the following:
b

Lemma C.3. For a strictly increasing sequence {A(Dn )}n in A, lim A(Dn ) is
−→n
complete.

Proof. Let {an }n be a Cauchy sequence of A. We will show that these exists a
divisor D such that {an }n ⊆ A(D). Assume that, on the contrary, for all divisors D,
{an }n * A(D). We claim that there exists a subsequence {akn }n of {an }, a (strictly
increasing) subsequence {Dkn }n of {Dn }n , and an open neighborhood U of 0 in
lim A(Dn ) such that
−→n

(i) akn ∈ A(Dkn )\A(Dkn−1 ) for all n ≥ 2,

(ii) ak1 , . . . , akn , · · · < U, and
(iii) aki+1 , . . . , akn , · · · < U + A(Di ), i ≥ 1.

If so, since akm < U + A(Dm ) and akn ∈ A(Dm ), for any n > m, akn − akm <
U. So, {akn }n is not a Cauchy sequence of lim A(Dn ). But, since {an }n is a
−→n
Cauchy sequence in A, with respect to the natural induced topology on lim A(Dn ),
−→n
{akn }n ⊆ lim A(Dn ) is a Cauchy sequence of lim A(Dn ). This is a contradiction.
−→n −→n
Therefore, there exists a divisor D such that {an }n ⊆ A(D). By Proposition 40,
A(D) is complete. So the Cauchy sequence {an }n is convergent in A(D) and hence
in A as well.
To prove the above claim, we select {akn }n and the corresponding Dkn ’s as
follows. To begin, let ak1 = a1 . Being an element of A, there always a divisor Dk1
such that ak1 ∈ A(Dk1 ). Since for all D, {an }n * A(D), there exists k2 and a divisor
Dk2 such that Dk2 > Dk1 and ak2 ∈ A(Dk2 ) − A(Dk1 ). By repeating this process, we
obtain a subsequence {akn }n of {an }, a (strictly increasing) subsequence {Dkn }n of
{Dn }n such that (i) above holds. Hence to verify the above claim, it suffices to find
an open subset U satisfying the conditions (ii) and (iii) above. This is the contents
of the following:
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 484

484 Five Essays on Arithmetic Cohomology

Claim C.1. Let {A(Dn )}n be a strictly increasing sequence and {an }n be a se-
quence of elements of A. Assume that an ∈ A(Dn ) − A(Dn−1 ) for all n ≥ 1.
Then there exists a open subset U of lim A(Dn ) such that a1 , . . . , an , · · · < U and
−→n
am+1 , . . . , an , · · · < U + A(Dm ) for all m < n.

Proof. We separate the finite and infinite adeles.

Finite Adeles Since A(D1 )/A(D0 ) is Hausdorff, there exists an open, and hence
closed, subgroup U1 ⊆ A(D1 ) such that a1 < U1 and U1 ⊃ A(D0 ). Since A(D1 )
is complete and U1 is closed in A(D1 ), U1 is complete as well. Now, viewing in
A(D2 ), since A(D2 ) is Hausdorff, U1 is a complete subspace, so U1 is closed in
A(D2 ). Hence A(D2 )/U1 is Hausdorff too. Therefore, there exists an open and
hence closed subgroup V2,0 of A(D2 ) such that a1 , a2 < V2,0 and V2,0 ⊃ U1 . In

addition, A(D2 ) A(D1 ) is Hausdorff, there exists an open subgroup V2,1 such that
a2 < V2,1 and V2,1 ⊃ A(D2 ). Consequently, if we set U2 = V2,0 ∩V2,1 , U2 is an open
hence closed subgroup of A(D2 ) such that a1 , a2 < U2 , a2 < U2 + A(D1 ) and U2 ⊃
U1 . So, inductively, we may assume that there exists an increasing sequence of
open subgroups U1 , . . . , Un−1 satisfying the properties required. In particular, the
following quotient spaces A(Dn ) Un−1 +A(D0 ) = A(Dn ) Un−1 , . . . , A(Dn ) Un−1 +
 

A(Dm ), . . . , A(Dn ) Un−1 +A(Dn−1 ) = A(Dn ) A(Dn−1 ) are Hausdorff. Hence, there
 

are open subgroups Vn,m , 0 ≤ m ≤ n − 1 of A(Dn ) such that am+1 , . . . , an < Vn,m
and Vn,m ⊃ Un−1 + A(Dm ). Define Un := ∩n−1 m=1 Vn,m . Then U n is an open subgroup
of A(Dn ) satisfying a1 , . . . , an < Un , am+1 , . . . , an < Un + A(Dm ), 1 ≤ m ≤ n −
1 and Un ⊃ Un−1 . Accordingly, if we let U = lim Un , by definition, U is an
−→n
open subgroup of lim A(Dn ), and from our construction, a1 , . . . , an , · · · < U and
−→n
am+1 , . . . , an , · · · < U + A(Dm ), m ≥ 1.

Infinite Adeles Since A(D1 ) is Hausdorff, there exists an open subset U1 of


A(D1 ) such that a1 < A(D1 ). Moreover, since A(D2 ) ' A(D2 ) A(D1 ) ⊕ A(D1 )

and A(D2 ) A(D1 ) is Hausdorff, there exists an open subset U2 of A(D2 ) such
that a1 , a2 < U2 and U2 ∩ A(D1 ) = U1 . In particular, a2 < U2 + A(D1 ).
Similarly, as above, with an inductive process, based on the fact that A(Dn ) '
 
A(Dn ) A(Dn−1 ) ⊕ A(Dn−1 ) and A(Dn ) A(Dn−1 ) is Hausdorff, there exists an open
subset Un of A(Dn ) such that a1 , . . . , an < Un and Un ∩ A(Dn ) = Un−1 . Conse-
quently, am+1 , . . . , an < Un + A(Dm ), 1 ≤ m ≤ n − 1. In this way, we obtain
an infinite increasing sequence of open subsets Un . Let U = lim Un . Then by
−→n
definition U is an open subset of lim A(Dn ) satisfying a1 , . . . , an , · · · < U and
−→n
am+1 , . . . , an , · · · < U + A(Dm ), m ≥ 1. This then proves the claim, the lemma and
hence also the proposition. 
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 485

Ind-Pro Topologies in Arithmetic Dimension Two 485

C.1.4 Adelic Spaces are Compactly Oriented

Proposition C.6. For an arithmetic surface X, its associated adelic spaces Afin X
and A∞X are compactly oriented. That is to say, for any compact subgroup, resp.
∞
a compact subset, K in Afin
X , resp. AX , there exists a divisor D on X, resp. on XF ,
such that K ⊆ AX,12 (D), resp. K ⊆ AXF (D).

Proof. As in the previous subsection, we treat both finite and infinite places uni-
formly. Assume that for all divisor D, K * A(D). Fix a suitable D0 . By our
assumption, K * A(D0 ). Since A is Hausdorff and A(D0 ) is closed, there exists a
certain divisor D1 and an element a1 ∈ K, such that D1 > D0 , a1 ∈ A(D1 )\A(D0 ).

Similarly, since A(D1 ) is closed, A A(D0 ) is Hausdorff, we can find an open sub-
group U10 ⊆ A A(D0 ) such that a1 + A(D0 ) * A A(D0 ). Consequently, there exists
 
an open subgroup U1 of A such that a1 < U1 and U1 ⊃ A(D0 ). Now use (D1 , A)
instead of (D0 , A), by repeating the above construction, we can find a divisor D2 ,


an open subgroup U2 ⊂ A and an element a2 ∈ K ∩ A(D2 )\A(D1 ) such that
D2 > D1 , a2 < U2 , U2 ⊃ U1 . In this way, we obtain a sequence of divisors Dn ,


a sequence of elements an ∈ K ∩ A(Dn )\A(Dn−1 ) and a sequence of open sub-
groups Un such that Dα > Dn−1 , an < Un , Un ⊃ Un−1 . Let U = lim Un . Then U is
−→n
an open, and hence closed, subgroup of A. Consequently, K ∩ U is compact. This
is a contradiction. Indeed, since a1 , . . . , an < Un for all n, the open covering {Un }n
of K ∩ U admits no finite sub-covering. 

C.1.5 Dual Adelic Spaces

Proposition C.7. As topological groups,

 ∨  ∨
lim AX,12 (D) AX,12 (E) ' lim AX,12 (D) AX,12 (E) ,
 
←−E:E≤D −→E:E≤D

∨  ∨
⊗Q R ,
 
lim AXF (D) AXF (E)b ⊗Q R ' lim AXF (D) AXF (E)b
←−E:E≤D −→E:E≤D
 ∨  ∨ (C.5)
lim AX,12 (D) ' lim AX,12 (D) ,
−→D ←−D

∨  ∨
lim AXF (D)b ⊗Q R ' lim AXF (D)b ⊗Q R ,
−→D ←−D
  
where, AXF (D)b⊗Q R := lim AXF (D) AXF (E)b ⊗Q R .
←−E:E≤D
In particular,
 ∨
X ' lim lim AX,12 (D) AX,12 (E) ,
dfin 
A
←−D −→E
E≤D
 ∨
A X ⊗Q R .
c∞ ' lim lim AXF (D)AXF (E)b
←−D −→E
E≤D
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 486

486 Five Essays on Arithmetic Cohomology

Proof. We apply Proposition C.1, resp. Proposition C.2 to prove the first, resp.
the second, group of two isomorphisms. We need to check the conditions listed
there.
As above, we treat finite adeles and infinite adeles simultaneously. So we use
A and A(D) as in §C.1.3. Then A(D) = lim A(D) A(E). Now, for E < E 0 ,

←−E:E≤D
A(D) A(E) = A(D) A(E 0 ) ⊕ A(D0 ) A(E) −→ A(D) A(E 0 ), and the first projections
   
coincide with structural maps πD/E,D/E 0 : A(D) A(E) → A(D) A(E 0 ). So πD/E,D/E 0
 
are surjective and open. Similarly, being a natural quotient map, πD,D/E : A(D) →

A(D) A(E) are surjective and open. So, by Proposition C.1, we get the first group
of two homeomorphisms for topological groups.

To treat the second group, recall that A(D) A(E) are complete. So, their pro-
jective limits A(D)’s are complete. This implies that A(D) are closed in A, since
A is Hausdorff. On the other hand, for D < D0 , AX,12 (D) ⊆ AX,12 (D0 ). So the
structural maps ιD,D0 : A(D) → A(D0 ) and ι : A(D) → A are injective and closed.
Thus, by Proposition C.2, it suffices to show that the inductive system {A(D)}D is
compact oriented. This is simply the contents of § C.1.3 and §C.1.4. All this then
completes our proof, since the last two homeomorphisms are direct consequences
of previous four. 

Corollary C.2. As topological groups,

[ [
X ' AX , X ' AX ,
d
Afin fin
A∞
d ∞
and hence car ' Aar .
d
A X X

Proof. Since A(D)/A(E) are locally compact and hence they are self dual. Thus,
to prove this double dual properties for our spaces, it suffices to check the con-
ditions listed in Proposition C.2 for inductive systems { lim A(D) [
A(E)}E and
−→E:E≤D
in Proposition C.1 for the projective system {A(D)}
[ D . With the above lengthy dis-
cussions, all this now becomes rather routine. For example, to verify that c A is
Hausdorff, we only need to recall that S1 is compact. Still, as careful examinations
would help understand the essences of our proof above, we suggest ambitious
readers to supply omitted details. 

C.2 Adelic Spaces and Their Duals

C.2.1 Continuity of Scalar Products

Now we prove that the scalar product maps on adelic spaces are continuous, even
adelic spaces are not topological rings.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 487

Ind-Pro Topologies in Arithmetic Dimension Two 487

∞
Proposition C.8. For a fixed element a of Afin
X , resp. of AX , the induced scalar
a× a×
product map φfin
a : AX −→ AX , resp. φa : AX −→ AX , is continuous.
fin fin ∞ ∞ ∞

Proof. If a = 0, there is nothing to prove. Assume, from now on, that a , 0 and
X∞
write a = (aC )C = hC (aiC )πCiC C ∈ Afin


X . Here, for each C, we assume that
iC =iC,0
aiC,0 , 0. To prove that ϕa is continuous, by our description (C.4) of the ind-pro
X , it suffices to show that for an open subgroup U = (UC )C , as
topology on Afin
an open neighborhood of 0, its inverse image ϕ−1 a (U) contains an open subgroup.
X∞ X∞
Write U = (UC ) = = hC (AC,1 (D jC ) πCiC + hC (AC,01 )πCiC ∩ Afin



X , and for
jC =−∞ jC =rC
later use, set IC := rC − iC,0 .
X∞
Let b = (bC )C = hC (bkC )πCkC C ∈ ϕ−1
a (U) ⊂ AX . Then, for each fixed C,
fin


kC =−∞
∞
X ∞
X
aC bC = hC (aiC )hC (blC −iC ) πClC . Recall that hC is simply the lifting map


lC =−∞ iC =iC,0
Y0 Y0 Y0
hC : AC,01 ' bC,x πC
O bC,x −lifting
O −−−→ bC,x . Thus if bk
O ∈ AC,01 , we al-
C
x:x∈C x:x∈C x:x∈C
X∞
hC AC,01 πCmC . Moreover, if


ways have hC (aiC )hC (blC −iC ) ∈ we write, as we
mC =0
can, that aiC ∈ AC,1 (FiC ), bkC ∈ AC,1 (EkC ) for some divisors FiC and EkC , then we
∞
X
hC AC,1 (FiC + ElC −iC ) πCmC .


have hC (aiC )hC (blC −iC ) ∈
mC =0

Now write bC = kC =−∞ + kC =IC hC (bkC )πC . We will construct the required
PIC −1
P∞
kC

open subgroup according to the range of the degree index kC .

∞
X Y0
hC (AC,01 )πCkC ∩



(i) If bC ∈ k(X)C,x , we have aC bC ∈ UC .
kC =IC x:x∈C

(ii) To extend the range including also the degree IC − 1, choose a divisor E IC −1
such that hC AC,1 (FiC,0 + E IC −1 ) ⊆ hC AC,1 (DrC −1 ) . Then, if we choose bC ∈



∞
X Y0
hC AC,1 (E IC −1 )πCIC −1 + hC (AC,01 )πCkC ∩



k(X)C,x , aC bC ∈ UC as well.
kC =IC x:x∈C

(iii) Similarly, to extend the range including the degree IC − 2, choose a divi-
sor E IC −2 such that hC AC,1 (FiC,0 + E IC −2 ) ⊆ hC AC,1 (DrC −2 ) ∩ hC AC,1 (DrC −1 )




and hC AC,1 (FiC,0 +1 + E IC −2 ) ⊆ hC AC,1 (DrC −1 ) . Consequently, if we choose



November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 488

488 Five Essays on Arithmetic Cohomology

C −1
IX ∞
X Y0
hC AC,1 (EkC )πCkC + hC (AC,01 )πCkC ∩



element bC ∈ k(X)C,x , we
kC =IC −2 kC =IC x:x∈C
have aC bC ∈ UC .
Continuing this process repeatedly, we then obtain divisors EkC ’s such that,
C −1
IX ∞
X Y0
hC AC,1 (EkC )πCkC + hC (AC,01 )πCkC ∩



for bC ∈ VC := k(X)C,x , we
kC =−∞ kC =IC x:x∈C
have aC bC ∈ UC .
Since, for all but finitely many C, rC ≤ 0 and iC,0 ≥ 0, or better, IC = rC −iC,0 ≤
0. Therefore, from above discussions, we conclude that C VC ∩ Afin
Q
X is an open
subgroup of Afin fin

φ
Q
X and a V
C C ∩ A X ⊆ U. In particular, a is continuous.
A similar proof works for φ∞ a . We leave details to the reader. 

C.2.2 Residue Maps are Continuous

We prove next that the residue map is continuous. Fix a non-zero rational differ-
ential ω on X. Then, for an element a of Afin ∞
X , resp. AX , induced from the natural
residue pairing h·, ·iω , we get a natural map ϕa := ha, ·iω : Afin
fin
X −→ R/Z, resp.
ϕ∞a := ha, ·iω : A∞
X −→ R/Z.
∞
Lemma C.4. Let a be a fix element in Afin X , resp. AX . Then the induced map
ϕa := ha, ·iω : AX −→ R/Z, resp. ϕa := ha, ·iω : AX −→ R/Z, is continuous. In
fin fin ∞ ∞

particular, the residue map on arithmetic adeles Aar

X is continuous.

Proof. We prove only for ϕfin a , as a similar proof works ϕa . Write AX =

∞ fin
Q0
F: 2−dim F. And, for each local field K, fix an element tF of K such that for equal
local field
characteristic field K, tF is a uniformizer of K, while for mixed characteristics field
K, tF is a lift of a uniformizer of its residue field. Since, by Proposition C.8, the
scalar product is continuous, to prove the continuity of ha, ·iω , it suffices to show
that the residue map Res : Afin X → R/Z, x = (xF ) 7→
P
F resF (xF dtF ) is continu-
fin
ous. (Note that, by the definition of AX , e.g. (C.4), the above summation is a finite
P 
iC =∞ hC (AC,1 (0) πC + i=0 hC (AC,01 )πC ∩ AX
−1
iC P∞ iC fin
sum.) Since the open subgroup
is contained, the kernel of the residue map is an open subgroup. 

C.2.3 Adelic Spaces are Self-Dual

∞
We will treat both Afin
X and AX simultaneously. So as before, we use A to represent
them. Recall that, for a fixed a ∈ A, the map ha, ·iω : A → S1 is continuous.
Accordingly, we define a map ϕ : A → A,
b a 7→ ϕa := ha, ·iω .
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 489

Ind-Pro Topologies in Arithmetic Dimension Two 489

Proposition C.9. For the map ϕ : A → A,

b a 7→ ϕa := ha, ·iω , we have

(1) ϕ is continuous;
(2) ϕ is injective;
(3) The image of ϕ is dense;
(4) ϕ is open.

∞
Proof. (1) Let A be Afin
X , resp. AX . For an open subset W(K, 0) of A , where
c K
is a compact subgroup, resp. a compact subset, of A, let U := ϕ W(K, 0) . By
−1

A = lim lim A(D)/A(E).
Proposition C.7, c [ So we may write χ0 := h1, ·iω
←−D −→E:E≤D
as lim lim χD/E with χD/E ∈ A(D)/A(E).
[ Accordingly, write AD/E :=
←−D −→E:E≤D  
A(D)/A(E), KD/E := K ∩ A(D) K ∩ A(E) and let U D/E := aD/E ∈ AD/E :
χD/E aD/E KD/E = {0} resp. an open subset V . Since, for a fixed divisor D, A(D)


is closed in A, K ∩ A(D) is a subgroup, resp. a subset, of A(D). So, for E ≤ D,
KD/E is compact in AD/E . Consequently, from the non-degeneracy of χD/E on
locally compact spaces, U D/E is an open subgroup, resp. an open subset, of A,
and U = lim lim U D/E . We claim that U is open. Indeed, by Proposition C.6,
←−D −→E:E≤D
A is compact oriented. So, for compact K, there exists a divisor D1 such that
K ⊆ A(D1 ). On the other hand, since χ0 is continuous, there exists a divisor D2
such that A(D1 + D2 ) ⊆ Ker(χ0 ). Hence U ⊃ A(D2 ). Thus, for a fixed D, with
respect to sufficiently small E ≤ D, we have U D/E = AD/E . This verifies that U
is open, and hence proves (1), since the topology of c
A is generated by the open
subsets of the form W(K, 0).

(2) is a direct consequence of the non-degeneracy of the residue pairing. So

we have (2).
ψ d
To prove (3), we use the fact that A ' cA , where, for a ∈ A, ψa is given by
ψa : A → S , χ 7→ χ(a). Thus to show that the image of ϕ is dense, it suffices
c 1


to show that the annihilator subgroup Ann Im(ϕ) of Im(ϕ) is zero. Let then x ∈
Ann Im(ϕ) be an annihilator of Im(ϕ). Then, by definition, {0} = ψx {ϕa : a ∈


A} = {ϕa (x) : a ∈ A}. That is to say, ha, xiω = 0 for all a ∈ A}. But the residue


pairing is non-degenerate. So, x = 0.

(4) This is the dual of (2). Indeed, let U ⊆ A be an open subgroup, resp.
an open subset, of A. Then U ∩ A(D) is open in A(D). Since A(D) is closed,

U D/E := U ∩ A(D) U ∩ A(E) is open in AD/E . This, together with the fact that
χ is non-degenerate on its locally compact base space, implies that KD/E :=
 D/E
aD/E ∈ AD/E : χD/E aD/E · U D/E = {0} resp. an open subset V is a compact


November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 490

490 Five Essays on Arithmetic Cohomology

subset, resp. a compact subset. Let K := lim lim KD/E . Since U is open,
←−D −→E:E≤D
there exists a divisor E such that A(E) ⊆ U. This implies that there exists a
divisor D such that K ⊆ A(D). Otherwise, assume that, for any D, K * A(D).
Then, there exists an element k ∈ K such that k < A(ω) − E). Hence we have
χ(k · A(E)) , {0}, a contradiction. This then completes the proof of (4), and hence
the proposition. 

We end this long discussions on ind-pro topology over adelic space Aar
X with
the following main theorem.

Theorem C.2. Let X be an arithmetic surface. Then, as topological groups,

X ' AX . X ' AX .
fin ∞ ar
[
Afin
X ' AX and A∞
d In particular, Aar
d

Proof. With all the preparations above, this now becomes rather direct. Indeed,
by Proposition C.9, we have an injective continuous open morphism ϕ : A → c A.
So it suffices to show that ϕ is surjective. But this is a direct consequence of the
fact that ϕ is dense, since both A and cA are complete and Hausdorff. 
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 491

Appendix D

Closeness of Adelic Subspaces and

1
Finiteness of Har for Arithmetic Surfaces

For an arithmetic surface X and a Weil divisor D, there are natural arithmetic
cohomology groups Hari (X, OX (D)) (i = 0, 1, 2). Using ind-pro topology on adelic
space Aar 0 1
X,012 , we show that Har (X, OX (D)) is discrete, Har (X, OX (D)) is finite, and
2
Har (X, OX (D)) is compact. Moreover, we prove that all possible summations of
canonical subspaces Aar X,i (D), AX,kl (D) (i, k, l = 0, 1, 2) are closed in AX,012 , and
ar ar

hence complete our proof of topological dualities of among Hari ’s.

D.1 Arithmetic Cohomology Groups: A Review

Let π : X → Spec (OK ) be an arithmetic surface over integer ring OK of a number

field K. We assume that π is projective and flat, that X is integral and regular.
Denote by XF its generic fiber, and k(X), resp. k(XF ), the field of rational functions
of X, resp. of XF . We have k(X) = k(XF ).
Let (X, C, x) be a flag of X, consisting of an integral curve C on X and a closed
point x of C). Denote by k(X)C,x its local ring. It is known that k(X)C,x is a direct
sum of two-dimensional local fields. Since X is a Noetherian scheme, following
Parshin-Beilinson ( [10, 91, 92]), we have a two-dimensional adelic space Afin X,012 ,
its level two subspaces AX,01 , AX,02 , AX,12 (D) and the associated level one sub-
fin fin fin

X,0 , AX,1 (D), AX,01 (D), for a Weil divisor D on X. These spaces can be
spaces Afin fin fin

roughly described as follows:

Y0 Y0 Y0
X = AX,012 := AX;012 (OX ) :=
Afin k(X)C,x ,
fin fin

k(X)C,x :=
(C,x): x∈C C x: x∈C

Afin
X,01 := {( fC )C,x ∈
AX,012 }, Afin:= {( f x )C,x ∈ AX,012 },
X,02
AX,12 (D) := ( fC,x )C,x ∈ AX,012 ordC ( fC,x ) + ordC (D) ≥ 0 ∀C ⊆ X ,
fin 

X,0 := AX,01 ∩ AX,02 , AX,1(D) := AX,01 ∩ AX,12 (D), AX,2 (D) := AX,02 ∩ AX,12 (D).
Afin fin fin fin fin fin fin fin fin

491
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 492

492 Five Essays on Arithmetic Cohomology

It is well-known that AX,012 admits a natural ind-pro structure

X;012 (OX ) = lim
Afin

−→
lim
←−
AX;12 (D1 ) AX;12 (D1 ). (D.1)
D1 D2 : D2 ≤D1

Furthermore, following [3], introduce the adelic space at infinity by

 
A∞ ,


X := AXF ⊗Q R := lim lim A X ;1 (D1 ) A X ;1 (D1 ) ⊗ R (D.2)
b F F Q
−→ ←−
D1 D2 :D2 ≤D1
and define
M
Aar X.
A∞
ar fin
X := AX;012 := AX (D.3)
Then, by §B.3.1, Aar
X,012 admits three level two subspaces X,01 ,
Aar X,02 ,
Aar Aar
X,12 (D),
which can be described by
X,01 = ( fC,x ) × ( fP ) ∈ AX ( fC,x )C,x = ( fC )C,x ∈ AX,01 , fE P = fP ∀P ∈ XF ,
Aar
 ar fin
M M
X,02 = AX,02
Aar ⊗Q R .
fin
⊗Q R, Aar fin

k(XF )b X,12 (D) := AX,12 (D) AXF ,1 (DF )b
Here, as usual, within the adelic space AXF ,01 for the curve XF /F, we have
AXF ,1 (DF ) := ( fP ) ∈ AXF ,01 ordP ( fP ) + ordP (DF ) ≥ 0 ∀P ∈ XF

(D.4)
and we, moreover, define
 
⊗Q R :=
AXF ,1 (DF ) b lim
←−
AXF ,1 (DF )/AXF ,1 (E F ) ⊗Q R , (D.5)
E F :E F ≤DF
using the natural diagonal embeddings k(X) ,→ AXF ,→ Aar
X . Similarly, we have
three level one canonical subspaces
X,0 := AX,01 ∩ AX,02 , AX,1 (D) := AX,01 ∩ AX,12 (D), AX,2 (D) := AX,02 ∩ AX,12 (D).
Aar ar ar ar ar ar ar ar ar

Accordingly, by (B.30), there are natural ind-pro structures on Aar

X,012
Aar = ar  ar
X lim
−→ ←−
lim A X,12 (D) A X,12 (E), (D.6)
D E: E≤D
and, by Corollary B.1 on Aar ar
X,01 and AX,02 ,

X,01 = lim
Aar lim Aar
 ar
−→ ←− X,1 (D) AX,1 (E),
D E: E≤D
(D.7)
Aar = lim lim Aar
 ar
X,02 −→ ←− X,2 (D) AX,2 (E).
D E: E≤D

Based on these genuine ind-pro structures on arithmetic adelic spaces, we may

introduce natural ind-pro topologies on these spaces, as what we do for locally
compact spaces. With topologies hence defined, following §C.1.2 and §C.1.3, par-
ticularly, Proposition C.4, we know that natural inclusions of the above canonical
level two and hence level one subspaces to the total space are all continuous with
closed images. Consequently, these subspaces are also Hausdorff, since Aar X,012 is
Hausdorff by Theorem C.1. For later use, we summarize this as
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 493

1 for Arithmetic Surfaces

Finiteness of Har 493

Proposition D.1 (Proposition C.4). Let D be a Weil divisor on X. Its canonical

01 , A02 , A12 (D), A0 , A1 (D), A2 (D) are closed and Hausdorff.
adelic spaces Aar ar ar ar ar ar

Following §B.4.1, we recall the following:

Definition D.1. Let D be a Weil divisor on X. The arithmetic cohomology groups

Hari (X, D) of D on X (i = 0, 1, 2) are defined by
Har0 (X, D) = Aar ar ar
X,01 ∩ AX,02 ∩ AX,12 (D),
 . 
Har1 (X, D) = Aar X,01 + AX,02 ∩ AX,12 (D)
ar
ar
X,01 ∩ AX,12 (D) + AX,02 ∩ AX,12 (D) ,
Aar ar ar ar

. 
Har2 (X, D) = Aar
X,012 AX,01 + AX,02 + AX,12 (D) .
ar ar ar

In the sequel, to simplify notations, when write Aar

X,∗ , we will omit X. For
example, we write Aar
X,012 simply as Aar
012 .

D.2 Closeness of Adelic Sub-Quotient Spaces

Lemma D.1. Let E be a Weil divisor on X. Then

(1) Aar ar

01 A1(E) is discrete;
(2) A12 (E) A1 (E)ar is compact.
ar

Proof. (1) Using inductive limit, we have

01 A1 (E) = lim
Aar ar
Aar ar
 
−→ 1 (D) A1 (E)
D: D≥E

where D runs over all Weil divisors on X such that D − E is effective. Note that
if D = E + C for an integral curve on X, either horizontal or vertical, we have
1 (D) A1 (E) ' k(C) = AC,0 , which
Aar
 ar ar ar
is discrete (say, in AC,01 ). Hence, for any
ar  ar
D ≥ E, the quotient space A1 (D) A1 (E) is discrete as there exist finitely many
Ci ’s such that D − E = i Ci . To complete proof, recall that topologies on our
P
adelic spaces are induced from the ind-pro one, that is, the strongest topology
making all inductive limits continuous. Consequently, as an inductive limit of
discrete spaces, Aar ar

01 A1 (E) is discrete.
(2) From exact sequence
 ar 0
Aar
12 (E) A12 (E )
A1 (E) A1 (E ) → A12 (E) A12 (E )  ar  ar 0 ,
ar  ar 0 ar  ar 0
(D.8)
A1 (E) A1 (E )
by taking projective limit on E 0 , we obtain an exact sequence
Aar ar ar  ar
1 (E) → A12 (E)  A12 (E) A1 (E). (D.9)
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 494

494 Five Essays on Arithmetic Cohomology


Aar (E) Aar (E 0 ) Aar
12 (E)
(One may also see this using first 12ar  12ar 0 ' Aar (E)+A ar 0 and then
A1 (E) A1 (E ) 1 12 (E )
ar
A12 (E)
12 (E) A1 (E) = lim
Aar
 ar ar  ar
ar ar 0 .) Hence to prove A12 (E) A1 (E) is compact,
←− A1 (E)+A12 (E )
0 0
E :E ≤E 
Aar (E) Aar (E 0 )
being the projective limit, it suffices to show that the quotient spaces 12ar  12ar 0
A1 (E) A1 (E )
are compact. On the other hand, as above, for any integral curve C on X, there is
a canonical exact sequence
ar ar ar  ar
0 → AC,0 → AC,01 → AC,01 AC,0 → 0 (D.10)

ar  ar Aar (E) Aar (E−C)
with AC,01 AC,0 compact. That is to say, 12ar  ar12 is compact, since exact
A1 (E) A1 (E−C)

 ar 0(D.8) and (D.10) are equivalent when E = E + C. Consequently,

0
sequences
Aar (E) (E )
12
 12ar 0 are compact for any Weil divisors E ≥ E 0 , by first writing E − E 0 =
A
Aar
1 (E) A1 (E )
P
i C i ≥ 0, then arguing in the same way as in proof of (1) above. 

Proposition D.2. Let E be a Weil divisor on X. Then, as subspaces of Aar

012 ,

01 (E) + A12 (E) is closed;

(1) Aar ar

(2) A0 + A1 (E) is closed;

ar ar

1 (E) + A2 (E) is closed.

(3) Aar ar

Proof. (1) Recall that A1 (E)ar := Aar ar

01 ∩ A12 (E), and by Lemma D.1(2), the
ar  ar
quotient A12 (E) A1 (E) is compact. Hence, its topological dual is discrete and
Aar +Aar ((ω)−E) Aar +Aar ((ω)−E)
given by 01Aar ((ω)−E) 12
. As a subspace, 01Aar ((ω)−E) 12
is then discrete. Since
12 12
ar
A12 ((ω) − E) is closed, and all quotient spaces involved here are Hausdorff,
01 + A12 ((ω) − E) is closed. So is A01 + A12 (E), since E is arbitrary above.
Aar ar ar ar

(2) Similarly, by Lemma D.1(1), the quotient Aar

 ar
01 A1 (E) is discrete. Hence
01 A0 + A1 (E) is discrete. Clearly, it is also Hausdorff. Thus with the fact
Aar
 ar ar

that A01 is closed, we have Aar 0 + A1 (E) is closed.
ar ar

(3) By definition, Har0 (X, OX (E)) = Aar 01 ∩ A02 ∩ A12 (E) = A01 ∩ A12 (E) ∩
ar ar ar ar

02 ∩ A12 (E) = A1 (E) ∩ A2 (E). Hence

Aar ar
ar ar
there exists a natural surjection
A2 (E) Har (X, OX (E))  A1 (E) + Aar
ar 0 ar
 ar
2 (E) A 1 (E).
By Lemma D.2 below,
Aar +
 0 ar
2 (E) H ar (X, O X (E)) is compact. So is A 1 (E) Aar ar
2 (E) A1 (E). It is also Haus-
dorff, since A1 (E) is closed and A012 is Hausdorff. Therefore, Aar 1 (E) + A2 (E) is
ar ar ar

closed. 

Lemma D.2. For a Weil divisor E on X, Aar

 0
2 (E) Har (X, OX (E)) is compact.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 495

1 for Arithmetic Surfaces

Finiteness of Har 495

Proof. This can be deduced from the natural exact sequence

0 (X, O (E)) + Afin (E) Aar Aar
Har X 2 2 (E) 2 (E)
0→ −→ −→ → 0.
0 (X, O (E))
Har X
0
Har (X, OX (E)) Har (X, OX (E)) + Afin
0
2 (E)
0
Har (X,OX (E))+Afin
2 (E)
Indeed, 0
Har (X,OX (E))
is nothing but Afin
2 (E), since, by definition,
Afin (E)
2 (E) = {0}. Moreover, A2 (E) = lim
Har0 (X, OX (E)) ∩ Afin fin 2
←− Afin (E 0 )
. Thus, by the fact
2
E
Afin (E) fin
that, for an integral curve C on X, Afin2(E−C) ' AC,1 (E|C ) is compact, we conclude
2
Afin (E) H 0 (X,OX (E))+Afin 2 (E)
that Afin2 (E 0 ) and hence ar H 0 (X,O are compact as well. On the other hand, by
2 ar X (E))

definition, AX,2 (E) = AX,2 (E) + H (X, OX (E)) ⊗Z R and Har0 (X, OX (E)) + Afin
ar fin 0
2 (E) =
ar fin 0
(E) A (E)+H (X,O X (E))⊗Z R
H 0 (X, OX (E)) + Afin = X,2Afin (E)+H 0 (X,O (E))
A 2
2 (E). Hence, we have Har 0
(X,OX (E))+Afin
2 (E) 2 X
0
is naturally isomorphic to H H(X,O X (E))⊗Z R
0 (X,O (E))
X
, a compact torus. 

Theorem D.1. For a Weil divisor D on X, Har1 (X, OX (D)) is finite.

01 ∩(A02 +A12 (D))

Aar ar ar
Proof. By proof of Proposition D.2(2), Aar +A ar
(D) = Har1 (X, OX (D)), is dis-
0 1
Aar
12 (D)
crete. On the other hand, is compact, since, by Lemma D.1(2),
A1 (D)ar +A2 (D)ar
Aar ar
D.2(3), resp. Proposition 1, A1 (D)ar +

12 (D) A1 (D) is compact. By Proposition
Aar
12 (D)
A2 (D)ar , resp. Aar
12 (D), is closed. So is Hausdorff, since Aar
A1 (D)ar +A2 (D)ar 012 is Haus-


01 +A12 (D)
ar ar ar
(D)∩
+
A A
dorff. By Proposition D.2(1), Aar 01 A ar
12 (D) is closed. Hence 12
A1 (D) +A2 (D)
ar ar is
closed and hence compact. But this later sub-quotient is another expression of
Har1 (X, OX (D)). Hence, Har1 (X, OX (D)) is both compact and discrete. Therefore, it
is finite. 

0 + A2 (E) is closed.
Corollary D.1. For a Weil divisor E on X, Aar ar

02 ∩ A01 +A12 (E)
Aar ar ar
1
Proof. Note that H (X, OX (E)) can also be written as Aar +A 2 (E) ar . By Propo-
0

01 + A12 (E) is closed. Hence, the numerator is also closed. But Har
sition D.2(1), Aar ar 1

0 + A2 (E) is closed. 
is finite and Hausdorff, hence the denominator, namely, Aar ar

Lemma D.3. Let E be a Weil divisor on X. Then we have

0 + A12 (E) is closed;

(1) Aar ar

1 (E) + A02 is closed;

(2) Aar ar

(3) A2 (E) + Aar

ar
01 is closed.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 496

496 Five Essays on Arithmetic Cohomology

Proof. (1) By Lemma D.1(2), Aar

 ar
ar ar
12 (E) A1 (E) is compact. With the natural surjec-
A12 (E)+A0
tion Aar
 ar
12 (E) A1 (E)  Aar 1 (E)+A0
ar , the later quotient space is also compact. Clearly,

it is also Hausdorff. By Proposition D.2(2), the denominator Aar ar

1 (E)+A0 is closed.
Hence the numerator A0 + A12 (E) is closed as well.
ar ar

ar 0
Aar ar
1 (D)+A02 ∩A12 (D )
(2) Since Aar 1 (D)+A ar
02 = lim lim
ar , it is sufficient for us to show
−→ ←− Aar ar
1 (D)+A02 ∩A12 (E)

D0 E:E≤D0
0
Aar (D)+Aar ar
02 ∩A12 (D )
1 (D) + A02 ∩ A12 (E) is closed.
that 1ar ar

ar , or more strongly, Aar ar
ar
A1 (D)+A02 ∩A12 (E)
We first assume that E ≥ D. Then, Aar 1 (D) + A02 ∩ A12 (E) =
ar
ar

Aar1 (D) + A2 (E).

ar
By Lemma D.2, Aar 0

2 (E) Har (X, OX (E)) is compact. This
1 Aar (D)+Aar (E)
2
implies that Aar (D)+H 0 is compact, since there is a natural surjection
1 ar (X,OX (E))
ar ar
A1 (D)+A2 (E)
Aar 0

2 (E) Har (X, OX (E))  Aar (D)+H 0 (X,O (E)) .
1 ar
Thus, it suffices to show that
X

1 (D) + Har (X, OX (E)) is closed. Now, by Lemma

Aar 0
D.1(2), the quotient space
01 +A12 (D)
Aar ar
ar  ar
A12 (D) A1 (D) is compact. Hence, ar ar

is compact, from
A01 +A12 (D) ∩ Aar 01 +A02 +A12 (E)
ar ar

A01 +A12 (D)

ar ar
surjection Aar
 ar
12 (D) A1 (D) 

ar ar ar
. Furthermore, with the
ar ar
A01 +A12 (D) ∩ A01 +A02 +A12 (E)
closeness of the spaces Aar 01 +A12 (D), resp. A01 +A02 +A12 (E), as proved in Propo-
ar ar ar ar

sition D.2(1), resp. Proposition D.3(3) below, whose proof is independent of this
Aar +Aar (D)
argument, we conclude that this latest quotient space ar ar
01 12ar ar ar

A01 +A12 (D) ∩ A01 +A02 +A12 (E)
is nothing but the topological dual of Aar (D) + H 0 (X, OX (E)) Aar


1 1 (D). So
Aar + 0 ar ar
+ 0


1 (D) H (X, O X (E)) A 1 (D) is discrete. Therefore, A 1 (D) H (X, OX (E))
is closed, as required.
To complete the proof, we next treat general E. Fix a Weil divisor D0 on X
such that D0 ≥ D, D0 ≥ E. By the special case just proved, we have Aar 1 (D) +
Aar ar 0 ar
+ ar
ar
= ar
+ ar



02 ∩ A 12 (D ) is closed. Since A 1 (D) A 02 ∩ A12 (E) A 1 (D) A 02 ∩
12 (D ) ∩ A12 (E), A1 (D) + A02 ∩ A12 (E) is closed.
0

Aar ar ar ar
ar

(3) By Lemma D.2, Aar (E) Har0 (X, OX (E)) is compact. With the natural surjection

2
2 (E) Har (X, OX (E))  A2 (E) + A01 A01 , this latest space is also compact.
Aar
 0 ar ar
 ar

Clearly, it is Hausdorff as well. Consequently, with denominator being closed, the

numerator Aar 2 (E) + A01 is closed.
ar


Proposition D.3. Let E be a Weil divisor on X. Then we have

01 + A02 is closed;
Aar ar
(1)
(2) A02 + Aar
ar
12 (E) is closed;
(3) A01 + Aar
ar
02 + A12 (E) is closed;
ar

(4) A0 + A1 (E) + Aar

ar ar
2 (E) is closed.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 497

1 for Arithmetic Surfaces

Finiteness of Har 497

Aar (E)∩ Aar +Aar
Proof. (1) By Theorem D.1, H 1 (X, OX (E)) = A121 (E)ar +A012 (E)02ar , is finite. Hence, the
numerator Aar 12 (E) ∩ A01 + A12 (E) is closed since the denominator A1 (E) +
ar ar
ar

A2 (E)ar is closed by Proposition D.2(3). Consequently, for any two Weil di-
visors D, E with D ≥ E, viewing as subspaces of Aar
 ar

12 (D) A12 (E), the space
12 (D)∩ A01 +A02
Aar ar ar

is closed. On the other hand, by definition,
A12 (E)∩ A01 +A02
ar ar ar

12 (D) ∩ A01 + A02

Aar ar ar

Aar + Aar = lim lim
.
12 (E) ∩ A01 + A02
01 02 −→D ←−E:E≤D Aar ar ar

Therefore, by definition of ind-pro topology, the subspace Aar 01 + A02 is closed.

ar

(2) By Lemma D.1(2), Aar ar


12 (E) A1 (E) is compact. With natural surjection
ar  ar A ar
+Aar
(E)
A12 (E) A1 (E)  Aar +Aar (E) , the latest space is compact as well. Thus as above,
02 12
02 1
since, by Lemma D.3(2), the denominator is closed, the numerator Aar 02 + A12 (E)
ar

is closed.
(3) By Lemma D.1(2), Aar (E) Aar

12 1 (E) is compact. With natural surjection
01 +A02 +A12 (E)
Aar ar ar
A12 (E) A1 (E)  Aar +Aar +Aar (E) , the latest space is compact as well. But Aar
ar  ar
01 02 1
01 ⊃
Aar
1 (E), hence the denominator is simply A ar
01 + A ar
02 . By (1), it is closed. Conse-
quently, as above, the numerator Aar 01 + A02 + A12 (E) is closed.
ar ar

(4) By Lemma D.2, Aar

 0
2 (E) Har (X, OX (E)) is compact. By definition, we have
Har0 (X, OX (E)) = Aar ar ar
0 ∩ A1 (E) ∩ arA2 (E). Hence, arfrom composition of natural
ar
A2 (E) A2 (E) A0 +Aar
1 +A2 (E)
ar
surjections 0 
ar  Aar +Aar (E) , we conclude that
Har (X,OX (E)) A0 +A1 (E) ∩A2 (E))
ar ar 0 1

0 +A1 (E)+A2 (E)

Aar ar ar

0 + A1 (E) is closed. Conse-

Aar ar
A0 +A1 (E)
ar ar is compact. By Proposition D.2(2),
quently, Aar 0 + A ar
1 (E) + Aar
2 (E) is closed as well. 

Theorem D.2. Let D be a Weil divisor on X. Then we have

(1) Aar 0
0 is discrete. In particular, Har (X, OX (D)) is discrete;
2
(2) H (X, OX (D)) is compact.

Proof. (1) By Lemma D.1(2), Aar (E) Aar


12
 1 (E) is compact. Taking its topolog-
ical dual, we have Aar 01 + Aar
12 (E) Aar
12 (E) is discrete, since Aar 1 (E) = A01 ∩
ar

Aar Proposition D.2(1), Aar 01 + A12 (E) is closed.

ar
12 (E) and, by Consequently,
A0 + A12 (E) A12 (E) is discrete, since we have a natural injection Aar
ar ar
0 +

 ar
Aar ,→ +

 ar ar ar
 ar ar  0
12 (E) A12 (E) A 01 A 12 (E) A 12 (E). In particular, A0 H (X, O X (E)) is
discrete since, by the second isomorphism theorem, there is a natural isomor-
phism Aar 0 + A12 (E) A12 (E) ' A0 H (X, OX (E)). By definition, H (X, OX (E)) ⊆
ar
 ar ar  0 0

H (XF , OXF (E F )), where OXF (E F ) denotes the restriction of OX (E) to the generic
0
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 498

498 Five Essays on Arithmetic Cohomology

fiber XF of arithmetic surface X → Spec OK . Hence, we may choose E with neg-

ative enough OXF (E F ) so as to get a vanishing H 0 (X, OX (E)). This then implies
that Aar 0
0 and hence also H (X, OX (D)) are discrete.
(2) Taking topological dual, we see that Aar012 A01 + A02 is compact, since, by
 ar ar

01 + A02 is closed. This then

Proposition D.3(1), Aar ar
implies that H 2 (X, OX (D)) is
compact, since we have a natural surjection A012 A01 +Aar
ar  ar
 Aar A01 +Aar02 +

 ar
02 012
A12 (D) = H (X, OX (D)).
ar 2




D.3 Topological Duality of Arithmetic Cohomology Groups

Now we are ready to prove the duality theorem.

Theorem D.3 (Theorem B.1). Let X be an arithmetic surface with a canonical

divisor KX and D be a Weil divisor on X. There are the following topological
dualities
Hari (X,
[ OX (D)) ' Har2−i (X, OX (KX − D)) i = 0, 1, 2. (D.11)

Recall that there is a natural global residue pairing on Aar

X,012 , introduced using
local residue theory. Indeed, by §B.2.1, for a fixed non-zero rational differential
ω on X, we can define a global pairing on Aar X,012 with respect to ω by

h·, ·iω : Aar ar

X,012 × AX,012 −→ S1
( fC,x , fP,σ ), (gC,x , gP,σ ) 7→ C⊆X,x∈C: π(x)∈S fin ResC,x ( fC,x gC,x ω)

P
+ P∈XF σ∈S ∞ ResP,σ ( fP,σ gP,σ ω).
P P

Here ResC,x , resp. ResP,σ , denotes the local pairing on X, resp. on Xσ . Moreover,
we have

Proposition D.4. Let X be an arithmetic surface, D a Weil divisor and ω a non-

zero rational differential on X. Then we have

(1) (Lemma B.1) The pairing h·, ·iω is well defined;

(2) (Proposition B.1) The pairing h·, ·iω is non- degenerate;
(3) (Proposition B.2) With respect to h·, ·iω ,

⊥ ar
⊥ ar
⊥ ar
X,01 = AX,01 ,
Aar X,02 = AX,02 ,
Aar X,12 (D) = AX,12 (ω) − D .
Aar

To apply this result, we use the following well-known

Lemma D.4. With respect to a continue, non-degenerate pairing h·, ·i on a topo-

logical space W, e.g. Aar
X,012 , we have
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 499

1 for Arithmetic Surfaces

Finiteness of Har 499

(1) If W1 and W2 are closed subgroups of W such that the spaces W1 + W2 and
W1⊥ + W2⊥ are closed, then
(W1 + W2 )⊥ = W1⊥ ∩ W2⊥ and (W1 ∩ W2 )⊥ = W1⊥ + W2⊥ ;
(2) If W is a closed subgroup of W, then, algebraically and topologically,
(W ⊥ )⊥ = W W .
[  ⊥
and W' W

With above preparation, we are ready to prove our theorem.

Proof. (A) Topological duality between Har0 and Har2

By Definition D.1 for Har2 , Lemma D.4, and Proposition D.4(3),
 ⊥
(ω) − D) ' Aar X,01 + AX,02 + AX,12 ((ω) − D)
ar ar
Har2 (X,[
 ⊥  ⊥  ⊥
' Aar X,01 ∩ AarX,02 ∩ Aar
X,12 ((ω) − D)

= Aar ar ar 0
X,01 ∩ AX,02 ∩ AX,12 (D) ' Har (D).

Indeed, by Proposition D.3(1), AarX,01 + AX,02 is closed, and by Proposition D.3(3),

ar

AX,01 + AX,02 + AX,12 ((ω) − D) is closed.

ar ar ar

(B) Topological duality among Har1

For Har1 , by Proposition B.3(2), there are the following group theoretic
isomorphisms
Har1 (X, D)

. ar 
' AarX,01 ∩ AX,02 + AX,12 (D) X,02 + AX,01 ∩ AX,12 (D)
ar ar
AX,01 ∩ Aar ar ar


. ar 
' AarX,02 ∩ AX,01 + AX,12 (D) X,02 + AX,02 ∩ AX,12 (D) .
ar ar
AX,01 ∩ Aar ar ar

Consequently,
X,02 ∩ AX,01 + AX,12 ((ω) − D)
Aar
 ar ar

c
Har1 (X,[
(ω) − D) =
Aar∩ Aar X,02 + AX,02 ∩ AX,12 ((ω) − D)
ar ar
X,01
⊥
⊥
Aar∩ Aar ∩ Aar ar


X,01 X,02 X,02 ∩ AX,12 ((ω) − D)
'
⊥
⊥
Aar
X,02 + Aar X,01 + AX,12 ((ω) − D)
ar

X,01 + AX,02 ∩ AX,02 + AX,12 (D)

Aar ar
ar ar

=
X,02 + AX,01 ∩ AX,12 (D)
Aar ar ar

X,01 + AX,02 ∩ AX,12 (D)

Aar ar
ar
' ' Har1 (X, D).
Aar
X,01 ∩ A ar
X,12 (D) + A ar
X,02 ∩ A ar
X,12 (D)
Indeed, the first equality and the last isomorphism are direct consequence of the
definition. To verify the second isomorphism and the third equality, we use Propo-
sition D.3(1) and (2), namely, AarX,01 + AX,02 and AX,02 + AX,12 (D) are closed, and
ar ar ar
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 500

500 Five Essays on Arithmetic Cohomology

Proposition D.3(3), namely, Aar X,01 + AX,02 + AX,12 ((ω) − D) is closed. As for
ar ar

the fourth morphisms, both associated quotients spaces are Hausdorff, since de-
nominators and numerators are all closed and Aar X,012 is Hausdorff. They are also
discrete since Har1 is always so. Therefore, this fourth morphism is a topological
homeomorphism, being a group isomorphism. 
January 26, 2018 9:4 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 501

Appendix E

Central Extensions and Reciprocity Laws

for Arithmetic Surfaces

About 50 years ago, Tate [114] developed a theory of residues for curves using
traces and adelic cohomologies. Tate’s work was integrated with the K2 cen-
tral extensions by Arbarello-De Concini-Kac [2]. Reciprocity laws for algebraic
surfaces were established by Parshin in [92]. Later, these reciprocity laws were
reproved by Osipov [85], based on Kapranov’s dimension theory [50]. Osipov
constructs dimension two central extensions and hence establishes the reciprocity
law for algebraic surfaces using Parshin’s adelic theory [91]. More recently, a
categorical proof of Parshin’s reciprocity law was found by Osipov-Zhu [88]. In
essence, Osipov’s construction may be viewed as K2 type theory of central exten-
sions developed by Brylinski-Deligne [19].
However, to establish reciprocity laws for arithmetic surfaces, algebraic K2
theory of central extensions is not sufficient: while it works for reciprocity laws
around points and along vertical curves, it fails when we treat horizontal curves.
To remedy this, instead, we first develop a new theory of arithmetic central ex-
tensions, based on the fact that for exact sequence of metrized R-vector spaces of
finite dimensional
V∗ : 0 → V 1 → V 2 → V 3 → 0,
there is a volume discrepancy γ(V ∗ ). Accordingly, for an R-vector space V, not
necessary to be finite dimensional, a subspace A, and commensurable subspaces
B, C, following [2], we have the group
GL(V, A) := {g ∈ Aut(V) : A ∼ gA},
the R-line
(A|B) := λ(A/A ∩ B)∗ ⊗ λ(B/A ∩ B),
and the natural contraction isomorphism
α := αA,B,C : (A|B) ⊗ (B|C) ' (A|C).

501
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 502

502 Five Essays on Arithmetic Cohomology

Note that A/A ∩ B and B/A ∩ B are finite dimensional R-spaces, we hence can
introduce metrics on them and hence obtaining the metrized R-line (A|B). In
general, the contraction map αA,B,C of (4.3) of [2] does not give an isometry.
We denote the corresponding discrepancy by γ(αA,B,C ) and hence the isometry
αA,B,C := αA,B,C · γ(αA,B,C ) so that we obtain a canonical isometry
α := αA,B,C : (A|B) ⊗ (B|C)  (A|C).
In parallel, for a pair of commensurable (metrized) subspaces A, B and A0 , B0 , the
natural isomorphism β introduced in §4.4) of [2] induced a canonical isometry
βA,B;A0 ,B0 : (A|B) ⊗ (A0 |B0 ) −→ (A ∩ A0 |B ∩ B0 ) ⊗ (A + A0 |B + B0 ).
The up-shot of this consideration then leads to the following construction of arith-
c Ar of GL is given by
metic central extension GL
c Ar := (g, a) : g ∈ GL, a ∈ (A|gA), a , 0
n o
GL
together with the multiplication
(g, a) ◦ (g0 , a0 ) := (gg0 , a ◦ g(a0 ))
where
ab := a ◦ b := αA,B,C (a ⊗ b) ∈ (A|C) ∀a ∈ (A|B), b ∈ (B|C).
Our first result is the following analogue of

Proposition I

c Ar , ◦, (e, 1)
forms a group;

(1) GL
(2) There is a canonical central extension of groups
ι Ar π
1 → R∗ −→ GL
c −→ GL → e.
c Ar , their commutator is given by
Consequently, for elements (g, a), (h, b) ∈ GL
[(g, a), (h, b)] = [g, h], a ◦ g(b) ◦ (ghg−1 )(a−1 ) ◦ [g, h](b−1 ).
In particular, from (2) of the above proposition, we have, if g and h commute,
hg, hi := hg, hiA := a ◦ g(b) ◦ ghg−1 (a−1 ) ◦ [g, h](b−1 ) ∈ R∗ .
We will show that hg, hi is well-defined. Moreover, we have the following

Proposition II (Arithmetic Reciprocity Law in Dimension One) Let A, B be

two subspaces of V. For g, h ∈ GL ∩ GL(V, B), we have
β hg, hiA ⊗ hg, hiB = hg, hiA∩B ⊗ hg, hiA+B .


November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 503

Central Extensions and Reciprocity Laws 503

In particular, if g and g commutes, we have

hg, hiA hg, hiB = hg, hiA∩B hg, hiA+B .

As a special case, if we assume that the metric system is rigid, that is to say, ,
we then recover the constructions of [2].
To go further, we consider the local field R((t)) of Laurent series with R
coefficients. For each finite dimensional sub-quotient space of R((t)), we may
identify it with a standard form Vm,n := ni=m Rti for suitable integer m, n ∈ Z.
P
We assign a metric on Vm,n based on the standard Euclidean metric of R with
(ti , t j ) = δi j , ∀m ≤ i, j ≤ n. Then, for any f (t), g(t) ∈ R((t)), if we write
f (t) = tνt ( f ) f0 (t), g(t) = tνt (g) g0 (t), then f (t) R[[t]] g(t) R[[t]] can be explicitly


calculated. As a direct consequence, we have
| f0 (0)νt (g) |
νR((t)) ( f, g) := logh f, giR[[t] = log .
|g0 (0)νt ( f ) |
The pairing h f, giR[[t] is in fact the reciprocity symbol along horizontal curve
at infinity. To introduce reciprocity symbol at finite places, we start with corre-
sponding 2 dimensional local fields L, kL ((u)), kL {{u}} with kL finite extensions of
Q P . On L, there exists a discrete valuation of rank 2: (ν1 , ν2 ) : L∗ → Z ⊕ Z. With
the help of this, then we can define a reciprocity symbol
ν1 ( f ) ν1 (g)
νL : K2 (L) → Z, ( f, g) 7→ .
ν2 ( f ) ν2 (g)

With all this, we are now ready to state our reciprocity laws. Let π : X →
SpecOF be a regular arithmetic surface defined over a number field F with generic
fiber XF . Let C be an irreducible curve on X and x a closed point of X. As usual,
see e.g., [92], we obtain an Artinian ring KC,x which is a finite direct sum of two-
dimensional local fields Li . Accordingly, we define

νC,x := ⊕i [ki : Fq ] · νLi .

Our main theorem of this appendix is the following:

Theorem (Reciprocity Law for Arithmetic Surfaces)

(1) For a fixed point x ∈ X,

X
νC,x ( f, g) = 0, ∀ f, g ∈ k(X)∗ ,
C: x∈C

where C run over all irreducible curves on X which pass through x;

November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 504

504 Five Essays on Arithmetic Cohomology

(2) For a fixed vertical prime divisor V on X,

X
νV,x ( f, g) = 0, ∀ f, g ∈ k(X)∗ ,
x: x∈V

where x run over all closed points of X which lie on V;

(3) For a fixed Horizontal prime divisor E P on X corresponding an algebraic
point P on XF . For each infinite place σ of F, denote by Pσ, j be corresponding
closed points on XFσ . Then
X XX
νEP ,x ( f, g) · log q x + Nσ · νPσ, j ( f, g) = 0, ∀ f, g ∈ k(X)∗ .
x: x∈E P σ j

Our proof of this theorem is as follows: for the first two, similar as in [84],
we first construct a K2 -central extension of k(X)∗ with the help of Kapranov’s
dimension theory, then we interpret νC,x as a commutator of lifting of elements in
some K2 central extension of the group K(X)∗ , and finally use the splitting of the
central extension of adeles.
During this process, we discover that the fundamental reason for our reci-
procity law along a vertical curve is the Riemann-Roch in dimension one and
the intersection in dimension two. To establish the reciprocity law for horizontal
curves, we need to construct a new arithmetic central extension, using Arakelov
intersection in dimension two and a refined arithmetic Riemann-Roch theorem for
arithmetic curves and our arithmetic adelic theory built up in [106], or the same,
Appendices A and B of this book.

E.1 Algebraic Aspect

E.1.1 Reciprocity Law around Point/Along Vertical Curve

Let K be a number field with OK its integer ring. By an arithmetic surface

π : X → Spec OK , we mean a regular two-dimensional Noetherian scheme X,
together with a proper, flat morphism π. Denote by k(X) its field of rational func-
tions. For an integral curve C on X and a closed point x of C, let Ci be a com-
ponent of a certain normalization of C and xi ∈ Ci a closed point which maps to
x, (i = 1, . . . , m). For each i, using the localization and completion, we obtain a
two-dimensional valuation ring and hence its fractional field KCi ,xi . By the classi-
fications of two-dimensional local field, depending on whether C is horizontal or
vertical, KCi ,xi can be realized as either a finite extension of Q p ((w)), or Q p {{w}}
for some local parameter w. For later use, we set KC,x := ⊕m i=1 Li with Li := KCi ,xi .
By Theorem A.1, KC,p is an Artin ring and uniquely determined by (C, x).
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 505

Central Extensions and Reciprocity Laws 505

For a two-dimensional local field L as above, denote by νc the standard valu-

ation on L determined by the curve C, and L its residue field, either of the form
Fq ((u)) or a p-adic field, depending whether C is vertical or horizontal. Moreover,
denote by νr the standard valuation on the one-dimensional local field L and kL its
residue field. Then we obtain a natural rank two valuation
(ν1 , ν2 ) : L∗−→ Z × Z, f 7→ νr ( f · u−νc ( f ) ), νc ( f ) , (E.1)

and hence a well-defined K2 -symbol

ν1 ( f ) ν1 (g)
νL : K2 (L) → Z, (K, g) 7→ . (E.2)
ν2 ( f ) ν2 (g)
From Proposition 5 of [85], νL does not depend on the choices used, and is bi-
multiplicative and skew-symmetric. Accordingly, if k(x) denotes the finite residue
field of x ∈ X, we define
νC,x := ⊕i kLi : k(x) · νLi .
 
(E.3)

Theorem E.1. Let π : X → Spec OK be an arithmetic surface. Then,

(1) (Reciprocity law around a point) For a closed point x of X,

X
νC,x (K, g) = 0, ∀ f, g ∈ k(X)∗ ,
C: C3x
where C runs over all integral curves on X which contain x.
(2) (Reciprocity law along a vertical curve) For an integral vertical curver C on
X,
X
νC,x (K, g) = 0, ∀ f, g ∈ k(X)∗ ,
x: x∈C
where x run over all closed points of X which lie on C.

For algebraic surfaces, a similar result is proved in [85]. The arguments there
works here since the objects we are treating are algebraic. Instead of a verbatim
copy, we will develop a much refined theory for the reciprocity laws along hor-
izontal curves, where both algebraic and arithmetic structures should be treated
simultaneously.

E.1.2 K2 -Central Extensions

The key is a construction of K2 -central extensions. For the reasons mentioned

above, details will be omitted in this section.
We start with numerations of one-dimensional local fields, which are locally
compact. Let then G be an additive locally compact group. For two open compact
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 506

506 Five Essays on Arithmetic Cohomology

B A
subgroups A and B, we define [A|B]1 := log # A∩B − log # A∩B . One checks eas-
ily that, for open compact subgroups A, B, C, we have [A|B]1 + [B|C]1 = [A|C]1 .
Motivated by this, we call a real valued function num on the set of open com-
pact subgroups of G a numeration if for any two open compact subgroups A, B,
num(B) = num(A) + [A|B]1 . Denote by Num(G) the collection of all numerations
on G. Clearly, for a fixed real number α, if num ∈ Num(G), num + α, defined by
(num + α)(A) = num(A) + α, also belongs to Num(G). Consequently, Num(G)
admits a natural R-torsor structure. Furthermore, for a short exact sequence of
locally compact subgroups 0 → G1 → G2 → G3 → 0, there exists a canonical
isomorphisms of R-torsors Num(G1 ) ⊗R Num(G3 ) ' Num(G2 ). For a proof, see,
e.g. Proposition 2 of [85].
Next, we work with numerations for two-dimensional local fields. We will
use the natural ind-pro structure to build our constructions from that of dimension
one. Let M = i∈I Li with I a certain index set and Li some two-dimensional
Q
local fields. Induced by pro-ind topology, there is a natural product topology
on M. Let W1 and W2 be two closed subspaces of M satisfying the condi-
tion that there exists a closed subspace W ⊆ W1 ∩ W2 such that Wi /W’s are
locally compact. Accordingly, we define an R-torsor by [W1 |W2 : W]2 :=
HomZ (Num(W1 /W), Num(W2 /W)). By Lemmas 4,5 of [85], for Wi , i = 1, 2, 3
and W as above, there is a canonical isomorphisms of R-torsor [W1 |W2 : W]2 ⊗
[W2 |W3 : W]2 ' [W1 |W3 : W]2 . Consequently, using the ind-pro topology, we
define [W1 |W2 ]2 := lim [W1 |W2 : W]2 , an R-torsor. Moreover, by Proposition D.2
←−W
of [85], there is a canonical isomorphism [W1 |W2 ]2 ⊗ [W2 |W3 ]2 ' [W1 |W3 ]2 .
We apply this construction a single two-dimensional local field L appeared
above. For a fixed n, set On := un OL where OL denotes the valuation ring of
L w.r.t. ν2 , and u a local parameter w.r.t. νc . Denote Autc (L) the automorphism
group of L with respect to the ind-pro topology. Similarly, as Corollary 2 of Propo-
sition 1 in [85], we have L∗ ⊆ Autc (L). Hence, if W is a subspace of L such that
On1 ⊆ W ⊂ On2 for n2 , n1 ∈ Z, then, for g ∈ Autc (L), there exists m2 ≤ m1 such
that Om1 ⊆ gW ⊆ Om2 , since g and its inverse are continuous w.r.t. the ind-pro
topology. In particular, for any g1 , g2 , we have an R-torsor [g1 W|g2 W]2 satis-
fying g[g1 W|g2 W]2 ' [gg1 W|gg2 W]2 for g ∈ Autc (L). Consequently, there are
canonical isomorphisms [W|g1 W]2 ⊗ g1 [W|g2 W]2 ' [W|g1 W]2 ⊗ [g1 W|g1 g2 W]2 '
[W|g1 g2 W]2 . For ni ∈ [W|gi W]2 , i = 1, 2, denote by n1 g1 (n2 ) ∈ [W|g1 g2 W]2 the
image of n1 ⊗ (g1 (n2 )).
With numerations in dimension two introduced, we are ready to construct a K2
center extension. By definition, a center extension Aut
dc (L)W of Autc (L) w.r.t W is
given by the following data: set theoretically, Autc (L) = (g, n) | g ∈ Autc (L), n ∈
d 
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 507

Central Extensions and Reciprocity Laws 507

[W|gW]2 ; group theoretically, (g1 , n1 )(g2 , c2 ) := (g1 g2 , n1 g1 (n2 )). Obviously, this
π
is a central extension 0 → R → Aut dc (L) → Autc (L) → 1. Denote the associated
2-cocycle by c. Moreover, by Proposition 4 of [85], if H is a subgroup of Autc (L)
such that HW ⊆ W, the central extension splits over H.
This central extension can be used to introduce a reciprocity symbol. For gi ∈
Autc (L) (i = 1, 2), let b dc (L) such that π(b
gi ∈ Aut gi ) = gi . Set hg1 , g2 iL := [b
g1 ,b
g2 ],
since it is well known that [b g1 ,bg2 ] depends only on g1 , g2 , and can be written as
hg1 , g2 iL = c(g1 , g2 ) − c(g2 , g1 ). (See e.g. Proposition 6 of [85].) The last relation
implies that h·, ·iL is bi-multiplicative and skew-symmetric. Hence if we use the
decomposition L∗ = L∗ × uZ × UL1 with UL1 the special unit group 1 + uOL , we
have, for f, g ∈ k(X)∗ ⊆ Autc (L), h f, gi = [k(p) : k p ] · vL (K, g), where k denotes
the residue field of the prime π(p) ∈ Spec OK and νL is the one introduced in §1.
For details, see e.g. Theorem 1 of [85].
We end this discussion on K2 -central extensions with a global additivity of
reciprocity pairing. As above, let M = i∈I Li with Li two-dimensional local
Q
fields. For each i, we have similarly the subspace Oi,n . Fix a subspace Wi satisfying
Q
Oi,n1 ⊆ Wi ⊆ Oi,n2 and set W := i∈I Wi . On the other hand, let H be a group
admitting embeddings H ⊆ Autc (Li ) for all i. Assume that, for each h ∈ H, there
exists m1 , m2 ∈ Z such that Oi,m1 ⊆ hWi ⊆ Oi,m2 for all i ∈ I, and hence R-torsors
[W|hW]2 for all h ∈ H. Similarly, with this data, we obtain a central extension
0 → R → H bW → H → 1 and hence h·, ·i M on H. Moreover, if I = I1 t I2 ,
Mk := i∈Ik Lk , Wk := i∈Ik Wi , k = 1, 2, and H := k(X)∗ , we have h f, gi M =
Q Q
h f, gi M1 + h f, gi M2 . (See e.g. Proposition 7 of [85].)

E.1.3 Sketch of Proof

We start with (1). Denote by OX,x the (completed) local ring of X at x and
KX,x := Frac OX,x (its fractional field). Clearly, there is a canonical correspon-
dence between prime ideals of OX,x and the branches of (normalizations of)
curves passing through x. For f, g ∈ k(X)∗ , we may and will choose free OX,x -
module Dk ⊆ KX,x (k = 1, 2) such that D1 ⊆ OX,x ⊆ D2 , D1 ⊆ f OX,x ⊆ D2 ,
and D1 ⊆ gOX,x ⊆ D2 . Furthermore, let I be the (finite) set of prime ideals
which are the supports of OX,x -module D2 /D1 and divisors of K and g. Let
M1 := P∈I KP,x , W1 := P∈I OP,x , M2 = P<I KP,x , W2 = P<I OP,x and
Q Q Q Q
M := M1 × M2 , W := W1 × W2 . From this data, we obtain three central ex-
tensions and reciprocity symbols. By additivity, for f, g ∈ k(X)∗ , h f, gi M =
h f, gi M1 + h f, gi M2 . Moreover, since f M2 = M2 and gM2 = M2 are invariant,
h f, gi M2 = 0. Thus, h f, gi M = h f, gi M1 = P∈I h f, giKP,x , by additivity again.
P
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 508

508 Five Essays on Arithmetic Cohomology

Similarly, for (2), let S f,g be set of closed points of C in supports of divisors
of K and g, M1 := x∈S f,g KC,x , W1 := x∈S f,g OC,x , M2 := x∈(CrS f,g ) KC,x , W2 :=
Q Q Q

x∈(CrS f,g ) OC,x , M := M1 × M2 = x∈C KC,x = M1 × M2 , and W = W1 × W2 .

Q Q
Similarly, by additivity, for f, g ∈ k(X)∗ , h f, gi M = x∈S f,g h f, giKC,x .
P

Hence, to prove Theorem 1, it suffices to prove the splitness of central ex-

tensions 0 → R → k(X) [∗ W → k(X)∗ → 1. This can be verified using Parshin-
Beilnson’s adelic cohomology theory, and will be omitted, since details can be
found in §§4, 5 of [85]: arguments there work well here as only algebraic part of
X is involved. Alternatively, the reader can find a proof below when we develop a
much more refined theory to establish reciprocity law along horizontal curves.

E.2 Arithmetic Aspect

E.2.1 Arithmetic Adelic Complex

Let π : X → Spec OK be an arithmetic surface with functional field k(X). Denote

its generic fiber by XF , let S ∞ be the set of infinite places of K, X∞ be the infinite
fibers of π, and let dµ be the fixed Arakelov volume form on X∞ ( [67]). For later
use, set Nσ := [Fσ : R] for σ ∈ S ∞ . Let P be an algebraic point of XF . Denote E P
the corresponding horizontal curve of X, IEP the ideal sheaf of E P ⊆ X, and E P
its Arakelov compactification. Let F be a coherent sheaf on X. We introduce an
arithmetic adelic complex A Ar (F) by AEAr,∗ (F) := lim Ar
(F ⊗ InEP ImEP ),

lim AX,∗
E P ,∗ P−→ ←−
n m: m≥n
Ar
where AX,∗ denotes the canonical arithmetic adelic functor. (For the reader who
is not familiar with arithmetic adelic cohomology, please refer to Appendix A,
particularly, §1.2.3. In fact, for the case we use below, every terms will be given
explicitly.) Since A Ar (F) is defined over an infinitesimal neighborhood of hori-
E P ,∗
zontal curve E P in X, the complex consists of three terms. For example, when L
is an invertible sheaf, a direct calculation implies that A Ar (L) is given by
E P ,∗
Y

⊗Q R −→ AEAr


k(X)EP ×
[ Bx ⊗ObX,x L × (BP ⊗ObX,x L) b
x∈E P P

where Bx := OEP ,x ((u)), BP = OXF P ((u)) with u a parameter defining E P in X, b ⊗R

denotes the ind-pro tensor product of Osipov-Parshin introduced in [87] (see also
Appendix C), and A Ar := 0x∈E k(E P ) x ((u)), consisting of elements a = (a x ) x∈E P ,
Q [
E P P
where a x = i ai,x u satisfying that, for any fixed i, (aix ) x∈E P is an adele of the
i
P

arithmetic curve E P . For example, A Ar (OX ) is given by

E P ,∗
Y
Y0
k(X)EP ×
[ OEP ,x ((u)) × OXF P ((u)) b ⊗Q R −→ [
k(E P ) x ((u)).
x∈E P x∈E P
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 509

Central Extensions and Reciprocity Laws 509

Consequently, if we fix a non-zero rational section s of L such that s does not

P
vanish on E P , we may and will write L|EP as OEP ( x n x [x]) and write LF := L|XF
as LF = OXF ( Q∈XF mQ [Q]). Here x n x [x] = div(s|EP ) and Q∈XF mQ [Q] =
P P P
div(s|XF ). Accordingly, we obtain a complex
Y Y Y0
m−n x ((u))×

dE ×
k(X) OXF (Q)⊗mQ P ((u)) b
⊗Q R −→ [
k(E P ) x ((u)).
P x∈E P E P ,x Q∈XF x∈E P

Here mEP ,x denotes the maximal ideal of the local field of E P at x, and we remind
the reader that OXF (P) can be identified with m−1
XF ,P .

E.2.2 Numerations in Terms of Arakelov Intersection

E.2.2.1 Case I: w/o Adjunction Formula

For a metrized line bundle L on X, we first define a canonical numeration of

second term of the above complex. For simplicity, in this paper, a metrized line
bundle L is always assume to be dµ-admissible, i.e. satisfying the condition that
their first Chern form equals deg(L) · dµ.
Write e for exp, let s , 0 be a rational section of L, and set
Y   Y Z
m−n
 
W ar := x
E ,x ((u)) × OXF (Q)⊗mQ P ((u)) b
⊗Q R × e − log kskdµ .
L,s P
X∞
x∈E P Q∈XF
R 
Here, we add unconventional factor e X − log kskdµ for the purpose that, when
∞
numerating W ar later, we can obtain a nice normalization. In particular, for a
L,s
metrized line bundle L1 satisfying L1 ,→ L with a rational section s1 , of L1
which does not vanish along E P , we have
Y −n x +n1,x   Y  Z ksk 
W ar W ar = OXF (Q)⊗mQ −m1Q P ((u)) b dµ .

mE ((u)) × ⊗Q R · e − log
L,s L1 ,s1 P ,x ks1 k1
x∈E P Q∈XF X∞

To numerate these spaces, with numerations for (products of) 2 dimensional

local fields defined in §1, it suffices to treat the spaces OXF (Q)|P ((u)). It consists
of two parts: one for OXF (Q)|P ’s, which we call the coefficients, another for the
power series in u. Obviously, it is rather easy to numerate the u-series part since
we can use ind-pro topology in the same way as in §E.1. So the only difficulty is
to numerate OXF (Q)|P . For this, we use Arakelov intersection (see e.g. [67]).

Definition E.1. We define the canonical additive numerations for the spaces
above by


(1) num0 mEP ,x := − log q x with q x the size of residue field of E P at x;


(2) num0 OXF (Q)|P := g∞ (Q, P), the Arakelov-Green function on X∞ evaluated
at the points of X∞ corresponding to P and Q;
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 510

510 Five Essays on Arithmetic Cohomology

(3) num0 A[[u]] un A[[u]] := n · num0 (A) for n ∈ Z and A locally compact and


numerable;
(4) num0 OXF (Q)k |P [[u]] um OXF (Q)k |P [[u]] := mk · g∞ (P, Q) for k, m ∈ Z.


Let Num W ar and Num W ar

 ar

W be the associated R-torsors.
L,s L1 ,s1 L,s

Proposition E.1. Let L1 , resp. L, be a metrized line bundle on X satisfying L ,→

L1 , and s1 and s01 , resp. s and s0 , be two non-zero rational sections of L1 , resp.
of L, which do not vanish along E P . Then there exist canonical isomorphisms of
R-torsors
ar
ar

Num WLar ,s WL,s .
 ar

 Num WLar ,s0 WL,s
 ar

Num WL,s  Num WL,s 0
and 0
1 1 1 1

Proof. With constructions above, our proofs for two isomorphisms are similar.
Hence we only give details for the first. By definition, it suffices to identify the
canonical numerations num0 for both sides. Since canonical numerations of Lau-
rent series parts for both sides are the same, we are led to identify num0 for the
coefficient part of both sides, i.e. to verify that
 Y   Y Z
m−n
 
num0 x
E ,x × OXF (Q)⊗mQ P
⊗Q R × e
b − log kskdµ
P
x∈E P Q∈XF X∞
 Y −n0x
  Y ⊗m0Q
 Z 
= num0 m E P ,x × OXF (Q) ⊗ R ×e
P Q
b − log ks0 kdµ .
x∈E P Q∈XF X∞

By definition, if we set G∞ (P, Q) = eg∞ (P,Q) , the

 R left hand side is equal to the
logarithm of x∈EP qnxx Q∈XF G∞ (P, div(sF )) · e X − log kskdµ , which, by the
Q Q
∞

Arakelov intersection theory [67], is simply degar (L|E P ), the Arakelov degree of
metrized line bundle L|E P on E P . Similarly, the right hand side is the logarithm of
n0x Q
R 
Q∈XF G ∞ (P, div(sF )) · e X − log ks kdµ , namely, degar (L|E P ).
Q 0 0
x∈E P q x 
∞

Definition E.2. For metrized line bundles Li on X with non-zero rational sec-
tions
 si , s0i (i = 1, 2), which do not vanish along E P , we define a R-torsor
W ar W ar by
L1 ,s1 L2 ,s2 2
   
WLar ,s WLar ,s := lim
←−
HomR Num W ar  ar

L ,s
WL,s
, Num W ar  ar

L ,s
WL,s
.
1 1 2 2 2 1 1 1 2
(L,s)
L,→L1 ,L2

Proposition E.2. There is a natural isomorphism of R-torsors W ar W ar 2 

 
L1 ,s1 L2 ,s2
W 0 W ar 0 2 . In particular, the space W ar W ar 2 is well-defined.
 ar   
L1 ,s1 L2 ,s2 L1 L2
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 511

Central Extensions and Reciprocity Laws 511

Proof. This is a direct consequence


of the following verifications 
= W , Num W ar
 ar
W ar HomR Num W ar
  ar
 ar

W lim W
L1 ,s1 L2 ,s2 2 ←− L1 ,s1 L,s L1 ,s1 L,s
(L,s)
L,→L1 ,L2
  ar

HomR Num W ar W 0 0 , Num W ar
 ar

 lim W 0 0
←− L1 ,s01 L ,s L2 ,s02 L ,s
0
(L ,s0 )
L ,→L1 ,L2
0

(by the second isomorphism of Proposition 2)

= W ar W ar 0 2 .
 
L1 ,s01 L2 ,s2


E.2.2.2 Case II: w/ Adjunction Formula

In §E.2.2.1, the condition that s does not vanish on E P plays a crucial role, even it
does not hold in general. We next develop a general theory using residues.
Let L be a metrized line bundle on X. If s , 0 is a rational section of L, let f s
be a rational function such that s0 := s · f s−1 does not vanish along E P .
To construct W ar , we write s = s0 · uνEP (s) · f0 with f0 a rational func-
L,s
tion, and u a local parameter defining E P in X. Then s0 is a section of the
line bundle L0 := L(−νEP (s) · E P ). By [67], there is a natural metric on L0
obtained from L ⊗ OX (E P )νEP (s) . Here, as usual, OX (E P ) denotes the line bun-
dle OX (E P ) equipped with the Arakelov metric defined by g∞ (P, ·) ( [67]). Our
idea of constructing W ar is to introduce a term-to-be-defined decomposition
L,s

(νEP (s))
W ar = W ar ⊗ W ar ⊗ W ar . Since s0 and f0 do not vanish along
L,s L0 ,s0 OX (E P ),u O X , f0
E P , the spaces W ar and WOX , f0 have already been defined. Therefore, we have
L0 ,s0
to define WOX (E P ),u , or WE P ,u shortly, and justify the decomposition.

To start with, we recall the Arakelov adjunction formula. Let Kπ be the canon-
ical divisor of arithmetic surface π : X → Y := Spac OK . Set E P = Spec A for a
certain order in Frac (A), a finite extension of K and let ω0 , 0 be a rational sec-
tion of Kπ such that ω0 (E P ) . 0. Then, see e.g. Corollary 5.5 at page 99 of [67],
2
globally, K π · E P + E P = dEP /Y + dλ (E P ). Here,

(a) d∞ (E P ) = σ∈S ∞ Nσ dg,σ (E P ) with dg,σ (E P ) := i≤ j gσ (Pi , P j ), where {Pi }

P P
denotes the set of conjugating points on the Riemann surface Xσ , the fiber
of π at σ ∈ S ∞ , corresponding to the algebraic point P of XF and gσ the
σ-component of g∞ ; and
(b) dEP /Y = − log (WEP /Y : OK ), with WEP /R the dualizing module of E P over Y,
defined by WEP /Y := {b ∈ F(P) : Tr(bA) ⊆ OK }. This is a fractional ideal of
F(P). For later use, we write WEP /Y := x∈EP mbExP ,x .
Q
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 512

512 Five Essays on Arithmetic Cohomology

Moreover, by the reside theorem, i.e. Theorem 4.1 of Chapter IV in [67], the
residue map induces a canonical isomorphism res : Kπ (E P )|EP ' WeEP /Y , where
WeEP /Y denotes the sheaf on affine curve E P associated to the A-module WEP /Y . Set
Y b x −ν x (ω0 |E P )
WEfinP ,u,ω0 := u · mE ((u)),
P ,x
x∈E P
Y
⊗−νQ (ω0 |X )
WE∞P ,u,ω0 := u · OXF Q F
P
⊗Q R,
((u))b
Q∈XF
Z
cE P ,ω := log kω0 kdµ − dλ (E P ).
0
X∞

With all this, we then may introduce W ar based on the decomposition

E P ,u
 
WEar ,u,ω = WEfinP ,u,ω0 × WE∞ ,u,ω · e cE P ,,ω0 .
P 0 P 0

Remark E.1. It is too naive to set

Y Y
WEfinP ,u = u · OEP ,x ((u)) × uνP ( f ) OXF |P ((u)).
x∈E P Q∈XF

A bit thought in terms of Arakelov intersection, in particular, the adjunction for-

mula and the associated residue theorem, leads to the use of the formula
WEar ,u ⊗ WK π ,ω0  WKar (E .
P π P ),u ω0

Since, by the residue theorem above, Kπ (E P )|EP = WeEP /Y . This is the reason that
we define WEfinP ,u as above, and, as will be seen below, then gives a nice decompo-
sition for W ar .
E P ,u,ω0

Such defined, we can follow Definition E.1 to introduce canonical numerations

num0 for above data, using the adjunction formula and the residue theorem above
again. Indeed, we have
Z
Y b x −ν x (ω0 |E P ) Y
⊗−νQ (ω0 |X )

num0 mE × OXF Q F ⊗ R·e log kω0 kdµ − dλ (E P )
P ,x P Q
b
x∈E P Q∈XF X∞

= − dE P /Y + K π · E P − dλ (E P ) = K π · E P − K π + E P · E P

= EP · EP.

All in all, we are ready to introduce the following

Definition E.3. Let L be a metrized line bundle on X with s a non-zero rational

section. Let ω0 , 0 be a rational section of Kπ such that ω0 (E P ) . 0. Then with
respect to a decompose s = s0 · uνEP (s) · f0 , we define
ar
:= WEar ,u,ω ⊗νP ( fs ) ⊗ W ar ⊗−νP ( fs ) .


WL,s,ω
0 P 0 L⊗E P , f0 s0
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 513

Central Extensions and Reciprocity Laws 513

From this definition, canonical numeration num0 for the coefficient part is
given by
 ⊗−νP ( f s )

νP ( f s ) E P + c1,ar L ⊗ E P + c1,ar (OX ) · E P = c1,ar (L) · E P = degar L|E P .

Here, as usual, c1,ar denotes the first arithmetic Chern class of a metrized line
bundle. This latest degree is independent of s, s0 , f0 and ω0 . Hence we have
proved the following

Proposition E.3. For non-zero rational sections s and s0 of L, and non-zero ra-
tional sections ω0 and ω00 of Kπ such that ω0 (E P ) . 0, ω00 (E P ) . 0, we have
Num W ar  Num W ar 0 0 . In particular, we may write Num W ar




simply
L,s,ω0 L,s ,ω0 L,s,ω0
as Num W ar

or Num W .
ar

L,s L

Definition E.4. For metrized line bundles Li with non-zero rational sections si , s0i
of Li (i = 1, 2), we define an R-torsor W ar W ar 2 by
 
L1 ,s1 L2 ,s2
  ar

, Num WLar ,s WL,s .
 ar
WL ,s WLar ,s 2 := lim HomR Num WLar ,s WL,s
  ar

1 1 2 2 ←−:(L,s) 1 1 1 2
L,→L1 ,L2

Same proof as that for Proposition D.2 in §E.2.2.1 then proves the following:

Proposition E.4. There is a natural isomorphism of R-torsors W ar W ar 2 '

 
L1 ,s1 L2 ,s2
W 0 W ar 0 2 . In particular, the space W ar W ar 2 is well-defined.
 ar   
L1 ,s1 L2 ,s2 L1 L2

E.3 Numerations for Arithmetic Adelic Cohomologies

With numerations in terms of intersection developed, next we introduce a nu-

meration in terms of one dimensional cohomology theory. As we will see later
reciprocity laws for arithmetic surfaces then can be proved using an arithmetic
Riemann-Roch theorem for curves (see e.g. Theorem 1.5 or [67]).

Proposition E.5. Let L be an invertible sheaf on X. Denote the i-th arithmetic

cohomology group of L on X by Hari E P , L|EP (i = 0, 1). Then cohomology


groups of the complex A Ar (L) is given by
E P ,∗
0
AEAr,∗ (L) =Har0 E P , L|EP ((u)) and H 1 AEAr,∗ (L) = Har1 E P , L|EP ((u)).





H
P P

Proof. This is a direct consequence of the construction of the arithmetic adelic

complex related, a special case treated in Appendix A. Indeed, for the invertible
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 514

514 Five Essays on Arithmetic Cohomology

sheaf L, by definition of the complex A Ar (L), its cohomology is given by

E P ,∗

lim Hari (E P , L ⊗ InEP ImEP ), i = 0, 1.


lim
−→ ←−
n m: m≥n

This then already implies H 0 A Ar (L) = Har0 E P , L|EP ((u)). On the other hand,



E P ,∗
to treat the 1-st cohomology group, note that Parshin-Beilinson’s H 1 (E P , ·) is al-
ways 0, since E P is affine. Hence, Parshin-Beilinson’s H 0 (E P , ·) is exact. This
implies that for horizontal curve E P , the maps

ϕl,n;m : Har1 (E P , L ⊗ InEP ImEP ) −→ Har1 (E P , L ⊗ IlEP ImEP )

 

are all injective. Therefore, H 1 A Ar (L) = Har1 E P , L|EP ((u)).


E P ,∗

Definition E.5. The canonical numeration num0 on cohomologies is defined by

(1) num0 Har0 E P , L|EP := h0ar E P , L|EP , the 0-th arithmetic cohomology for the




metrized invertible sheaf L|EP on the curve E P ;
(2) num0 Har1 E P , L|EP := h1ar E P , L|E P , the 1-st arithmetic cohomology for the

metrized invertible sheaf L|E P on the curve E P ;

(3) num0 A[[u]] un A[[u]] := n · num0 (A) for n ∈ Z and A locally compact and


numerable.

Therefore, by shifting with real constants, we obtain two R-torsors

Num H 0 AEAr,∗ (L) Num H 1 AEAr,∗ (L) .

and
P P

Definition E.6. The R-torsor Num A Ar (L) is defined by

E P ,∗



AEAr,∗ (L) := HomR Num H 1 AEAr,∗ (L) , Num H 0 AEAr,∗ (L) .




Num
P P P

With all this, we are now ready to state one of our main results. We will use it
in our proof of the splitness of an arithmetic centra extension below.

Theorem E.2. Let Li be metrized line bundle on X (i = 1, 2). There is a canonical

isomorphism of R-torsors


WL WL 2  HomR Num AEAr,∗ (L1 ) , Num AEAr,∗ (L2 ) .
 ar ar 

1 2 P P

Proof. It suffices to identify canonical numerations of both sides. For this, first,
by Definitions E.1 and E.5, we conclude that the Laurent series parts for both sides
are numerated in the same way. Therefore, it suffices to treat the corresponding
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 515

Central Extensions and Reciprocity Laws 515

coefficients for both sides. By Definition E.3 and the discussion after it, we know
that num0 for the left hand side gives

degar L1 |EP − degar L2 |EP = degar (L1 − L2 )|EP .

(E.4)

On the other hand, by Definitions E.5 and E.6, num0 for the coefficient of
Num A Ar (L1 ) equals


E P ,∗

h1ar E P , L1 |EP − h0ar E P , L1 |EP = −χar (E P , L1 |EP ).

Here, χar denotes Arakelov-Euler characteristic. Hence, by definition, num0 for

the coefficient part of the right hand equals χar (E P , L1 |EP ) − χar (E P , L2 |EP ), which
by the Arakelov-Riemann-Roch is simply

1  
1 
degar (L1 )|E P − log |∆k(P) | − degar (L2 )|E P − log |∆k(P) | = degar (L1 − L2 )|E P .


2 2
This coincides with the canonical numeration num0 for the coefficient of the space
 ar ar 
W W 2 given in (E.4), obtained using Arakelov intersection. 
L1 L2

E.4 Arithmetic Central Extension

We here construct a central extension of k(X)∗ by R, using the spaces in §E.2.2.

First, we introduce an action of f ∈ k(X)∗ on WOarX =
Q
x∈E P OE P ,x ((u)) ×
Q
O
Q∈XF XF ,Q P | ((u)). Assume, for the time being, that f (E P ) . 0. Write
div( f |EP ) = x n x [x], div( f |XF ) = Q mQ [Q]. Then, it is natural to define
P P
Y Y
f · WOarX := m−n x
E P ,x ((u)) × OXF (Q)mQ |P ((u)) =: WOarX (div( f )) .
x∈E P Q∈XF

In parallel, arithmetically, we define an action of K on W ar by

OX ,1
Y Y Z
m−n

f W ar := x
E ,x ((u)) × OXF (Q)mQ |P ((u)) · e − log k f kdµ .
OX ,1 x∈E P P Q∈XF X∞

This is simply W ar , or W ar shortly, where OX (divar ( f )) denotes

OX (divar ( f )), f OX (divar ( f ))
metrized line bundle for the Arakelov divisor divar ( f ) of K.
Motivated this discussion, for f ∈ k(X)∗ , which may vanish along E P , we
define an action of K on W ar by f · W ar = W ar , and hence an action
OX ,1 OX ,1 OX (divar ( f )), f
of k(X) on W , since divar ( f g) = divar ( f ) + divar (g).
∗ ar
OX ,1

Proposition E.6. For f, g ∈ k(X)∗ , we have natural isomorphism

Π
[W ar : f W ar ]2 ⊗ [ f W ar : f gW ar ]2  [W ar : f gW ar ]2 .
OX OX OX OX OX OX
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 516

516 Five Essays on Arithmetic Cohomology

Proof. We prove this by the following verifications.

[W ar : W ar ]2 ⊗ [W ar : W ar ]2
OX OX (divar ( f )) OX (divar ( f )) OX (divar ( f g))
  ar

= HomR Num W ar W , Num W ar
 ar

lim W
←− OX ,1 L,s OX (divar ( f )), f L,s
(L,s), L,→OX
L,→OX (divar ( f ))
  ar

HomR Num W ar W , Num W ar
 ar

⊗ lim W
←− OX (divar ( f )), f L,s OX (divar ( f g)), f g L,s
(L,s) L,→OX (divar ( f ))
L,→OX (divar ( f g))
   ar

HomR Num W ar W ar , Num W ar


 lim W
←− OX ,1 L,s OX (divar ( f )), f L,s
(L,s) L,→OX ,OX (divar ( f ))
L,→OX (divar ( f g))
  ar

⊗ HomR Num W ar W , Num W ar
 ar

W
OX (divar ( f )), f L,s OX (divar ( f g)), f g L,s
 
W ar W ar , Num W ar

 ar

 lim HomR Num W
←− OX ,1 L,s OX (divar ( f g)), f g L,s
(L,s)
L,→OX ,OX (divar ( f g))

= [W ar : W ar ]2 . 
OX OX (divar ( f g))

In particular, if α ∈ [W ar : f W ar ]2 , β ∈ [W ar : gW ar ]2 , then f (β) ∈ [ f W ar :

OX OX
OX OX OX
f gW ar ]2 , and α ◦ f (β) := Π α ⊗ f (β) ∈ [W ar : f gW ar ]2 .
OX OX OX

With all this, we are ready to introduce the following main definition.

Definition E.7. We define a central extension k(X)∗W ar of k(X)∗ by R with respect

OX
to W ar by the following data:
OX

(a) As a set, k(X)∗W ar := ( f, α) | f ∈ k(X)∗ , α ∈ [W ar : f W ar ]2 ;


O OX
OX

X
(b) For group law, ( f, α) ◦ (g, β) := f g, α ◦ f (β) .

f, b
Furthermore, for f, g ∈ k(X)∗ , let b g be their inverse images in k(X)∗W ar with
OX

f ,b
respect to the natural projection. Then [ b g] = b
f ◦b
g◦ bf −1 ◦ b
g−1 depends only on
f, g, and belongs to the center R since f g = g f . Define a reciprocity symbol by
f ,b
[ f, g]W ar := [ b g].
OX

E.5 Splitness

Theorem E.3. With the same notation as above,

(1) k(X)∗W ar is a central extension of k(X)∗ by R via the exact sequence

OX

0 → R −→ k(X)∗W ar −→ k(X)∗ → 1.
OX
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 517

Central Extensions and Reciprocity Laws 517

(2) The exact sequence (1) splits, and hence the reciprocity symbol degenerates.

Proof. (1) follows directly from the definition. As for (2), we introduce one more
0
central extension k(X)∗W ar of the group k(X)∗ by R as follows:
OX

(a) As a set, its elements are given by pairs ( f, α0 ), where f ∈ k(X)∗ and α0 ∈
HomR Num(A Ar (OX ), Num(A Ar ( f OX )) ;


E P ,∗ E P ,∗
(b) For the group law, ( f, α0 ) ◦ (g, β0 ) := f g, α0 ◦ f (β0 ) .

0
By Theorem E.2, k(X)∗W ar ' k(X)∗W ar . In addition, since h ∈ k(X)∗ takes
OX OX

Num(A Ar (OX )) to Num(A Ar (OX (divar (h)))), and hence the exact sequence ad-
E P ,∗ E P ,∗
mits a natural section induced by the element h ∈ HomR Num(A Ar (OX )),
E P ,∗
0
Num(A Ar (OX (divar (h)))) . Therefore, the central extension k(X)∗W ar splits.



E P ,∗ OX

E.6 Reciprocity Law along Horizontal Curve

Let P be an algebraic point on XF , with E P the corresponding (completed) hor-

izontal curve. For any closed point x ∈ E P , resp. Q ∈ XF , denote by KEP ,x ,
resp. KXF ,Q its local field. Set M ar := x∈EP KEP ,x × Q∈XF KXF ,Q Pb
Q Q
⊗Q R, and
EP
ar Q Q ar
W := x∈EP OEP ,x × Q∈XF OXF ,Q P ⊗Q R, which is simply W . Here, as above,
b
EP OX
⊗ denotes the Osipov-Parshin ind-pro tensor. Moreover, for f, g ∈ k(X)∗ , Let S EP
b
be the supports of div( f |EP ) and div(g|EP ), and let S XF be the supports of div( f |XF )
and div(g|XF ). Finally, set M arf,g := x∈S EP KEP ,x × Q∈S XF KXF ,Q Pb
Q Q
⊗Q R, a factor
of M ar , and let Marf,g be its cofactor in M ar . Similarly, we have W ar f,g
f,g , War . By
EP EP
definition, M ar = M arf,g × Marf,g , W ar = W ar f,g
f,g × War .
EP EP
To go further, we use the central extension k(X)∗W ar of k(X)∗ with respect to
OX
W ar . Because of the splitness proved in Theorem E.3, the reciprocity symbol
OX
[·, ·]W ar on k(X)∗ × k(X)∗ vanishes. Moreover, in a similar way, we can also
OX

introduce central extensions of k(X)∗ with respect to M arf,g and Marf,g to get first
the groups k(X)∗Mar and k(X)∗ f,g and hence the associated reciprocity symbols
f,g Mar
[∗, ∗] Marf,g and [∗, ∗] Marf,g . Based on the techniques in §E.3, it is rather direct to show
[∗, ∗] Mar = [∗, ∗] Marf,g + [∗, ∗] Marf,g . (E.5)
EP

Furthermore, since both f, g keep Marf,g (unchanged), we have for any f, g ∈

k(X)∗ , [ f, g] Marf,g = 0. Indeed, for any factor subspace M of W ar , we can con-
OX
struct its associated central extension k(X)∗M and hence obtaining the reciprocity
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 518

518 Five Essays on Arithmetic Cohomology

pairing [∗, ∗] M such that [∗, ∗] is additive, and, if f, g ∈ k(X)∗ keep M stable, then,
with the proof of Proposition 4(4) of [85], we have [ f, g] M = 0. Consequently,
by (E.5), [ f, g] Marf,g = [ f, g] Marf,g + [ f, g] Marf,g = [ f, g]W ar = 0. Using the additivity
OX
again, we have 0 = x∈EP [ f, g]KEP ,x + Q∈XF [ f, g]KX ,Qb⊗Q R . This then proves
P P
F

Theorem E.4. (Reciprocity Law along Horizontal Curve) Let π : X → Spec OK

be an arithmetic surface. For an algebraic point P on XK , let E P be associated
horizontal curve on X. Then
X X
[ f, g]KEP ,x + ⊗Q R = 0.
[ f, g]KXF ,Q b
x∈E P Q∈XF

We mention that reciprocity symbols above can be written down explicitly.

Indeed, by §E.1, [ f, g]KEP ,x = νEP ,x (K, g) log q x . In addition, for each infinite place
σ of K, if we set Fσ the σ- completion of K, and Pσ, j be closed points on XFσ
corresponding to P,
X X X k f (P j )kνP (g)
[ f, g]KXF ,Q = − [Fσ : R] · log νP ( f )
, ∀ f, g ∈ k(X)∗ . (E.6)
Q∈X σ j
kg(P j )k
F

A proof may be given using discrepancy for exact sequence of metrized local free
sheaves based on §6 of [2]. In fact, an experienced reader should already see this
f (P)νP (g)
from the facts that standard tame symbol on XF is given by (−1)νP ( f )νP (g) g(P) νP ( f )

and that gσ (P, div( f )) + X − log k f kdµ = − log k f (P)k. This formula is also one of
R
σ R 
the reasons whey we add the factor e X − log kskdµ in the definition of W ar .
∞ L,s
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 519

Bibliography

[1] A. Ash, D. Mumford, M. Rapoport, Y. Tai, Smooth compactification of locally sym-

metric varieties, Cambridge University Press, 2010.
[2] E. Arbarello, C. De Concini, and V.G. Kac, The infinite wedge representation and
the reciprocity law for algebraic curves, Proc. of Symp in Pure Math, 49 (1989),
171-190.
[3] J. Arthur, A trace formula for reductive groups. I. Terms associated to classes in
G(Q). Duke Math. J. 45 (1978) 911-952.
[4] J. Arthur, A trace formula for reductive groups. II. Applications of a truncation
operator. Compos. Math. 40 (1980) 87-121.
[5] J. Arthur, A measure on the unipotent variety, Canad. J. Math 37 (1985) 1237-1274.
[6] J. Arthur, An Introduction to the Trace Formula, Harmonic analysis, the trace for-
mula, and Shimura varieties, Clay Math. Proc. 4 (2005) 1-263.
[7] M.F. Atiyah and R. Bott, The Yang-Mille equations over Riemann surfaces, Phil.
Trans. R. Soc. Long. A 308 (1983) 523-615.
[8] W.L. Baily and A. Borel, On the compactification of arithmetically defined quotients
of bounded symmetric domains, Bull. Amer. Math. Soc., 70 (1964) 588-593.
[9] W.L. Baily and A. Borel, Compactification of arithmetic quotients of bounded sym-
metric domains, Ann. Math. 84 (3), 442-528, 1966.
[10] A.A. Beilinson, Residues and adeles, Funct. Anal. Pril. 14 (1980) 44-45, English
transl. in Func. Anal. Appl. 14 (1980), 34-35.
[11] G. Birkhoff, Three Observations on Linear Algebra, Univ. Nac. Tucuman. Rev. Ser.
A 5 (1946) 147-151.
[12] A. Borel, Some finiteness properties of adele groups over number fields, Publ.
Math., IHES, 16 (1963) 5-30.
[13] A. Borel, Introduction aux groupes arithmetictiques, Hermann, 1969.
[14] A. Borel, Linear Algebraic Groups, GTM 126, Springer, 1991.
[15] A. Borel and J. P. Serre, Corners and arithmetic groups, Comment. Math. Helv. 48
(1973) 436-491.
[16] N. Bourbaki, N. Elements de mathematique, groupes et algebras de Lie, Chapters
4-6, Hermann, 1968.
[17] N. Bourbaki, Elements of Mathematics: General Topology, Springer, 1987.
[18] N.G. de Bruijn, The roots of trigonometric integrals, Duke Math. J. 17 (1950) 197-
226.

519
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 520

520 Bibliography

[19] J.-L. Brylinski and P. Deligne, Central extensions of reductive groups by K2, Publ.
Math. Inst. Hautes Etudes Sci. 94 (2001) 5-85.
[20] W. Casselman, Introduction to the Theory of Admissible Representations of p-adic
Reductive Groups, unpublished notes.
[21] B. Conrad, Reductive group schemes, manuscripts, 2011.
[22] B. Diehl, Die analytische Fortsetzung der Eisensteinreihe zur Siegelschen Modul-
gruppe, J. reine angew. Math. 317 (1980) 40-73.
[23] H.M. Edwards, Riemann’s Zeta Function, Dover Pub., 1974.
[24] I.Y. Efrat, The Selberg Trace Formula for PSL2 (R)n , Memoirs of AMS, 359, 1987.
[25] J. Elstrodt, F. Grunewald and J.L. Mennicke, Groups Acting on Hyperbolic Space:
Harmonic Analysis and Number Theory, Springer, 1997.
[26] G.B. Folland, A course in abstract harmonic analysis, Studies in advanced mathe-
matics, CRC Press, 1995.
[27] P. Francini, The size function h0 for quadratic number fields. Journal de Theorie des
Nombres de Bordeaux 13 (2001) 125-135.
[28] R. Friedman and J.W. Morgan, Holomorphic principal bundles over elliptic curves.
II. The parabolic construction. J. Differ. Geom. 56 (2000) 301-379.
[29] A. Fröhlich and M.J. Taylor, Algebraic Number Theory, Cambridge studies in ad-
vanced mathematics 27, Cambridge Univ. Press, 1991.
[30] S.A. Gaal, Point set topology, Pure and Applied Mathematics 16, Academic Press,
1964.
[31] R. Gangolli and V.S. Varadarajan, Harmonic analysis of spherical functions on real
reductive groups, Springer, 1988.
[32] G. van der Geer & R. Schoof, Effectivity of Arakelov Divisors and the Theta Divisor
of a Number Field, Sel. Math., New ser. 6 (2000) 377-398.
[33] D.R. Grayson, Reduction theory using semistability. Comment. Math. Helv. 59
(1984) 600-634.
[34] D.R. Grayson, Reduction theory using semistability. II. Comment. Math. Helv. 61
(1986) 661-676.
[35] R.P. Groenewegen, An arithmetic analogue of Clifford’s Theorem. Journal de The-
orie des Nombres de Bordeaux 13 (2001) 143-156.
[36] SGA3, Seminaire de Geometrie Algebrique du Bois Marie, Schemas en groupes
1962-1964, Lecture Notes in Mathematics 151, 152 and 153, 1970.
[37] S.D. Gupta, On the Rankin-Selberg Method for functions not of rapid decay on
congruence subgroups, J Number Theory 120 (1997) 95-103.
[38] T. Hayashi, Computation of Weng’s rank 2 zeta function over an algebraic number
fields, J. number Theory 125 (2007) 473-527.
[39] D.A. Hejhal, On a result of Selberg concerning zeros of linear combinations of
Lfunctions. Int. Math. Res. Not. 11 (2000) 551-577.
[40] D.A. Hejhal, On the horizontal distribution of zeros of linear combinations of Euler
products. C. R. Math. Acad. Sci. Paris 338 (2004) 755-758.
[41] W. Hoffmann, Geometric estimates for the trace formula, Ann. Global Analysis and
Geometry 34 (2008) 233-261.
[42] A. Horn, Doubly stochastic matrices and the diagonal of a rotation matrix, Amer. J.
Math. 76 (1954) 620-630.
[43] A. Huber, On the Parshin-Beilinson adeles for schemes, Abh. Math. Sem. Univ.
Hamburg, 61 (1991) 249-273.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 521

Bibliography 521

[44] J.E. Humphreys, Introduction to Lie Algebras and Representation Theory, GTM 9
Springer, 1978.
[45] K. Iwasawa, Letter to Dieudonné (April 8, 1952), Advanced Studies in Pure Math.
21 (1992) 445-450.
[46] K. Iwasawa, Lecture notes at Princeton university, notes in Japanese, 1973.
[47] H. Jacquet, E. Lapid, and J. Rogawski, Periods of automorphic forms. J. Am. Math.
Soc. 12 (1999) 173-240.
[48] A. Javanpeykar and D. Loughran, Good reduction of algebraic groups and flag va-
rieties, Archiv Mathematik 104 (2015), 133-143.
[49] V.G. Kac, Infinite-Dimensional Lie Algebras, Cambridge University Press, 1990.
[50] M. Kapranov, Semi-infinite symmetric powers, arXiv:math.QA/0107089.
[51] S. Katayama, Weyl group of E6 and Weng zeta function, Thesis, Kyushu Univ. (in
Japanese) 2016.
[52] K. Kato, Existence theorem for higher local fields, Geom. Top. Monograph. 3,
Geom. Topol. Publ. (2000) 165-195.
[53] V.G. Kazakevich and A.K. Stavrova, Subgroups normalized by the commutator
group of the Levi subgroup, Zapiski Nauchnykh Seminarov POMI 319 (2004) 199-
215.
[54] G. Kempf, F. Knudsen, D. Mumford and B. Saint-Donat, Toroidal Embeddings I,
Lect. Notes in Math. 339, Springer, 1972.
[55] H. Ki, On the zeros of Weng’s zeta functions, IMRN 13 (2010) 2367-2393.
[56] H. Ki, Y. Komori and M. Suzuki, On the zeros of Weng zeta functions for Chevalley
groups, Manuscript Math. 148 (2015) 119-176.
[57] H. Kim and L. Weng, Volume of truncated fundamental domains, Proc. AMS 135
(2007) 1681-1688.
[58] H. Kim and L. Weng, Functional equation for zeta functions of (SLn , Pn−1,1 ),
preprint.
[59] Y. Komori, Functional equations for Weng’s zeta functions for (G, P)/Q, Amer. J.
Math. 135 (2013) 1019-1038.
[60] B. Konstant, On convexity, the Weyl group and the Iwasawa decomposition, Ann.
Sci. Ec. Norm. Super. 6 (1973) 413-455.
[61] M. Kontsevich and Y. Soibelman, Stability structures, motivic Donaldson-Thomas
invariants and cluster transformations, arXiv:0811.2435.
[62] T. Kubota, Elementary theory of Eisenstein series. Kodansha scientific books, Ko-
dansha, 1973.
[63] L. Lafforgue, Chtoucas de Drinfeld et conjecture de Ramanujan-Petersson. Aster-
isque 243, 1997.
[64] J.C. Lagarias, On a positivity property of the Riemann ξ-function, Acta Arithmetic
89 (1999) 217-234.
[65] J.C. Lagarias and M. Suzuki, The Riemann hypothesis for certain integrals of Eisen-
stein series, J. Number Theory 118 (2006) 98-122.
[66] S. Lang, Algebraic Number Theory, GTM 110, Springer, 1986.
[67] S. Lang, Introduction to Arakelov Theory, Springer, 1988.
[68] R. Langlands, the volume of the fundamental domain for some arithmetical sub-
groups of Chevalley groups, Proc. Symp. Pure Math. 9 (1966) 143-148.
[69] R. Langlands, Euler products, Yale Mathematical Monographs, 1. Yale University
Press, 1971.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 522

522 Bibliography

[70] R. Langlands, On the functional equations satisfied by Eisenstein series, Lecture

Notes in Math. 544, Springer, 1976.
[71] G. Laumon and M. Rapoport, The Langlands lemma and the Betti numbers of stacks
of G-bundles on a curve. Internat. J. Math. 7 (1996) 29-45.
[72] A. I. Madunts and I. B. Zhukov, Multidimensional complete fields: topology and
other basic constructions, Amer. Math. Soc. Transl. (Ser. 2) 165 (1995) 1-34.
[73] L. Manivel, The canonical strip phenomenon for complete intersections in homoge-
neous spaces, arXiv:0904.2470.
[74] H. Minkowski, Geometrie der Zahlen, Teubner, 1896.
[75] C. Moeglin and J.-L. Waldspurger, Spectral decomposition and Eisenstein series.
Cambridge Tracts in Math 113, Cambridge Univ Press, 1995.
[76] H.L. Montgomery, The pair correlation of the zeta function, Proc. Symp. Pure Math.
24 (1973) 181-193.
[77] C. Moreno, Algebraic curves over finite fields. Cambridge Tracts in Math. 97, Cam-
bridge University Press, 1991.
[78] M. Morrow, An explicit approach to residues on and dualizing sheaves of arithmetic
surfaces. New York J. Math. 16 (2010) 575-627.
[79] M. Morrow, Grothendieck’s trace map for arithmetic surfaces via residues and
higher adeles, Algebraic and number theory 7 (2012) 1503-1536.
[80] D. Mumford, Geometric Invariant Theory, Springer, 1994.
[81] M.S. Narasimhan and C.S. Seshadri, Stable and unitary vector bundles on a compact
Riemann surface, Ann. Math. 82 (1965) 540-567.
[82] J. Neukirch, Algebraic Number Theory, Grundlehren der Math. Wissenschaften 322,
Springer, 1999.
[83] A.M. Odlyzko, On the distribution of spacings between zeros of the zeta function.
Math. Comput. 48 (1987) 273-308.
[84] D.V. Osipov, Adeles on n-dimensional schemes and categories Cn , Inter. J. Math. 18
(2007) 269-279.
[85] D.V. Osipov, Central extensions and reciprocity laws on algebraic surfaces, Sb.
Math. 196 (2005) 1503-1527.
[86] D.V. Osipov, Unramified two-dimensional Langlands correspondence, Izv. Math. 77
(2013) 714-741.
[87] D.V. Osipov and A.N. Parshin, Harmonic analysis on local fields and adelic spaces
II, Izv. Math. 75 (2011) 749-814.
[88] D.V. Osipov and X.W. Zhu, A categorical proof of the Parshin reciprocity laws on
algebraic surfaces, Algebra Number Theory 5 (2011) 289-337.
[89] M.S. Osborne and G. Warner, The Selberg trace formula. II. Partition, reduction,
truncation. Pacific J. Math. 106 (1983) 307-496.
[90] M.S. Osborne and G. Warner, The Theory of Eisenstein Systems, Academic Press,
1981.
[91] A.N. Parshin, On the arithmetic of two-dimensional schemes. I. Distributions and
residues. Izv. Akad. Nauk SSSR Ser. Mat. 40 (1976) 736-773.
[92] A.N. Parshin, Chern classes, adeles and L-functions. J. Reine Angew. Math. 341
(1983) 174-192.
[93] S. Ramanan and A. Ramanathan, Some remarks on the instability flag. Tohoku
Math. J. 36 (1984) 269-291.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 523

Bibliography 523

[94] I. Satake, On the compactification of the Siegel space, J. Indian Math. Soc. 20 (1956)
259-281.
[95] I. Satake, On representations and compactifications of symmetric Riemannian
spaces, Ann. of Math. 71 (1960) 77-110.
[96] I. Satake, On compactifications of the quotient spaces for arithmetically defined
discontinuous groups, Ann. of Math. 72 (1960) 555-580.
[97] A. Selberg, Old and new conjectures and results about a class of Dirichlet series,
Proceedings of the Amalfi Conference on Analytic Number Theory (Maiori, 1989)
(Salerno), (1992) 367-385.
[98] J.-P. Serre, Algebraic Groups and Class Fields, GTM 117, Springer, 1988.
[99] U. Shapira and B. Weiss, Stable lattices and the diagonal group, J. Euro. Math. Soc.
18 (2016) 1753-1767.
[100] G. Shimura, Introduction to Arithmetic Theory of Automorphic Functions, Publica-
tions of the Mathematical Society of Japan 11, Math. Soc. Japan, 1971.
[101] C.L. Siegel, Lectures on the geometry of numbers, Springer, 1989.
[102] D.M. Snow, Homogeneous vector bundles. Group actions and invariant theory, CMS
Conference Proceedings 10, (1989) 193-205.
[103] T.A. Springer, Linear Algebraic Groups, Birkhäuser, 1998.
[104] U. Stuhler, Eine Bemerkung zur Reduktionstheorie quadratischer Formen. Arch.
Math. 27 (1976) 604-610.
[105] U. Stuhler, Eine Bemerkung zur Reduktionstheorie quadratischer Formen. II. Arch.
Math. 28 (1977) 611-619.
[106] K. Sugahara and L. Weng, Arithmetic cohomology groups, arXiv: 1507.06074.
[107] K. Sugahara and L. Weng, Har1 for arithmetic surface is finite, arXiv:1603.02353.
[108] K. Sugahara and L. Weng, Arithmetic central extensions and reciprocity laws for
arithmetic surfaces, arXiv:1603.02355.
[109] M. Suzuki, A proof of the Riemann hypothesis for the Weng zeta function of rank
3 for the rationals. in The Conference on L-Functions, 175–199, World Sci. Publ.
2007.
[110] M. Suzuki, The Riemann hypothesis for Weng’s zeta function of Sp(4) over Q, with
an appendix by L. Weng, J. Number Theory 129 (2009) 551-579.
[111] M. Suzuki and L. Weng, private communications.
[112] M. Suzuki and L. Weng, Zeta functions for G2 and their zeros, IMRN (2009) 241-
290.
[113] J. Tate, Fourier analysis in number fields and Hecke’s zeta functions, Thesis, Prince-
ton University, 1950.
[114] J. Tate, Residues of differential forms on curves, Ann. Sci. Eco. Norm, Sup., 4
(1968) 149-159.
[115] A. Terras, Harmonic analysis on symmetric spaces and applications II, Springer,
1988.
[116] E.C. Titchmarsh, The theory of the Riemann zeta-function, The Clarendon Press
Oxford University Press, 1986.
[117] G.N. Watson, A Treatise on the Theory of Bessel Functions, Cambridge University
Press, 1922.
[118] A. Weil, Basic Number Theory, Springer, 1973.
[119] L. Weng, Geometry of numbers, arXiv:1102.1302.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 524

524 Bibliography

[120] L. Weng, Non-abelian zeta functions for function fields. Amer. J. Math. 127 (2005)
973-1017.
[121] L. Weng, Rank Two Non-Abelian Zeta and its Zeros, arXiv:math/0412009.
[122] L. Weng, A rank two zeta and its zeros. J. Ramanujan Math. Soc. 21 (2006) 205-266.
[123] L. Weng, Automorphic Forms, Eisenstein Series and Spectral Decompositions, in
Arithmetic Geometry and Number Theory, 123-210, World Sci. 2006.
[124] L. Weng, Geometric arithmetic: a program, in Arithmetic geometry and number
theory, 211-400, World Sci., 2006.
[125] L. Weng, A geometric approach to L-functions. in The Conference on L-Functions,
219–370, World Sci., 2007.
[126] L. Weng, Symmetries and the Riemann Hypothesis, in Advanced Studies in Pure
Mathematics 58, Japan Math. Soc., (2010) 173-224.
[127] L. Weng, Stability and Arithmetic, in Advanced Studies in Pure Mathematics 58,
Japan Math. Soc., (2010) 225-360.
[128] L. Weng, Parabolic Reduction, Stability and the Masses I: Special Linear Groups, at
https://repository.kulib.kyoto-u.ac.jp/dspace/bitstream/2433/194754/1/1826-16.pdf.
[129] L. Weng, Automorphic Forms and Automorphic L-Functions, RIMS Kokyuroku
1826, (2013) 168-179.
[130] L. Weng, Motivic Euler product and its applications, available at
http://www2.math.kyushu-u.ac.jp/ weng/writings.html.
[131] L. Weng, Distributions of zeros for non-abelian zeta functions, available at
http://www2.math.kyushu-u.ac.jp/ weng/writings.html.
[132] L. Weng, Higher rank zeta functions and Riemann Hypothesis for elliptic curves,
talk at Arithmetic and Algebraic Geometry 2013, Tokyo.
[133] L. Weng, Zeros of zeta functions for exceptional groups of type E, available at
http://www2.math.kyushu-u.ac.jp/ weng/writings.html.
[134] L. Weng and D. Zagier, Higher rank zeta functions for elliptic curves, available at
http://www2.math.kyushu-u.ac.jp/ weng/writings.html.
[135] L. Weng and D. Zagier, Higher rank zeta functions and SLn -zeta functions for
curves, available at http://www2.math.kyushu-u.ac.jp/ weng/writings.html.
[136] A. Yekutieli, An Explicit construction of the Grothendieck residue complex, Aster-
isque 208 (1992).
[137] D. Zagier, The Rankin-Selberg method for automorphic functions which are not of
rapid decay. J. Fac. Sci. Univ. Tokyo 28 (1982) 415-437.
[138] S. Zhang, Positive line bundles on arithmetic surfaces, Ann. Math.136 (1992) 569-
587.
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 525

Index

adelic ring ν-canonical type, 415

∞, 453 ν-parabolic reduction, 415
finite, 453 canonical type, 415
adelic space parabolic reduction, 415
arithmetic, 455 semi-stable, 414
finite, infinite, 455 arithmetic adelic ring
algebraic group arithmetic surface, 453
F-torus, 161 arithmetic adelic space
split extension, 161 k-th, 457
height function, 172 m-th reduced, 456
multiplicative, 161 type I reduced, 456
torus arithmetic cohomology group, 471
F-rank, 161 i-th, 457
absolute rank, 161 arithmetic parabolic group scheme
rank, 161 admissible, 413
split field, 161 arithmetical cohomology group, 112
unipotent, 161 Arthur
unipotent element, 161 analytic truncation, 190
unipotent radical, 161 level P-analytic truncation, 191
Arakelov period, 194
Arakelov-Euler characteristic, 11 automorphic function, 114, 173
degree, 11 constant term, 173, 190
Picard group, 4 automorphic representation, 175
arithmetic G torsor, 411 cuspidal datum, 175
slope, 411 equivalence, 175
arithmetic P torsor
admissible, 413 Bessel K-function, 116
arithmetic P- torsor
canonical reduction, 415 canonical
arithmetic P-torsor, 413 (infinitesimal) OF -structure, 406
ν-canonical reduction, 415 filtration, 370

525
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 526

526 Index

polygon, 370 flag, 447

parabolic subgroup, 370 reduced, 447
canonical form, 403 Fourier coefficient, 114
canonical polygon, 179 Fourier expansion, 114
Cartan involution, 403, 404 function
central extension, 516 z-finite, 173
character z p (s), a p (s), 307
associated to a normalized p, 375 K-finite, 173
characteristic function #-integration, 198
+ G
aG+
P , aP , 390 exponential polynomial, 195
1(p ≤ p), 370
∗
exponent, 195
cohomology group moderate growth, 172
arithmetic count, 32 rapidly decreasing, 173
arithmetic numeration, 32 smooth, 173
cohomology group 0-th, 5 type (C), 196
complete Dedekind zeta function, 5 fundamental domain, 69, 71, 81, 83
cone, 195 boundary point, 91
#-integrable, 196, 197 inner point, 90
convex cone, 391
side, 82
convex polygon, 370
vertex, 82
comparison, 370
vertices, 82
normalized, 375
fundamental parallelepiped, 23
count function
`±p,w , 302
general linear group
n p,w (k, h), n p (k, h),M p (k, h), 300
cohomology group, 30
cusp, 66, 69, 70, 79
element equivalence, 30
associated lattice, 79
influence sphere, 89, 99 partial order, 30
semi-reduced modular point, 90 group
cusp form, 174 adelic, 28
cuspidal representation, 175 discontinuous, 68, 70
discrete, 68, 70
distance to cusp, 95, 99 first kind Fuchsian, 70
Fuchsian, 69
Eisenstein period maximal abelian quotient, 387
geometric, 195 paraboli celement, 71
Eisenstein series, 115, 117, 119, 122, 126, parabolic element, 70
127, 176 special orthogonal, 18
Siegel, 211, 218 special unitary, 18
P1,...,1 , 212
Pn−1,1 , 212 Haar measure
Siegel-Maaβ, 217 compatibility condition, 412
entire function of order one, 153 compatibility condition, 171
envelop of a submodule, 13 Hilbert modular group, 71
horoball, 84
fiberwise admissible metric, 411 horosphere, 84
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 527

Index 527

idele narrow Hilbert modular group, 64

positivity, 39 narrow regulator of K, 25
idele group, 4 nine diagram, 31
intertwining operator, 176 norm, 4, 74
invariant norm map, 4
d p,w , 307 normalized transform, 134
h±k , 315 number field
isogeny, 162, 387 adelic ring, 27
arithmetical cohomology groups, 28
lattice, 9 canonical idele, 30
µ-invariant, 12 canonical lattice, 35
0th cohomolog group, 35 finite adelic ring, 27
1st cohomolog group, 35 Gamma invariant, 36
1st canonical sublattice, 15 totally real, 63
2nd canonical sublattice, 15 numeration, 506
Arakelov-Euler characteristic, 11 additive, 509
canonical filtration, 16 cohomologies, 514
canonical polygon, 16
canonical volume, 10 ordered partition, 166
first Minkowski successive minimum,
37 period
induced from submodule, 11 (G, P), 244, 249
isometry, 9 (Φ, ∆), 337
Lebesgue (co)volume, 10 SLn , 215
polygon, 16 Sp2n , 217
principal, 406 polynomial
quotient, 11 R p (s), 305
restriction of scalars, 10 power function, 212, 219
semi-stable, 11 projective module, 7
stable, 11 standard form, 9
sublattice, 11
Levi decomposition, 162 rank of projective module, 7
Levi subgroup, 162 rational function
standard, 162 ra (s), 284
reciprocity symbol, 516
matrix reduced point, 76
doubly stochastic, 394 reductive group, 161
stochastic, 394 F-subgroup, 162
metrized G-torsor, 406 T -period, 203
Minkowski embedding, 17 Langlands decomposition, 170
module logarithmic map, 170
invertible, 7 logarithmic morphism, 169
moduli spaces maximal split quotient torus, 162
complexity, 44 maximal split quotient torus, 387
minimal and maximal values, 44 maximal split torus, 162, 387
multiple L-function, 182 parabolic subgroup, 162, 387
November 15, 2017 15:15 ws-book9x6 BC: 10723 - Zeta Functions of Reductive Groups and Their Zeros WengZeta page 528

528 Index

Levi decomposition, 387 compact oriented, 479

Levi subgroup, 387, 402 inductive, 477
standard, 162, 169, 387 projective, 477
period, 203 trace, 75
regularized period, 199 truncated domain, 192
root system, 163, 336
Σ p (k), Σ p (k, h), 310 upper half plane, 56
≤, 399 fractional action, 56
(acute) Weyl chamber, 185, 390 hyperbolic line, 56
relative, 397 hyperbolic metric, 56
(obtuse) Weyl cone, 185, 390 Laplace-Beltrami operator, 56
relative, 397 upper half space, 57
≤, 390 fractional action, 58
Dynkin diagram Hamilton’s quaternions, 58
D5 , 350 hyperbolic line, 58
E6 , E7 , E8 , 349 hyperbolic metric, 58
highest root, 314 hyperbolic plane, 58
lowest root, 314 Laplace-Beltrami operator, 58
positive, 422
positive chamber, 186 Weyl group
positive cone, 186 W 0 , 302
positive root, 163 W p , 306
positive Weyl chamber, 390 maximal discrepancy, 306
positive Weyl cone, 390 p-level, 315
pseudo, 335 length of element, 248
restricted, 336 special element, 249, 300
root hyperplane, 183 special permutation, 255
simple coroot, 164
simple coweight, 164 zeta function
simple root, 164 (G, P), 245
D5 , 350 discriminant, 285, 300
E6 , E7 , E8 , 349 rational factor, 250
simple weight, 164 zeta factor, 250
wall, 399 G2 , 226, 227
SLn , 215, 216
Siegel set, 172, 188 Sp2n , 218–220
Siegel upper half space, 217 Epstein, 113, 217
slope, 413 Koecher, 211
slow growth, 118, 134 normalization process, 225
space of characters, 162 partial, 122, 284
stabilizer group, 69, 70 rank two, 113, 139
reduced, 69 Weng, 244, 296, 356, 386
sufficiently regular, 188 zeta zeros
sufficiently small, 188 Delta pair correlation, 435, 437, 438
pair correlation, 434
topology Riemann hypothesis, 241, 386, 432