This page intentionally left blank

CAMBRIDGE STUDIES IN ADVANCED MATHEMATICS

Editorial Board
B. Bollobas, W. Fulton, A. Katok, F. Kirwan, P. Sarnak, B. Simon, B. Totaro

AUTOMORPHIC FORMS AND L-FUNCTIONS FOR

THE GROUP GL (n, R)

L-functions associated with automorphic forms encode all classical num-

ber theoretic information. They are akin to elementary particles in physics.
This book provides an entirely self-contained introduction to the theory of
L-functions in a style accessible to graduate students with a basic knowledge
of classical analysis, complex variable theory, and algebra. Also within the
volume are many new results not yet found in the literature. The exposition
provides complete detailed proofs of results in an easy-to-read format using
many examples and without the need to know and remember many complex
definitions. The main themes of the book are first worked out for GL(2,R) and
GL(3,R), and then for the general case of GL(n,R). In an appendix to the book,
a set of Mathematica® functions is presented, designed to allow the reader to
explore the theory from a computational point of view.
CAMBRIDGE STUDIES IN ADVANCED MATHEMATICS
Editorial Board:
B. Bollobas, W. Fulton, A. Katok, F. Kirwan, P. Sarnak, B. Simon, B. Totaro
Already published
30 D.J. Benson Representations and cohomology I
31 D.J. Benson Representations and cohomology II
32 C. Allday & V. Puppe Cohomological methods in transformation groups
33 C. Soule et al. Lectures on Arakelov geometry
34 A. Ambrosetti & G. Prodi A primer of nonlinear analysis
35 J. Palis & F. Takens Hyperbolicity, stability and chaos at homoclinic bifurcations
37 Y. Meyer Wavelets and operators I
38 C. Weibel An introduction to homological algebra
39 W. Bruns & J. Herzog Cohen-Macaulay rings
40 V. Snaith Explicit Brauer induction
41 G. Laumon Cohomology of Drinfeld modular varieties I
42 E.B. Davies Spectral theory and differential operators
43 J. Diestel, H. Jarchow, & A. Tonge Absolutely summing operators
44 P. Mattila Geometry of sets and measures in Euclidean spaces
45 R. Pinsky Positive harmonic functions and diffusion
46 G. Tenenbaum Introduction to analytic and probabilistic number theory
47 C. Peskine An algebraic introduction to complex projective geometry
48 Y. Meyer & R. Coifman Wavelets
49 R. Stanley Enumerative combinatorics I
50 I. Porteous Clifford algebras and the classical groups
51 M. Audin Spinning tops
52 V. Jurdjevic Geometric control theory
53 H. Volklein Groups as Galois groups
54 J. Le Potier Lectures on vector bundles
55 D. Bump Automorphic forms and representations
56 G. Laumon Cohomology of Drinfeld modular varieties II
57 D.M. Clark & B.A. Davey Natural dualities for the working algebraist
58 J. McCleary A user’s guide to spectral sequences II
59 P. Taylor Practical foundations of mathematics
60 M.P. Brodmann & R.Y. Sharp Local cohomology
61 J.D. Dixon et al. Analytic pro-P groups
62 R. Stanley Enumerative combinatorics II
63 R.M. Dudley Uniform central limit theorems
64 J. Jost & X. Li-Jost Calculus of variations
65 A.J. Berrick & M.E. Keating An introduction to rings and modules
66 S. Morosawa Holomorphic dynamics
67 A.J. Berrick & M.E. Keating Categories and modules with K-theory in view
68 K. Sato Levy processes and infinitely divisible distributions
69 H. Hida Modular forms and Galois cohomology
70 R. Iorio & V. Iorio Fourier analysis and partial differential equations
71 R. Blei Analysis in integer and fractional dimensions
72 F. Borceaux & G. Janelidze Galois theories
73 B. Bollobas Random graphs
74 R.M. Dudley Real analysis and probability
75 T. Sheil-Small Complex polynomials
76 C. Voisin Hodge theory and complex algebraic geometry I
77 C. Voisin Hodge theory and complex algebraic geometry II
78 V. Paulsen Completely bounded maps and operator algebras
79 F. Gesztesy & H. Holden Soliton Equations and their Algebro-Geometric Solutions Volume 1
81 Shigeru Mukai An Introduction to Invariants and Moduli
82 G. Tourlakis Lectures in logic and set theory I
83 G. Tourlakis Lectures in logic and set theory II
84 R.A. Bailey Association Schemes
85 James Carlson, Stefan Müller-Stach, & Chris Peters Period Mappings and Period Domains
86 J.J. Duistermaat & J.A.C. Kolk Multidimensional Real Analysis I
87 J.J. Duistermaat & J.A.C. Kolk Multidimensional Real Analysis II
89 M. Golumbic & A.N. Trenk Tolerance Graphs
90 L.H. Harper Global Methods for Combinatorial Isoperimetric Problems
91 I. Moerdijk & J. Mrcun Introduction to Foliations and Lie Groupoids
92 János Kollár, Karen E. Smith, & Alessio Corti Rational and Nearly Rational Varieties
93 David Applebaum Lévy Processes and Stochastic Calculus
95 Martin Schechter An Introduction to Nonlinear Analysis

See http:www.cambridge.org for a complete list of books available in this series

Automorphic Forms and L-Functions for
the Group GL (n, R)

DORIAN GOLDFELD
Columbia University

With an Appendix by Kevin A. Broughan

University of Waikato
cambridge university press
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo

Cambridge University Press

The Edinburgh Building, Cambridge cb2 2ru, UK
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
Information on this title: www.cambridge.org/9780521837712

© D. Goldfeld 2006

This publication is in copyright. Subject to statutory exception and to the provision of

relevant collective licensing agreements, no reproduction of any part may take place
without the written permission of Cambridge University Press.

First published in print format 2006

isbn-13 978-0-511-22101-9 eBook (NetLibrary)

isbn-10 0-511-22101-0 eBook (NetLibrary)

isbn-13 978-0-521-83771-2 hardback

isbn-10 0-521-83771-5 hardback

Cambridge University Press has no responsibility for the persistence or accuracy of urls
for external or third-party internet websites referred to in this publication, and does not
guarantee that any content on such websites is, or will remain, accurate or appropriate.
Dedicated to Ada, Dahlia, and Iris
 Contents

Introduction page xi
1 Discrete group actions 1
1.1 Action of a group on a topological space 3
1.2 Iwasawa decomposition 8
1.3 Siegel sets 15
1.4 Haar measure 19
1.5 Invariant measure on coset spaces 23
1.6 Volume of S L(n, Z)\S L(n, R)/S O(n, R) 27
2 Invariant differential operators 38
2.1 Lie algebras 39
2.2 Universal enveloping algebra of gl(n, R) 42
2.3 The center of the universal enveloping algebra of gl(n, R) 46
2.4 Eigenfunctions of invariant differential operators 50
3 Automorphic forms and L–functions for S L(2, Z) 54
3.1 Eisenstein series 55
3.2 Hyperbolic Fourier expansion of Eisenstein series 59
3.3 Maass forms 62
3.4 Whittaker expansions and multiplicity one for G L(2, R) 63
3.5 Fourier–Whittaker expansions on G L(2, R) 67
3.6 Ramanujan–Petersson conjecture 68
3.7 Selberg eigenvalue conjecture 70
3.8 Finite dimensionality of the eigenspaces 71
3.9 Even and odd Maass forms 73
3.10 Hecke operators 74
3.11 Hermite and Smith normal forms 77
3.12 Hecke operators for L2 (S L(2, Z))\h2 80

vii
viii Contents

3.13 L–functions associated to Maass forms 84

3.14 L-functions associated to Eisenstein series 89
3.15 Converse theorems for S L(2, Z) 91
3.16 The Selberg spectral decomposition 94
4 Existence of Maass forms 99
4.1 The infinitude of odd Maass forms for S L(2, Z) 100
4.2 Integral operators 101
4.3 The endomorphism ♥ 105
4.4 How to interpret ♥: an explicit operator with purely
cuspidal image 106
4.5 There exist infinitely many even cusp forms for S L(2, Z) 108
4.6 A weak Weyl law 110
4.7 Interpretation via wave equation and the role of finite
propagation speed 111
4.8 Interpretation via wave equation: higher rank case 111
5 Maass forms and Whittaker functions for S L(n, Z) 114
5.1 Maass forms 114
5.2 Whittaker functions associated to Maass forms 116
5.3 Fourier expansions on S L(n, Z)\hn 118
5.4 Whittaker functions for S L(n, R) 128
5.5 Jacquet’s Whittaker function 129
5.6 The exterior power of a vector space 134
5.7 Construction of the Iν function using wedge products 138
5.8 Convergence of Jacquet’s Whittaker function 141
5.9 Functional equations of Jacquet’s Whittaker function 144
5.10 Degenerate Whittaker functions 150
6 Automorphic forms and L-functions for S L(3, Z) 153
6.1 Whittaker functions and multiplicity one for S L(3, Z) 153
6.2 Maass forms for S L(3, Z) 159
6.3 The dual and symmetric Maass forms 161
6.4 Hecke operators for S L(3, Z) 163
6.5 The Godement–Jacquet L-function 172
6.6 Bump’s double Dirichlet series 186
7 The Gelbart–Jacquet lift 194
7.1 Converse theorem for S L(3, Z) 194
7.2 Rankin–Selberg convolution for GL(2) 210
7.3 Statement and proof of the Gelbart–Jacquet lift 213
7.4 Rankin–Selberg convolution for G L(3) 223
 Contents ix

8 Bounds for L-functions and Siegel zeros 235

8.1 The Selberg class 235
8.2 Convexity bounds for the Selberg class 238
8.3 Approximate functional equations 241
8.4 Siegel zeros in the Selberg class 245
8.5 Siegel’s theorem 249
8.6 The Siegel zero lemma 251
8.7 Non-existence of Siegel zeros for Gelbart–Jacquet lifts 252
8.8 Non-existence of Siegel zeros on G L(n) 256
9 The Godement–Jacquet L-function 259
9.1 Maass forms for S L(n, Z) 259
9.2 The dual and symmetric Maass forms 261
9.3 Hecke operators for S L(n, Z) 266
9.4 The Godement–Jacquet L-function 277
10 Langlands Eisenstein series 285
10.1 Parabolic subgroups 286
10.2 Langlands decomposition of parabolic subgroups 288
10.3 Bruhat decomposition 292
10.4 Minimal, maximal, and general parabolic Eisenstein series 295
10.5 Eisenstein series twisted by Maass forms 301
10.6 Fourier expansion of minimal parabolic Eisenstein series 303
10.7 Meromorphic continuation and functional equation of
maximal parabolic Eisenstein series 307
10.8 The L-function associated to a minimal parabolic
Eisenstein series 310
10.9 Fourier coefficients of Eisenstein series twisted by
Maass forms 315
10.10 The constant term 319
10.11 The constant term of S L(3, Z) Eisenstein series twisted by
S L(2, Z)-Maass forms 321
10.12 An application of the theory of Eisenstein series to the
non-vanishing of L-functions on the line (s) = 1 322
10.13 Langlands spectral decomposition for S L(3, Z)\h3 324
11 Poincaré series and Kloosterman sums 337
11.1 Poincaré series for S L(n, Z) 337
11.2 Kloosterman sums 339
11.3 Plücker coordinates and the evaluation of Kloosterman sums 343
11.4 Properties of Kloosterman sums 350
x Contents

11.5 Fourier expansion of Poincaré series 352

11.6 Kuznetsov’s trace formula for S L(n, Z) 354
12 Rankin–Selberg convolutions 365
12.1 The GL(n) × GL(n) convolution 366
12.2 The GL(n) × GL(n + 1) convolution 372
12.3 The G L(n) × G L(n  ) convolution with n < n  376
12.4 Generalized Ramanujan conjecture 381
12.5 The Luo–Rudnick–Sarnak bound for the generalized
Ramanujan conjecture 384
12.6 Strong multiplicity one theorem 393
13 Langlands conjectures 395
13.1 Artin L-functions 397
13.2 Langlands functoriality 402

List of symbols 407

Appendix The GL(n)pack Manual Kevin A. Broughan 409
A.1 Introduction 409
A.2 Functions for GL(n)pack 413
A.3 Function descriptions and examples 416
References 473
Index 485
 Introduction

The theory of automorphic forms and L-functions for the group of n × n invert-
ible real matrices (denoted G L(n, R)) with n ≥ 3 is a relatively new subject.
The current literature is rife with 150+ page papers requiring knowledge of a
large breadth of modern mathematics making it difficult for a novice to begin
working in the subject. The main aim of this book is to provide an essentially
self-contained introduction to the subject that can be read by someone with a
mathematical background consisting only of classical analysis, complex vari-
able theory, and basic algebra – groups, rings, fields. Preparation in selected
topics from advanced linear algebra (such as wedge products) and from the
theory of differential forms would be helpful, but is not strictly necessary for
a successful reading of the text. Any Lie or representation theory required is
developed from first principles.
This is a low definition text which means that it is not necessary for the reader
to memorize a large number of definitions. While there are many definitions,
they are repeated over and over again; in fact, the book is designed so that a
reader can open to almost any page and understand the material at hand without
having to backtrack and awkwardly hunt for definitions of symbols and terms.
The philosophy of the exposition is to demonstrate the theory by simple, fully
worked out examples. Thus, the book is restricted to the action of the discrete
group S L(n, Z) (the group of invertible n × n matrices with integer coefficients)
acting on G L(n, R). The main themes are first developed for S L(2, Z) then
repeated again for S L(3, Z), and yet again repeated in the more general case of
S L(n, Z) with n ≥ 2 arbitrary. All of the proofs are carefully worked out over
the real numbers R, but the knowledgeable reader will see that the proofs will
generalize to any local field. In line with the philosophy of understanding by
simple example, we have avoided the use of adeles, and as much as possible
the theory of representations of Lie groups. This very explicit language appears

xi
xii Introduction

particularly useful for analytic number theory where precise growth estimates
of L-functions and automorphic forms play a major role.
The theory of L-functions and automorphic forms is an old subject with roots
going back to Gauss, Dirichlet, and Riemann. An L-function is a Dirichlet series


∞
an
n=1
ns
where the coefficients an , n = 1, 2, . . . , are interesting number theoretic func-
tions. A simple example is where an is the number of representations of n as a
sum of two squares. If we knew a lot about this series as an analytic function of
s then we would obtain deep knowledge about the statistical distribution of the
values of an . An automorphic form is a function that satisfies a certain differ-
ential equation and also satisfies a group of periodicity relations. An example
is given by the exponential function e2πi x which is periodic (i.e., it has the
same value if we transform x → x + 1) and it satisfies the differential equa-
2
tion ddx 2 e2πi x = −4π 2 e2πi x . In this example the group of periodicity relations
is just the infinite additive group of integers, denoted Z. Remarkably, a vast
theory has been developed exposing the relationship between L-functions and
automorphic forms associated to various infinite dimensional Lie groups such
as G L(n, R).
The choice of material covered is very much guided by the beautiful paper
(Jacquet, 1981), titled Dirichlet series for the group G L(n), a presentation of
which I heard in person in Bombay, 1979, where a classical outline of the theory
of L-functions for the group G L(n, R) is presented, but without any proofs. Our
aim has been to fill in the gaps and to give detailed proofs. Another motivating
factor has been the grand vision of Langlands’ philosophy wherein L-functions
are akin to elementary particles which can be combined in the same way as
one combines representations of Lie groups. The entire book builds upon this
underlying hidden theme which then explodes in the last chapter.
In the appendix a set of Mathematica functions is presented. These have
been designed to assist the reader to explore many of the concepts and results
contained in the chapters that go before. The software can be downloaded by
going to the website given in the appendix.
This book could not have been written without the help I have received from
many people. I am particularly grateful to Qiao Zhang for his painstaking read-
ing of the entire manuscript. Hervé Jacquet, Daniel Bump, and Adrian Diaconu
have provided invaluable help to me in clarifying many points in the theory.
I would also like to express my deep gratitude to Xiaoqing Li, Elon Linden-
strauss, Meera Thillainatesan, and Akshay Venkatesh for allowing me to include
their original material as sections in the text. I would like to especially thank
 Introduction xiii

Dan Bump, Kevin Broughan, Sol Friedberg, Jeff Hoffstein, Alex Kontorovich,
Wenzhi Luo, Carlos Moreno, Yannan Qiu, Ian Florian Sprung, C. J. Mozzochi,
Peter Sarnak, Freydoon Shahidi, Meera Thillainatesan, Qiao Zhang, Alberto
Perelli and Steve Miller, for clarifying and improving various proofs, defini-
tions, and historical remarks in the book. Finally, Kevin Broughan has provided
an invaluable service to the mathematical community by creating computer
code for many of the functions studied in this book.

Dorian Goldfeld
 1
Discrete group actions

The genesis of analytic number theory formally began with the epoch making
memoir of Riemann (1859) where he introduced the zeta function,


∞
ζ (s) := n −s , ((s) > 1),
n=1

and obtained its meromorphic continuation and functional equation

s    ∞
−s/2 −(1−s)/2 1−s du
π  ζ (s) = π  ζ (1 − s), (s) = e−u u s .
2 2 u
0

Riemann showed that the Euler product representation

 
1 −1
ζ (s) = 1− s ,
p p

together with precise knowledge of the analytic behavior of ζ (s) could be used
to obtain deep information on the distribution of prime numbers.
One of Riemann’s original proofs of the functional equation is based on the
Poisson summation formula



f (ny) = y −1 fˆ(ny −1 ),
n∈Z n∈Z

where f is a function with rapid decay as y → ∞ and

 ∞
fˆ(y) = f (t)e−2πit y dt,
−∞

is the Fourier transform of f . This is proved by expanding the periodic function

F(x) = f (x + n)
n∈Z

1
2 Discrete group actions

in a Fourier series. If f is an even function, the Poisson summation formula

may be rewritten as


∞

∞
1
f (ny −1 ) = y fˆ(ny) − (y fˆ(0) − f (0)),
n=1 n=1
2

from which it follows that for (s) > 1,

 ∞  ∞
∞
dy dy
ζ (s) f (y)y s = f (ny)y s
0 y 0 n=1
y
 ∞
∞
 dy
= f (ny)y s + f (ny −1 )y −s
1 n=1
y
 ∞


∞
 dy 1 f (0) ˆ(0)
f
= f (ny)y s + fˆ(ny)y 1−s − + .
1 n=1
y 2 s 1−s

If f (y) and fˆ(y) have sufficient decay as y → ∞, then the integral above
converges absolutely for all complex s and, therefore, defines an entire function
of s. Let
 ∞
dy
f˜(s) = f (y)y s
0 y
denote the Mellin transform of f , then we see from the above integral rep-
resentation and the fact that fˆˆ(y) = f (−y) = f (y) (for an even function f )
that

ζ (s) f˜(s) = ζ (1 − s) f˜ˆ (1 − s).

Choosing f (y) = e−π y , a function with the property that it is invariant under
2

Fourier transform, we obtain Riemann’s original form of the functional equa-

tion. This idea of introducing an arbitrary test function f in the proof of the
functional equation first appeared in Tate’s thesis (Tate, 1950).
A more profound understanding of the above proof did not emerge until
much later. If we choose f (y) = e−π y in the Poisson summation formula, then
2

since fˆ(y) = f (y), one observes that for y > 0,

∞
1
∞
e−πn e−π n /y .
2 2
y
=√
n=−∞ y n=−∞

This identity is at the heart of the functional equation of the Riemann zeta
function, and is a known transformation formula for Jacobi’s theta function


∞
2
θ(z) = e2πin z ,
n=−∞
 1.1 Action of a group on a topological space 3

 
a b
where z = x + i y with x ∈ R and y > 0. If is a matrix with integer
c d
coefficients a, b, c, d satisfiying ad − bc = 1, c ≡ 0 (mod 4), c = 0, then the
Poisson summation formula can be used to obtain the more general transfor-
mation formula (Shimura, 1973)
 
az + b
= d−1 χc (d)(cz + d) 2 θ(z).
1
θ
cz + d
Here χc is the primitive character of order ≤ 2 corresponding to the field exten-
1
sion Q(c 2 )/Q,

1 if d ≡ 1 (mod 4)
d =
i if d ≡ −1 (mod 4),
1
and (cz + d) 2 is the “principal determination” of the square root of cz + d, i.e.,
the one whose real part is > 0.
It is now well understood that underlying the functional equation of the
Riemann zeta function are the above transformation formulae for θ (z). These
transformation
  formulae are induced from the action of a group of matrices
a b
on the upper half-plane h = {x + i y | x ∈ R, y > 0} given by
c d
az + b
z → .
cz + d
The concept of a group acting on a topological space appears to be absolutely
fundamental in analytic number theory and should be the starting point for any
serious investigations.

1.1 Action of a group on a topological space

Definition 1.1.1 Given a topological space X and a group G, we say that G
acts continuously on X (on the left) if there exists a map ◦ : G → Func(X → X )
(functions from X to X ), g → g◦ which satisfies:
r x → g ◦ x is a continuous function of x for all g ∈ G;
r g ◦ (g  ◦ x) = (g · g  ) ◦ x, for all g, g  ∈ G, x ∈ X where · denotes the
internal operation in the group G;
r e ◦ x = x, for all x ∈ X and e = identity element in G.

Example 1.1.2 Let G denote the additive group of integers Z. Then it is easy
to verify that the group Z acts continuously on the real numbers R with group
4 Discrete group actions

action ◦ defined by

n ◦ x := n + x,

for all n ∈ Z, x ∈ R. In this case e = 0.

+
 1.1.3 Let G = G L(2, R) denote the group of 2 × 2 matrices
Example

a b
with a, b, c, d ∈ R and determinant ad − bc > 0. Let
c d
 
h := x + i y  x ∈ R, y > 0
 
a b
denote the upper half-plane. For g = ∈ G L(2, R)+ and z ∈ h define:
c d

az + b
g ◦ z := .
cz + d
Since

az + b ac|z|2 + (ad + bc)x + bd (ad − bc) · y

= +i ·
cz + d |cz + d|2 |cz + d|2

it immediately follows that g ◦ z ∈ h. We leave as an exercise to the reader, the

verification that ◦ satisfies the additional axioms of a continuous action. One
usually extends this action to the larger space h∗ = h ∪ {∞}, by defining
  
a b a/c if c = 0,
◦∞=
c d ∞ if c = 0.

Assume that a group G acts continously on a topological space X . Two

elements x1 , x2 ∈ X are said to be equivalent (mod G) if there exists g ∈ G
such that x2 = g ◦ x1 . We define
 
Gx := g ◦ x  g ∈ G

to be the equivalence class or orbit of x, and let G\X denote the set of equiva-
lence classes.

Definition 1.1.4 Let a group G act continuously on a topological space X .

We say a subset  ⊂ G is discrete if for any two compact subsets A, B ⊂ X ,
there are only finitely many g ∈  such that (g ◦ A) ∩ B = φ, where φ denotes
the empty set.
 1.1 Action of a group on a topological space 5

Example 1.1.5 The discrete subgroup S L(2, Z). Let

   
a b 
 = S L(2, Z) := a, b, c, d ∈ Z, ad − bc = 1 ,
c d 
and let
   
1 m 
∞ := m∈Z
0 1 

be the subgroup of  whichfixes ∞.  Note that ∞ \ is just a set of coset

a b
representatives of the form where for each pair of relatively prime
c d
integers (c, d) = 1 we choose a unique a, b satisfying ad − bc = 1. This fol-
lows immediately from the identity
     
1 m a b a + mc b + md
· = .
0 1 c d c d
The fact that S L(2, Z) is discrete will be deduced from the following lemma.
Lemma 1.1.6 Fix real numbers 0 < r, 0 < δ < 1. Let Rr,δ denote the
rectangle
 
Rr,δ = x + i y  − r ≤ x ≤ r, 0 < δ ≤ y ≤ δ −1 .
Then for every  > 0, and any fixed set S of coset representatives for
∞ \S L(2, Z), there are at most 4 + (4(r + 1)/δ) elements g ∈ S such that
Im(g ◦ z) >  holds for some z ∈ Rr,δ .
 
a b
Proof Let g = . Then for z ∈ Rr,δ ,
c d
y
Im(g ◦ z) = <
c2 y 2 + (cx + d)2
if |c| > (y)− 2 . On the other hand, for |c| ≤ (y)− 2 ≤ (δ)− 2 , we have
1 1 1

y
<
(cx + d)2
if the following inequalities hold:
|d| > |c|r + (y −1 ) 2 ≥ |c|r + (δ)− 2 .
1 1

Consequently, Im(g ◦ z) >  only if

|c| ≤ (δ)− 2 and |d| ≤ (δ)− 2 (r + 1),
1 1

and the total number of such pairs (not counting (c, d) = (0, ±1), (±1, 0)) is at
most 4(δ)−1 (r + 1).

6 Discrete group actions

It follows from Lemma 1.1.6 that  = S L(2, Z) is a discrete subgroup of

S L(2, R). This is because:

(1) it is enough to show that for any compact subset A ⊂ h there are only
finitely many g ∈ S L(2, Z) such that (g ◦ A) ∩ A = φ;
(2) every compact subset of A ⊂ h is contained in a rectangle Rr,δ for some
r > 0 and 0 < δ < δ −1 ;
(3) ((αg) ◦ Rr,δ ) ∩ Rr,δ = φ, except for finitely many α ∈ ∞ , g ∈ ∞ \.

To prove (3), note that Lemma 1.1.6 implies that (g ◦ Rr,δ ) ∩ Rr,δ = φ except
for finitely many g ∈ ∞ \. Let S ⊂ ∞ \ denote this finite set of such ele-
ments g. If g ∈ S, then Lemma 1.1.6 tells us that it is because Im(gz) < δ for all
z ∈ Rr,δ . Since Im(αgz) = Im(gz) for α ∈ ∞ , it is enough to show that for each
g ∈ S, there are only finitely many α ∈ ∞ such that ((αg) ◦ Rr,δ ) ∩ Rr,δ = φ.
This last statement follows from the fact that g ◦ Rr,δ itself
 lies in some other
1 m
rectangle Rr  ,δ , and every α ∈ ∞ is of the form α = (m ∈ Z), so
0 1
that
 −1 
α ◦ Rr  ,δ = x + i y  − r  + m ≤ x ≤ r  + m, 0 < δ  ≤ δ  ,

which implies (α ◦ Rr  ,δ ) ∩ Rr,δ = φ for |m| sufficiently large.

Definition 1.1.7 Suppose the group G acts continuously on a connected topo-

logical space X . A fundamental domain for G\X is a connected region D ⊂ X
such that every x ∈ X is equivalent (mod G) to a point in D and such that no
two points in D are equivalent to each other.

Example 1.1.8 A fundamental domain for the action of Z on R of

Example 1.1.2 is given by

Z\R = {0 ≤ x < 1 | x ∈ R}.

The proof of this is left as an easy exercise for the reader.

Example 1.1.9 A fundamental domain for S L(2, Z)\h can be given as the
region D ⊂ h where
  
 1 1

D = z  − ≤ Re(z) ≤ , |z| ≥ 1 ,
2 2

with congruent boundary points symmetric with respect to the imaginary axis.
 1.1 Action of a group on a topological space 7

-1 -1/2 0 1/2 1

 √ 
Note that the vertical line V  := − 12 + i y  y ≥ 23 is equivalent to the
 √ 
vertical line V := 12 + i y  y ≥ 23 under the transformation
 z → z + 1.
 
 z − 2 ≤ Re(z) < 0, |z| = 1 is equivalent to
1
Furthermore, the arc A :=
the reflected arc A := z  0 < Re(z) ≤ 12 , |z| = 1 , under the transformation
z → −1/z. To show that D is a fundamental domain, we must prove:

(1) For any z ∈ h, there exists g ∈ S L(2, Z) such that g ◦ z ∈ D;

(2) If two distinct points z, z  ∈ D are congruent (mod S L(2, Z)) then
Re(z) = ± 12 and z  = z ± 1, or |z| = 1 and z  = −1/z.

We first prove (1). Fix z ∈ h. It follows from Lemma 1.1.6 that for every
 > 0, there are at most finitely many g ∈ S L(2, Z) such that g ◦ z lies in the
strip
  
 1 1
D := w  − ≤ Re(w) ≤ ,  ≤ Im(w) .
2 2

Let B denote the finite set of such g ∈ S L(2, Z). Clearly, for sufficiently small
, the set B contains at least one element. We will show that there is at least
one g ∈ B such that g ◦ z ∈ D. Among these finitely many g ∈ B ,choose one 
0 −1
such that Im(g ◦ z) is maximal in D . If |g ◦ z| < 1, then for S = ,
1 0
8 Discrete group actions

 
1 1
T = , and any m ∈ Z,
0 1
 
−1 Im(g ◦ z)
Im(T Sg ◦ z) = Im
m
= > Im(g ◦ z).
g◦z |g ◦ z|2
This is a contradiction because we can always choose m so that T m Sg ◦ z ∈ D .
So in fact, g ◦ z must be in D.
To complete the verification that D is a fundamental
  domain, it only remains
a b
to prove the assertion (2). Let z ∈ D, g = ∈ S L(2, Z), and assume
c d
that g ◦ z ∈ D. Without loss of generality, we may assume that
y
Im(g ◦ z) = ≥ Im(z),
|cz + d|2
(otherwise just interchange z and g ◦ z and√ use g −1 ). This implies that
|cz + d| ≤ 1 which implies that 1 ≥ |cy| ≥ 23 |c|. This is clearly impossi-
ble if |c| ≥ 2. So we only have to consider the cases c = 0, ±1. If c = 0
then d = ±1 and g is a translation by b. Since − 12 ≤ Re(z), Re(g ◦ z) ≤ 12 ,
this implies that either b = 0 and z = g ◦ z or else b = ±1 and Re(z) = ± 12
while Re(g ◦ z) = ∓ 12 . If c = 1, then |z + d| ≤ 1 implies that d = 0 unless
z = e2πi/3 and d = 0, 1 or z = eπi/3 and d = 0, −1. The case d = 0 implies
that |z| ≤ 1 which implies |z| = 1. Also, in this case, c = 1, d = 0, we
must have b = −1 because ad − bc = 1. Then g ◦ z = a − 1z . It follows that
a = 0. If z = e2πi/3 and d = 1, then we must have a − b = 1. It follows that
g ◦ e2πi/3 = a − 1+e12πi/3 = a + e2πi/3 , which implies that a = 0 or 1. A similar
argument holds when z = eπi/3 and d = −1. Finally, the case c = −1 can be
reduced to the previous case c = 1 by reversing the signs of a, b, c, d.

1.2 Iwasawa decomposition

This monograph focusses on the general linear group G L(n, R) with n ≥ 2.
This is the multiplicative group of all n × n matrices with coefficients in R
and non-zero determinant. We will show that every matrix in G L(n, R) can be
written as an upper triangular matrix times an orthogonal matrix. This is called
the Iwasawa decomposition (Iwasawa, 1949).
The Iwasawa decomposition, in the special case of G L(2, R), states that
every g ∈ G L(2, R) can be written in the form:
   
y x α β d 0
g= (1.2.1)
0 1 γ δ 0 d
 1.2 Iwasawa decomposition 9

where y > 0, x, d ∈ R with d =

 0 and
 
α β
∈ O(2, R),
γ δ
where
 
O(n, R) = g ∈ G L(n, R)  g · gt = I

 on G L(n, R) and g
t
is the orthogonal group. Here I denotes the identity
 matrix
y x
denotes the transpose of the matrix g. The matrix in the decomposition
0 1
 
α β
(1.2.1) is actually uniquely determined. Furthermore, the matrices
γ δ
   
d 0 ±1 0
and are uniquely determined up to multiplication by .
0 d 0 ±1
Note that explicitly,
   
± cos t − sin t 
O(2, R) =  0 ≤ t ≤ 2π .
± sin t cos t
We shall shortly give a detailed proof of (1.2.1) for G L(n, R) with n ≥ 2.
The decomposition (1.2.1) allows us to realize the upper half-plane
 
h = x + i y  x ∈ R, y > 0

as the set of two by two matrices of type

   
y x 
x ∈ R, y > 0 ,
0 1 
or by the isomorphism
 
h ≡ G L(2, R) O(2, R), Z 2 , (1.2.2)

where
⎧⎛ ⎞  ⎫
⎪  ⎪
⎨ d 0  ⎬
⎜ .. ⎟ 
Zn = ⎝ . ⎠  d ∈ R, d = 0
⎪
⎩  ⎪
⎭
0 d 

is the center of G L(n, R), and O(2, R), Z 2  denotes the group generated by
O(2, R) and Z 2 .
The isomorphism (1.2.2) is the starting point for generalizing the classical
theory of modular forms on G L(2, R) to G L(n, R) with n > 2. Accordingly,
we define the generalized upper half-plane hn associated to G L(n, R).
10 Discrete group actions

Definition 1.2.3 Let n ≥ 2. The generalized upper half-plane hn associated

to G L(n, R) is defined to be the set of all n × n matrices of the form z = x · y
where
⎛ ⎞ ⎛  ⎞
1 x1,2 x1,3 · · · x1,n yn−1
⎜ 1 x2,3 · · · x2,n ⎟ ⎜ 
yn−2 ⎟
⎜ ⎟ ⎜ ⎟
⎜ . . ⎟ ⎜ . ⎟
x =⎜ .. .. ⎟ , y=⎜ .. ⎟,
⎜ ⎟ ⎜ ⎟
⎝ 1 xn−1,n ⎠ ⎝ y  ⎠
1
1 1
yi
with xi, j ∈ R for 1 ≤ i < j ≤ n and > 0 for 1 ≤ i ≤ n − 1.
To simplify later formulae and notation in this book, we will always express
y in the form:
⎛ ⎞
y1 y2 · · · yn−1
⎜ y1 y2 · · · yn−2 ⎟
⎜ ⎟
⎜ .. ⎟
y=⎜ . ⎟,
⎜ ⎟
⎝ y1 ⎠
1
with yi > 0 for 1 ≤ i ≤ n − 1. Note that this can always be done since yi = 0
for 1 ≤ i ≤ n − 1.
Explicitly, x is an upper triangular matrix with 1s on the diagonal and y
is a diagonal matrix beginning with a 1 in the lowest right entry. Note that x
is parameterized by n · (n − 1)/2 real variables xi, j and y is parameterized by
n − 1 positive real variables yi .
Example 1.2.4 The generalized upper half plane h3 is the set of all matrices
z = x · y with
⎛ ⎞ ⎛ ⎞
1 x1,2 x1,3 y1 y2 0 0
x = ⎝0 1 x2,3 ⎠ , y = ⎝ 0 y1 0⎠,
0 0 1 0 0 1
where x1,2 , x1,3 , x2,3 ∈ R, y1 , y2 > 0. Explicitly, every z ∈ h3 can be written
in the form
⎛ ⎞
y1 y2 x1,2 y1 x1,3
z=⎝ 0 y1 x2,3 ⎠ .
0 0 1
Remark 1.2.5 The generalized upper half-plane h3 does not have a com-
plex structure. Thus h3 is quite different from h2 , which does have a complex
structure.
 1.2 Iwasawa decomposition 11

Proposition 1.2.6 Fix n ≥ 2. Then we have the Iwasawa decomposition:

G L(n, R) = hn · O(n, R) · Z n ,
i.e., every g ∈ G L(n, R) may be expressed in the form

g = z · k · d, (· denotes matrix multiplication)

where z ∈ hn is uniquely determined, k ∈ O(n, R), and d ∈ Z n is a non-zero

diagonal matrix which lies in the center of G L(n, R). Further, k and d are also
uniquely determined up to multiplication by ±I where I is the identity matrix
on G L(n, R).

Remark Note that for every n = 1, 2, 3, . . . , we have Z n ∼

= R× . We shall,
henceforth, write
hn ∼
= G L(n, R)/(O(n, R) · R× ).

Proof Let g ∈ G L(n, R). Then g · tg is a positive definite non–singular

matrix. We claim there exists u, ∈ G L(n, R), where u is upper triangular with
1s on the diagonal and is lower triangular with 1s on the diagonal, such that

u · g · tg = · d (1.2.7)

with
⎛ ⎞
d1
⎜ .. ⎟
d=⎝ . ⎠, d1 , . . . , dn > 0.
dn
 
a b
For example, consider n = 2, and g = . Then
c d
     
a b a c a 2 + b2 ac + bd
g· g= t
· = .
c d b d ac + bd c2 + d 2
 
1 t
If we set u = , then u satisfies (1.2.7) if
0 1
   2   
1 t a + b2 ac + bd ∗ 0
· = ,
0 1 ac + bd c2 + d 2 ∗ ∗

so that we may take t = (−ac − bd)/(c2 + d 2 ). More generally, the upper

triangular matrix u will have n(n − 1)/2 free variables, and we will have to
12 Discrete group actions

solve n(n − 1)/2 equations to satisfy (1.2.7). This system of linear equations
has a unique solution because its matrix g · tg is non–singular.
It immediately follows from (1.2.7) that u −1 d = g · tg = d · t (t u)−1 , or
equivalently

" · d#$· u% = "u · #$

d · t %
t
= d.
lower  upper 

The above follows from the fact that a lower triangular matrix can only equal an
upper triangular matrix if it is diagonal, and that this diagonal matrix must be
d by comparing diagonal entries. The entries di > 0 because g · tg is positive
definite.
Consequently d = d(t u)−1 . Substituting this into (1.2.7) gives

u · g · tg · t u = d = a −1 · (t a)−1

for
⎛ − 12
⎞
d1
⎜ ⎟
a=⎜
⎝
..
. ⎟.
⎠
− 12
dn

Hence aug · (tg · t u · t a) = I so that aug ∈ O(n, R). Thus, we have expressed
g in the form

g = (au)−1 · (aug),

from which the Iwasawa decomposition immediately follows after dividing and
−1
multiplying by the scalar dn 2 to arrange the bottom right entry of (au)−1 to
be 1.
It only remains to show the uniqueness of the Iwasawa decomposition.
Suppose that zkd = z  k  d  with z, z  ∈ hn , k, k  ∈ O(n, R), d, d  ∈ Z n . Then,
since the only matrices in hn and O(n, R) which lie in Z n are ±I where I is the
identity matrix, it follows that d  = ±d. Further, the only matrix in hn ∩ O(n, R)
is I . Consequently z = z  and k = ±k  .

We shall now work out some important instances of the Iwasawa decompo-
sition which will be useful later.
 1.2 Iwasawa decomposition 13

Proposition 1.2.8 Let I denote the identity matrix on G L(n, R), and for every
1 ≤ j < i ≤ n, let E i, j denote the matrix with a 1 at the {i, j}th position and
zeros elsewhere. Then, for an arbitrary real number t, we have
⎛ ⎞
1
⎜ . ⎟
⎜ .. ⎟
⎜ ⎟
⎜ ⎟
⎜ 1
··· 2 t
⎟
⎜ ⎟ 
1 1
(t 2 + 1) 2 (t + 1) 2
⎜ ⎟
I + t E i, j = ⎜ ..
.
.. ⎟ mod (O(n, R) · R× ) ,
⎜ . ⎟
⎜ 1 ⎟
⎜ (t 2 + 1) 2 ⎟
⎜ ⎟
⎜ .. ⎟
⎝ . ⎠
1

1
where, in the above matrix, 1
1 occurs at position { j, j}, (t 2 + 1) 2 occurs
(t 2 + 1) 2
at position {i, i}, all other diagonal entries are ones, t
1 occurs at position
(t 2 + 1) 2
{ j, i}, and, otherwise, all other entries are zero.

Proof Let g = I + t E i, j . Then

g · tg = (I + t E i, j ) · (I + t E j,i ) = I + t E i, j + t E j,i + t 2 E i,i .

If we define a matrix u = I − (t/(t 2 + 1))E j,i , then u · g · tg · t u must be a

diagonal matrix d. Setting d = a −1 · (t a)−1 , we may directly compute:

t2
u · g · tg · t u = I + t 2 E i,i − E j, j ,
t2 +1
t
u −1 = I + E j,i ,
t2 + 1
 
1 &
a −1 =I+ √ − 1 E j, j + t 2 + 1 − 1 E i,i .
t +1
2

Therefore,
 
1 & t
u −1 a −1 = I + √ − 1 E j, j + t 2 + 1 − 1 E i,i + √ E j,i .
t +1
2 t +1
2

As in the proof of Proposition 1.2.6, we have g = u −1 · a −1

(mod (O(n, R), R× )).

14 Discrete group actions

Proposition 1.2.9 Let n ≥ 2, and let z = x y ∈ hn have the form given in

Definition 1.2.3. For i = 1, 2, . . . , n − 1, define
⎛ ⎞
1
⎜ .. ⎟
⎜ . ⎟
⎜ ⎟
⎜ ⎟
⎜ 0 1 ⎟
ωi = ⎜ ⎟,
⎜ 1 0 ⎟
⎜ ⎟
⎜ .. ⎟
⎝ . ⎠
1

the n ×n identity matrix except for the ith and (i + 1)th rows where we
to be 
0 1
have on the diagonal. Then
1 0
⎛    ⎞⎛ ⎞
1 x1,2 x1,3 ··· x1,n y1 y2 · · · yn−1


⎜ 1 
x2,3 ··· 
x2,n ⎟⎜ y1 y2 · · · yn−2
 ⎟
⎜ ⎟⎜ ⎟
⎜ .. .. ⎟⎜ .. ⎟
ωi z ≡ ⎜ . . ⎟·⎜ . ⎟
⎜ ⎟⎜ ⎟
⎝ 1 
xn−1,n ⎠⎝ y1 ⎠
1 1

mod (O(n, R) · R× ) , where yk = yk except for k = n − i + 1, n − i, n − i − 1,
in which case
yn−i '
 
yn−i = 2 , yn−i±1 = yn−i±1 · 2
xi,i+1 + yn−i
2
,
xi,i+1 + yn−i
2


and xk, = xk, except for = i, i + 1, in which case

 
2
xi− j,i yn−i + xi− j,i+1 xi,i+1
xi− j,i = x i− j,i+1 − x i− j,i x i,i+1 , xi− j,i+1 = ,
2
xi,i+1 + yn−i
2

for j = 1, 2, . . . , i − 2.

Proof Brute force computation which is omitted.

Proposition 1.2.10 The group G L(n, Z) acts on hn .

Proof Recall the definition of a group acting on a topological space given in

Definition 1.1.1. The fact that G L(n, Z) acts on G L(n, R) follows immediately
from the fact that G L(n, Z) acts on the left on G L(n, R) by matrix multiplication
and that we have the realization hn = G L(n, R)/(O(n, R) · R× ), as a set of
cosets, by the Iwasawa decomposition given in Proposition 1.2.6.

 1.3 Siegel sets 15

1.3 Siegel sets

We would like to show that  n = G L(n, Z) acts discretely on the generalized
upper half-plane hn defined in Definition 1.2.3. This was already proved for
n = 2 in Lemma 1.1.6, but the generalization to n > 2 requires more subtle
arguments. In order to find an approximation to a fundamental domain for
G L(n, Z)\hn , we shall introduce for every t, u ≥ 0 the Siegel set t,u .

Definition 1.3.1 Let a, b ≥ 0 be fixed. We define the Siegel set a,b ⊂ hn to

be the set of all
⎛ ⎞ ⎛ ⎞
1 x1,2 x1,3 ··· x1,n y1 y2 · · · yn−1
⎜ 1 x2,3 ··· x2,n ⎟ ⎜ y1 y2 · · · yn−2 ⎟
⎜ ⎟ ⎜ ⎟
⎜ .. .. ⎟ ⎜ .. ⎟
⎜ . . ⎟·⎜ . ⎟
⎜ ⎟ ⎜ ⎟
⎝ 1 xn−1,n ⎠ ⎝ y1 ⎠
1 1

with |xi, j | ≤ b for 1 ≤ i < j ≤ n, and yi > a for 1 ≤ i ≤ n − 1.

Let  n = G L(n, Z) and ∞ n

⊂  n denote the subgroup of upper triangular
matrices with 1s on the diagonal. We have shown in Proposition 1.2.10 that  n
acts on hn . For g ∈  n and z ∈ hn , we shall denote this action by g ◦ z. The
following proposition proves that the action is discrete and that  √3 , 1 is a good
2 2
approximation to a fundamental domain.

Proposition 1.3.2 Fix an integer n ≥ 2. For any z ∈ hn there are only finitely
many g ∈  n such that g ◦ z ∈  √3 , 1 . Furthermore,
2 2

(
G L(n, R) = g ◦  √3 , 1 . (1.3.3)
2 2
g∈ n
√
Remarks The bound 23 is implicit in the work of Hermite, and a proof can
be found in (Korkine and Zolotareff, 1873). The first part of Proposition 1.3.2
is a well known theorem due to Siegel (1939). For the proof, we follow the
exposition of Borel and Harish-Chandra (1962).

Proof of Proposition 1.3.2 In order to prove (1.3.3), it is enough to show that

(
S L(n, R) = g ◦  ∗√3 1 , (1.3.4)
2 ,2
g∈S L(n,Z)

∗
where t,u denotes the subset of matrices t,u · Z n which have determinant 1
∗ ∗
and ◦ denotes the action of S L(n, Z) on 0,∞ . Note that every element in a,b
16 Discrete group actions

is of the form
⎛ ⎞ ⎛ ⎞
1 x1,2 x1,3 ··· x1,n dy1 y2 · · · yn−1
⎜ 1 x2,3 ··· x2,n ⎟ ⎜ dy1 y2 · · · yn−2 ⎟
⎜ ⎟ ⎜ ⎟
⎜ .. .. ⎟ ⎜ . ⎟
⎜ . . ⎟·⎜ .. ⎟
⎜ ⎟ ⎜ ⎟
⎝ 1 xn−1,n ⎠ ⎝ dy1 ⎠
1 d
(1.3.5)
where the determinant
⎛ ⎞
dy1 y2 · · · yn−1
⎜ dy1 y2 · · · yn−2 ⎟
⎜ ⎟
⎜ .. ⎟
Det ⎜ . ⎟ = 1,
⎜ ⎟
⎝ dy1 ⎠
d
so that

−1/n

n−1
d= yin−i .
i=1

In view of the Iwasawa decomposition of Proposition 1.2.6, we may identify

∗
0,∞ as the set of coset representatives S L(n, R)/S O(n, R), where S O(n, R)
denotes the subgroup O(n, R) ∩ S L(n, R).

In order to prove (1.3.4), we first introduce some basic notation. Let

e1 = (1, 0, . . . , 0), e2 = (0, 1, . . . , 0), ..., en = (0, 0, . . . , 1),

denote the canonical basis for Rn . For 1 ≤ i ≤ n and any matrix g ∈

G L(n, R), let ei · g denote the usual multiplication of a 1 × n matrix with
an n × n'matrix. For an arbitrary v = (v1 , v2 , . . . , vn ) ∈ Rn , define the norm:
||v|| := v12 + v22 + · · · + vn2 . We now introduce a function

φ : S L(n, R) → R>0

from S L(n, R) to the positive real numbers. For all g = (gi, j )1≤i, j≤n in S L(n, R)
we define
'
φ(g) := ||en · g|| = gn,12
+ gn,2
2
+ · · · + gn,n
2 .

Claim The function φ is well defined on the quotient space

S L(n, R)/S O(n, R).
 1.3 Siegel sets 17

To verify the claim, note that for k ∈ S O(n, R), and v ∈ Rn , we have
& √ √
||v · k|| = (v · k) · t (v · k) = v · k · t k · t v = v · t v = ||v||.

This immediately implies that φ(gk) = φ(g), i.e., the claim is true.
∗
Note that if z ∈ 0,∞ is of the form (1.3.5), then

−1/n

n−1
φ(z) = d = yi(n−i) . (1.3.6)
i=1

∗
Now, if z ∈ 0,∞ is fixed, then

en · S L(n, Z) · z ⊂ Ze1 + · · · + Zen − {(0, 0, . . . , 0)} · z, (1.3.7)

where · denotes matrix multiplication. The right-hand side of (1.3.7) consists

of non–zero points of a lattice in Rn . This implies that φ achieves a positive
minimum on the coset S L(n, Z) · z. The key to the proof of Proposition 1.3.2
will be the following lemma from which Proposition 1.3.2 follows immediately.
∗
Lemma 1.3.8 Let z ∈ 0,∞ . Then the minimum of φ on S L(n, Z) ◦ z is
∗√
achieved at a point of  3 1 .
2 ,2

Proof It is enough to prove that the minimum of φ is achieved at a point of

 ∗√3 because we can always translate by an upper triangular matrix
2 ,∞

⎛ ⎞
1 u 1,2 u 1,3 ··· u 1,n
⎜ 1 u 2,3 ··· u 2,n ⎟
⎜ ⎟
⎜ .. .. ⎟
u=⎜ . . ⎟ ∈ S L(n, Z)
⎜ ⎟
⎝ 1 u n−1,n ⎠
1

to arrange that the minimum of φ lies in  ∗√3 1 . This does not change the
2 ,2
value of φ because of the identity φ(u · z) = ||en · u · z|| = ||en · z||. We shall
use induction on n. We have already proved a stronger statement for n = 2
in Example 1.1.9. Fix γ ∈ S L(n, Z) such that φ(γ ◦ z) is minimized. We set
γ ◦ z = x · y with
⎛ ⎞ ⎛ ⎞
1 x1,2 x1,3 · · · x1,n dy1 y2 · · · yn−1
⎜ 1 x2,3 · · · x2,n ⎟ ⎜ dy1 y2 · · · yn−2 ⎟
⎜ ⎟ ⎜ ⎟
⎜ .. . ⎟ ⎜ .. ⎟
x =⎜ . .. ⎟ , y = ⎜ . ⎟,
⎜ ⎟ ⎜ ⎟
⎝ 1 xn−1,n ⎠ ⎝ dy1 ⎠
1 d
18 Discrete group actions

)n−1 n−i −1/n √

with d = ( i=1 yi ) as before. We must show yi ≥ 2
3
for i = 1, 2, . . . ,
n − 1. The proof proceeds in 3 steps.
√
Step 1 y1 ≥ 3
.
2 ⎛ ⎞
In−2
This follows from the action of α := ⎝ 0 −1 ⎠ on γ ◦ z. Here In−2
1 0
denotes the identity (n − 2) × (n − 2)–matrix. First of all

φ(α ◦ γ ◦ z) = ||en · α ◦ γ ◦ z|| = ||en−1 · x · y|| = ||(en−1 + xn−1,n en ) · y||

'
= d y12 + xn−1,n
2
.

Since |xn−1,n | ≤ 12 we see that φ(αγ z)2 ≤ d 2 (y12 +14 ). On the other hand, the
assumption of minimality forces φ(γ z)2 = d 2 ≤ d 2 y12 + 14 . This implies that
√
y1 ≥ 2
3
.
 
 g 1
Step 2 Let g ∈ S L(n − 1, Z), g = . Then φ(gγ z) = φ(γ z).
0 1
This follows immediately from the fact that en · g = en .
√
Step 3 yi ≥ 3
for i = 2, 3, . . . , n − 1.
2    
z ·d ∗
Let us write γ ◦ z = with z  ∈ S L(n − 1, R) and d  ∈ Z n−1
d
a suitable diagonal matrix. By induction, there exists g  ∈ S L(n − 1, Z) such
that g  ◦ z  = x  · y  ∈  ∗√3 1 ⊂ hn−1 , the Siegel set for G L(n − 1, R). This is
2 ,2
equivalent to the fact that
⎛ ⎞
an−1
⎜ an−2 ⎟
⎜ ⎟
y = ⎜ .. ⎟
⎝ . ⎠
a1

and
√
a j+1 3
≥ for j = 1, 2, . . . , n − 2. (1.3.9)
aj 2
 
g 0
Define g := ∈ S L(n, Z). Then
0 1
       
g 0 z ·d ∗ g ◦ z · d  ∗
g◦γ ◦z = ◦ = = x  · y  ,
0 1 d 0 d
 1.4 Haar measure 19

   
yd  0 x ∗
where y  = , x  = . The inequalities (1.3.9) applied to
0 d 0 1
⎛ ⎞
y1 y2 · · · yn−1 d
  ⎜ .. ⎟
yd  0 ⎜ . ⎟
y  = =⎜ ⎟,
0 d ⎝ y1 d ⎠
d
√
imply that yi ≥ 3
for i = 2, 3, . . . , n − 1. Step 2 insures that multiplying by
2 √
g on the left does not change the value of φ(γ z). Step 1 gives y1 ≥ 2
3
.

1.4 Haar measure

Let n ≥ 2. The discrete subgroup S L(n, Z) acts on S L(n, R) by left multipli-
cation. The quotient space S L(n, Z)\S L(n, R) turns out to be of fundamental
importance in number theory. Now, we turn our attention to a theory of inte-
gration on this quotient space.
We briefly review the theory of Haar measure and integration on locally
compact Hausdorff topological groups. Good references for this material are
(Halmos, 1974), (Lang, 1969), (Hewitt and Ross, 1979). Excellent introductary
books on matrix groups and elementary Lie theory are (Curtis, 1984), (Baker,
2002), (Lang, 2002).
Recall that a topological group G is a topological space G where G is also
a group and the map

(g, h) → g · h −1

of G × G onto G is continuous in both variables. Here · again denotes the

internal group operation and h −1 denotes the inverse of the element h. The
assumption that G is locally compact means that every point has a compact
neighborhood. Recall that G is termed Hausdorff provided every pair of distinct
elements in G lie in disjoint open sets.

Example 1.4.1 The general linear group G L(n, R) is a locally compact

Hausdorff topological group.

Let gl(n, R) denote the Lie algebra of G L(n, R). Viewed as a set, gl(n, R)
is just the set of all n × n matrices with coefficients in R. We assign a topology
20 Discrete group actions

to gl(n, R) by identifying every matrix

⎛ ⎞
g1,1 g1,2 ··· g1,n
⎜ g2,1 g2,2 ··· g2,n ⎟
⎜ ⎟
g=⎜ . .. ⎟
⎝ .. ··· . ⎠
gn,1 gn,2 ··· gn,n
with a point
2
(g1,1 , g1,2 , . . . , g1,n , g2,1 , g2,2 , . . . , g2,n , . . . , gn,n ) ∈ Rn .

This identification is a one–to–one correspondence. One checks that gl(n, R) is

a locally compact Hausdorff topological space under the usual Euclidean topol-
2
ogy on Rn . The determinant function Det : gl(n, R) → R is clearly continuous.
It follows that

G L(n, R) = gl(n, R) − Det−1 (0)

must be an open set since {0} is closed. Also, the operations of addition and
multiplication of matrices in gl(n, R) are continuous maps from

gl(n, R) × gl(n, R) → gl(n, R).

The inverse map

Inv : G L(n, R) → G L(n, R),

given by Inv(g) = g −1 for all g ∈ G L(n, R), is also continuous since each entry
of g −1 is a polynomial in the entries of g divided by Det(g). Thus, G L(n, R)
is a topological subspace of gl(n, R) and we may view G L(n, R) × G L(n, R)
as the product space. Since the multiplication and inversion maps: G L(n, R) ×
G L(n, R) → G L(n, R) are continuous, it follows that G L(n, R) is a topological
group.
By a left Haar measure on a locally compact Hausdorff topological group
G, we mean a positive Borel measure (Halmos, 1974)

µ : {measurable subsets of G} → R+ ,

which is left invariant under the action of G on G via left multiplication. This
means that for every measurable set E ⊂ G and every g ∈ G, we have

µ(g E) = µ(E).

In a similar manner, one may define a right Haar measure. If every left invariant
Haar measure on G is also a right invariant Haar measure, then we say that G
is unimodular.
 1.4 Haar measure 21

Given a left invariant Haar measure µ on G, one may define (in the usual
manner) a differential one-form dµ(g), and for compactly supported functions
f : G → C an integral

f (g) dµ(g),
G

which is characterized by the fact that


dµ(g) = µ(E)
E

for every measurable set E. We shall also refer to dµ(g) as a Haar measure.
The fundamental theorem in the subject is due to Haar.

Theorem 1.4.2 (Haar) Let G be a locally compact Hausdorff topological

group. Then there exists a left Haar measure on G. Further, any two such Haar
measures must be positive real multiples of each other.

We shall not need this general existence theorem, because in the situations
we are interested in, we can explicitly construct the Haar measure and Haar
integral. For unimodular groups, the uniqueness of Haar measure follows easily
from Fubini’s theorem. The proof goes as follows. Assume we have two Haar
measures µ, ν on G, which are both left and right invariant. Let h : G → C be
a compactly supported function satisfying

h(g) dµ(g) = 1.
G

Then for an arbitrary compactly supported function f : G → C,

  
f (g)dν(g) = h(g  )dµ(g  ) f (g)dν(g)
G
G  G

= h(g  ) f (g)dν(g)dµ(g  )
 
G G

= h(g  ) f (g · g  ) dν(g) dµ(g  )

G G
= h(g  ) f (g · g  ) dµ(g  ) dν(g)
 
G G

= h(g −1 · g  ) f (g  ) dµ(g  ) dν(g)

 
G G

= h(g −1 · g  ) f (g  ) dν(g) dµ(g  )


G G

=c· f (g  ) dµ(g  )
G
* −1
where c = G h(g )dν(g).
22 Discrete group actions

Proposition 1.4.3 For n = 1, 2, . . . , let

⎛ ⎞
g1,1 g1,2 · · · g1,n
⎜ g2,1 g2,2 · · · g2,n ⎟
⎜ ⎟
g=⎜ . .. ⎟ ∈ G L(n, R),
⎝ .. ··· . ⎠
gn,1 gn,2 ··· gn,n
where g1,1 , g1,2 , . . . , g1,n , g2,1 , . . . , gn,n are n 2 real variables. Define
)
dgi, j
1≤i, j≤n
dµ(g) := , (wedge product of differential one-forms)
Det(g)n
where dgi, j denotes the usual differential one–form on R and Det(g) denotes
the determinant of the matrix g. Then dµ(g) is the unique left–right invariant
Haar measure on G L(n, R).
Proof Every matrix in G L(n, R) may be expressed as a product of a diagonal
matrix in Z n and matrices of the form x̃r,s (with 1 ≤ r, s ≤ n) where x̃r,s denotes
the matrix with the real number xr,s at position r, s, and, otherwise, has 1s on
the diagonal and zeros off the diagonal. It is easy to see that
dµ(g) = dµ(ag)
for a ∈ Z n . To complete the proof, it is, therefore, enough to check that
dµ(x̃r,s · g) = dµ(g · x̃r,s ) = dµ(g),
for all 1 ≤ r, s ≤ n. We check the left invariance and leave the right invariance
to the reader.
It follows from the definition that in the case r = s,
⎛ ⎞


⎜ ) ⎟ )
⎝ dgi, j ⎠ d(gr, j + gs, j xr,s )
1≤i, j≤n 1≤ j≤n
i=r
dµ(x̃r,s · g) = .
Det(x̃r,s · g)n
First of all,
Det(x̃r,s · g) = Det(x̃r,s ) · Det(g) = Det(g)
because Det(x̃r,s ) = 1.
Second, for any 1 ≤ j ≤ n,
⎛ ⎞
⎜  ⎟
⎝ dgi, j ⎠ ∧ dgs, j = 0
1≤i, j≤n
i=r
 1.5 Invariant measure on coset spaces 23

⎛ ⎞
⎜ ) ⎟
because gs, j also occurs in the product ⎝ dgi, j ⎠ and dgs, j ∧ dgs, j = 0.
1≤i< j≤n
i=r
Consequently, the measure is invariant under left multiplication by x̃r,s .

On the other hand, if r = s, then

⎛ ⎞


⎜ ) ⎟ )
⎝ dgi, j ⎠ (xr,s · dgr, j )
1≤i, j≤n 1≤ j≤n
i=r
dµ(x̃r,s · g) =
Det(x̃r,s · g)n
n
xr,s
= dµ(g) ·
Det(x̃r,s )n
= dµ(g).

1.5 Invariant measure on coset spaces

This monograph focusses on the coset space
G L(n, R)/(O(n, R) · R× ).
We need to establish explicit invariant measures on this space. The basic prin-
ciple which allows us to define invariant measures on coset spaces, in general,
is given in the following theorem.
Theorem 1.5.1 Let G be a locally compact Hausdorff topological group, and
let H be a compact closed subgroup of G. Let µ be
* a Haar measure on G, and let
ν be a Haar measure on H , normalized so that H dν(h) = 1. Then there exists
a unique (up to scalar multiple) quotient measure µ̃ on G/H . Furthermore
   
f (g) dµ(g) = f (gh) dν(h) d µ̃(g H ),
G G/H H

for all integrable functions f : G → C.

Proof For a proof see (Halmos, 1974). We indicate, however, why the formula
in Theorem 1.5.1 holds. First of all note that if f : G → C, is an integrable
function on G, and if we define a new function, f H : G → C, by the recipe

f H (g) := f (gh) dν(h),
H

then f (gh) = f (g) for all h ∈ H. Thus, f H is well defined on the coset
H H

space G/H. We write f H (g) = f H (g H ), to stress that f H is a function on

24 Discrete group actions

the coset space. For any measurable subset E ⊂ G/H , we may easily choose
a measurable function δ E : G → C so that

1 if g H ∈ E,
δ E (g) = δ EH (g H ) =
0 if g H ∈ E.

We may then define an H –invariant quotient measure µ̃ satisfying:

 
µ̃(E) = δ E (g) dµ(g) = δ EH (g H ) d µ̃(g H ),
G G/H

and
 
f (g) dµ(g) = f H (g H ) d µ̃(g H ),
G G/H

for all integrable functions f : G → C.

Remarks There is an analogous version of Theorem 1.5.1 for left coset spaces
H \G. Note that we are not assuming that H is a normal subgroup of G. Thus
G/H (respectively H \G) may not be a group.

Example 1.5.2 (Left invariant measure on G L(n, R)/(O(n, R) · R× ))

For n ≥ 2, we now explicitly construct a left invariant measure on the

generalized upper half-plane hn = G L(n, R)/(O(n, R) · R× ). Returning to the
Iwasawa decomposition (Proposition 1.2.6), every z ∈ hn has a representation
in the form z = x y with
⎛ ⎞ ⎛ ⎞
1 x1,2 x1,3 · · · x1,n y1 y2 · · · yn−1
⎜ 1 x2,3 · · · x2,n ⎟ ⎜ y1 y2 · · · yn−2 ⎟
⎜ ⎟ ⎜ ⎟
⎜ .. .. ⎟ , ⎜ .. ⎟
x =⎜ . . ⎟ y=⎜ . ⎟,
⎜ ⎟ ⎜ ⎟
⎝ 1 xn−1,n ⎠ ⎝ y1 ⎠
1 1

with xi, j ∈ R for 1 ≤ i < j ≤ n and yi > 0 for 1 ≤ i ≤ n − 1. Let d ∗ z denote

the left invariant measure on hn . Then d ∗ z has the property that

d ∗ (gz) = d ∗ z

for all g ∈ G L(n, R).

 1.5 Invariant measure on coset spaces 25

Proposition 1.5.3 The left invariant G L(n, R)–measure d ∗ z on hn can be

given explicitly by the formula
d∗z = d∗x d∗ y
where
 
n−1
d∗x = d xi, j , d∗ y = yk−k(n−k)−1 dyk . (1.5.4)
1≤i< j≤n k=1

 
y x
For example, for n = 2, with z = , we have d ∗ z = d xd y
y2
, while for
0 1
n = 3 with
⎛ ⎞
y1 y2 x1,2 y1 x1,3
z= ⎝ 0 y1 x2,3 ⎠ ,
0 0 1

we have
dy1 dy2
d ∗ z = d x1,2 d x1,3 d x2,3 .
(y1 y2 )3

Proof We sketch the proof. The group G L(n, R) is generated by diagonal

matrices, upper triangular matrices with 1s on the diagonal, and the Weyl group
Wn which consists of all n × n matrices with exactly one 1 in each row and
column and zeros everywhere else. For example,
   
1 0 0 1
W2 = , ,
0 1 1 0
⎧⎛ ⎞ ⎛ ⎞ ⎛ ⎞
⎨ 1 0 0 1 0 0 0 1 0
W3 = ⎝ 0 1 0 ⎠ , ⎝ 0 0 1 ⎠ , ⎝ 1 0 0 ⎠ ,
⎩
0 0 1 0 1 0 0 0 1
⎛ ⎞ ⎛ ⎞ ⎛ ⎞⎫
0 1 0 0 0 1 0 0 1 ⎬
⎝0 0 1⎠,⎝1 0 0⎠,⎝0 1 0⎠ .
⎭
1 0 0 0 1 0 1 0 0

Note that the Weyl group Wn has order n! and is simply the symmetric group on
n symbols. It is clear that d ∗ (gz) = d ∗ z if g is an upper triangular matrix with
1s on the diagonal. This is because the measures d xi, j (with 1 ≤ i < j ≤ n) are
all invariant under translation. It is clear that the differential d ∗ z is Z n -invariant
where Z n ∼= R× denotes the center of G L(n, R). So, without loss of generality,
26 Discrete group actions

we may define a diagonal matrix a with its lower-right entry to be one:

⎛ ⎞
a1 a2 · · · an−1
⎜ a1 a2 · · · an−2 ⎟
⎜ ⎟
⎜ .. ⎟
a=⎜ . ⎟.
⎜ ⎟
⎝ a1 ⎠
1

Then

az = ax y = (axa −1 ) · ay
⎛ ⎞
1 an−1 x1,2 an−1 an−2 x1,3 · · · an−1 · · · a1 x1,n
⎜ 1 an−2 x2,3 · · · an−2 · · · a1 x2,n ⎟
⎜ ⎟
⎜ .. .. ⎟
=⎜ . . ⎟
⎜ ⎟
⎝ 1 a1 xn−1,n ⎠
1
⎛ ⎞
a1 y1 · · · an−1 yn−1
⎜ .. ⎟
⎜ . ⎟
×⎜ ⎟.
⎝ a1 y1 ⎠
1
n−1 
∗ −1
)
Thus d (axa ) = akk(n−k) d ∗ x. It easily follows that
k=1

d ∗ (az) = d ∗ (axa −1 · ay) = d ∗ z.

It remains to check the invariance of d ∗ z under the Weyl group Wn . Now, if

w ∈ Wn and
⎛ ⎞
dn
⎜ dn−1 ⎟
⎜ ⎟
d=⎜ .. ⎟ ∈ G L(n, R)
⎝ . ⎠
d1

is a diagonal matrix, then wdw −1 is again a diagonal matrix whose diagonal

entries are a permutation of {d1 , d2 , . . . , dn }. The Weyl group is generated
by the transpositions ωi (i = 1, 2, . . . n − 1) given in Proposition 1.2.9 which
interchange (transpose) di and di+1 when d is conjugated by ωi . After a tedious
calculation using Proposition 1.2.9 one checks that d ∗ (ωi z) = d ∗ z.

 1.6 Volume of S L(n, Z)\S L(n, R)/S O(n, R) 27

1.6 Volume of S L(n, Z)\S L(n, R)/S O(n, R)

Following earlier work of Minkowski, Siegel (1936) showed that the volume
of

S L(n, Z)\S L(n, R)/S O(n, R) ∼

= S L(n, Z)\G L(n, R)/(O(n, R) · R× )
∼
= S L(n, Z)\hn ,
can be given in terms of

ζ (2) · ζ (3) · · · ζ (n)

where ζ (s) is the Riemann zeta function. The fact that the special values (taken
at integral points) of the Riemann zeta function appear in the formula for the
volume is remarkable. Later, Weil (1946) found another method to prove such
results based on a direct application of the Poisson summation formula. A vast
generalization of Siegel’s computation of fundamental domains for the case of
arithmetic subgroups acting on Chevalley groups was obtained by Langlands
(1966). See also (Terras, 1988) for interesting discussions on the history of this
subject.
The main aim of this section is to explicitly compute the volume

d ∗ z,
S L(n,Z)\S L(n,R)/S O(n,R)
∗
where d z is the left–invariant measure given in Proposition 1.5.3. We follow
the exposition of Garret (2002).

Theorem 1.6.1 Let n ≥ 2. As in Proposition 1.5.3, fix

 
n−1
d∗z = d xi, j yk−k(n−k)−1 dyk
1≤i< j≤n k=1

to be the left S L(n, R)–invariant measure on hn = S L(n, R)/S O(n, R). Then
 n
ζ ( )
d ∗ z = n 2n−1 · ,
S L(n,Z)\hn =2
Vol(S −1 )

where
√
2( π)
Vol(S −1 ) =
 ( /2)
+
∞
denotes the volume of the ( − 1)–dimensional sphere S −1 and ζ ( ) = n −
n=1
denotes the Riemann zeta function.
28 Discrete group actions

Proof for the case of S L(2, R) We first prove the theorem for S L(2, R). The
more general result will follow by induction. Let K = S O(2, R) denote the
maximal compact subgroup of S L(2, R). We use the Iwasawa decomposition
which says that

  
  ,
1 
1 x y2 0 
S L(2, R)/K ∼
= z =  x ∈ R, y > 0 .
y− 2 
1
0 1 0

Let f : R2 /K → C be an arbitrary smooth compactly supported function.

Then, by definition, f ((u, v) · k) = f ((u, v)) for all (u, v) ∈ R2 and all k ∈ K .
We can define a function F : S L(2, R)/K → C by letting

F(z) := f ((m, n) · z).
(m,n)∈Z2

 
a b
If γ = ∈ S L 2(Z), then
c d

   
a b
F(γ z) = f (m, n) · ·z
c d
(m,n)∈Z2


= f (ma + nc, mb + nd) · z
(m,n)∈Z2
= F(z).

Thus, F(z) is S L(2, Z)–invariant.

Note that we may express

-  .

{(m, n) ∈ Z2 } = (0, 0) ∪ · (0, 1) · γ  0 < ∈ Z, γ ∈ ∞ \S L(2, Z) ,
(1.6.2)

where
   
1 r 
∞ =  r ∈Z .
0 1 

We now integrate F over \h2 , where h2 = S L(2, R)/K ,  = S L(2, Z),

and d xd y/y 2 is the invariant measure on h2 given in Proposition 1.5.3. It
 1.6 Volume of S L(n, Z)\S L(n, R)/S O(n, R) 29

immediately follows from (1.6.2) that


d xd y
F(z) 2 = f ((0, 0)) · Vol(\h2 )
y
\h2



 d xd y
+ f (0, 1) · γ · z
>0 γ ∈∞ \
y2
\h2

  d xd y
= f ((0, 0)) · Vol(\h2 ) + 2 f (0, 1) · z .
>0
y2
∞ \h2

 
The factor 2 occurs because −1 −1 acts trivially on h2 . We easily observe
that


1 
 y2 0  1

f (0, 1) · z = f (0, 1) · − 12
= f 0, y − 2 .
0 y

It follows, after making the elementary transformations

y → 2 y, y → y −2

that
 ∞
d xd y
F(z) 2 = f ((0, 0)) · Vol(\h2 ) + 22 ζ (2) f ((0, y)) ydy. (1.6.3)
y
\h2 0

Now, the function

 f ((u, v)) isinvariant under multiplication by k ∈ K on the
sin θ − cos θ
right. Since ∈ K , we see that
cos θ sin θ

f ((0, y)) = f ((y cos θ, y sin θ ))

for any 0 ≤ θ ≤ 2π. Consequently

 ∞  2π  ∞
1
f ((0, y)) ydy = f ((y cos θ, y sin θ)) dθ ydy
2π 0
0
 0
1
= f ((u, v)) dudv
2π
R2
1 ˆ
= f ((0, 0)). (1.6.4)
2π
30 Discrete group actions

Here fˆ denotes the Fourier transform of f in R2 . If we now combine (1.6.3)

and (1.6.4), we obtain

d xd y 2ζ (2) ˆ
F(z) 2 = f ((0, 0)) · Vol(\h2 ) + f ((0, 0)). (1.6.5)
y π
\h2

To complete the proof, we make use of the Poisson summation formula (see
appendix) which states that for any z ∈ G L(2, R)

1

F(z) = f ((m, n)z) = fˆ (m, n) · (t z)−1
(m,n)∈Z2
|Det(z)| (m,n)∈Z2


= fˆ (m, n) · (t z)−1 ,
(m,n)∈Z2


y− 2 x
1 1
y2
since z = and Det(z) = 1. We now repeat all our computations
y− 2
1
0
with the roles of f and fˆ reversed. Since the group  is stable under transpose–
inverse, one easily sees (from the Poisson summation formula above), by letting
z → (t z)−1 , that the integral

d xd y
F(z) 2
y
\h2

is unchanged if we replace f by fˆ.

Also, since fˆˆ (x) = f (−x), the formula (1.6.5) now becomes

d xd y 2ζ (2)
F(z) 2 = fˆ((0, 0)) · Vol(\h2 ) + f ((0, 0)). (1.6.6)
y π
\h2

If we combine (1.6.5) and (1.6.6) and solve for the volume, we obtain
  2ζ (2)
f ((0, 0)) − fˆ((0, 0)) · vol(\h2 ) = f ((0, 0)) − fˆ((0, 0)) · .
π
Since f is arbitrary, we can choose f so that f ((0, 0)) − fˆ((0, 0)) = 0. It
follows that
2ζ (2) π
Vol(\h2 ) = = .

π 3
Proof for the case of S L(n, R) We shall now complete the proof of
Theorem 1.6.1 using induction on n.

The proof of Theorem 1.6.1 requires two preliminary lemmas which we

straightaway state and prove. For n > 2, let Un (R) (respectively Un (Z)) denote
 1.6 Volume of S L(n, Z)\S L(n, R)/S O(n, R) 31

the group of all matrices of the form

⎛ ⎞
1 u1
⎜ .. .. ⎟
⎜ . . ⎟
⎜ ⎟
⎝ 1 u n−1 ⎠
1
with u i ∈ R (respectively, u i ∈ Z), for i = 1, 2, . . . , n − 1.
Lemma 1.6.7 Let n > 2 and fix an element γ ∈ S L(n − 1, Z). Consider
 of Un (Z) on R
n−1
the action
 given by left matrix multiplication of Un (Z) on
γ 0
· Un (R). Then a fundamental domain for this action is given by the
0 1
set of all matrices
⎛ ⎞
1 u1
  ⎜ .. .. ⎟
γ 0 ⎜ . . ⎟
·⎜ ⎟
0 1 ⎝ 1 u n−1 ⎠
1
with 0 ≤ u i < 1 for 1 ≤ i ≤ n − 1. In particular,
/ 
γ 0
Un (Z) · Un (R) ∼
= (Z\R)n−1 .
0 1

Proof of Lemma 1.6.7 Let m be a column vector with (m 1 , m 2 , . . . , m n−1 ) as

entries. Then one easily checks that
       
In−1 m γ γ I γ −1 m
· = · n−1 ,
1 1 1 1
where In−1 denotes the (n − 1) × (n − 1) identity matrix. It follows that

 
(  In−1 m  γ I (Z\R)n−1

· · n−1
1 1 1
m∈Zn−1
(      
γ In−1 γ −1 m In−1 (Z\R)n−1
= · ·
1 1 1
m∈Zn−1
  (  
γ In−1 (Z\R)n−1 + γ −1 m
= ·
1 1
m∈Zn−1
 
γ
= · Un (R).
1
It is also clear that the above union is over non-overlapping sets. This is because
γ −1 Zn−1 = Zn−1 for γ ∈ S L(n − 1, Z).

The second lemma we need is a generalization of the identity (1.6.4).

32 Discrete group actions

Lemma 1.6.8 Let n > 2 and let f : Rn → C be a smooth function, with

sufficient decay at ∞, which satisfies f (u 1 , . . . , u n ) = f (v1 , . . . , vn ) whenever
u 21 + · · · + u 2n = v12 + · · · + vn2 . Then
 ∞ 
1
f (0, . . . , 0, t) t n−1 dt = f (x1 , . . . , xn ) d x1 · · · d xn
0 Vol(S n−1 ) Rn
fˆ(0)
= ,
Vol(S n−1 )
where
√
2( π )n
Vol(S n−1
)=
 (n/2)

denotes the volume of the (n − 1)–dimensional sphere S n−1 .

Proof of Lemma 1.6.8 For n ≥ 2 consider the spherical coordinates:
x1 = t · sin θn−1 · · · sin θ2 sin θ1 ,
x2 = t · sin θn−1 · · · sin θ2 cos θ1 ,
x3 = t · sin θn−1 · · · sin θ3 cos θ2 , (1.6.9)
..
.
xn−1 = t · sin θn−1 cos θn−2 ,
xn = t · cos θn−1 ,
with
0 < t < ∞, 0 ≤ θ1 < 2π, 0 ≤ θ j < π, (1 < j < n).
Clearly x12 + · · · + xn2 = t 2 . One may also show that the invariant measure on
the sphere S n−1 is given by

dµ(θ) = (sin θ j ) j−1 dθ j ,
1≤ j<n

and that d x1 d x2 · · · d xn = t n−1 dt dµ(θ). Then the volume of the unit sphere,
Vol(S n−1 ), is given by
 √
2( π)n
Vol(S ) =
n−1
dµ(θ ) = .
S n−1  (n/2)
Since f is a rotationally invariant function, it follows that

1
f (0, . . . , 0, t) = f (x1 , . . . , xn ) dµ(θ)
Vol(S n−1 ) S n−1
 1.6 Volume of S L(n, Z)\S L(n, R)/S O(n, R) 33

with x1 , . . . , xn given by (1.6.9). Consequently

 ∞  ∞
1
f (0, . . . , 0, t) t n−1 dt = f (x1 , . . . , xn )t n−1 dµ(θ )dt
0 Vol(S n−1 ) 0 S n−1

1
= f (x1 , . . . , xn ) d x1 · · · d xn .
Vol(S n−1 ) Rn

We now return to the proof of Theorem 1.6.1. Let K n = S O(n, R) denote

the maximal compact subgroup of S L(n, R). In this case, the Iwasawa decom-
position (Proposition 1.2.6) says that every z ∈ S L(n, R)/K n is of the form
z = x y with
⎛ ⎞
1 x1,2 x1,3 ··· x1,n
⎜ 1 x2,3 ··· x2,n⎟
⎜ ⎟
⎜ .. .. ⎟
x=⎜ . . ⎟,
⎜ ⎟
⎝ 1 xn−1,n ⎠
1
⎛ ⎞ (1.6.10)
y1 y2 · · · yn−1 t
⎜ y1 y2 · · · yn−2 t ⎟
⎜ ⎟
⎜ .. ⎟
y=⎜ . ⎟,
⎜ ⎟
⎝ y1 t ⎠
t

n−1 −1/n
)
with t = Det(y)−1/n = yin−i .
i=1
In analogy to the previous proof for S L(2, R) we let f : Rn /K n → C be an
arbitrary smooth compactly supported function. We shall also define a function
F : S L(n, R)/K n → C by letting


F(z) := f (m · z).
m∈Zn

As before, the function F(z) will be invariant under left multiplication by

S L(n, Z).
Let
 
∗
Pn = ∈ S L(n, Z)
0 0 ··· 0 1
34 Discrete group actions

denote the set of all n × n matrices in S L(n, Z) with last row (0, 0, . . . , 0, 1).
Let n = S L(n, Z). Then we have as before:



F(z) = f (0) + f ( en · γ · z),
0< ∈Z γ ∈Pn \n

where f (0) denotes f ((0, 0, . . . , 0)) and en = (0, 0, . . . , 0, 1).

We now integrate F(z) over a fundamental domain for n \hn . It follows that


∗
F(z) d z = f (0) · Vol(n \h ) + 2
n
f ( en · z) d ∗ z. (1.6.11)
>0
n \hn Pn \hn

The factor 2 occurs because −In (In = n × n identity matrix) acts trivially on
hn . The computation of the integral above requires some preparations.
We may express z ∈ hn in the form
⎛ ⎞
y1 y2 · · · yn−1 t
⎜ y1 y2 · · · yn−2 t ⎟    − 1 
⎜ ⎟ 1
⎜ .. ⎟ t n − 1 · In−1 t n − 1 · In−1
z = x ·⎜ . ⎟· · ,
⎜ ⎟ t −1 t
⎝ y1 t ⎠
t

where x and t are given by (1.6.10). It follows that

⎛ ⎞ ⎛ ⎞
1 x1,n 1 x1,2 x1,3 ··· x1,n−1 0
⎜ 1 ⎟
x2,n ⎟ ⎜ ⎜ 1 x2,3 ··· x2,n−1 0⎟
⎜ ⎟
⎜ . . ⎟ ⎜ .. .. .. ⎟
z=⎜ . . . ⎟
. ·⎜ . . .⎟
⎜ ⎟ ⎜ ⎟
⎝ 1 xn−1,n ⎠ ⎝ 1 0⎠
1 1
⎛ n ⎞
y1 y2 · · · yn−1 · t n−1

⎜ y1 y2 · · · yn−2 · t n − 1
n
⎟  
⎜ ⎟ 1
⎜ .. ⎟ t − n − 1 · In−1
×⎜ . ⎟·
⎜ ⎟ t
⎝ n
y1 · t n − 1 ⎠
1
⎛ ⎞
1 x1,n
⎜ 1 x2,n ⎟  
− 1 
⎜ ⎟
⎜ .. .. ⎟ z t n − 1 · In−1
=⎜ . . ⎟· · ,
⎜ ⎟ 1 t
⎝ 1 xn−1,n ⎠
1
(1.6.12)
 1.6 Volume of S L(n, Z)\S L(n, R)/S O(n, R) 35

where
⎛ ⎞
1 x1,2 x1,3 ··· x1,n−1
⎜ 1 x2,3 ··· x2,n−1 ⎟
⎜ ⎟
⎜ .. .. ⎟
z =⎜ . . ⎟
⎜ ⎟
⎝ 1 xn−2,n−1 ⎠
1
⎛ n ⎞
y1 y2 · · · yn−1 · t n−1

⎜ n
y1 y2 · · · yn−2 · t n−1 ⎟
⎜ ⎟
×⎜ .. ⎟.
⎝ . ⎠
n
y1 · t n−1

Now z  represents the Iwasawa coordinate for S L(n − 1, R)/S O(n − 1, R) =

hn−1 , and the Haar measure d ∗ z  can be computed using Proposition 1.5.3 and
is given by
 
n−2
−k(n−1−k)−1
d ∗ z = d xi, j yk+1 dyk+1 .
1≤i< j≤n−1 k=1

If we compare this with

 
n−1
d∗z = d xi, j yk−k(n−k)−1 dyk
1≤i< j≤n k=1

 
n−2
−(k+1)(n−1−k)−1
= d xi, j yk+1 dyk ,
1≤i< j≤n k=0

we see that

n−1
dy1
d ∗ z = d ∗ z d x j,n t n . (1.6.13)
j=1
y1

Here, the product of differentials is understood as a wedge product satisfying

the usual rule: du ∧ du = 0, given by the theory of differential forms. Since

−(n−1)/n

n−1
−(n−i)/n
t = y1 yi ,
i=2

we see that
dt n − 1 dy1
=− + ,
t n y1
where  is a differential form involving dy j for each j = 2, 3, . . . , n − 1, but
36 Discrete group actions

not involving dy1 . It follows from (1.6.13) that

n 
n−1
dt
d∗z = − d ∗ z d x j,n t n . (1.6.14)
n−1 j=1
t

We also note that, by (1.6.12), we have


t − n−1 · In−1
1

f ( en · z) = f en · = f ( t en ). (1.6.15)
t
The last thing we need to do is to construct a fundamental domain for the
action of Pn on hn . Every p ∈ Pn can be written in the form
 
γ b
p= , (with γ ∈ S L(n − 1, Z), b ∈ Zn−1 ).
1
By (1.6.12), we may express z ∈ hn in the form
⎛ ⎞
x1,n
 
− 1 
⎜ ⎟
z u t n−1 · In−1 ⎜ x2,n ⎟
z= · , u=⎜ .. ⎟ ∈ Rn−1 .
1 t ⎝ . ⎠
xn−1,n
It follows that
 
− 1 
γ z γ ·u+b t n−1 · In−1
p·z = · ,
1 t
from which one deduces from Lemma 1.6.7 that
Pn \hn ∼
= S L(n − 1, Z)\hn−1 × (R/Z)n−1 × (0, ∞),
With these preliminaries, we can now continue the calculation of (1.6.11).
It follows from (1.6.14), (1.6.15), and Lemma 1.6.8 that


2 f ( en · z) d ∗ z
>0
Pn \hn
⎛ ⎞ ⎛ ⎞
  ∞
2n
⎜ ∗ ⎟
⎜ 
n−1
⎟ dt
= ⎝ d z⎠ ⎝ d x j,n ⎠ f ( t en ) t n
n − 1 >0 j=1
t
n−1 \hn−1 (R/Z)n−1 0
∞
2n  dt
= ζ (n) Vol n−1 \hn−1 · f (t en ) t n
n−1 t
t=0

2n  fˆ(0)
= ζ (n) Vol n−1 \hn−1 · . (1.6.16)
n−1 Vol(S n−1 )
 1.6 Volume of S L(n, Z)\S L(n, R)/S O(n, R) 37

Here, we used the facts that

  
 n−1
d ∗ z  = Vol n−1 \hn−1 , d x j,n = 1.
j=1
n−1 \hn−1 (R/Z)n−1

Combining (1.6.11) and (1.6.16) gives

 
∗ 2n ζ (n) Vol n−1 \hn−1
F(z) d z = f (0) · Vol(n \h ) + f (0) ·
n ˆ .
(n − 1)Vol(S n−1 )
n \hn
(1.6.17)
As before, we make use of the Poisson summation formula



F(z) = f (m · z) = fˆ(m · (t z)−1 ),
m∈Zn m∈Zn

which holds for Det(z) = 1. Since the group n is stable under transpose–
inverse, we can repeat all our computations with the roles of f and fˆ reversed,
and the integral

F(z) d ∗ z
n \hn

again remains unchanged. The formula (1.6.17) now becomes


2n ζ (n) Vol n−1 \hn−1
f (0) · Vol(n \h ) + f (0) ·
n ˆ
(n − 1)Vol(S n−1 )

2n ζ (n) Vol n−1 \hn−1
= fˆ(0) · Vol(n \h ) + f (0) ·
n
.
(n − 1)Vol(S n−1 )
Taking f so that f (0) = fˆ(0), we obtain

2n ζ (n) Vol n−1 \hn−1
Vol(n \h ) =
n
.
(n − 1)Vol(S n−1 )
Theorem 1.6.1 immediately follows from this by induction.

GL(n)pack functions The following GL(n)pack functions, described in the

appendix, relate to the material in this chapter:

IwasawaForm IwasawaXMatrix IwasawaXVariables

IwasawaYMatrix IwasawaYVariables IwasawaQ
MakeXMatrix MakeXVariables MakeYMatrix
MakeYVariables MakeZMatrix MakeZVariables
VolumeFormDiagonal VolumeFormHn VolumeFormUnimodular
VolumeHn Wedge d.
 2
Invariant differential operators

It has been shown in the previous chapter that discrete group actions can give
rise to functional equations associated to important number theoretic objects
such as the Riemann zeta function. Thus, there is great motivation for studying
discrete group actions from all points of view. Let us explore this situation in
one of the simplest cases. Consider the additive group of integers Z acting on
the real line R by translation as in Example 1.1.2. The quotient space Z\R is just
the circle S 1 . One may study S 1 by considering the space of all possible smooth
functions f : S 1 → C. These are the periodic functions that arise in classical
Fourier theory. The Fourier theorem says that every smooth periodic function
f : S 1 → C can be written as a linear combination


f (x) = an e2πinx
n∈Z

where
 1
an = f (x)e−2πinx d x,
0

for all n ∈ Z. In other words, the basic periodic functions, e2πinx with n ∈ Z,
form a basis for the space L2 (Z\R) . It is clear that a deeper understanding of
this space is an important question in number theory. We shall approach this
question from the viewpoint of differential operators and obtain a fresh and
illuminating perspective thereby.
A basis for the space L2 (Z\R) may be easily described by using the Laplace
2
operator ddx 2 . One sees that the basic periodic functions e2πinx (with n ∈ Z) are
all eigenfunctions of this operator with eigenvalue −4π 2 n 2 , i.e.,

d 2 2πinx
e = −4π 2 n 2 · e2πinx .
dx2

38
 2.1 Lie algebras 39

Consequently, the space L2 (Z\R) can be realized as the space generated by

the eigenfunctions of the Laplacian. What we have pointed out here is a simple
example of spectral theory. Good references for spectral theory are (Aupetit,
1991), (Arveson, 2002).
In the higher-dimensional setting, we shall investigate smooth functions
invariant under discrete group actions by studying invariant differential opera-
tors. These are operators that do not change under discrete group actions. For
2
example, the classical Laplace operator ddx 2 does not change under the action
x → x + n for any fixed integer n. It is not so obvious that the operator
 2 
∂ ∂2
−y 2 +
∂x2 ∂ y2
is invariant under the discrete group actions
az + b
z → ,
cz + d
 
a b
with ∈ G L(2, R). The main aim of this chapter is to develop a general
c d
theory of invariant differential operators and the best framework to do this in
is the setting of elementary Lie theory. For introductary texts on differential
operators, see (Boothby, 1986), (Munkres, 1991). The classic reference on Lie
groups and Lie algebras is (Bourbaki, 1998b), but see also (Bump, 2004). In
order to give a self-contained exposition, we begin with some basic definitions.

2.1 Lie algebras

Definition 2.1.1 An associative algebra A over a field K is a vector space
A over K with an associative product ◦, which satisfies for all a, b, c ∈ A, the
following conditions:
r a ◦ b is uniquely defined and a ◦ b ∈ A,
r a ◦ (b ◦ c) = (a ◦ b) ◦ c, (associative law)
r (a + b) ◦ c = a ◦ c + b ◦ c,
c ◦ (a + b) = c ◦ a + c ◦ b, (distributive law).
Note that in a vector space we can either add vectors or multiply them by
scalars (elements of A). The associative product gives a way of multiplying
vectors themselves.

Example Let R be the field of real numbers and let

A = M(n, R)
40 Invariant differential operators

denote the set of n × n matrices with coefficients in R. Then A = M(n, R) is

an associative algebra over R where ◦ denotes matrix multiplication. A basis
for the vector space A over R is given by the set of n 2 vectors E i, j where E i, j
denotes the matrix with a 1 at position {i, j} and zeros everywhere else.

Definition 2.1.2 A Lie algebra L over a field K is a vector space L over K

together with a bilinear map [, ] (pronounced bracket), of L into itself, which
satisfies for all a, b, c ∈ L:
r [a, b] is uniquely defined and [a, b] ∈ L;
r [a, βb + γ c] = β[a, b] + γ [a, c], ∀β, γ ∈ K ;
r [a, a] = 0;
r [a, b] = −[b, a], (skew symmetry);
r [a, [b, c]] + [b, [c, a]] + [c, [a, b]] = 0, (Jacobi identity).

Example Every associative algebra A (with associative product ◦) can be

made into a Lie algebra (denoted Lie(A)) by defining a bracket on A:

[a, b] = a ◦ b − b ◦ a, ∀ a, b ∈ A.

This is easily proved since [a, b] is clearly bilinear, [a, a] = 0 and

[[a, b], c] + [[b, c], a] + [[c, a], b]

= (a ◦ b − b ◦ a) ◦ c − c ◦ (a ◦ b − b ◦ a)
+ (b ◦ c − c ◦ b) ◦ a − a ◦ (b ◦ c − c ◦ b)
+ (c ◦ a − a ◦ c) ◦ b − b ◦ (c ◦ a − a ◦ c).

We now show that it is also possible to go in the other direction. Namely,

given a Lie algebra L, we show how to construct an associative algebra U (L)
(called the universal enveloping algebra) where

L ⊆ Lie(U (L)).

In order to construct the universal enveloping algebra, we remind the reader of

some basic concepts and notation in the theory of vector spaces. Let V denote
a vector space over a field K with basis vectors v1 , v2 , . . . Then we may write

V = ⊕i K vi .

Similarly, if W is another vector space with basis vectors w1 , w2 , . . . such that

wi ∈ V for i = 1, 2, . . . , then we may form the vector space

V ⊕W
 2.1 Lie algebras 41

(defined over K ) with basis vectors v1 , w1 , v2 , w2 , . . . Similarly, we can also

define higher direct sums

⊕ V

of a set of linearly independent vector spaces V . We shall also consider the

tensor product V ⊗ W which is the vector space with basis vectors

vi ⊗ w j , (i = 1, 2, . . . , j = 1, 2, . . .)

and the higher tensor products ⊗k V (for k = 0, 1, 2, 3, . . .) where ⊗0 V = K

and for k ≥ 1, ⊗k V denotes the vector space with basis vectors

vi1 ⊗ vi2 ⊗ · · · ⊗ vik

where i j = 1, 2, . . . for 1 ≤ j ≤ k. If L is a Lie algebra, then when we take

direct sums or tensor products of L the convention is to forget the Lie bracket
and simply consider L as a vector space.
Definition 2.1.3 Let L be a Lie algebra with bracket [, ]. Define
∞
T (L) = ⊕ ⊗k L,
k=0

and define I (L) to be the two–sided ideal of T (L) generated by all the ten-
sors (linear combinations of tensor products), X ⊗ Y − Y ⊗ X − [X, Y ], with
X, Y ∈ L. The universal enveloping algebra U (L) of L is defined to be

U (L) = T (L)/I (L)

with an associative multiplication ◦ given by

η◦ξ =η⊗ξ (mod I (L)).

Example 2.1.4 Let L = M(3, R), the Lie algebra of 3 × 3 matrices with
coefficients in R and with Lie bracket
[X, Y ] = X · Y − Y · X
where · denotes matrix multiplication. We shall now exhibit two examples of
multiplication ◦ in U (L). First:
⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞
1 0 0 0 1 0 1 0 0 0 1 0
⎝ 0 0 0 ⎠ ◦ ⎝ 0 0 0 ⎠ = ⎝ 0 0 0 ⎠ ⊗ ⎝ 0 0 0 ⎠ (mod I (L)).
0 0 0 0 0 0 0 0 0 0 0 0
42 Invariant differential operators

The second example is:

⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞
1 0 0 0 1 0 0 1 0 1 0 0
⎝0 0 0⎠ ◦ ⎝0 0 0⎠ − ⎝0 0 0⎠ ◦ ⎝0 0 0⎠
0 0 0 0 0 0 0 0 0 0 0 0
⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞
1 0 0 0 1 0 0 1 0 1 0 0
= ⎝0 0 0⎠ ⊗ ⎝0 0 0⎠ − ⎝0 0 0⎠ ⊗ ⎝0 0 0 ⎠ (mod I (L))
0 0 0 0 0 0 0 0 0 0 0 0
⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞
1 0 0 0 1 0 0 1 0 1 0 0
= ⎝0 0 0⎠ · ⎝0 0 0⎠ − ⎝0 0 0⎠ · ⎝0 0 0 ⎠ (mod I (L))
0 0 0 0 0 0 0 0 0 0 0 0
⎛ ⎞
0 1 0
= ⎝ 0 0 0 ⎠ (mod I (L)).
0 0 0
The latter example shows that in general

X ◦Y −Y ◦ X = X ·Y −Y · X (mod I (L)),

which is easily proved.

2.2 Universal enveloping algebra of gl(n, R)

The group G L(n, R) is a Lie group and there is a standard procedure to pass
from a Lie group to a Lie algebra. We shall not need this construction because
the Lie algebra of G L(n, R) is very simply described. Let gl(n, R) be the Lie
algebra of G L(n, R). Then gl(n, R) is the additive vector space (over R) of all
n × n matrices with coefficients in R with Lie bracket given by

[α, β] = α · β − β · α

for all α, β ∈ gl(n, R), and where · denotes matrix multiplication. We shall
find an explicit realization of the universal enveloping algebra of gl(n, R) as an
algebra of differential operators. We shall consider the space S consisting of
smooth (infinitely differentiable) functions F : G L(n, R) → C.

Definition 2.2.1 Let α ∈ gl(n, R) and F ∈ S. Then we define a differential

operator Dα acting on F by the rule:
∂  ∂ 
 
Dα F(g) := F(g · exp(tα))  = F(g + t(g · α))  .
∂t t=0 ∂t t=0
 2.2 Universal enveloping algebra of gl(n, R) 43

+
Remark Recall that exp(tα) = I + ∞ k=1 (tα) /k!, where I denotes the
k

identity matrix on gl(n, R). Since we are differentiating with respect to t and
then setting t = 0, only the first two terms in the Taylor series for exp(tα)
matter.

The differential operator Dα satisfies the usual properties of a derivation:

Dα (F(g) · G(g)) = Dα F(g) · G(g) + F(g) · Dα G(g), (product rule),

Dα F((G(g)) = (Dα F)(G(g)) · Dα G(g), (chain rule),

for all F, G ∈ S, and g ∈ G L(n, R).

a b
Example 2.2.2 Let g= , F (g) := 2a + a 2 + b + d + d 3 ,
c d
0 1
α= . Then we have
0 0       
∂ a b 1 t  ∂ a at + b 
Dα F(g) =  
∂t
F
c d 0 1  = ∂t F c ct + d 
t=0 t=0
∂  

= 2a + a 2 + at + b + ct + d + (ct + d)3 
∂t t=0
= a + c + 3cd . 2

The differential operators Dα with α ∈ gl(n, R) generate an associative alge-

bra Dn defined over R. Then every element of Dn is a linear combination
(with coefficients in R) of differential operators Dα1 ◦ Dα2 ◦ · · · ◦ Dαk with
α1 , α2 , . . . , αk ∈ gl(n, R), where ◦ denotes multiplication in Dn , which is
explicitly given by composition (repeated iteration) of differential operators.

Proposition 2.2.3 Fix n ≥ 2. Let Dα , Dβ ∈ Dn with α, β ∈ gl(n, R). Then

Dα+β = Dα + Dβ ,
Dα ◦ Dβ − Dβ ◦ Dα = D[α,β] ,

where [α, β] = α · β − β · α, denotes the Lie bracket in gl(n, R), i.e., · denotes
matrix multiplication.

Proof Let g = (gi, j )1≤i≤n,1≤ j≤n ∈ G L(n, R). A smooth complex–valued

function F(g), defined on G L(n, R), can be thought of as a function of n 2
real variables gi, j with 1 ≤ i ≤ n, 1 ≤ j ≤ n. It immediately follows from the
chain rule for functions of several real variables and Definition 2.2.1 that

n
∂
Dα F(g) = (g · α)i, j · F(g). (2.2.4)
i, j=1
∂gi, j
44 Invariant differential operators

Here, (g · α)i, j denotes the i, j entry of the matrix g · α. It immediately follows

from (2.2.4) that Dα+β = Dα + Dβ . If we now apply Dβ to the above expression
(2.2.4), we see that

0 1
n
∂ ∂ 
Dβ ◦ Dα F(g) = ((g + t(g · β)) · α)i, j · F(g + t(g · β)) 
i, j=1
∂t ∂g i, j t=0


n
∂F
= (g · β · α)i, j ·
i, j=1
∂gi, j


n
∂2 F
+ (g · β)i  , j  (g · α)i, j · .
i  , j  =1
∂gi, j ∂gi  , j 

Consequently,


n
∂F
(Dβ ◦ Dα − Dα ◦ Dβ )F(g) = (g · (β · α − α · β))i, j ·
i, j=1
∂gi, j
= D[β,α] F(g),

which completes the proof of the proposition.

Proposition 2.2.3 shows that the ring of differential operators Dn is a real-

ization of the universal enveloping algebra of the Lie algebra gl(n, R).

Corollary 2.2.5 If Dα ◦ Dβ = Dβ ◦ Dα for α, β ∈ gl(n, R), then

Dα·β = Dβ·α .

Proof It follows from Proposition 2.2.3 that

0 = Dα ◦ Dβ − Dβ ◦ Dα = Dα·β−β·α = Dα·β − Dβ·α .

Define
⎛ ⎞
−1
⎜ 1 ⎟
⎜ ⎟
δ1 := ⎜ .. ⎟.
⎝ . ⎠
1
Proposition 2.2.6 For n ≥ 2, let f : G L(n, R) → C be a smooth function
which is left invariant by G L(n, Z), and right invariant by the center Z n . Then
for all D ∈ Dn , D f is also left invariant by G L(n, Z), right invariant by Z n ,
and right invariant by the element δ1 .

Proof It is enough to consider the case when

D = Dα1 ◦ Dα2 ◦ · · · ◦ Dαm

 2.2 Universal enveloping algebra of gl(n, R) 45

with m ≥ 1, and αi ∈ gl(n, R) for i = 1, 2, . . . , m. Since f is left invariant by

G L(n, Z), i.e., f (γ · g) = f (g) for all g ∈ G L(n, R), it immediately follows
from Definition 2.2.1 that
∂ ∂ ∂  
D f (γ g) = ··· f γ get1 α1 +···+tm αm 
∂t1 ∂t2 ∂tm t1 =0,...,tm =0

∂ ∂ ∂  t1 α1 +···+tm αm 
= ··· f ge 
∂t1 ∂t2 ∂tm t1 =0,...,tm =0

= D f (g).

Similarly, let δ ∈ Z n . Then δg = gδ for all g ∈ G L(n, R). It follows, as

above, that
∂ ∂ ∂  t1 α1 +···+tm αm 
D f (gδ) = ··· f gδe 
∂t1 ∂t2 ∂tm t1 =0,...,tm =0

∂ ∂ ∂  t1 α1 +···+tm αm  
= ··· f ge δ 
∂t1 ∂t2 ∂tm t1 =0,...,tm =0

∂ ∂ ∂  

= ··· f get1 α1 +···+tm αm 
∂t1 ∂t2 ∂tm t1 =0,...,tm =0
= D f (g).

Finally, we must show that D f is right invariant by δ1 . Note that for all
g ∈ G L(n, R),

(D f )((δ1 · g · δ1 ) · δ1 ) = (D f )(δ1 · g) = (D f )(g) = D( f (g))

= D( f (δ1 · g · δ1 )),

because δ1 ∈ O(n, R) ∩ G L(n, Z). Thus, (D f )(g1 · δ1 ) = D( f (g1 )) with

g1 = δ1 · g · δ1 . The proposition follows because the map g → g1 is an iso-
morphism of hn .

The associative algebra Dn can also be made into a Lie algebra by defining
a bracket [D, D  ] = D ◦ D  − D  ◦ D for all D, D  ∈ Dn . There is a useful
identity given in the next proposition.

Proposition 2.2.7 For n ≥ 2, let α, β ∈ gl(n, R) and D ∈ Dn . Then

[Dα , Dβ ◦ D] = [Dα , Dβ ] ◦ D + Dβ ◦ [Dα , D].

Proof We have [Dα , Dβ ◦ D] = Dα ◦ Dβ ◦ D − Dβ ◦ D ◦ Dα . On the other

hand, [Dα , Dβ ] ◦ D + Dβ ◦ [Dα , D] = (Dα ◦ Dβ − Dβ ◦ Dα ) ◦ D + Dβ ◦
(Dα ◦ D − D ◦ Dα ). It is obvious that these expressions are the same.

46 Invariant differential operators

2.3 The center of the universal enveloping algebra

of gl(n, R)
Let n ≥ 2. We now consider the center Dn of Dn . Every D ∈ Dn satisfies
D ◦ D  = D  ◦ D for all D  ∈ Dn .

Proposition 2.3.1 Let n ≥ 2 and let D ∈ Dn lie in the center of Dn . Then D

is well defined on the space of smooth functions

f : G L(n, Z)\G L(n, R)/ (O(n, R)Z n ) → C,

i.e.,

(D f )(γ · g · k · δ) = D f (g),

for all g ∈ G L(n, R), γ ∈ G L(n, Z), δ ∈ Z n , and k ∈ O(n, R).

Proof Proposition 2.2.6 proves Proposition 2.3.1 for the left action of
G L(n, Z) and the right action by the center Z n . It only remains to show that
(D f )(g · k) = D f (g) for k ∈ O(n, R) and g ∈ G L(n, R).
Fix the function f , the differential operator D ∈ Dn ,the matrix
g ∈ G L(n, R), and, in addition, fix a matrix h ∈ gl(n, R) which satisfies
h + t h = 0. Given f, D, g, h, we define a function {φ f,D,g,h = φ} : R → C as
follows:
 
φ(u) := D f (g · exp(uh)) − (D f ) g · exp(uh) .

Clearly φ(0) = 0. We will now show that dφ/du = 0, which implies by ele-
mentary calculus that φ(u) is identically zero. The proof that φ  (u) = 0 goes as
follows:
∂ 

φ  (u) = φ(u + t) 
∂t t=0
∂   
= D f (g · exp((u + t) · h) − (D f ) g · exp((u + t) · h) 
∂t t=0
∂   
= D f (g · exp(uh) · exp(th) − (D f ) g · exp(uh) · exp(th) 
∂t t=0
= (D ◦ Dh )( f (g · exp(uh))) − ((Dh ◦ D) f )(g · exp(uh))
= 0,

because D ◦ Dh = Dh ◦ D.
It follows, as explained before, that φ(u) = 0. Now, the elements exp(uh)
with h + t h = 0 generate O(n, R)+ (the elements of the orthogonal group with
 2.3 The center of the universal enveloping algebra of gl(n, R) 47

positive determinant). This is because Det(exp(uh)) > 0 and

exp(uh) · t (exp(uh)) = exp(u(h + t h)) = I,

the identity matrix. Consequently,

 
0 = φ(u) = D f (g · exp(uh)) − (D f ) g · exp(uh)
 
= D f (g) − (D f ) g · exp(uh) .

Thus, D f is invariant on the right by O(n, R)+ . On the other hand, we already
know by Proposition 2.2.6 that D f is invariant on the right by the special
element δ1 of determinant −1. It follows that D f must be right–invariant by
the entire orthogonal group O(n, R). This proves the proposition.

We now show how to explicitly construct certain differential operators (called

Casimir operators) that lie in Dn , the center of the universal enveloping algebra
of gl(n, R). For 1 ≤ i ≤ n, 1 ≤ j ≤ n let E i, j ∈ gl(n, R) denote the matrix with
a 1 at the i, jth component and zeros elsewhere. Then, computing the bracket
of two such elements, we easily see that

[E i, j , E i  , j  ] = E i, j · E i  , j  − E i  , j  · E i, j (2.3.2)
=δ i , j E i, j  −δ
i, j  E i , j ,

1 if i = j
where δi, j = is Kronecker’s delta function.
0 otherwise,

Proposition 2.3.3 Let n ≥ 2 and E i, j (with 1 ≤ i, j ≤ n) be as above. Define

Di, j = D Ei, j with D Ei, j given by Definition 2.2.1. Then for 2 ≤ m ≤ n, the
differential operator (Casimir operator)


n

n

n
··· Di1 ,i2 ◦ Di2 ,i3 ◦ · · · ◦ Dim ,i1
i 1 =1 i 2 =1 i m =1

lies in Dn , the center of the universal enveloping algebra of gl(n, R).

Proof Let


n

n

n
D= ··· Di1 ,i2 ◦ Di2 ,i3 ◦ · · · ◦ Dim ,i1 .
i 1 =1 i 2 =1 i m =1

It is enough to show that [Dr,s , D] = 0 for all integers 1 ≤ r ≤ n, 1 ≤ s ≤ n.

We shall give the proof for m = 2. The case of general m follows by induction.
48 Invariant differential operators

It follows from Proposition 2.2.7, (2.3.2) and Proposition 2.2.3 that

n

n

[Dr,s , D] = [Dr,s , Di1 ,i2 ] ◦ Di2 ,i1 + Di1 ,i2 ◦ [Dr,s , Di2 ,i1 ]
i 1 =1 i 2 =1
n 


n

  
= δi1 ,s Dr,i2 − δr,i2 Di1 ,s ◦ Di2 ,i1 + Di1 ,i2 ◦ δi2 ,s Dr,i1 − δr,i1 Di2 ,s
i 1 =1 i 2 =1

n

n

n

n
= Dr,i2 ◦ Di2 ,s − Di1 ,s ◦ Dr,i1 + Di1 ,s ◦ Dr,i1 − Dr,i2 ◦ Di2 ,s
i 2 =1 i 1 =1 i 1 =1 i 2 =1
= 0.

Example 2.3.4 (The Casimir operator for gl(2, R))  We use the notation
y x
of Proposition 2.3.3 in the case n = 2, and let z = ∈ G L(2, R). By
0 1
Definition 2.2.1 we have the following explicit differential operators acting on
smooth functions f : h2 → C.
     
∂ y x y x 1 0 
D1,1 f (z) := f +t 
∂t 0 1 0 1 0 0 
t=0
  
∂ y(1 + t) x 
= f 
∂t 0 1 
t=0
∂
= y f (z).
∂y
     
∂ y x y x 0 1 
D1,2 f (z) := f +t 
∂t 0 1 0 1 0 0 
t=0
  
∂ y x + ty 
= f 
∂t 0 1 
t=0
∂
= y f (z).
∂x
     
∂ y x y x 0 0 
D2,1 f (z) := f +t 
∂t 0 1 0 1 1 0 
t=0
  
∂ y + tx x 
= f 
∂t t 1 

t=0

∂ y xt 2
+yt+x 

= f t +1
2 t +1
2

∂t 0 1 
t=0
∂
= y f (z)
∂x
 2.3 The center of the universal enveloping algebra of gl(n, R) 49

     
∂ y x y x 0 0 
D2,2 f (z) := f +t 
∂t 0 1 0 1 0 1 
t=0
  
∂ y x + tx 
= f 
∂t 0 1+t 
t=0
 y  
∂ x 
= f 1+t 
∂t 0 1 
t=0
∂
= −y f (z).
∂y

Warning Although f : h2 → C satisfies f (z · k) = f (z) for all

k ∈ O(2, R) · Z 2 , it is not the case that Di, j f is well defined on h2 for
1 ≤ i, j ≤ 2. For example, although D1,2 = D2,1 = y∂/∂ x, it is not true that
D2,1 ◦ D1,2 = (y∂/∂ x)2 .
We compute the second order differential operators as follows.
      
∂ ∂ y x1 + t1 0 1 + t2 0 
D1,1 ◦ D1,1 f (z) = f · · 
∂t1 ∂t2 0 1 0 1 0 1 
t1 =0, t2 =0
   
∂ ∂ y x 1 + t1 + t2 + t1 t2 0 
= f 
∂t1 ∂t2 0 1 0 1 
t1 =0, t2 =0
  
∂ ∂ y(1 + t1 + t2 + t1 t2 ) x 
= f 
∂t1 ∂t2 0 1 
t1 =0, t2 =0
 2 
∂ ∂
= y + y 2 2 f (z).
∂y ∂y
      
∂ ∂ y x 1 0 1 t2 
D1,2 ◦ D2,1 f (z) = f · · 
∂t1 ∂t2 0 1 t1 1 0 1 
t1 =0, t2 =0
    
∂ ∂ y x 1 t2 
= f · 
∂t1 ∂t2 0 1 t1 1 + t1 t2 
t1 =0, t2 =0
  
∂ ∂ y + xt1 yt2 + x(1 + t1 t2 ) 
= f 
∂t1 ∂t2 t1 1 + t1 t2 
t1 =0, t2 =0
⎛⎛ ⎞⎞
x + t (t t 2 +2t +t )+1 
y ( t1 t22 +t2 +t1 ) y
∂ ∂ ⎜⎜ t 1 (t 1 t2
2
+2t 2 +t 1 )+1 1 1 2 2 1 ⎟⎟
= f ⎝⎝ ⎠⎠
∂t1 ∂t2 
0 1 
t1 =0, t2 =0
 
∂ ∂ 2
= −2y + y2 2 f (z).
∂y ∂x
50 Invariant differential operators

      
∂ ∂ y x 1 t1 1 0 
D2,1 ◦ D1,2 f (z) = f · · 
∂t1 ∂t2 0 1 0 1 t2 1 
t1 =0, t2 =0
  
∂ ∂ t2 x + y(1 + t1 t2 ) x + t1 y 
= f 
∂t1 ∂t2 t2 1 
t1 =0, t2 =0

 
t1 t22 +t2 +t1
∂ ∂ y
x+( )y 

= f t22 +1 t22 +1

∂t1 ∂t2 0 1 t1 =0, t2 =0

∂2
= y2 f (z).
∂x2
 
∂ ∂2
D2,2 ◦ D2,2 f (z) = y + y2 2 f (z).
∂y ∂y

The Casimir operator is then given by

∂2 ∂2
D1,1 ◦ D1,1 + D1,2 ◦ D2,1 + D2,1 ◦ D1,2 + D2,2 ◦ D2,2 = 2y 2 + 2y 2 2 .
∂y 2 ∂x

The following proposition is the basic result in the subject. As pointed out in
(Borel, 2001), it was first proved by Capelli (1890).

Proposition 2.3.5 Every differential operator which lies in Dn (the center of

the universal enveloping algebra of gl(n, R)) can be expressed as a polynomial
(with coefficients in R) in the Casimir operators defined in Example 2.3.4.
Furthermore, Dn is a polynomial algebra of rank n − 1.

2.4 Eigenfunctions of invariant differential operators

We would like to construct an eigenfunction of all differential operators
D ∈ Dn , where Dn denotes the center of the universal enveloping algebra of
gl(n, R). Here gl(n, R) is the Lie algebra of G L(n, R), i.e., the vector space
of all n × n matrices with coefficients in R. We would like the eigenfunction
f to be a smooth function f : hn → C, where hn = G L(n, R)/(O(n, R) · R× ).
Then, we say f is an eigenfunction of D ∈ Dn , if there exists a complex number
λ D such that

D f (z) = λ D f (z)
 2.4 Eigenfunctions of invariant differential operators 51

for all z ∈ hn . Let z = x · y ∈ hn where

⎛ ⎞ ⎛ ⎞
1 x1,2 x1,3 · · · x1,n y1 y2 · · · yn−1
⎜ 1 x2,3 · · · x2,n ⎟ ⎜ y1 y2 · · · yn−2 ⎟
⎜ ⎟ ⎜ ⎟
x =⎜⎜ . . ⎟ ⎜ .. ⎟,
. . . ⎟, y = ⎜
. . ⎟
⎝ 1 xn−1,n ⎠ ⎝ y1 ⎠
1 1
with xi, j ∈ R for 1 ≤ i < j ≤ n and yi > 0 for 1 ≤ i ≤ n − 1.
We shall now define the important Is –function, which is a generalization
of the imaginary part function (raised to a complex power s) on the classical
upper half-plane. It will be shown that the Is –function is an eigenfunction
of Dn .

Definition 2.4.1 For n ≥ 2, s = (s1 , s2 , . . . , sn−1 ) ∈ Cn−1 , and z = x · y ∈ hn ,

as above, we define the function, Is : hn → C, by the condition:

 n−1
n−1  b s
Is (z) := yi i, j j ,
i=1 j=1

where

ij if i + j ≤ n,
bi, j =
(n − i)(n − j) if i + j ≥ n.

The coefficients bi, j are incorporated into the definition because they make
later formulae simpler. Note that since Is (z) is defined on the generalized upper
half-plane, hn , it must satisfy

Is (z · k · a) = Is (z)

for all k ∈ O(n, R), a ∈ R× .

Example
  (Eigenfunction for G L(2, R)) In this example, we may take
2.4.2
y x
z= , s ∈ C, and Is (z) = y s . Then, we have shown in Example 2.3.4
0 1
 2 
∂ ∂2
that  = y 2 + is a generator of D2 . Clearly,
∂x2 ∂ y2

Is (z) = s(s − 1)Is (z).

52 Invariant differential operators

Proposition 2.4.3 Let n ≥ 2, and let Is (z) be as given in Definition 2.4.1.

Define Di, j = D Ei, j where E i, j ∈ gl(n, R) is the matrix with a 1 at the i, jth
component and zeros elsewhere. Then for 1 ≤ i, j ≤ n, and k = 1, 2, . . . , we
have

s k · Is (z) if i = j
Di,k j Is (z) = n−i
0 otherwise,

where Di,k j = Di, j ◦ · · · ◦ Di, j denotes composition of differential operators

iterated k times.
Proof Note that the function Is (z) satisfies Is (x · y) = Is (y) for all x, y of the
form
⎛ ⎞ ⎛ ⎞
1 x1,2 x1,3 · · · x1,n y1 y2 · · · yn−1
⎜ 1 x2,3 · · · x2,n ⎟ ⎜ y1 y2 · · · yn−2 ⎟
⎜ ⎟ ⎜ ⎟
⎜ .. . ⎟ ⎜ .. ⎟
x =⎜ . .. ⎟ , y = ⎜ . ⎟,
⎜ ⎟ ⎜ ⎟
⎝ 1 xn−1,n ⎠ ⎝ y1 ⎠
1 1
with xi, j ∈ R for 1 ≤ i < j ≤ n and yi > 0 for 1 ≤ i ≤ n − 1. It easily follows
that Di, j Is (x · y) = Di, j Is (y). If i < j, then by Definition 2.2.1, we have
∂  ∂

Di, j Is (y) = Is (y + t y · E i, j ) = y1 y2 · · · yn−i Is (y) = 0.
∂t t=0 ∂ xi, j
If i = j, then
∂ 

Di,i Is (y) = Is (y + t y · E i,i )
∂t

t=0

∂
n−1
∂
= yn−i − y Is (y)
∂ yn−i =n−i+1 ∂ y
= sn−i · Is (y).
In a similar manner,
 k
∂  
k
Di,i Is (y) = Is y · et Ei,i  = sn−i
k
Is (y).
∂t t=0

If i > j, then the argument is more complicated. We make use of Proposition

1.2.8. It follows as before that
∂  
Di, j Is (y) = Is y · (I + t E i, j  ,
∂t t=0

where I is the identity matrix. By Proposition 1.2.8, I + t E i, j is a matrix with

either 1, (t 2 + 1) 2 , or (t 2 + 1)− 2 on the diagonal. When you take the derivative
1 1
 2.4 Eigenfunctions of invariant differential operators 53

of any of these with respect to t and then set t = 0 you must get zero as an
1
answer. So the only contribution comes from the off diagonal entry t/(t 2 + 1) 2 .
Consequently
∂
Di, j Is (y) = y1 y2 · · · yn− j Is (y) = 0.

∂ x j,i
GL(n)pack functions The following GL(n)pack functions, described in the
appendix, relate to the material in this chapter:
ApplyCasimirOperator GetCasimirOperator IFun.
 3
Automorphic forms and L–functions for
S L(2, Z)

The spectral theory of non-holomorphic automorphic forms formally began

with Maass (1949). His book (Maass, 1964) has been a source of inspiration
to many. Some other references for this material are (Hejhal, 1976), (Venkov,
1981), (Sarnak, 1990), (Terras, 1985), (Iwaniec-Kowalski, 2004).
Maass gave examples of non-holomorphic forms for congruence subgroups
of S L(2, Z) and took the very modern viewpoint, originally due to Hecke (1936),
that automorphicity should be equivalent to the existence of functional equa-
tions for the associated L-functions. This is the famous converse theorem given
in Section 3.15, and is a central theme of this entire book. The first converse
theorem was proved by Hamburger (1921) and states that any Dirichlet series
satisfying the functional equation of the Riemann zeta function ζ (s) (and suit-
able regularity criteria) must actually be a multiple of ζ (s).
Hyperbolic Fourier expansions of automorphic forms were first introduced
in (Neunhöffer, 1973). In (Siegel, 1980), the hyperbolic Fourier expansion of
G L(2) Eisenstein series is used to obtain the functional equation of certain
Hecke L-functions of real quadratic fields with Grössencharakter (Hecke, 1920).
When this is combined with the converse theorem, it gives explicit examples
of Maass forms. These ideas are worked out in Sections 3.2 and 3.15.
Another important theme of this chapter is the theory of Hecke opera-
tors (Hecke, 1937a,b). We follow the beautiful exposition of Shimura (1971),
but reduce the key computations to the Hermite and Smith normal forms
(Cohen, 1993), a method which easily generalizes to S L(n, Z) with n ≥ 2.
The Hecke operators map automorphic forms to automorphic forms. Hecke
proved the remarkable theorem that if an automorphic form is an eigenfunction
of all the Hecke operators then its associated L-function has an Euler product
expansion.
Finally, the chapter concludes with the Selberg spectral decomposition
(Selberg, 1956) which has played such a pivotal role in modern number theory.

54
 3.1 Eisenstein series 55

3.1 Eisenstein series

Let
   
y x 
h =
2  x ∈ R, y > 0
0 1 

be the upper half-plane associated to G L(2, R), i.e., h2 = G L(2, R)/(O(2, R) ·

R× ). It is clear that for s ∈ C and z ∈ h, the function
 2 Is (z) = y is an eigen-
s

∂ ∂2
function of the hyperbolic Laplacian  = −y ∂ x 2 + ∂ y 2 , with eigenvalue
2

s(1 − s). If (s) ≥ 12 , the function Is (z) is neither automorphic for S L(2, Z)
nor is it square integrable with respect to the G L + (2, R) invariant measure
d xd y/y 2 (over the standard fundamental domain S L(2, Z)\h given in Exam-
ple 1.1.9). The fact that  is an invariant differential operator does imply that

Is (gz) = s(1 − s)Is (gz) (3.1.1)

for any g ∈ G L + (2, R). An automorphic function for S L(2, Z) is a smooth
function f : S L(2, Z)h2 → C. One way to construct an automorphic function
for S L(2, Z) which is also an eigenfunction of the Laplacian  is to average
over the group. Since Is (αz) = Is (z) for any
   
1 m 
α ∈ ∞ := m∈Z ,
0 1 

and ∞ is an infinite group, it is necessary to factor out by this subgroup. The

cosets ∞ \S L(2, Z) are determined by the bottom row of a representative
       
a b ∗ ∗ u −v 
∞ = =  du + cv = 1 .
c d c d c d
Each relatively prime pair (c, d) determines a coset.
Definition 3.1.2 Let z ∈ h2 , (s) > 1. We define the Eisenstein series:

Is (γ z) 1
ys
E(z, s) := = .
γ ∈ \S L(2,Z)
2 2 c,d∈Z |cz + d|2s
∞
(c,d)=1

Proposition 3.1.3 The Eisenstein series E(z, s) converges absolutely and

uniformly on compact sets for z ∈ h2 and (s) > 1. It is real analytic in z and
complex analytic in s.
In addition, we have the following:
(1) Let  > 0. For σ = (s) ≥ 1 +  > 1, there exists a constant c() such that
|E(z, s) − y s | ≤ c()y − , for y ≥ 1.
56 Automorphic forms and L–functions for S L(2, Z)

 
 az+b a b
(2) E cz+d
, s = E(z, s) for all ∈ S L(2, Z).
c d
(3) E(z, s) = s(1 − s)E(z, s).

Proof First of all, for y ≥ 1, we have

1 yσ
|E(z, s) − y s | ≤ · 
(c,d)=1
c 2σ z + d 2σ
c
c>0



c
1
1
= yσ   .
c≥1 r =1
c m∈Z z + + m 2σ
2σ r
(r,c)=1 c

Since the set {|z + (r/c) + m| | m ∈ Z, 1 ≤ r ≤ c, (r, c) = 1} forms a set of

points spaced by (1/c), we may majorize each term so that
 
 r 
≤ x + + m  < + 1
c
for some integer . There are at most φ(c) such terms for each . It follows
that


∞
φ(c)
1
|E(z, s) − y s | ≤ y σ
c=1
c2σ ∈Z
( 2 + y 2 )σ
ζ (2σ − 1)
∞
1
≤ 2y σ
ζ (2σ ) =0 ( + y 2 )σ
2
  ∞ 
ζ (2σ − 1) du
≤ 2y σ y −2σ +
ζ (2σ ) 0 (u 2 + y 2 )σ
 y 1−σ .

The second statement of Proposition 3.1.3 follows easily from the fact that for
every γ ∈ S L(2, Z), we have γ (∞ \S L(2, Z)) = (∞ \S L(2, Z)) . The third
statement is an easy consequence of (3.1.1) and the definition of E(z, s).

We now determine the Fourier expansion of the Eisenstein series. This

requires some computations involving Ramanujan sums whose definitions and
theory we briefly review.

Definition 3.1.4 For fixed integers n, c with c ≥ 1, the Ramanujan sum is

the exponential sum


c
r
S(n; c) = e2πin c .
r =1
(r,c)=1
 3.1 Eisenstein series 57

The Ramanujan sum can be explicitly evaluated using the Moebius function
µ(n) which is defined by the conditions:
µ(1) = 1, (3.1.5)


1 if n = 1
µ(d) = (3.1.6)
d|n 0 if n > 1,
or equivalently by the identity
1
∞
µ(n) 
= = (1 − p −s ).
ζ (s) n=1
n s
p

Proposition 3.1.7 We have

c
S(n; c) = µ ,
|n, |c

and for (s) > 1,

∞
S(n; c) σ1−s (n)
= ,
c=1
c s ζ (s)
+
where σz (n) = d|n d z is the divisor function.
Proof Using the properties (3.1.5), (3.1.6) of the Moebius function, we sift
out those integers r relatively prime to c as follows:


c
r


S(n; c) = e2πin c µ(d)
r =1 d|c,d|r



c
r
= µ(d) e2πin c
d|c r =1
r ≡ 0 (mod d)
c



d
md
= µ(d) e2πin c

d|c m=1

c
c
= µ(d) = µ .
d
d |c, d |n
c |n, |c

+
q nm
Here, we have use the fact that the sum e2πi q is zero unless q|n, in which
m=1
case it is q.
For the second part of the proposition, we calculate

S(n; c)

∞ ∞
c

∞
µ(m)

−s
= c µ = = 1−s · ζ (s)−1 .
c=1
c s
c=1 |n, |c
|n m=1
(m ) s
|n

It is easily verified that all the above sums converge absolutely for
(s) > 1.

58 Automorphic forms and L–functions for S L(2, Z)

 
y x
Theorem 3.1.8 Let (s) > 1 and z = ∈ h2 . The Eisenstein series
0 1
E(z, s) has the Fourier expansion
√
2π s y
1
E(z, s) = y s + φ(s)y 1−s + σ1−2s (n)|n|s− 2 K s− 12 (2π |n|y)e2πinx
(s)ζ (2s) n=0

where

√  s − 12 ζ (2s − 1)
φ(s) = π ,
(s) ζ (2s)


σs (n) = ds,
d|n
d>0

and
 ∞
1 du
e− 2 y (u+ u ) u s
1 1
K s (y) = .
2 0 u

Proof First note that

ys
ζ (2s)E(z, s) = ζ (2s)y s + .
c>0 d∈Z
|cz + d|2s

1 n=0
If we let δn,0 = and d = mc + r , it follows that
0 n = 0,
 1
ζ (2s) E(z, s)e−2πinx d x
0
c

1


∞
y s e−2πinx
−2s
= ζ (2s)y δn,0 + s
c   dx
c=1 r =1 m∈Z
z + m + r 2s
0 c
1+m+ rc


∞

c

y s e−2πin (x− c )
r

= ζ (2s)y s δn,0 + c−2s dx

c=1 r =1 m∈Z |z|2s
m+ rc

∞
∞
−2s
c
2πinr y s e−2πinx
= ζ (2s)y δn,0 + s
c e c d x.
c=1 r =1
(x 2 + y 2 )s
−∞

Since



c
2πinr c c|n
e c =
r =1 0 c  |n,
 3.2 Hyperbolic Fourier expansion of Eisenstein series 59

it is clear that
 1  ∞
e−2πinx y
ζ (2s) E(z, s)e−2πinx d x = ζ (2s)y s δn,0 + σ1−2s (n)y 1−s d x,
0 −∞ (x 2 + 1)s
with the understanding that σ1−2s (0) = ζ (1 − 2s). The proof of the theorem
now immediately follows from the well-known Fourier transform:
⎧√
 ∞ −2πi x y ⎨ π (s− 2 )
1
if y = 0,
e (s)
dx = (3.1.9)
−∞ (x + 1)
2 s ⎩ 2π |y| K 1 (2π|y|)
s s− 21
if y =
 0.
(s) s− 2

We may easily prove (3.1.9) as follows.

 ∞ −2πi x y  ∞ ∞  s
e −u−2πi x y u du
(s) dx = e dx
−∞ (x + 1) 1+x
2 s 2 u
0 −∞
 ∞  ∞
du
e−u u s e−ux e−2πi x y d x
2
=
0 −∞ u
 ∞
√ π 2 y2 1 du
= π e−u− u u s− 2 ,
0 u
since e−π x is its own Fourier transform.
2

Theorem 3.1.10 Let z ∈ h2 and s ∈ C with (s) > 1. The Eisenstein series
E(z, s) and the function φ(s) appearing in the constant term of the Fourier
expansion of E(z, s) can be continued to meromorphic functions on C satisfying
the functional equations:
(1) φ(s)φ(1 − s) = 1;
(2) E(z, s) = φ(s)E(z, 1 − s).
The modified function E ∗ (z, s) = π −s (s)ζ (2s)E(z, s) is regular except
for simple poles at s = 0, 1 and satisfies the functional equation
E ∗ (z, s) = E ∗ (z, 1 − s). Furthermore the residue of the pole at s = 1 is given
by
3
Res E(z, s) =
s=1 π
for all z ∈ h .2

3.2 Hyperbolic Fourier expansion of Eisenstein series

In this section we use the classical upper half-plane
 model
 for h2 . Thus
a b
z = x + i y with y > 0, x ∈ R, and the action of ∈ S L(2, R) on z
c d
is given by (az + b)/(cz + d).
60 Automorphic forms and L–functions for S L(2, Z)

 
α β
Let ρ = ∈ S L(2, Z) with trace |α + δ| > 2 and γ > 0. Such a
γ δ
matrix is termed hyperbolic. Set D = (α + δ)2 − 4. Then, a point w ∈ C is
termed fixed under ρ if (αw + β)/(γ w + δ) = w. It is easily seen that ρ has
exactly two real fixed points
√ √
α−δ+ D  α−δ− D
ω= , ω = .
2γ 2γ
   
1 −ω −1 
Now, define κ = . Then κρκ = is a diagonal
1 −ω  −1
matrix with action on z ∈ h2 given by  2 z. Since conjugation preserves the √trace,
we see that  +  −1 = α + δ. Consequently,
√  = γ ω 
+ δ = (α + δ − D)/2
is a unit in the quadratic field
√ Q( D). We shall assume that it is a fundamental
unit, i.e., every unit in Q( D) is, up to ±1, an integral power of  > 0. The
Eisenstein series E(κ −1 z, s) is invariant under z →  2 z. This is because
E(κ −1 z, s) = E(ρκ −1 z, s) = E(κ −1 ( 2 z), s).
Therefore, on the positive imaginary axis (i.e., choosing z = iv), the Eisenstein
series ζ (2s) · E(κ −1 z, s) (for (s) > 1) has a Fourier expansion

πin
ζ (2s) · E(κ −1 (iv), s) = bn (s) v log  , (3.2.1)
n∈Z

with
 2
1 πin dv
bn (s) = ζ (2s) · E(κ −1 (iv), s) v − log  .
2 log  1 v
A direct computation shows that

v s · (ω − ω )s
ζ (2s) · E(κ −1 (iv), s) =  s .
c,d ∈ Z (cω + d)2 v 2 + (cω + d)2
{c,d}={0,0}

The reason for multiplying by ζ (2s) on the left is to have the sum go over all
c, d ∈ Z ({c, d} = {0, 0}) and not just coprime pairs of c, d. Thus,

2 β
 − πin  · β  2 s
(ω − ω )
 s  β  log  v πin dv
bn (s) = N (β)−s    v − log  ,
2 log  β=0 β v 2 +1 v
β
β

where the sum goes over all non-zero β = cω + d, β  = cω + d with c, d

∈ Z and√{c, d} = {0, 0}. These elements β lie in an √ ideal b where ω − ω
= N (b) D, and where N denotes the norm from Q( D) to Q. The above
integral can be √
further simplified by using an idea of Hecke. For an algebraic
integer β ∈ Q( D), let (β) denote the principal ideal generated by β. Two
 3.2 Hyperbolic Fourier expansion of Eisenstein series 61

√
integers β1 , β2 ∈ Q( D) satisfy (β1 ) = (β2 ) if and only if β1 =  m β2 for some
integer m. Consequently
√  − πin
(N (b) D)s
  log 
−s  β 
bn (s) = N (β)   
2 log  b|(β)=0 β
 m
 2 · ββ m

  s
v2 πin dv
× v − log 
m∈Z
v +1
2 v
β  m
β m
√  − πin
(N (b) D)s
 β  log 
= N (β)−s   
2 log  b|(β)=0 β

 −2m+2 · ββ

  s
v2 πin dv
× v − log 
m∈Z
v +1
2 v

 −2m · ββ
√  − πin ∞  s
(N (b) D)s
  log 
−s  β  v2 πin dv
= N (β)    v − log 
2 log  b|(β)=0 β v2 + 1 v
0
 πin   πin 
s− log  s+ log 
  √
2 2 (N (b) D)s
= ·
(s) 2 log 

 − πin
 β  log 
×   N (β)−s .
 β 
b|(β)=0

√
For a principal ideal (β) of Q( D), we define the Hecke grössencharakter
 − πi
 β  log 
ψ((β)) :=    ,
β

and for an ideal b, we define the the Hecke L–function

  πin

 β − log 
L b(s, ψ ) =
n   N (β)−s . (3.2.2)
 β 
b|(β)=0

It now immediately follows from (3.2.1) that

 √ s
πin

πin

∗ −1 N (b) D
s − log 
s + log 
E (κ (iv), s) =  
2π s log  n∈Z 2 2
πin
× L b(s, ψ n ) · v log  , (3.2.3)
62 Automorphic forms and L–functions for S L(2, Z)

where E ∗ (z, s) = π −s (s)ζ (2s)E(z, s) = E ∗ (z, 1 − s). The expansion (3.2.3)

is termed the hyperbolic Fourier expansion of the Eisenstein series. An imme-
diate consequence of this expansion is the following proposition.
√
Proposition 3.2.4 Let K = Q( D) be a real quadratic field with ring of inte-
gers O(K ). For an ideal b in O(K ), let L b (s, ψ n ) denote the Hecke L–function
with grössencharakter given in (3.2.2). Then L b (s, ψ n ) has a meromorphic
continuation to all s with at most a simple pole at s = 1, and satisfies the
functional equation

√ s
πin

πin

N (b) D s − log 
s + log 
b(s) :=
n
  L b(s, ψ n )
π 2 2
= nb(1 − s).

3.3 Maass forms

We shall study the vector space L2 (S L(2, Z)\h2 ) (defined over C) which
is the completion of the subspace consisting of all smooth functions
f : S L(2, Z)\h2 → C satisfying the L2 condition

d xd y
| f (z)|2 < ∞.
y2
S L(2,Z)\h2

The space L (S L(2, Z)\h2 ) is actually a Hilbert space with inner product given
2

by

d xd y
 f, g := f (z)g(z) 2
y
S L(2,Z)\h2

for all f, g ∈ L2 (S L(2, Z)\h2 ). This inner product was first introduced by
Petersson.
Definition 3.3.1 Let ν ∈ C. A Maass form of type ν for S L(2, Z) is a non–
zero function f ∈ L2 (S L(2, Z)\h2 ) which satisfies:
 
r f (γ z) = f (z), for all γ ∈ S L(2, Z), z = y x ∈ h2 ;
r  f = ν(1 − ν) f ; 0 1
r * 1
0 f (z)d x = 0.

Proposition 3.3.2 Let f be a Maass form of type ν for S L(2, Z). Then ν(1 − ν)
is real and ≥ 0.
Proof The proof is based on the fact that the eigenvalues of a symmetric
operator on a Hilbert space are real. We have by Green’s theorem (integration
 3.4 Whittaker expansions and multiplicity one for G L(2, R) 63

by parts) that
ν(1 − ν) f, f  =  f, f 
  2  
∂ ∂2
= − + 2 f (z) · f (z) d xd y
∂x2 ∂y
2
S L(2,Z)\h

    
 ∂ f 2  ∂ f 2
=    
 ∂ x  +  ∂ y  d xd y
S L(2,Z)\h2
=  f,  f  = ν(1 − ν)  f, f .
The positivity of  f, f  and the inner integral above implies that ν(1 − ν) is
real and non–negative.

Proposition 3.3.3 A Maass form of type 0 or 1 for S L(2, Z) must be a constant

function.
Proof Let f be a Maass form of type 0 or 1. Then f (z) is a harmonic
function because  f = 0. Furthermore, since f is a Maass form it is bounded
as F(z) → ∞. The only harmonic functions on S L(2, Z)\h2 which are bounded
at infinity are the constant functions.

3.4 Whittaker expansions and multiplicity one for

G L(2, R)

 form of type ν for S L(2, Z), as in Definition 3.3.1. Since the

Let f bea Maass
1 1
element is in S L(2, Z) it follows that a Maass form f (z) satisfies
0 1
      
y x 1 1 y x y x +1
f = f = f .
0 1 0 1 0 1 0 1
Thus f (z) is a periodic function of x and must have a Fourier expansion of type


f (z) = Am (y)e2πimx . (3.4.1)
m∈Z

Define Wm (z) = Am (y)e 2πimx

. Then Wm (z) satisfies the following two
conditions:
Wm (z) = ν(1 − ν)Wm (z),
  
1 u
Wm · z = Wm (z)e2πimu .
0 1
We call such a function a Whittaker function of type ν associated to the additive
character e2πimx . Recall that an additive character ψ : R → U , where U denotes
64 Automorphic forms and L–functions for S L(2, Z)

the unit circle, is characterized by the fact that ψ(x + x  ) = ψ(x)ψ(x  ), for all
x, x  ∈ R. Formally, we have the following definition.
Definition 3.4.2 A Whittaker function of type ν associated to an additive
character ψ : R → U is a smooth non–zero function W : h2 → C which
satisfies the following two conditions:
W (z) = ν(1 − ν)W (z),
  
1 u
W · z = W (z)ψ(u).
0 1

Remark 3.4.3 A Whittaker function W (z), of type ν and character ψ, can

always be written in the form
W (z) = Aψ (y) · ψ(x)
where Aψ (y) is a function of y only. This is because the function W (z)/ψ(x) is
invariant under translations x → x + u for any u ∈ R and, hence, must be the
constant function for any fixed y.
Whittaker functions can be constructed explicitly. We know that the func-
tion Iν (z) = y ν satisfies Iν (z) = ν(1 − ν)Iν (z), which is the first condition a
Whittaker function must satisfy. In order to impose the second condition, we
need the following simple lemma.
Lemma 3.4.4 Let h : R → C be a smooth *∞L1 function. Let ψ be an additive
character of R. Then the function H (x) := −∞ h(u 1 + x)ψ(−u 1 ) du 1 satisfies
H (u + x) = ψ(u)H (x)
for all u ∈ R.
Proof Just make the change of variables u 1 + u → u 1 in the integral for
H (u + x).

Now, Iν (γ z) = ν(1 − ν)Iν (γ z) for any γ ∈ G L(2, R) because  is an

invariant differential operator. It follows from this and Lemma 3.4.4 that the
function
 ∞     
0 −1 1 u
W (z, ν, ψ) := Iν · · z ψ(−u) du
−∞ 1 0 0 1
 ∞ ν
y
= ψ(−u) du
−∞  (u +  x)2 + y 2
∞ ν
y
= ψ(x) 2 + y2
ψ(−u) du (3.4.5)
−∞ u
 3.4 Whittaker expansions and multiplicity one for G L(2, R) 65

must be a Whittaker function of type  ν associated

 to ψ. Actually, we may use
0 −1
any matrix in G L + (2, R) instead of in the integral on the right-hand
1 0
side of (3.4.5). All that is required is that the integral converges absolutely,
which happens in our case provided (ν) > 12 .
The G L(2) theory of Whittaker functions is considerably simplified because
the one Whittaker function which we can construct, W (z, ν, ψ), can be evaluated
exactly in terms of classical Bessel functions. Unfortunately, we are not able to
obtain such explicit realizations for Whittaker functions on G L(n) if n > 3, and
this will lead to considerable complications in the development of the theory in
the higher-rank case.

Proposition 3.4.6 Let ψm (u) = e2πimu , and let W (z, ν, ψm ) be the Whittaker
function (3.4.5). Then we have

√ (π |m|)ν− 12 &
W (z, ν, ψm ) = 2 2π y K ν− 12 (2π |m|y) · e2πimx ,
(ν)

where

∞
1 du
e− 2 y (u+ u ) u ν
1 1
K ν (y) =
2 u
0

is the classical K -Bessel function.

Proof It follows from Remark 3.4.3 and (3.4.5) that

W (z, ν, ψm ) = W (y, ν, ψm ) · e2πimx

where

∞  ν
y
W (y, ν, ψm ) = e−2πium du
u + y2
2
−∞
∞
e−2πiuym
=y 1−ν
du.
(u 2 + 1)ν
−∞

Note that we made the transformation u → y · u to identify the above integrals.

The result now follows from (3.1.9).

66 Automorphic forms and L–functions for S L(2, Z)

Definition 3.4.7 Let f : h2 → C be a smooth function.  Wesay that f is of

y x
polynomial growth at ∞ if for fixed x ∈ R and z = ∈ h2 , we have
0 1
f (z) is bounded by a fixed polynomial in y as y → ∞. We say f is of rapid
decay if for any fixed N > 1, |y N f (z)| → 0 as y → ∞. Similarly, we say f is
of rapid growth if for any fixed N > 1, |y −N f (z)| → ∞ as y → ∞.
We will now state and prove the multiplicity one theorem for Whittaker
functions on G L(2, R). This theorem is a cornerstone of the entire theory
and provides the basis for the Fourier–Whittaker expansions given in the next
section.
Theorem 3.4.8 (Multiplicity one) Let (z) be an S L(2, Z)–Whittaker
function of type ν = 0, 1, associated to an additive character ψ, which has
rapid decay at ∞. Then
(z) = aW (z, ν, ψ)
for some a ∈ C with W (z, ν, ψ) given by (3.4.5). If ψ ≡ 1 is trivial, then a = 0.
Proof Let (z) = (y)ψ(x) be a Whittaker function of type ν associated to
ψ. We may assume that ψ(x) = e2πimx for some m ∈ Z, because every additive
character is of this form. It follows from Definition 3.4.2 that the differential
equation
 2 
∂ ∂2 
((y)e 2πimx
) = −y 2
+ 2 (y)e2πimx = ν(1 − ν)(y)e2πimx
∂x 2 ∂y
implies that (y) satisfies the differential equation
 
 ν(1 − ν)
 (y) − 4π m − 2 2
(y) = 0. (3.4.9)
y2
By the classical theory of differential equations, (3.4.9) will have exactly two
linearly independent solutions over C.
If ψ ≡ 1 is trivial, then m = 0. Assume ν = 12 . Then there are precisely two
solutions, to the above differential equation, namely: y ν , y 1−ν . Thus
(y) = ay ν + by 1−ν
for certain complex constants a, b. The assumption that (y) is of
rapid decay imples that a = b = 0. Similarly, if m = 0 and ν = 12 , then
1 1
(y) = ay 2 + by 2 log y and a = b = 0 as before.
If ν(1 − ν) = 0, then the equation becomes   (y) = 4π 2 m 2 (y), which
has the general solution (y) = ae−2π my + be2πmy for complex constants
a, b ∈ C, but this case does not come up in our theorem.
 3.5 Fourier–Whittaker expansions on G L(2, R) 67

For ν(1 − ν) = 0, m = 0, the differential equation (3.4.9) has pre-

cisely two smooth solutions (see (Whittaker and Watson, 1935)), namely:
√ √
2π |m|y · K ν− 12 (2π|m|y), 2π|m|y · Iν− 12 (2π|m|y), where
 1 ν+2k


∞
y
Iν (y) = 2
,
k=0
k! (k + ν + 1)

and
 ∞
π I−ν (y) − Iν (y) 1 dt
e− 2 y (t+ t ) t ν
1 1
K ν (y) = · =
2 sin πν 2 0 t
are classical Bessel functions which have the following asymptotic behavior:
√ 1
lim y Iν (y)e−y = √ ,
y→∞ 2π
2
√ π
lim y K ν (y)e y = .
y→∞ 2
The assumption that (y) has polynomial growth at ∞ forces
&
(y) = a 2π|m|y · K ν− 12 (2π|m|y), (a ∈ C)

which gives us multiplicity one.

3.5 Fourier–Whittaker expansions on G L(2, R)

We shall now show that every non–constant Maass form for S L(2, Z) can be
expressed as an infinite sum of Whittaker functions of type (3.4.5).

Proposition 3.5.1 Let f be a non–constant Maass form of type ν for S L(2, Z).
Then for z ∈ h2 we have the Whittaker expansion

&
f (z) = an 2π y · K ν− 12 (2π|n|y) · e2πinx
n=0

for complex coefficients an (n ∈ Z).

Proof Recall (3.4.1) which says that the fact that f (z) is periodic in x implies
that f has a Fourier expansion of type


f (z) = An (y)e2πinx .
n∈Z

Since  f = ν(1 − ν) f it follows that

(An (y)e2πinx ) = ν(1 − ν)An (y)e2πinx .

68 Automorphic forms and L–functions for S L(2, Z)

Consequently, An (y)e2πinx must be a Whittaker function of type ν associ-

ated to e2πinx . The assumption that f is not the constant function implies, by
Proposition 3.3.3, that ν = 0, 1. Since a Maass form f is an L2 function, i.e.,

d xd y
| f (z)|2 < ∞,
y2
S L(2,Z)\h2

it easily follows that An (y)e2πinx must have polynomial growth at ∞. The proof
of Proposition 3.5.1 is now an immediate consequence of the multiplicity one
Theorem 3.4.8.

3.6 Ramanujan–Petersson conjecture

Ramanujan had great interest in the Fourier coefficients of the  function,
which is defined by the product

∞

∞
(z) = e2πi z (1 − e2πinz )24 = τ (n)e2πinz
n=1 n=1

for z = x + i y with x ∈ R, y > 0. One may easily compute that

τ (1) = 1, τ (2) = −24, τ (3) = 252, τ (4) = −1472, τ (5) = 4830,

τ (6) = −6048, τ (7) = −16744, τ (8) = 84480, . . .

and Ramanujan conjectured that

11
τ (n) ≤ n 2 d(n)

for all n = 1, 2, 3, . . . The  function is a cusp form of weight 12 for S L(2, Z).
This means that
 
az + b
 = (cz + d)12 (z)
cz + d
 
a b
for all ∈ S L(2, Z). Petersson generalized Ramanujan’s conjecture to
c d
holomorphic cusp forms f of weight k which satisfy
 
az + b
f = (cz + d)k f (z)
cz + d
 
a b
for all ∈ S L(2, Z). Petersson conjectured that the nth Fourier coef-
c d

ficient of a weight k cusp form is bounded by O n (k−1)/2 d(n) . This explains
 3.6 Ramanujan–Petersson conjecture 69

the 11
2
in Ramanujan’s conjecture. We remark that Petersson’s conjecture has
been proved by Deligne for all holomorphic cusp forms (of even integral weight)
associated to congruence subgroups of S L(2, Z). Deligne’s proof is based on the
very deep fact that the coefficients of holomorphic cusp forms can be expressed
in terms of the number of points on certain varieties defined over certain finite
fields, and that optimal error terms for the number of points on a variety over a
finite field are a consequence of the Riemann hypothesis (proved by Deligne)
for such varieties.
Non–constant Maass forms for S L(2, Z) can be thought of as non–
holomorphic automorphic functions of weight zero. One might be tempted,
by analogy with the classical theory of holomorphic modular forms, to make
a Petersson type conjecture about the growth of the Fourier coefficients in the
Fourier–Whittaker expansion in Proposition 3.5.1. Remarkably, all evidence
points to the truth of such a conjecture.
We have shown in Proposition 3.5.1 that every non–constant S L(2, Z)–
Maass form of type ν has a Fourier expansion of type

&
f (z) = an 2π y · K ν− 12 (2π|n|y) · e2πinx . (3.6.1)
n=0

We now state the famous Ramanujan–Petersson conjecture for Maass forms.

Conjecture 3.6.2 (Ramanujan–Petersson) The Fourier–Whittaker

coefficients an (n ∈ Z, n = 0), occurring in the expansion (3.6.1), satisfy the
growth condition

|an | = O (d(n))
+
where d(n) = d|n 1 denotes the number of divisors of n, and the O–constant
depends only on the Petersson norm of f .

It is not hard to show that the nth Fourier–Whittaker coefficient of a non–

√
constant Maass form for S L(2, Z) is bounded by |n|.

Proposition 3.6.3 Let f (z) be a non–constant Maass form of type ν for

S L(2, Z), normalized to have Petersson norm equal to 1, i.e.,

d xd y
 f, f  = | f (z)|2 = 1.
y2
S L(2,Z)\h2

Then the Fourier–Whittaker coefficients an (n ∈ Z, n = 0), occurring in the

expansion (3.6.1), satisfy the growth condition
&
|an | = Oν |n| .
70 Automorphic forms and L–functions for S L(2, Z)

Proof It follows from Proposition 3.5.1 that for any fixed Y > 0, and any
fixed n = 0,
 ∞  1
 ∞
d xd y dy
| f (z)|
2
2
= 2π |am | ·
2
|K ν− 12 (y)|2
Y 0 y m=0 2π|m|Y y
 ∞
dy
 |an |2 · |K ν− 12 (y)|2 . (3.6.4)
2π |n|Y y

Now, it is a simple consequence of Lemma 1.1.6 that

⎛ ⎞
 ∞ 1 
d xd y ⎜ d xd y ⎟
| f (z)|2 2
= O ⎝Y −1 · | f (z)|2 −1
⎠ = O(Y ).
Y 0 y y2
S L(2,Z)\h2

If we now choose Y = |n|−1 and combine this estimate with (3.6.4) the propo-
sition immediately follows.

3.7 Selberg eigenvalue conjecture

A non–constant Maass form f of type ν for S L(2, Z) satisfies the partial dif-
ferential equation  f (z) = ν(1 − ν) f (z). It is, therefore, an eigenfunction of
the Laplace operator  with eigenvalue λ = ν(1 − ν). It follows from Propo-
sitions 3.3.2, 3.3.3 that λ > 0. The question arises as to how small λ can be?
Selberg proved that λ ≥ 14 for S L(2, Z), and conjectured that the smallest eigen-
value is greater or equal to 14 for Maass forms associated to any congruence
subgroup of S L(2, Z). Recall that a congruence subgroup  of S L(2, Z) is
a subgroup which contains the so–called principal congruence subgroup of
level N :
       
a b  a b 1 0
(N ) = 
∈ S L(2, Z)  ≡ (mod N ) ,
c d c d 0 1

for some integer N ≥ 1. A Maass form of type ν for  is a smooth non–zero

function f : h2 → C which is automorphic for , i.e., f (γ z) = f (z) for all
γ ∈ , z ∈ h2 , is square integrable on a fundamental domain \h2 , all constant
terms in Fourier expansions at cusps (which are rational numbers or ∞) vanish,
and satisfies  f = ν(1 − ν) f.

Conjecture 3.7.1 (Selberg) Let f be a Maass form of type ν for a congruence

subgroup  ⊂ S L(2, Z). Then ν(1 − ν) ≥ 14 , or equivalently, (ν) = 12 .
 3.8 Finite dimensionality of the eigenspaces 71

We shall now prove this conjecture for S L(2, Z) with a much better lower
bound than 14 . It should be remarked that the eigenvalue 14 can occur for con-
gruence subgroups (see (3.15.1)).
Theorem 3.7.2 (M–F Vigneras) Let f be a Maass form of type ν for
S L(2, Z). Then ν(1 − ν) ≥ 3π 2 /2.
Proof Let D = S L(2, Z)\h2 be the standard fundamental domain (see
Example 1.1.9) for the action of S L(2, Z) on h2 . We also let D ∗ denote the
transform by z → −1/z of D. Now, f is a Maass form of type ν with Fourier–
Whittaker expansion given by (3.6.1):

&
f (z) = an 2π y · K ν− 12 (2π|n|y) · e2πinx .
n=0

Set λ = ν(1 − ν). Then we calculate

  
 2  2 
∂ f   
2λ f, f  =
d xd y
 f (z) · f (z) 2 =   +  ∂ f  d xd y
y  ∂x   ∂y 
D∪D ∗ D∪D ∗

   2  2
1
∞  2
∂ f  ∂ f 
≥     d xd y
 ∂ x  d xd y ≥ ∂x 
√
D∪D ∗ 3 − 12
2

∞
& 2
 
= |an |2 · 4π 2 n 2 ·  2π y · K ν− 12 (2π |n|y) dy
√ n=0
3
2
1
∞  2
3 d xd y
≥ · 4π 2 | f (z)|2
4 √
y2
3 − 12
2

≥ 3π 2  f, f .
Hence, it follows that λ ≥ 3π 2 /2.

3.8 Finite dimensionality of the eigenspaces

Let Sλ denote subspace of all f ∈ L2 (S L(2, Z)\h2 ) which are Maass forms of
type ν with λ = ν(1 − ν).
Theorem 3.8.1 (Maass) For any λ ≥ 0, the space Sλ is finite dimensional.
Remark We already know that the space S0 is one-dimensional and just
contains the constant functions.
72 Automorphic forms and L–functions for S L(2, Z)

Proof By Proposition 3.5.1, every f ∈ Sλ has a Fourier–Whittaker expansion

of type

&
f (z) = an 2π y · K ν− 12 (2π |n|y) · e2πinx (3.8.2)
n=0

with λ = ν(1 − ν). If the dimension of Sλ were infinite, it would be possible,

for every integer n 0 > 1, to construct a non–zero finite linear combination of
Maass forms of type ν which had a Fourier–Whittaker expansion of the form
(3.8.2) where an = 0 for all |n| < n 0 . To complete the proof of Theorem 3.8.1
it is enough to prove the following lemma.

Lemma 3.8.3 Assume that the function

&
f (z) = an 2π y · K ν− 12 (2π|n|y)e2πinx
|n|>n 0

is a Maass form of type ν for S L(2, Z). Then, if n 0 is sufficiently large, it follows
that f (z) = 0 for all z ∈ h2 .

Proof Without loss of generality, we may assume that the Petersson norm
 f, f  = 1. We have

d xd y
1= | f (z)|2
y2
S L(2,Z)\h2

∞  2 
& 2
1
 d xd y
 
≤  an 2π y · K ν− 12 (2π|n|y)e2πinx 
√
 |n|>n 0
 y2
3 − 12
2


∞
& 2 dy
= |an |2 ·  2π y · K ν− 12 (2π |n|y) 2
|n|>n 0 √
y
3
2


∞
 2 dy
 |an |2 ·  K ν− 12 (2π|n|y) . (3.8.4)
|n|>n 0 √
y
3
2

√
We now make use of Proposition 3.6.3 which says that an is bounded by |n|.
We also make use of the well-known asymptotic formula
√ √π
lim y K ν (y)e y = 2
y→∞

for the K –Bessel function. It immediately follows from these remarks and
 3.9 Even and odd Maass forms 73

(3.8.4) that


∞
dy
√
1 |n| · e−4π |n|y  |n|e −2 3π|n|
 e−2n 0 ,
|n|>n 0 √
y2 |n|>n 0
3
2

which is a contradition for n 0 sufficiently large.

3.9 Even and odd Maass forms

We introduce the operator T−1 which maps Maass forms to Maass forms. It is
defined as follows. Let f be a Maass form of type ν for S L(2, Z). Then we
define
   
y x y −x
T−1 f := f .
0 1 0 1
The notation T−1 is used at this stage to conform to standard notation for Hecke
operators which will be defined later in this book. Note that (T−1 )2 is the identity
transformation, so that the eigenvalues of T−1 can only be ±1.
Since the Laplace operator  = −y 2 ((∂ 2 /∂ x 2 ) + (∂ 2 /∂ y 2 )) is invariant
under the transformation x → −x, it is easy to see that T−1 maps Maass forms
of type ν to Maass forms of type ν.
Definition 3.9.1 A Maass form f of type ν for S L(2, Z) is said to be even if
T−1 f = f. It is said to be odd if T−1 f = − f.
Proposition 3.9.2 Let f be a Maass form of type ν for S L(2, Z) with Fourier–
Whittaker expansion

&
f (z) = a(n) 2π y · K ν− 12 (2π|n|y) · e2πinx ,
n=0

as in Proposition 3.5.1. Then a(n) = a(−n) if f is an even Maass form and

a(n) = −a(−n) if f is an odd Maass form.
Proof We have
&  1
a(n) 2π y · K ν− 2 (2π|n|y) =
1 f (z)e−2πinx d x
& 0 1
a(−n) 2π y · K ν− 12 (2π|n|y) = f (z)e2πinx d x
0
 1
= (T−1 f (z))e−2πinx d x,
0

after making the transformation x → −x. The result immediately follows.

74 Automorphic forms and L–functions for S L(2, Z)

Finally, we remark that if f is an arbitrary Maass form of type ν for S L(2, Z),
then
13 4 13 4
f (z) = f (z) + T−1 f (z) + f (z) − T−1 f (z)
2 2
3 4 3 4
where 12 f (z) + T−1 f (z) is an even Maass form and 12 f (z) − T−1 f (z) is
an odd Maass form.

3.10 Hecke operators

We shall first define Hecke operators in a quite general setting, and then return
to the specific case of L2 S L(2, Z)\h2 .
Let G be a group which acts continuously on a topological space X as in
Definition 1.1.1. Let  ⊂ G be a discrete subgroup of G as in Definition 1.1.4.
Assume the quotient space \X has a left -invariant measure d x and define
the C-vector space:
   

L (\X ) = f : \X → C 
2  | f (x)| d x < ∞ .
2
(3.10.1)
\X

The commensurator of , denoted C G (), defined by

 
C G () := g ∈ G  (g −1 g) ∩  has finite index in both  and g −1 g ,
(3.10.2)

is of fundamental importance in the theory of Hecke operators. Here, for exam-

ple, g −1 g denotes the set of all elements of the form g −1 γ g with γ ∈  and we
have identities of type  ·  = . This notation and its obvious generalizations
are used in all that follows.
For every fixed g ∈ C G (), the group  can be expressed as a disjoint union
of right cosets

(
d

= (g −1 g) ∩  δi , (3.10.3)
i=1

where d is the index of (g −1 g) ∩  in . Note also that (3.10.3) may be

5
rewritten as g −1 g = i g −1 gδi which is equivalent to

(
d
g = gδi . (3.10.4)
i=1
 3.10 Hecke operators 75

Definition 3.10.5 (Hecke operators) Let a group G act continuously on

a topological space X and let  be a discrete subgroup of G as in Defini-
tion 1.1.4. For each g ∈ C G (), with C G () given by (3.10.2), we define a Hecke
operator

Tg : L2 (\X ) → L2 (\X )

by the formula

d
Tg ( f (x)) = f (gδi x),
i=1

for all f ∈ L2 (\X ), x ∈ X and δi , (i = 1, 2, . . . , d) given by (3.10.4).

In order for Tg to be well defined, we just need to check that

Tg ( f ) ∈ L2 (\X ). Now, for every γ ∈ ,

d
Tg ( f (γ x)) = f (gδi γ x).
i=1

But δi γ = δi δσ (i) for some permutation σ of {1, 2, . . . , d} and some

δi ∈ (g −1 g) ∩ . Also, note that gδi γ = gδi δσ (i) = δi gδσ (i) for some other
δi ∈ . It follows that

d

d
Tg ( f (γ x)) = f δi gδσ (i) x = f (gδi x) = Tg ( f (x)),
i=1 i=1

so Tg is well defined.
We also consider for any integer m, the multiple mTg , which acts on
f ∈ L2 (\X ) by the canonical formula (mTg )( f (x)) = m · Tg ( f (x)). In this
+
manner, one constructs the Hecke ring of all formal sums m k Tgk with
k∈Z
m k ∈ Z.
We now define a way of multiplying Hecke operators so that the product
of two Hecke operators is a sum of other Hecke operators. For g, h ∈ C G (),
consider the coset decompositions
( (
g = αi , h = β j , (3.10.6)
i j

as in (3.10.4). Then
( ( ( (
(g) · (h) = gβ j = αi β j = w = w.
j i, j w ⊂ gh w ⊂ gh
76 Automorphic forms and L–functions for S L(2, Z)

One may then define, for g, h ∈ C G (), the product of the Hecke operators,
Tg Th , by the formula


Tg Th = m(g, h, w)Tw , (3.10.7)
w ⊂ gh

where m(g, h, w) denotes the number of i, j such that αi β j = w with αi , β j

given by (3.10.6). One checks that this multiplication law is associative.
Definition 3.10.8 (The Hecke ring) Let a group G act continuously on a
topological space X and let  be a discrete subgroup of G as in Definition 1.1.4.
Fix any semigroup  such that  ⊂  ⊂ C G (). The Hecke ring R, is
defined to be the set of all formal sums


ck Tgk
k
with ck ∈ Z, gk ∈ . The multiplication law in this ring is induced from (3.10.7).
Definition 3.10.9 (Antiautomorphism) By an antiautomorphism of a
group G we mean a map g → g ∗ (for g ∈ G) satisfying (gh)∗ = h ∗ g ∗ for
all g, h ∈ G.
The following theorem is of supreme importance in the theory of auto-
morphic forms. It is the key to understanding Euler products in the theory of
L-functions.
Theorem 3.10.10 (Commutativity of the Hecke ring) Let R, be the
Hecke ring as in Definition 3.10.8. If there exists an antiautomorphism g → g ∗
of C G () such that  ∗ =  and (g)∗ = g for every g ∈ , then R, is
a commutative ring.
Proof Let g ∈ . Since (g)∗ = g, it immediately follows that if we
decompose g into either left or right cosets of , then the number of left
cosets must be the same as the number of right cosets. Let α ⊂ g be a right
coset and let β ⊂ g be a left coset. Then α ∈ g = β. Consequently
α = γβγ  with γ , γ  ∈ . Then γ −1 α = βγ  must be both a left and right coset
representative. It easily follows that there exists a common set of representatives
{αi } such that
( (
g = αi = αi .
i i
Similarly, for any other h ∈ , we have
( (
h = β j = βj
j j

for some set of common representatives {β j }.

 3.11 Hermite and Smith normal forms 77

Now, by our assumptions about the antiautomorphism ∗ we immedi-

ately see that (g)∗ = g ∗  and (h)∗ = h ∗ . We also know that
5
gh = w w which implies that (gh)∗ = gh. It follows that
(
(h) · (g) = (h ∗ ) · (g ∗ ) = (gh)∗ = w.
w

This tells us that

Tg · Th = m w w,
w


Th · Tg = m w w,
w

with the same components w but with possibly different integers m w , m w . To

complete the proof, we must prove that m w = m w . But
 
  # i, j  αi β j  = w

m w = # i, j αi β j = w =  .
# u ∈ w/   u ⊂ w
Similarly,
 
# i, j  β j αi  = w
m w =  .
# u ∈ w/   u ⊂ w
It only remains to show that
   
# i, j  αi β j  = w = # i, j  β j αi  = w . (3.10.11)
5 5 5
Now g = (g)∗ = αi∗ and h = (h)∗ = β ∗j = β ∗j .
Therefore, we have
(
(g) · (h) = αi∗ β ∗j .
i, j

So (3.10.11) will follow if

   
# i, j  αi∗ β ∗j  = w = # i, j  β j αi  = w . (3.10.12)
Clearly, (3.10.12) holds, as one easily sees by applying the antiautomor-
phism ∗.

3.11 Hermite and Smith normal forms

We have seen in the previous Section 3.10 that Hecke operators are defined by
expressing a double coset as a union of right cosets. In the classical literature, if
two matrices are in the same right coset then they are said to be right equivalent,
while if they are in the same double coset, then they are said to be equivalent.
78 Automorphic forms and L–functions for S L(2, Z)

Hermite found a canonical or normal form for right or left equivalent matrices
while Smith found a normal form for equivalent matrices. The Hermite and
Smith normal forms play such an important role in Hecke theory that we have
decided to give a self-contained exposition at this point.

Theorem 3.11.1 (Hermite normal form) For n ≥ 2, let A be an n × n

matrix with integer coefficients and Det(A) > 0. Then there exists a unique
upper triangular matrix B ∈ G L(n, Z) which is left equivalent to A, (i.e.,
B = γ A with γ ∈ S L(n, Z)) such that the diagonal entries of B are positive
and each element above the main diagonal lies in a prescribed complete set of
residues modulo the diagonal element below it.

Note that B takes the form:

⎛ ⎞
d1 α2,1 α3,1 · · · αn−1,1 αn,1
⎜ α3,2 · · · αn−1,2 ⎟
αn,2
⎜ d2 ⎟
⎜ ⎟
⎜ d3 · · · αn−1,3 αn,3
⎟
B=⎜ ⎜ .. .. ..⎟
⎟
⎜ . . .⎟
⎜ ⎟
⎝ dn−1 αn,n−1 ⎠
dn

where each αk, j satisfies 0 ≤ αk, j < dk .

Proof We first prove that every matrix in G L(n, Z), with n ≥ 1, is left equiva-
lent to an upper triangular matrix. The result is obvious if n = 1. We proceed by
induction on n. Let t (a1 , a2 , . . . , an ) denote the first column of A. Either every
ai = 0, (i = 1, 2, . . . , n) or some element is non-zero. Suppose the latter and set
δ = 0 to be the greatest common divisor of the elements ai , (i = 1, 2, . . . , n).
+
n
Then there exist coprime integers γ1 , γ2 , . . . , γn such that ai γi = δ. Thus,
i=1
there exists a matrix γ  ∈ S L(n, Z) with first row (γ1 , γ2 , . . . , γn ) such that the
matrix γ  A has δ in the (1, 1) position and all the remaining elements of the
first column are multiples of δ. By successively multiplying γ  A on the left by
matrices which have 1s on the diagonal, zeros everywhere else except for one
element in the first row one may easily show that there exists γ ∈ S L(n, Z)
such that γ A takes the form
⎛ ⎞
δ ∗ ∗ ··· ∗
⎜0 ⎟
⎜ ⎟
⎜  ⎟
γ A = ⎜0 A ⎟
⎜. ⎟
⎝ .. ⎠
0
 3.11 Hermite and Smith normal forms 79

where A is an (n − 1) × (n − 1) matrix. Byinduction, 

 we may bring A to upper
1
triangular form by left multiplication by , with M ∈ S L(n − 1, Z).
M
This proves that A is left equivalent to an upper triangular matrix T . It is clear
that we can make the diagonal elements of T positive by multiplying on the
left by a suitable diagonal matrix with entries ±1. Let T  denote this upper
triangular matrix with positive diagonal entries.
In order to make each element of T  above the main diagonal lie in a pre-
scribed complete set of residues modulo the diagonal element which lies below
it, we repeatedly multiply T  on the left by matrices of the form I + m i, j E i, j
(with suitable integers m i, j ) where I is the n × n identity matrix and E i, j is the
n × n matrix with zeros everywhere except at the (i, j) position where there is
a 1. For example, if ti, j is the ( j, i) entry of T  and di is the ith diagonal entry
of T  , then (I + m i, j E i, j )T  has a new ( j, i) entry which is ti, j + di · m i, j .
We leave the proof of the uniqueness of the Hermite normal form to the
reader.

Theorem 3.11.2 (Smith normal form) For n ≥ 2, let A be an n × n integer

matrix with Det(A) > 0. Then there exists a unique diagonal matrix D of the
form
⎛ ⎞
dn
⎜ .. ⎟
⎜ . ⎟
D=⎜ ⎟
⎝ d2 ⎠
d1
where 0 < d1 |d2 , d2 |d3 , . . . , dn−1 |dn , and A = γ1 Dγ2 for γ1 , γ2 ∈ S L(n, Z).
Proof We may assume that A contains a non-zero element which may be
brought to the (n, n) position by suitable row and column interchanges. As
in the proof of Theorem 3.11.1, this element may be replaced by the greatest
common divisor of the last column and the last row, and will divide every
element of the last row and column. By further elementary row and column
operations, we may obtain a new matrix B where every element of B (except
at the (n, n) position) in the last row and column is zero. Thus
⎛ ⎞
0
⎜ B 0 ⎟
⎜ ⎟
⎜ .. ⎟
B=⎜ . ⎟ (3.11.3)
⎜ ⎟
⎝ 0 ⎠
0 0 · · · 0 bn,n
where bn,n = 0 and B  is an (n − 1) × (n − 1) matrix.
80 Automorphic forms and L–functions for S L(2, Z)

Suppose that the submatrix B  contains an element bi, j which is not divisible
by bn,n . If we add column j to column n, then the new column n will be of the
form
t
(b1, j , b2, j , . . . , bi, j , . . . , bn−1, j , bn,n ).

We may then repeat the previous process and replace bn,n with a proper
divisor of itself. Continuing in this manner, we obtain a matrix B of the form
(3.11.3) which is equivalent to A and where bn,n divides every element of the
matrix B  .
The entire previous process can then be repeated on B  . Continuing induc-
tively, we prove our theorem. Again, we leave the proof of uniqueness to the
reader.

3.12 Hecke operators for L2 (S L(2, Z)\h2 )

We shall now work out the theory of Hecke operators for L2 (S L(2, Z)) \h2 . In
this case let

G = G L(2, R),  = S L(2, Z), X = G L(2, R)/(O(2, R) · R× ) = h2 .

 
n0n1 0
For integers n 0 , n 1 ≥ 1, it is easily seen that the matrix ∈ C G ().
0 n0
 
n0n1 0
Let  denote the semigroup generated by the matrices ,
0 n0
(n 0 , n 1 ≥ 1) and the modular group . In this situation, we have the antiau-
tomorphism

g → t g, g ∈ ,

where t g denotes the transpose of the matrix g. Since diagonal matrices are
always invariant under transposition, one immediately sees that the conditions
of Theorem 3.10.10 are satisfied so that the Hecke ring R, is commutative.

Lemma 3.12.1 Fix n ≥ 1. Define the set,

   
a b 
Sn := ad = n, 0 ≤ b < d .
0 d 
Then one has the disjoint partition
(   (
m0m1 0
  = α. (3.12.2)
0 m0 α∈Sn
m 0 m 1 =n
2
 3.12 Hecke operators for L2 (S L(2, Z)\h2 ) 81

Proof It is easy to see that the decomposition is disjoint, because if

     
r s a b a b
=
t u 0 d 0 d
then we must have t = 0, and, therefore, r = u = 1, s = 0.
Next, by Theorem 3.11.2, every element on the right-hand side of (3.12.2)
can be put into Smith normal form, so must occur as an element on the left-hand
side of (3.12.2). Similarly, by Theorem 3.11.1, every element on the left-hand
side of (3.12.2) can be put into Hermite normal form, so must occur as an
element on the right-hand side of (3.12.2). This proves the equality of the two
sides of (3.12.2).

Now, two double cosets are either the same or totally disjoint and different.
Thus a union of double cosets can be viewed as an element in the Hecke ring
as in Definition 3.10.8. It follows that for each integer n ≥ 1, we have a Hecke
operator Tn acting on the space of square integrable automorphic forms f (z)
with z ∈ h2 . The action is given by the formula
 
1
az + b
Tn f (z) = √ f . (3.12.3)
n ad=n d
0≤b<d

Clearly, T1 is just the identity operator. Note that we have introduced the nor-
√
malizing factor of 1/ n to simplify later formulae.
The C-vector space L2 (\h2 ) has a natural inner product (called the
Petersson inner product), denoted  , , and defined by

d xd y
 f, g = f (z)g(z) ,
y2
\h2

for all f, g ∈ L2 (\h2 ), where d xd y

y2
denotes the left invariant measure as given
in Proposition 1.5.3.

Theorem 3.12.4 (Hecke operators are self-adjoint) The Hecke operators

Tn , (n = 1, 2, . . . ) defined in (3.12.3) satisfy

Tn f, g =  f, Tn g,

for all f, g ∈ L2 (\h2 ).

Proof Lemma 3.12.1 says that

(   (
m0m1 0
  = α.
0 m0 α∈Sn
2
m 0 m 1 =n
82 Automorphic forms and L–functions for S L(2, Z)

Since the left union above is invariant on the right by any σ ∈ , it follows that
for any σ1 , σ2 ∈  :
(   (
m0m1 0
 = σ1 ασ2 .
0 m0 α∈S
m 0 m 1 =n
2 n

Further, since diagonal matrices are invariant under transposition, we see that
(   (
m0m1 0
 =  σ1 · t α · σ2 .
0 m0 α∈S
m 0 m 1 =n
2 n

But the action of the Hecke operator is independent of the choice of right coset
decomposition. Consequently


1 d xd y
Tn f, g = √ f (σ1 · t α · σ2 z)g(z)
n α∈Sn
y2
\h2


1 d xd y
=√ f (z) g σ2−1 · t α −1 · σ1−1 z , (3.12.5)
n α∈Sn
y2
\h2

after making the change of variables z → σ2−1 · t α −1 · σ1−1 z.

   
0 −1 −1 0 1
We shall now choose σ1 = and σ2 = σ1 = . A
1 0 −1 0
 
a b
simple computation shows that for α = ,
0 d
 
1 a b
σ2−1 · t α −1 · σ1−1 = .
ad 0 d
 
a b
Since the action of on z ∈ h2 is the same as the action of
0 d
 
a b
(1/ad) , we immediately get that
0 d
1
 −1 t −1 −1
√ g σ2 · α · σ1 z = Tn g(z).
n α∈Sn

Plugging this into (3.12.5) completes the proof of the theorem.

Theorem 3.12.6 The Hecke operators Tn , (n = 1, 2, . . . ) as defined in

(3.12.3) commute with each other, commute with theoperator T−1 in Definition
3.9.1, and commute with the Laplacian  = −y 2 ∂∂x 2 + ∂∂y 2 . Furthermore
2 2

T−1 commutes with .

 3.12 Hecke operators for L2 (S L(2, Z)\h2 ) 83

Proof We have already pointed out at the very beginning of this section that
matrix transposition satisfies the conditions of Theorem 3.10.10 which implies
that the Hecke operators commute with each other, i.e.,

Tm Tn = Tn Tm , ∀m, n ≥ 1.

That the Hecke operators also commute with T−1 and , and T−1 and  com-
mute with each other is a fairly straightforward calculation which we leave to
the reader.

By standard methods in functional analysis, it follows from Theorems 3.12.4

and 3.12.6 that the space L2 (\h2 ) may be simultaneously
 5 diagonaliazed by
the set of operators T := Tn | n = −1, n = 1, 2, . . . {}. We may, there-
fore, consider Maass forms f which are simultaneous eigenfunctions of T . If
 f = ν(1 − ν) f , then f will be a Maass form of type ν as in Definition 3.3.1.
It will be even or odd depending on whether T−1 f = f or T−1 f = − f as in
Definition 3.9.1. Further, there will exist real numbers λn such that

Tn f = λn f, (3.12.7)

for all n = 1, 2, . . .

Theorem 3.12.8 (Muliplicativity of the Fourier coefficients) Consider

&
f (z) = a(n) 2π y · K ν− 12 (2π|n|y) · e2πinx ,
n=0

a Maass form of type ν, as in Proposition 3.5.1, which is an eigenfunction

of all the Hecke operators, i.e., (3.12.7) holds. If a(1) = 0, then f vanishes
identically. Assume f = 0, and it is normalized so that a(1) = 1. Then

Tn f = a(n) · f, ∀n = 1, 2, . . .

Furthermore, we have the following multiplicativity relations:

a(m)a(n) = a(mn), if (m, n) = 1,

 mn 
a(m)a(n) = a ,
d|(m,n)
d2
a( pr +1 ) = a( p)a( pr ) − a( pr −1 ),

for all primes p, and all integers r ≥ 1.

Proof It follows from Proposition 3.5.1 that

&
f (z) = a(M) 2π y · K ν− 12 (2π|M|y) · e2πi M x .
M=0
84 Automorphic forms and L–functions for S L(2, Z)

It is convenient to rewrite (3.12.3) in the form

 
1
dz + a
Tn f (z) = √ f .
n bd=n b
0≤a<b

It follows that
2  



2πdy dy d x+a
Tn f (z) = a(M) · K ν− 12 2π |M| · e2πi M b .
M=0 bd=n 0≤a<b
nb b

The sum over 0 ≤ a < b is zero unless b|M, in which case the sum is b.
Consequently, if we let M = bm  , and afterwards m = dm  , then we see that
2


2πdby 

Tn f (z) = a(bm ) · K ν− 12 (2π|m  |dy)e2πidm x
m  =0 bd=n
n


 mb  &
= a · 2π y · K ν− 12 (2π |m|y) e2πimx
m=0 bd=n
d
d|m


 mn  &
= a · 2π y · K ν− 12 (2π|m|y) e2πimx .
m=0 d|m, d|n
d2

Since Tn f = λn f , it immediately follows that

 mn 
λn a(m) = a . (3.12.9)
d|(m,n)
d2

For m = 1, the identity (3.12.9) gives

a(n) = λn a(1) (3.12.10)

for all values of n. Consequently, if a(1) = 0 the Maass form f would have to
vanish identically. Thus, we may assume that if f = 0, then it is normalized
so that a(1) = 1. In this case, the identity (3.12.10) shows that λn = a(n). The
other claims of Theorem 3.12.8 follow easily.

3.13 L–functions associated to Maass forms

Let

&
f (z) = a(n) 2π y · K ν− 12 (2π|n|y) · e2πinx (3.13.1)
n=0

be a non-zero Maass form of type ν for S L(2, Z) which is an eigenfunction of

all the Hecke operators Tn , (n = 1, 2, . . . ) given by (3.12.3), and in addition, is
 3.13 L–functions associated to Maass forms 85

also an eigenfunction of T−1 given in Section 3.9. Assume f is normalized so

that a(1) = 1.
We have shown in Theorem 3.12.8 that
Tn f = a(n) · f
and that the eigenvalues a(n) satisfy multiplicative relations. It follows from
√
the bound a(n) = O( n) given in Proposition 3.6.3 that for (s) > 32 , the
+
∞
series a(n)n −s converges absolutely. Theorem 3.12.8 tells us that the Fourier
n=1
coefficients a(n) are multiplicative and satisfy a(m)a(n) = a(mn) if (m, n) = 1.
Consequently, we may write



∞ 
∞
a( p
)
a(n)n −s = .
n=1 p =0
p s
Let us evaluate a typical factor


∞
a( p )
φ p (s) := .
=0
p s
We compute (with the convention that a( p −1 ) = 0)
a( p)
∞
a( p )a( p)
∞
a( p +1 ) + a( p −1 )
φ p (s) = =
ps =0
p ( +1)s =0
p ( +1)s
= φ p (s) − 1 + p −2s φ p (s),
and, therefore,
φ p (s) = (1 − a( p) p −s + p −2s )−1 .
It immediately follows that


∞ 
a(n)n −s = (1 − α p p −s )−1 (1 − β p p −s )−1 (3.13.2)
n=1 p

= (1 − a( p) p −s + p −2s )−1 ,
p

where for each prime p, we have α p · β p = 1, α p + β p = a( p).

Thus, the Dirichlet series (3.13.2) has an Euler product which is very similar
to the Euler product


∞ 
n −s = (1 − p −s )−1
n=1 p

of the Riemann zeta function, the major difference being that there are two
Euler factors for each prime instead of one. We shall show that (3.13.2) also
86 Automorphic forms and L–functions for S L(2, Z)

satisfies a functional equation s → 1 − s. It is natural, therefore, to make the

following definition.
Definition 3.13.3 Let s ∈ C with (s) > 32 , and let f (z) be a Maass form
given by (3.13.1). We define the L–function L f (s) (termed the L–function
associated to f ) by the absolutely convergent series


∞
L f (s) = a(n)n −s . (3.13.4)
n=1

We now show that the L–function, L f (s), satisfies a functional equation

s → 1 − s. The only additional complication is the fact that there are two
distinct functional equations depending on whether f is an even or odd Maass
form as defined in Section 3.9.
Proposition 3.13.5 Let f be a Maass form of type ν for S L(2, Z). Then the
L–function, L f (s), (given in Definition 3.13.3) has a holomorphic continuation
to all s ∈ C and satisfies the functional equation



−s
s +  − 12 + ν s +  + 12 − ν
 f (s) := π   L f (s)
2 2
= (−1) ·  f (1 − s),
where  = 0 if f is even (T−1 f = f ) and  = 1 if f is odd (T−1 f = − f ).
Proof We follow the line of reasoning in Riemann’s original proof of the
functional equation for ζ (s), which is to set x = 0 and take the Mellin transform
in y of f (z). We shall need the well-known transform
 ∞    
s+ 12 dy − 32 +s 1+s−ν s+ν
K ν− 12 (y) y =2   ,
0 y 2 2
which is valid for (s − ν) > −1.
If f is an even Maass form, the computation goes as follows. For (s) > 1,
the following integrals converge absolutely.
∞   ∞

∞ &
y 0 dy dy
f ys =2 an 2π y · K ν− 12 (2π |n|y) y s
0 1 y n=1
y
0 0

 ∞ 1 dy
= 2(2π)−s L f s + 12 K ν− 12 (y) y s+ 2
y
  0 
s + ν 1 + s − ν 
= 2− 2 π −s 
1
 L f s + 12 .
2 2
(3.13.6)
 3.13 L–functions associated to Maass forms 87

 
y 0
On the other hand, since f is invariant under y → y −1 , it follows
0 1
that
∞  
y 0 dy
f ys
0 1 y
0
1   ∞  
y −1 0 dy y 0 dy
= f y s
+ f ys
0 1 y 0 1 y
0 1
∞     dy
y 0
= f y s + y −s . (3.13.7)
0 1 y
1
 
y 0
Since f has exponential decay in y as y → ∞, the above integral
0 1
is easily seen to converge for all s ∈ C, and thus defines an entire function. It
is also invariant under the transformation s → −s. This gives the functional
equation for even Maass forms.
We now consider the case when f is an odd Maass form. The above argu-
ment does not work because by Proposition 3.9.2, you would have an = −a−n ,
+
and, therefore, an · |n|−s = 0 which implies by the calculation (3.13.6)
n=0
that
∞  
y 0 dy
f ys = 0.
0 1 y
0

To get around this difficulty, we consider

   
∂ y x ∂
&
f = an 2π y · K ν− 12 (2π |n|y) · e2πinx
∂x 0 1 ∂ x n=0

&
= 2πi an · n · 2π y · K ν− 12 (2π |n|y) · e2πinx .
n=0

Hence,


∂ & 
an 2π y · K ν− 12 (2π|n|y) · e 2πinx 
∂x 
n=0 x=0

&
= 2πi an · n · 2π y · K ν− 12 (2π |n|y).
n=0
88 Automorphic forms and L–functions for S L(2, Z)

It follows as in (3.13.6), that

∞ 

∂ &  dy
an 2π y · K ν− 12 (2π|n|y)e2πinx  ys
∂x  y
n=0 x=0
0
   
√ s+ν 1+s−ν
= 2 i · π 1−s   L f (s − 12 ). (3.13.8)
2 2

Now, because f is automorphic, we have:

    
y x 0 −1 y x
f = f
0 1 1 0 0 1


y −x
= f x 2 +y 2 x 2 +y 2 .
0 1

Consequently
  

 
∂ y x  ∂ y  −x
f  = f x 2 +y 2  x 2 +y 2
∂x 0 1  ∂x 0 1 
x=0 x=0
 −1  
1 ∂ y x 
=− · f 
y2 ∂x 0 1 
x=0
 −1 
1 ∂f y 0
=− .
y2 ∂x 0 1
 
y x
Here, if we let f = f (y, x, z, w), then d
f = f (0,1,0,0) . It then
z w dx

follows as in (3.13.7) that

∞  
∂f y 0 dy
ys
∂x 0 1 y
0
1   ∞  
∂f y −1 0 dy ∂f y 0 dy
=− y s−2 + ys
∂x 0 1 y ∂x 0 1 y
0 1
 ∞   6 7 dy
∂f y 0
= y s − y 2−s . (3.13.9)
1 ∂x 0 1 y

The functional equation for odd Maass forms is an immediate consequence of

(3.13.8) and (3.13.9).

 3.14 L-functions associated to Eisenstein series 89

3.14 L-functions associated to Eisenstein series

Let w ∈ C with (w) > 1. The Eisenstein series
1
yw
E(z, w) = ,
2 c,d∈Z |cz + d|2w
(c,d)=1

defined in Definition 3.1.2 has the Fourier–Whittaker expansion

E(z, w) = y w + φ(w)y 1−w

2 2 π w− 2
&
1 1

+ σ1−2w (n)|n|w−1 2π|n|y · K w− 12 (2π |n|y)e2πinx ,

(w)ζ (2w) n=0

given in Theorem 3.1.8. Then E(z, w) is an even automorphic form, i.e., it is

invariant under the transformation x → −x. In an analogous manner to Defi-
nition 3.13.3, we may define the L–function associated to an Eisenstein series.

Definition 3.14.1 We define the L–function associated to the Eisenstein

series E(z, w) to be


∞
σ1−2w (n) · n w− 2 −s .
1
L E(∗,w) (s) =
n=1

The following elementary computation shows that L E(∗,w) (s) is just a product
of two Riemann zeta functions at shifted arguments.


∞
σ1−2w (n) · n w− 2 −s
1
L E(∗,w) (s) =
n=1


∞


∞

∞
n w−s− 2 (md)w−s− 2
1 1
= d 1−2w = d 1−2w
n=1 d|n d=1 m=1
 
= ζ s + w − ζ s − w + 12 .1
2

Consequently, if we define


s+w− 21 s + w − 1

 E(∗,w) (s) := π − 2  2
ζ s + w − 12
2


s−w+ 1
− 2 2
s − w + 12 
×π  ζ s − w + 12
2



s + w − 1
s − w + 1
 
= π −s  2
 2
ζ s + w − 12 ζ s − w+ 12 ,
2 2
90 Automorphic forms and L–functions for S L(2, Z)

then the functional equation,

s   
−s/2 −(1−s)/2 1−s
π  ζ (s) = π  ζ (1 − s),
2 2
of the Riemann zeta function immediately implies that

 E(∗,w) (s) =  E(∗,w) (1 − s).

Note that this matches perfectly the functional equation of an even Maass form
of type w (as in Proposition 3.13.5) as it should be.
We shall now show directly that the Eisenstein series is an eigenfunction
of all the Hecke operators. This explains why the L-function associated to the
Eisenstein series has an Euler product.

Proposition 3.14.2 The Eisenstein series E(z, s) is an eigenfunction of all

the Hecke operators. For n ≥ 1, let Tn denote the Hecke operator (3.12.3).
Then
1
Tn E(z, s) = n s− 2 σ1−2s (n) · E(z, s).

Proof For n ≥ 1, let

   
a b 
n =  ad − bc = n .
c d 

Then the set Sn given in Lemma 3.12.1 is just a set of coset representatives for
1 \n . If R is any set of coset representatives for ∞ \1 then naturally R Sn is
a set of coset representatives for ∞ \n . On the other hand, Sn R is also a set
of coset representatives for ∞ \n . It follows that
1

Tn E(z, s) = √ E(αz, s)
n α∈∞ \n
1


=√ (γ αz)s
n γ ∈∞ \1 α∈∞ \n
1


=√ (αγ z)s
n γ ∈∞ \1 α∈∞ \n
1
1−s  n s
=√ d E(z, s)
n d|n d
1
= n s− 2 σ1−2s (n)E(z, s).

 3.15 Converse theorems for SL(2, Z) 91

3.15 Converse theorems for SL(2, Z)

We have shown that the L–function associated to a Maass form for S L(2, Z) is
an entire function which satisfies a simple functional equation (see Proposition
3.13.5). The converse theorem of Maass–Hecke states that if a Dirichlet series is
entire and bounded in vertical strips and satisfies the same functional equation
as the L-function of a Maass form on S L(2, Z), then it must be an L–function
coming from a Maass form for S L(2, Z).
Now, L–functions have been studied by number theorists for a long time, and
many different L–functions satisfying all sorts of functional equations have been
discovered. Surprisingly, not a single L–function has been found which satisfies
exactly the right functional equation associated to a Maass form on S L(2, Z).
The closest examples known are the Hecke L–functions with grössencharakter
of real quadratic fields discussed in Section 3.2. If we compare the functional
equation of the L–function of an even Maass form given in Proposition 3.13.5:



−s
s − 12 + ν s + 12 − ν
 f (s) := π   L f (s) =  f (1 − s)
2 2

with the functional equation of the Hecke L–function with grössencharakter

given in Proposition 3.2.4:
 s
s − πin 
s + πin 
A log  log 
b(s) :=
n
  L b(s, ψ n )
π 2 2
= nb(1 − s). (3.15.1)
√
with A = N (b) D, we see that the two functional equations would match
πin
up if ν = 12 + log 
and A = 1. Unfortunately, A can never equal 1. It turns
out, however, that the functional equation (3.15.1) does match the functional
equation
1 of a Maass form φb for a congruence subgroup of S L(2, Z) and φb =
π 2 n2
4
+ (log )2
· φb. The converse theorem for congruence subgroups of S L(2, Z)
was first discovered by A. Weil. It requires a family of functional equations
(for twists of the original L–function by Dirichlet characters) instead of just
one functional equation. This is because a congruence subgroup of S L(2, Z),
considered as a finitely generated group will, in general, be generated by several
(more than 2) matrices. Weil’s converse theorem can be used to prove that the
Hecke L–function L b (s, ψ n ) is, in fact, associated to the Maass form φb of type
πin
1
2
+ log 
.
We now state and prove the converse theorem for Maass forms for S L(2, Z).
In order to simplify the exposition, we define what it means for a function of a
complex variable to be EBV (entire and bounded in vertical strips).
92 Automorphic forms and L–functions for S L(2, Z)

Definition 3.15.2 A function f : C → C is said to be EBV (entire and

bounded on vertical strips) if
r f (s) is holomorphic for all s ∈ C.
r For fixed A < B, there exists c > 0 such that | f (s)| < c for A ≤ (s) ≤ B.

+
∞
Theorem 3.15.3 (Hecke–Maass converse theorem) Let L(s) = (a(n)/n s )
n=1
(with a(n) ∈ C) be a given Dirichlet series which converges absolutely for (s)
sufficiently large. Assume that for fixed ν ∈ C, L(s) satisfies the functional
equation



ν −s
s +  − 12 + ν s +  + 12 − ν
 (s) := π   L(s) = (−1) ν (1 − s),
2 2
with  = 0 (respectively  = 1), where ν (s) is EBV. Then

&
a(n) 2π y · K ν− 12 (2π|n|y)e2πinx
n=0

must be an even (respectively odd) Maass form of type ν for S L(2, Z), where
we have defined a(n) = (−1) · a(−n) for n < 0.
+ √
Proof For z ∈ h2 , define f (z) = a(n) 2π y · K ν− 12 (2π |n|y)e2πinx , which
n=0
by our assumptions is an absolutely convergent series. Then clearly
 f (z) = ν(1 − ν) f (z),
√
because the Whittaker function 2π y K ν− 12 (2π y)e2πi x is an eigenfunction of
 with eigenvalue ν(1 − ν). We also get for free the fact that f (z) is periodic
in x. This implies that
  
1 1
f z = f (z).
0 1
Since S L(2, Z) is generated by the two matrices
   
1 1 0 −1
,
0 1 1 0
it follows that all that is left to be done to prove the converse theorem is to check
that
    
y x 0 −1 y x
f = f
0 1 1 0 0 1


y −x
= f x 2 +y 2 x 2 +y 2 . (3.15.4)
0 1

In order to accomplish this, we need a lemma.

 3.15 Converse theorems for SL(2, Z) 93

Lemma 3.15.5 Let F : h2 → C be a smooth eigenfunction of the Laplacian

 with eigenvalue λ, i.e., F = λF. Assume that
    
y 0 ∂F y x 
F =  = 0. (3.15.6)
0 1 ∂x 0 1 
x=0

Then F(z) is identically zero on h2 .

Proof Since F = λF it follows that F is real analytic. Thus, it has a power

+
series expansion in x of the form F(z) = ∞ n=0 bn (y)x . The eigenfunction
n

equation implies that

0 = F(z) − λF(z)


∞
3 4
= − (n + 2)(n + 1)y 2 bn+2 (y) − y 2 bn (y) − λbn (y) x n .
n=0

Now, the initial conditions (3.15.6) tell us that b0 (y) = b1 (y) = 0, and it imme-
diately follows from the recurrence relation
y 2 bn (y) + λbn (y)
bn+2 (y) = −
(n + 1)(n + 2)y 2
that bn (y) = 0 for all integers n ≥ 0.
Lemma 3.15.5 implies that we can prove (3.15.4) if the function
 


y −x
y x x +y x +y
F(z) = f − f
2 2 2 2

0 1 0 1

satisfies the initial conditions (3.15.6). These can be written

   −1 
y 0 y 0
f − f = 0, (3.15.7)
0 1 0 1
   −1 
∂f y 0 −2 ∂ f y 0
+ y = 0. (3.15.8)
∂x 0 1 ∂x 0 1
First note that if f is even (i.e., a(n) = a(−n)), then it is enough to show that
(3.15.7) holds. This is due to the fact that
 

∂f y 0 &
= 2πin a(n) 2π y · K ν− 12 (2π |n|y) = 0,
∂x 0 1 n=0

and similarly
 
−2 ∂f y −1 0
y = 0.
∂x 0 1
In an analogous manner, if f is odd, it is enough to prove (3.15.8).
94 Automorphic forms and L–functions for S L(2, Z)

We shall prove (3.15.7), (3.15.8) for even and odd f , respectively, using the
Mellin inversion formulae
∞
dy
h̃(s) = h(y)y s (3.15.9)
y
0
σ+i∞
1
h(y) = h̃(s)y −s ds. (3.15.10)
2πi
σ −i∞

These formulae hold for any smooth function h : R+ → C and any fixed real
σ provided h̃(s) is EBV.
We apply (3.15.7), (3.15.8) with h̃(s) = ν (s). It follows from our previous
calculations that we have the Mellin transform pair:
 ∞   1
ν ∂ 1 dy
 (s) = f (z) y s− 2 ,
0 ∂x x=0 y

  1 σ+i∞
∂ 1 dy
ν (s) y 2 −s
1
f (z) = . (3.15.11)
∂x x=0 2πi y
σ −i∞

The functional equation,

ν (s) = (−1) ν (1 − s),
combined with (3.15.11) immediately prove (3.15.7) and (3.15.8) for even and
odd f , respectively. This completes the proof of the converse Theorem 3.15.3.

3.16 The Selberg spectral decomposition

Our main goal of this section is the Selberg spectral decomposition for S L(2, Z)
which states that
  
L2 S L(2, Z)\h2 = C ⊕ L2cusp S L(2, Z)\h2 ⊕ L2cont S L(2, Z)\h2 ,
where C is the one–dimensional space of constant functions,
L 2cusp S L(2, Z)\h2 represents the Hilbert space of square integrable
functions on h2 whose constant term is zero, and L 2cont S L(2, Z)\h2 repre-
sents all square integrable functions on h2 which are representable as integrals
of the Eisenstein series. The reason for the terminology L2cusp , L2cont is because
the classical definition of cusp form, introduced by Hecke, requires that
 3.16 The Selberg spectral decomposition 95

the constant term in the Fourier expansion around any cusp (a real number
equivalent to ∞ under the discrete group) be zero, and also because the
Eisenstein series is in the continuous spectrum of the Laplace operator. The
latter means that E(z, s) = s(1 − s)E(z, s), or that s(1 − s) is an eigenvalue
of  for any complex number s.
Let η j (z), ( j = 1, 2, . . . ) be an orthonormal basis of Maass forms for
S L(2, Z). We may assume as in Theorem 3.12.8 that each η j is an eigen-
function of all the Hecke operators, so that its L-function has an Euler product.
We shall also adopt the convention that
2
3
η0 (z) = ,
π
is the constant function of norm 1. The Selberg spectral decomposition is given
in the following theorem.

Theorem 3.16.1 (Selberg spectral decomposition) Let f ∈ L2 (S L(2, Z)\h2 ).

Then we have
2 +i∞
1

∞ 
1
f (z) =  f, η j  η j (z) +  f, E(∗, s) E(z, s) ds,
j=0
4πi
2 −i∞
1

where

d xd y
 f, g = f (z)g(z)
y2
S L(2,Z)\h2

denotes the Petersson inner product on L2 S L(2, Z)\h2 .

We shall not give a complete proof of this theorem, but will only sketch one
of the key ideas of the proof which is contained in the following proposition.

Proposition 3.16.2 Let f (z) ∈ L2 S L(2, Z)\h2 be orthogonal to the
constant function, i.e.,

d xd y
 f, 1 = f (z) 2 = 0.
y
S L(2,Z)\h2

Assume that f is of sufficiently rapid decay so that the inner product


  d xd y
f, E(∗, s̄) = f (z)E(z, s) 2
y
S L(2,Z)\h2
96 Automorphic forms and L–functions for S L(2, Z)

converges absolutely for (s) > 1. Then

2 +i∞
1

1
f (z) = f 0 (z) +  f, E(∗, s)E(z, s) ds,
4πi
2 −i∞
1

where f 0 (z) is automorphic for S L(2, Z) with constant term in its Fourier
*1
expansion equal to zero, i.e., f 0 (z) d x = 0 or f 0 ∈ L2cusp .
0

Proof The main idea of the proof is based on Mellin inversion. Recall that if
h(y) is a smooth complex valued function for y ≥ 0 then the Mellin transform
of h is
 ∞
dy
h̃(s) = h(y)y s .
0 y
The transform is well defined provided there exists c ∈ R such that the integral
converges absolutely for (s) ≥ c, and in this case, h̃(s) is analytic for (s) ≥ c.
The inverse transform is given by

c+i∞
1
h(y) = h̃(s)y −s ds.
2πi
c−i∞

The proof of Proposition 3.16.2 consists of two steps. In the first step it is
shown that the inner product  f, E(∗, s) is the Mellin transform of the constant
term of f (z). In the second step, it is shown that the constant term of
2 +i∞
1

1
 f, E(∗, s)E(z, s) ds
4πi
2 −i∞
1

is the inverse Mellin transform of  f, E(∗, s) which brings you back precisely
to the constant term of f (z). Thus,
2 +i∞
1

1
f (z) −  f, E(∗, s)E(z, s) ds
4πi
2 −i∞
1

is automorphic with constant term equal to zero.

Step 1 Let


∞
f (z) = An (y)e2πinx
n=−∞
 3.16 The Selberg spectral decomposition 97

denote the Fourier expansion of f . Recall that for z ∈ h2 , Is (z) = y s , and ◦

denotes the action of S L(2, Z) on h2 . Since the function f and the measure
d xd y/y 2 are invariant under the action ◦, it follows that for (s) > 1,

1 d xd y
 f, E(∗, s̄) = f (z) Is (γ ◦ z) 2 (3.16.3)
γ ∈ \S L(2,Z)
2 y
∞
S L(2,Z)\h2


1 d xd y
= f (z)Is (z)
2 γ ∈ \S L(2,Z)  y2
∞
γ ◦ S L(2,Z)\h2
 ∞  1
d xd y
= f (z)y s
y2
0 ∞ 0
dy
= A0 (y)y s 2
0 y
= Ã0 (s − 1).

The assumption that f is orthogonal to the constant function implies that the
residue at s = 1 of  f, E(∗, s̄) is zero. Further, E(z, s) = E ∗ (z, s)/ζ (2s) which
implies (by the fact that ζ (1 + it) = 0 for real t) that Ã0 (s − 1) is holomorphic
for (s) ≥ 12 . The functional equation (Theorem 3.1.10) of the Eisenstein series
tells us that

 f, E(∗, s̄) = φ(s) f, E(∗, 1 − s̄),

or equivalently that

Ã0 (s − 1) = φ(s) Ã0 (−s).

Ã0 (−s) = φ(1 − s) Ã0 (s − 1). (3.16.4)

Step 2 By Mellin inversion it follows that for c > 1,


c+i∞
1
A0 (y) = Ã0 (s − 1) y 1−s ds.
2πi
c−i∞

Since Ã0 (s − 1) is holomorphic for (s) ≥ 12 , this implies that we may shift
the above line of integration to (s) = 12 . It follows from the transformation
s → 1 − s that
2 +i∞ 2 +i∞
1 1
 
1 1
A0 (y) = Ã0 (s − 1) y 1−s ds = Ã0 (−s) y s ds.
2πi 2πi
2 −i∞ 2 −i∞
1 1
98 Automorphic forms and L–functions for S L(2, Z)

If we now make use of the functional equation (3.16.4), we easily obtain

2 +i∞
1

1
A0 (y) = Ã0 (−s)(y s + φ(s)y 1−s ) ds. (3.16.5)
4πi
2 −i∞
1

But for (s) = 1

2
we have Ã0 (−s) = Ã0 (s̄ − 1), and by (3.16.3), we have
Ã0 (s̄ − 1) =  f, E(∗, s).
Plugging this into (3.16.5) completes the proof of the theorem.

GL(n)pack functions The following GL(n)pack functions, described in the

appendix, relate to the material in this chapter:

EisensteinFourierCoefficient EisensteinSeriesTerm HeckeCoefficientSum

HeckeEigenvalues HeckeMultiplicativeSplit HeckeOperator
HermiteFormLower HermiteFormUpper SmithForm
SmithElementaryDivisors SmithInvariantFactors Whittaker
WhittakerStar.
 4
Existence of Maass forms

Maass forms for S L(2, Z) were introduced in Section 3.3. An important objec-
tive of this book is to generalize these functions to the higher-rank group
S L(n, Z) with n ≥ 3. It is a highly non-trivial problem to show that infinitely
many even Maass forms for S L(2, Z) exist. The first proof was given by Selberg
(1956) where he introduced the trace formula as a tool to obtain Weyl’s law,
which in this context gives an asymptotic count (as x → ∞) for the number
of Maass forms of type ν with |ν| ≤ x. Selberg’s methods were extended by
Miller (2001), who obtain Weyl’s law for Maass forms on S L(3, Z) and Müller
(2004), who obtained Weyl’s law for Maass forms on S L(n, Z).
A rather startling revelation was made by Phillips and Sarnak (1985) where it
was conjectured that Maass forms should not exist for generic non-congruence
subgroups of S L(2, Z), except for certain situations where their existence is
ensured by symmetry considerations, see Section 4.1. Up to now no one has
found a single example of a Maass form for S L(2, Z), although Maass (1949)
discovered some examples for congruence subgroups (see Section 3.15). So it
seemed as if Maass forms for S L(2, Z) were elusive mysterious objects and the
non-constructive proof of their existence (Selberg, 1956) suggested that they
may be unconstructible.
Recently, Lindenstrauss and Venkatesh (to appear) found a new, short, and
essentially elementary proof which shows the existence of infinitely many
Maass forms on G(Z)\G(R)/K ∞ where G is a split semisimple group over
Z and K ∞ is the maximal compact subgroup. Lindenstrauss and Venkatesh
were also able to obtain Weyl’s law in a very broad context. Their method
works whenever one has Hecke operators. Although in the case of S L(2, Z),
the proof is perhaps not much simpler than the trace formula, it has the advan-
tage of being much more explicit (it allows one to write down an even cus-
pidal function). However, a much bigger advantage is that it generalizes in a

99
100 Existence of Maass forms

relatively straightforward way to higher rank (unlike the trace formula, where
one encounters formidable technical obstacles). I would very much like to thank
Elon and Akshay for preparing and allowing me to incorporate a preliminary
manuscript which formed the basis of this chapter.

4.1 The infinitude of odd Maass forms for SL(2, Z)

 2 
Let  = −y 2 ∂∂x 2 + ∂∂y 2 be the hyperbolic Laplacian on h2 . It was shown in
2

Theorem 3.16.1 that

  
L2 S L(2, Z)\h2 = C ⊕ L2cusp S L(2, Z)\h2 ⊕ L2cont S L(2, Z)\h2 ,

where L2cont (S L(2, Z)\h2 ) is spanned by the continuous spectrum  of , explic-

itly given by Eisenstein series E(z, 12 + ir ) with r ∈ R, and L2cusp S L(2, Z)\h2
is spanned by Maass forms.
Recall from Section 3.9 that a Maass form

&
f (z) = an 2π y K ν− 12 (2π |n|y)e2πinx
n=0

of type ν for S L(2, Z) (as in Proposition 3.5.1) is even or odd according to

whether
   
y x y −x
T−1 f (z) = T−1 f := f
0 1 0 1
is equal to f (z) or − f (z), respectively.
The following proposition is almost obvious.

Proposition 4.1.1 There are infinitely many odd Maass forms for S L(2, Z).

Proof The image of the endomorphism

 
J : L2 S L(2, Z)\h2 → L2 S L(2, Z)\h2 ,

defined by

J f (z) := f (z) − T−1 f (z),

is purely*cuspidal. This is due to the fact that the constant term of f − J f ,

1
given by 0 f (x + i y) − f (−x + i y) d x = 0 for all y > 0. We leave it to the
reader to show that the image of J is non-trivial.

The rest of this chapter will be devoted to showing that the space of even
Maass forms for S L(2, Z) is also infinite dimensional. The only other known
 4.2 Integral operators 101

proof of this fact uses the trace formula (Selberg, 1956), see also (Hejhal, 1976),
as we already mentioned earlier.

4.2 Integral operators

In this section we shall adopt the classical model of the upper half-plane h2
and consider  z ∈ h2 in
 the form z = x + i y with x ∈ R, y > 0. Note that
y x
every matrix can be put in this form by simply letting it act on
0 1
√
i = −1.
Let d(z, z  ) denote the hyperbolic distance between two points z, z  ∈
h , which is characterized by the property that d(αz, αz  ) = d(z, z  ) for all
2

α ∈ S L(2, R) and z, z  ∈ h2 . It is easy to check that

|z − z  |2
u(z, z  ) = (4.2.1)
4(z)(z  )
satisfies this property, and we have the relation
 
d(z, z  )
u(z, z  ) = sinh2 ,
2
because the hyperbolic distance between the points i and i y0 , with y0 > 1, is
given by
 y0
dy
= log(y0 ).
1 y
Definition 4.2.2 (The Abel transform) The Abel transform F(x) of an
integrable function f (x) on [0, +∞), is given by
∞  
√  f (v)dv
∞
ξ2
F(x) = f x+ dξ = 2 √ .
2 v−x
−∞ x

Proposition 4.2.3 Let f (x) be a continuously differentiable function on

[0, +∞). If F(x) denotes the Abel transform in Definition 4.2.2, then
∞  
1  η2
f (x) = − F x+ dη
2π 2
−∞

is the inverse Abel transform.

102 Existence of Maass forms

Proof First differentiate under the integral sign, and then convert to polar
coordinates, to obtain
∞   ∞ ∞  
 η2 ξ2 η2
F x+ dη = f x + + dξ dη
2 2 2
−∞ −∞ −∞
2π ∞  
 r2
= f x+ r dr dθ
2
0 0
∞
= 2π f  (x + w) dw,
0

where w = r /2 and dw = r dr. This equals −2π f (x + w)|w=0 or −2π f (x)

2

which proves the proposition.

Let g(x) be an even smooth function of compact support on the real line with
Fourier transform
∞
h(t) = g(x)eit x d x.
−∞
Following Selberg (1956), we define a variation of the Abel transform, denoted
k, as follows.
∞
1
(v − u)− 2 dq(v),
1
k(u) = − (4.2.4)
π
u

where
1  √ √
q(v) := g 2 log v + 1 + v .
2
The relation between k and h is called the Selberg transform. It is clear from
the definitions that k is compactly supported and continuous. Indeed, if g is
supported in [−M, M], then k(u) vanishes for u > sinh2 (M/2).
For z, w ∈ h2 , let u(z, w) be given by (4.2.1). Then u(z, w) is real valued
and positive. Define
u := u(z, w). (4.2.5)
Definition 4.2.6 (Point pair invariant) Let g : R → C be an even smooth
function of compact support. The point pair invariant K : h2 × h2 → C
associated to g is the function defined by
K (z, w) = k(u(z, w)) = k(u), (for all z, w ∈ h2 ),
where k is given by (4.2.4) and u is given by (4.2.5).
 4.2 Integral operators 103

It is clear that K is a continuous function on h2 × h2 . Moreover, it fol-

lows from the fact that k is compactly supported, that K (z, w) is supported
in d(z, w) ≤ R for some R that depends only on the support of g. Since the
point pair invariant K is a function of the hyperbolic distance, it is plain that
K (αz, αw) = K (z, w) for all α ∈ S L(2, R), and z, w ∈ h2 .
The point pair invariant K can be used to define an integral operator which
acts on functions in L1 (S L(2, Z)\h2 ).

Definition 4.2.7 (Integral operator) For any function f ∈ L1 S L(2, Z)\h2
and any point pair invariant K (z, w), define the integral operator

K ∗ f (z) := K (z, w) f (w) d ∗ w
h2
∞ ∞
dµdν
= K (z, µ + iν) f (µ + iν) ,
ν2
0 −∞

where for w = µ + iν ∈ h2 , the invariant measure d ∗ w = dµdν/ν 2 , as in

Proposition 1.5.3.

It is easy to check that K ∗ f is also invariant by S L(2, Z), and,

indeed, that f → K ∗ f defines a self-adjoint continuous endomorphism of
L2 (S L(2, Z)\h2 ). The key property of K that we will need is given in the
following lemma.
*∞
Lemma 4.2.8 We have K (i, t + ie x ) dt = e x/2 g(x).
−∞

Proof It follows from (4.2.1) that

(e x − 1)2 + t 2
u(i, t + ie x ) = ,
4e x
so that the integral is exactly equal to
∞    
2 x t2
k sinh + x dt.
2 4e
−∞

But, by (Iwaniec, 1995), we have

∞
k(u)(u − sinh2 (x/2))− 2 du,
1
g(x) = 2
sinh2 (x/2)

from which the lemma follows.

104 Existence of Maass forms

Remark It is a basic fact (although we will not need it for the proof, it
provides valuable intuition) that, if φλ is an eigenfunction of the hyperbolic
Laplacian  with eigenvalue λ = 14 + r 2 , then
K ∗ φλ = h(r )φλ . (4.2.9)

This can be readily established by computing K ∗ y 2 +ir .

1

Now, suppose we take g = g ( j) to be a Dirac δ−sequence with j → ∞. By

this we mean that

lim g ( j) (x) d x = 1,
j→∞ R

and the support of g ( j) shrinks to {0} as j → ∞. In this situation, the Fourier

transform h ( j) approaches the constant function, so we expect, in view of (4.2.9),
that the associated point pair operators
f → K ( j) ∗ f
will approach the identity endomorphism. In fact, we have already seen that as
the support of g ( j) shrinks to zero, then the support of K ( j) shrinks to zero also.
Moreover, it follows from Lemma 4.2.8, after making the substitution ν = e x ,
that
 ∞ ∞
∗ dµdν
K (i, w) d w =
( j)
K ( j) (i, µ + iν)
ν2
h2 0 −∞
∞ ∞
= K ( j) (i, µ + ie x ) · e−x dµ d x
0 −∞
∞
= e−x/2 g ( j) (x) d x −→ 1.
−∞

On the other hand, by the point pair property, and the fact that d ∗ w is an
S L(2, R)-invariant measure, we have for any α ∈ S L(2, R) that
 
K (i, w) d w = K ( j) (αi, αw) d ∗ w
( j) ∗

h2 h2

= K ( j) (αi, αw) d ∗ (αw)

h2

= K ( j) (αi, w) d ∗ w.
h2
 4.3 The endomorphism ♥ 105

Now, for any z ∈ h2 , we may choose α ∈ S L(2, R) such that z = αi. It follows
that

K ( j) (z, w) d ∗ w −→ 1
h2

for any z ∈ h2 , as j → ∞.
Consequently, for any continuous function f , we have

K ( j) ∗ f (z) = K ( j) (z, w) f (w) d ∗ w
h2
 
= K ( j) (z, w) f (z) d ∗ w + K ( j) (z, w)( f (w) − f (z)) d ∗ w
h2 h2

−→ f (z), (4.2.10)

as j → ∞. So the corresponding operators K ( j) will satisfy

K ( j) ∗ f (z) → f (z)

for any continuous funtion f and all z ∈ h2 .

4.3 The endomorphism ♥

In Section 4.1, we showed that there are infinitely many odd Maass forms
for S L(2, Z) by showing that the endomorphism J , given in Proposition
4.1.1, has a purely cuspidal image. The key idea of the present approach
is to construct an explicit endomorphism ♥ of L2cusp = L2cusp (S L(2, Z\h2 )
whose image is purely cuspidal. The endomorphism ♥, however, will use
the arithmetic structure of S L(2, Z) in a much more essential way than J
did.
Recall that for any rational prime p, we have the Hecke operator T p , given
in (3.12.3), which acts on functions f on S L(2, Z)\h2 via the rule

 
1

p−1
z+k
T p f (z) = √ f ( pz) + f . (4.3.1)
p k=0
p

The Hecke operators T p commute with , and we showed in Propositions 3.1.3

and 3.14.2 that the Eisenstein series E(z, 12 + ir ), defined in Definition 3.1.2,
106 Existence of Maass forms

satisfies
     
1 1 1
E z, + ir = + r · E z, + ir
2
(4.3.2)
2 4 2
   
1 −ir 1
T p E z, + ir = ( p + p ) · E z, + ir .
ir
(4.3.3)
2 2
We proceed formally for now. From (4.3.2) and (4.3.3), the operator
√1 √1
♥ := T p − p 4 − − p − 4 −

annihilates E(z, 12 + ir ). Similarly (again at the formal level) ♥ also annihilates

the constant function. The operator ♥ may be given a rigorous interpretation
either in terms of the wave equation or using convolution operators. (In fact, for
technical simplicity in our treatment, we will use not ♥ but a certain smoothed
version.) For the time being, let us accept – as is indeed the case – that ♥ may
be given a rigorous interpretation and that it is a self-adjoint endomorphism of
the space, L2+ , of even square integrable automorphic functions.
Since ♥ kills the continuous spectrum and is self-adjoint, it has cuspidal
image. To show that there exist even Maass forms for S L(2, Z), it suffices to
find a single non-constant function in L2+ not annihilated by ♥; this we do by
choosing an appropriate test function supported high in the cusp.
Although we have appealed to the theory of Eisenstein series, this is not really
necessary: it is possible to prove that the image of ♥ is cuspidal directly from
the definition, and in fact the proof we present will be completely independent
of any knowledge of Eisenstein series.

4.4 How to interpret ♥: an explicit operator with purely

cuspidal image
Let h be the Fourier transform of g. Proceeding formally for a moment, let us
also note that if g were the sum of a δ-mass at x = log( p) and at x = − log( p),
then h(r ) = pir + p −ir , and so in this case (4.2.9) says – if we can make sense
of√it – that f √ → K ∗ f has the properties we would expect of the operator
− 14 − − 14
p +p . In this section we will mildly modify this construction
(because we prefer not to deal with the technicalities that arise by taking g to
be a distribution).
Now, let g0 be an even smooth function of compact support on R, and
for p ≥ 1 put g p (x) := g0 (x + log( p)) + g0 (x − log( p)). Starting from g p for
p ≥ 0, we define k p (u) and K p (z, w) as in Definition 4.2.6.
 4.4 How to interpret ♥ 107

Let us pause to explain the connection of this with the vague idea that
we described at the start of this section. Work formally for a moment and
suppose that g0 were the “delta-function at 0.” Then, formally speaking, the
Fourier transform h 0 is the constant function 1, whereas the Fourier trans-
form h p (r ) = pir + p −ir . Using (4.2.9) we see (formally speaking – this is not
intended to be a rigorous proof!) that the map f → K 0 ∗ f is just the identity
endomorphism
√ √and that the operator f → K p ∗ f is essentially the operator
p − 4 + p − − 4 . So f → K p ∗ f − T p (K 0 ∗ f ), formally speaking, gives
1 1

an interpretation to the operator ♥ that we discussed earlier. In practice, to avoid

certain technical complications, we just take g0 to be smooth rather than the
δ√function, and√K 0 (resp. K p ) approximates the identity endomorphism (resp.
p − 4 + p − − 4 ); in fact, we rigorously proved at the end of Section 4.2
1 1

that, as g0 varies through a δ sequence, the operator K 0 approaches the identity

endomorphism in an appropriate sense.

Lemma 4.4.1 Let f ∈ L1 S L(2, Z)\h2 . Then K p ∗ f − T p (K 0 ∗ f ) defines
a cuspidal function on S L(2, Z)\h2 .

Proof For any function F on S L(2, Z)\h2 , we define for y > 0, the constant
term (denoted FC T ):

FC T (y) = F(x + i y) d x.
R/Z

Using the explicit definition of T p given in (4.3.1), we see that

(T p F)C T (y) = p − 2 FC T ( py) + p 2 FC T ( p −1 y)

1 1
(4.4.2)

holds for any F on S L(2, Z)\h2 . To prove Lemma 4.4.1, we need to check that,
for any y > 0, (K p ∗ f )C T = (T p (K 0 ∗ f ))C T . So we just need to check that

(K p ∗ f )C T (y) = p −1/2 (K 0 ∗ f )C T ( py) + p 1/2 (K 0 ∗ f )C T ( p −1 y). (4.4.3)

Now, for any p ≥ 0, we have

 
(K p ∗ f )C T (y) = K p (x + i y, w) f (w) d ∗ w d x (4.4.4)
x∈R/Z w∈h2
 
= K p (x + i y, w) f (w) d ∗ w d x,
x∈R w∈h2
0≤(w)≤1

where we have used the fact that f (w) = f (w + 1) to unfold the integral over
x ∈ R/Z, at the cost of restricting the w-integration from h2 to h2 /{w → w + 1}.
108 Existence of Maass forms

Now, using the fact that K (gz, gw) = K (z, w) for g ∈ S L(2, R), we see that
that if w = xw + i yw , then
 
K p (x + i y, w) d x = K p (x + i y, i yw ) d x (4.4.5)
x∈R x∈R


= yw K p x + i yyw−1 , i d x
x∈R
 
= (yyw )1/2 g p log yyw−1 .

So we get
∞
dy
y0 y −1/2 g p (log(y0 y −1 )) f C T (y)
1/2
(K p ∗ f )C T (y0 ) = .
y
0

From this and the fact that g p (x) := g0 (x + log( p)) + g0 (x − log( p)), (4.4.3)
follows by a simple computation.

4.5 There exist infinitely many even cusp forms for

S L(2, Z)
Let notations be as in Section 4.4 and let ♥ be the self-adjoint endomorphism
of L2 (S L(2, Z)\h2 ) defined by f → K p ∗ f − T p (K 0 ∗ f ). It is easy to check
that ♥ preserves L2cusp,+ (in fact, all operators in sight do). To show that there
exist even cusp forms, we must show that ♥ = 0 on L2cusp,+ . The idea, in words,
is the following. Let ∞ be the stabilizer of the cusp at ∞ in S L(2, Z), that is
to say, the group generated by z → z + 1.
High in the cusp, S L(2, Z)\h2 looks like the cylinder ∞ \h2 . This cylinder
has rotational symmetry, i.e., it admits the action z → z + t of the group R/Z.
It turns out that the maps f → K p ∗ f and f → T p (K 0 ∗ f ) behave totally
differently with respect to this action; so this incompatibility forces ♥ to be
non-zero.
Let T ≥ 1 and let
 
S(T ) = T, 12 = ∞ \ z ∈ h2  (z) > T ,

be the Siegel set as in Definition 1.3.1. Then the natural projection: S(T ) to
S L(2, Z)\h2 is a homeomorphism onto an open subset. We can, therefore,
regard Cc∞ (S(T )), the space of smooth compactly supported functions on
S(T ), as a subset of Cc∞ (S L(2, Z) \h2 ); similarly L2 (S(T )) is a subset of
 4.5 There exist infinitely many even cusp forms for S L(2, Z) 109

L2 (S L(2, Z)\h2 ). We will make these identifications throughout the rest of this
argument.
If f ∈ C ∞ (S(T )), we define for n ∈ Z, the nth Fourier coefficient an, f (y) to
*1
be the function on (T, ∞) defined by the rule an, f (y) = f (x + i y)e−2πinx d x.
0
Now, let R be so large that k0 (z, w) and k p (z, w) are supported in d(z, w) ≤
R, and let Y ≥ pe R . Then, one sees from (4.3.2) and Definition 4.2.7 that ♥
maps C ∞ (S(Y )) into C ∞ (S(1)). Indeed, it is enough to check that this is true
for f → K p ∗ f and f → T p (K 0 ∗ f ); we deal with the first and leave the
second to the reader. It is clear from Definition 4.2.7 that K p ∗ f is supported
in an R-neighborhood of the support of f . But an R-neighborhood of S(Y ) is
contained in S(Y/R), thus the claim.
Moreover, if the nth Fourier coefficient an, f (y) vanishes identically, then so
does an,K p ∗ f (y). This follows from (4.4.4):
 
−2πinx
an,K p ∗ f (y) := e K p (x + i y, w) f (w) d ∗ w d x
x∈Z\R w∈h2
 
= e−2πinx K p (i y, w − x) f (w) d ∗ w d x
x∈Z\R w∈h2
 
= K p (i y, w) f (w + x)e−2πinx d x d ∗ w,
w∈h2 x∈Z\R

and the final integral clearly vanishes if an, f vanishes identically.

Fix an arbitrary integer N ≡ 0 (mod p). Let

f ∈ C+∞ (S(Y )) := Cc∞ (S(Y )) ∩ L2cusp,+

be a non-zero even function so that an, f vanishes identically for all n = ±N .

Then an,K p ∗ f vanishes identically for n = ±N . On the other hand, we see from
Lemma 4.4.1 that, for z ∈ h2 , we have

T p (K 0 ∗ f )(z) = K 0 ∗ f ( pz),

so (by the same argument as before), an,T p (K 0 ∗ f ) vanishes identically for

n = ± pN .
It follows that ♥ f is a non-zero even cuspidal function, as long as
K 0 ∗ f = 0. But we are still free to choose the function g0 that entered in
the definition of K 0 , and it is clear from the discussion of Section 4.2 that, as
we let g0 approximate the δ function, K 0 ∗ f will approach f pointwise; in
particular, it will be non-zero.
110 Existence of Maass forms

We have therefore shown that – for appropriate choice of g0 – the function

F = ♥ f is non-zero, cuspidal, belongs to C+∞ (S(1)), and, moreover, has the
property (as is clear from our discussion above) that an,F (y) vanishes unless
n ∈ {N , −N , pN , − pN }. Since N could be any integer not divisible by p, we
conclude from this that L2cusp,+ is infinite dimensional, so there are infinitely
many even cusp forms.

4.6 A weak Weyl law

The proof given above shows there exist infinitely many cusp forms. It is easy
to make this quantitative. Here we will just explain how to prove a weak
version of the Weyl law and we will say a few words about how to prove
the full Weyl law in Section 4.8. We first recall the following “variational
principle”:

Lemma 4.6.1 Suppose H is a Hilbert space and A a non-negative self-adjoint

(possibly unbounded) operator on H with discrete spectrum λ1 ≤ λ2 ≤ . . .
Suppose V ⊂ H is a finite-dimensional subspace and  is such that
Av ≤  v whenever v ∈ V . Then #{λi ≤ } ≥ dim(V ).

Proof Let vi be the eigenvector corresponding to the eigenvalue λi . Let W

be the space spanned by all the vi ’s with λi ≤ . If the claim is false, then
dim(W ) < dim(V ), so there is a vector in V perpendicular to W . Such a vector
+
must have the form v = j c j v j , where the sum is taken only over eigenvec-
tors v j with eigenvalue > . But it is clear that such a vector cannot satisfy
Av ≤  v , contradiction.

Proposition 4.6.2 Let N () be the number of eigenfunctions of the Laplacian

in L2cusp,+ with eigenvalue ≤ . Then there exists c > 0 such that N () ≥ c
for all  ≥ 1.

Proof (Sketch only) We follow the notations of the previous section. Fix a
non-zero smooth function h on the real line, supported in (0, 1). Fix an integer
N ≥ 1. For each pair of positive integers j, k satisfying 1 ≤ j, k ≤ N and so that
p does not divide j, we put f jk (x + i y) = h ((N (y − Y )/Y ) − k) cos(2π j x),
regarded as an element of C ∞ (S(Y ))+ ⊂ L2cusp,+ . Let W be the span of f jk , so
an N 2 − N [N / p] dimensional subspace of L2cusp,+ . Also, let V = ♥(W ), where
we take the function g0 entering in the definition of ♥ to be an approximation
to a δ function. Now apply the previous lemma to V .

 4.8 Interpretation via wave equation: higher rank case 111

4.7 Interpretation via wave equation and the role of finite

propagation speed
We now comment how ♥ may be understood in terms of the wave equation.
Equations (4.3.2) and (4.3.3) admit a nice interpretation in terms of the auto-
morphic wave equation
u
u tt = −u + . (4.7.1)
4
A solution u = u(x + i y, t) to (4.7.1) may be regarded as describing the ampli-
tude of a wave propagating in the hyperbolic plane. The low order term of u/4
is natural for the hyperbolic Laplacian (see (Lax and Phillips, 1976)).
For every t ∈ R we can define a linear endomorphism Ut of
 
L2 S L(2, Z)\h2 ∩ C ∞ S L(2, Z)\h2

to itself, taking a function f (x + i y) to 2u(x + i y, t), where u is the solu-

tion to (4.7.1) with u|t=0 = f , u t |t=0 = 0. One may show that this operator
is well defined in a standard way; moreover, it is self-adjoint
√ 1 (“time reversal
√1
symmetry”). Formally speaking, one may write Ut = p t 4 − + p −t 4 − ;
in fact, Ut gives a rigorous meaning to the right-hand side.

4.8 Interpretation via wave equation: higher rank case

√ 1 √ 1
In this section, we briefly detail how the operator p − 4 + p − − 4 may be
viewed in terms of the wave equation. We then conclude by discussing how the
considerations of this section generalize. For further details, we refer the reader
to (Lindenstrauss and Venkatesh, to appear).
The automorphic wave equation
u
u tt = −u + (4.8.1)
4
describes the propagation of waves on the hyperbolic plane. A solution

u = u(x + i y, t)

to (4.8.1) may be regarded as describing the amplitude (at time t and position
x + i y) of a wave propagating in the hyperbolic plane. The low order term of
u/4 is natural for the hyperbolic Laplacian (see (Lax and Phillips, 1976)).
For every t ∈ R we can define a linear endomorphism Ut of
 
L2 S L(2, Z)\h2 ∩ C∞ S L(2, Z)\h2
112 Existence of Maass forms

to itself, taking a function f (x + i y) to 2u(x + i y, t), where u is the solu-

tion to (4.8.1) with u|t=0 = f , u t |t=0 = 0. One may show that this operator
is well defined in a standard way; moreover, it is self-adjoint
√ (“time√reversal
t − 14
+ e−t − 4 ; in
1
symmetry”). Formally speaking, one may write Ut = e
fact, Ut gives a rigorous meaning to the right-hand side.
Moreover, for any function f ∈ C∞ (h2 ), the value of Ut f at z ∈ h2 depends
only on the values of f at points w with d(z, w) ≤ t; this fact expresses the
finite propagation speed of waves in the hyperbolic plane, and corresponds to
the fact that the point pair invariants K we used earlier were supported within
d(z, w) ≤ R for some R. √ 1 √ 1
In this fashion the operator p − 4 + p − − 4 can be regarded as the oper-
ator on functions that corresponds to propagating a wave for a time log p. One
can thereby rephrase our previous arguments using the wave equation.
Finally, we note that the methods of this section can be extended to the
higher-rank case, e.g. existence of cusp forms on S L(n, Z)\S L(n, R); and
moreover, a more careful analysis gives not merely the existence of cusp forms
but the full Weyl law, that is to say, the correct asymptotic for the number
of cusp forms of eigenvalue ≤ . The idea is that, again, one may construct
operators like “♥” by combining Hecke operators and integral convolution
operators.
To write such an operator down explicitly for S L(3, Z)\S L(3, R) would
be rather a painful process! However, it is not too difficult to convince your-
self that they do exist: in the S L(2, Z) case, the crucial point was that both
the Hecke eigenvalue pit + p −it and the Laplacian eigenvalue 1/4 + t 2 of the
Eisenstein series E(z, 12 + it) were controlled by just one parameter t ∈ R, and
so it is not too surprising that we can concoct a combination of these param-
eters that always vanishes. A similar phenomenon occurs for higher-rank: the
Eisenstein series is controlled by too few parameters for the archimedean and
Hecke eigenvalues to be completely independent. This is perhaps a bit sur-
prising since, on S L(3, Z)\S L(3, R), there exist Eisenstein series indexed not
merely by a complex parameter t but also by Maass forms (see Section 10.5)
on S L(2, Z)\S L(2, R)!
In the higher-rank case one uses a slightly different approach to see that
the equivalent of “♥” is non-zero. Let us describe this approach in the
S L(2, Z)\h2 case; the higher-rank case proceeds analogously but using higher-
rank Whittaker functions. The idea is again to explicitly write down a function
which ♥ does not annihilate; but we will use instead a somewhat more com-
plicated function than before. The payoff will be that ♥ will act on it in a very
simple way.
 4.8 Interpretation via wave equation: higher rank case 113

Choose θ, r ∈ R and define for k ≥ 0 the complex numbers

ak := e−ikθ + e−i(k−2)θ + · · · + eikθ .
+ √
Put f (z) = ak y K ir ( p k y) cos(2π p k x). Recall that for T ≥ 1 we defined
k≥0
 
S(T ) = T, 12 = ∞ \ z ∈ h2  (z) > T
to be the Siegel set as in Definition 1.3.1. If we restrict f to S(T ),
we may, thereby, regard it as belonging to L2 (S L(2, Z)\h2 ). One veri-
fies that, if T   T is sufficiently large, then for any w ∈ S(T  ), we have
♥ f (w) = h 0 (ir )( pir + p −ir − eiθ − e−iθ ) f (w); here h 0 is the Fourier trans-
form of the basic function g0 that was chosen at the start of Section 4.4. In other
words, for this particular function f , “high in the cusp,” ♥ actually acts on f by
a scalar, namely h 0 (ir )( pir + p −ir − eiθ − e−iθ ). In particular, we can choose
r and θ so that ♥ f = 0.
It is possible to reproduce this behavior in any rank using higher-rank
Whittaker functions, and this allows one to show that the relevant convolu-
tion operators are non-zero.
 5
Maass forms and Whittaker functions for
S L(n, Z)

5.1 Maass forms

Maass forms for S L(2, Z) were introduced in Section 3.3. We want to generalize
the theory to S L(n, Z) with n > 2. Accordingly, we will define, for n ≥ 2, a
Maass form as a smooth complex valued cuspidal function on

hn = G L(n, R)/(O(n, R) · R× )

which is invariant under the discrete group S L(n, Z) and which is also an eigen-
function of every invariant differential operator in Dn , the center of the universal
enveloping algebra as defined in Section 2.3. A cuspidal function (or cuspform)
on h2 was defined by the condition that the constant term in its Fourier expansion
vanishes which in turn is equivalent to the condition that φ(z) has exponential
decay as y → ∞. These notions are generalized in the formal Definition 5.1.3.
Harish-Chandra was the first to systematically study spaces of automor-
phic forms in a much more general situation than G L(n). He proved (Harish-
Chandra, 1959, 1966, 1968) that the space of automorphic functions of a cer-
tain type (characterized by a cuspidality condition, eigenfunction condition,
and good growth) is finite dimensional. Godement (1966) explains why Maass
forms on G L(n) are rapidly decreasing. It was not at all clear at that time if
a theory of L-functions, analogous to the G L(2) theory could be developed
for G L(n) with n > 2. The first important breakthrough came in (Piatetski-
Shapiro, 1975), and independently in (Shalika, 1973, 1974), where the Fourier
expansion of a Maass form for S L(n, Z) was obtained for the first time. The
Fourier expansion involved Whittaker functions. In his thesis, Jacquet intro-
duced and obtained the meromorphic continuation and functional equations
of Whittaker functions on an arbitrary Chevalley group (see (Jacquet, 1967)).
These papers provided the cornerstone for an arithmetic theory of L-functions
in the higher-rank situation.

114
 5.1 Maass forms 115

Recall that for n ≥ 2, an element z ∈ hn takes the form z = x · y where

⎛ ⎞ ⎛ ⎞
1 x1,2 x1,3 · · · x1,n y1 y2 · · · yn−1
⎜ 1 x2,3 · · · x2,n ⎟ ⎜ y1 y2 · · · yn−2 ⎟
⎜ ⎟ ⎜ ⎟
⎜ . . ⎟ ⎜ . ⎟
x =⎜ .. .. ⎟, y = ⎜ .. ⎟,
⎜ ⎟ ⎜ ⎟
⎝ 1 xn−1,n ⎠ ⎝ y1 ⎠
1 1
with xi, j ∈ R for 1 ≤ i < j ≤ n and yi > 0 for 1 ≤ i ≤ n − 1.
Let ν = (ν1 , ν2 , . . . νn−1 ) ∈ Cn−1 . We have shown in Section 2.4 that the
function
 n−1
n−1  b νj
Iν (z) = yi i, j (5.1.1)
i=1 j=1

with

ij if i + j ≤ n,
bi, j =
(n − i)(n − j) if i + j ≥ n,
is an eigenfunction of every D ∈ Dn . Let us write
D Iν (z) = λ D · Iν (z) for everyD ∈ Dn . (5.1.2)
The function λ D (viewed as a function of D) is a character of Dn because it
satisfies
λ D1 ·D2 = λ D1 · λ D2
for all D1 , D2 ∈ Dn . It is sometimes called the Harish–Chandra character.
Definition 5.1.3 Let n ≥ 2, and let ν = (ν1 , ν2 , . . . νn−1 ) ∈ Cn−1 . A Maass
form for S L(n, Z) of type ν is a smooth function f ∈ L2 (S L(n, Z)\hn ) which
satisfies
(1) f (γ z) = f (z), for all γ ∈ S L(n, Z), z ∈ hn ,
 = λ D f (z), for all D ∈ D , with λ D given by (5.1.2),
n
(2) D f (z)
(3) f (uz) du = 0,
(S L(n,Z)∩U )\U

for all upper triangular groups U of the form

⎧⎛ ⎞⎫
⎪
⎪ Ir1 ⎪
⎪
⎪⎜
⎨ ⎟⎬⎪
⎜ Ir2 ∗ ⎟
U= ⎜ . ⎟ ,
⎪
⎪ ⎝ .. ⎠⎪⎪
⎪
⎩ ⎪
⎭
Irb
116 Maass forms and Whittaker functions for S L(n, Z)

with r1 + r2 + · · · + rb = n. Here Ir denotes the r × r identity matrix, and ∗

denotes arbitrary real entries.

5.2 Whittaker functions associated to Maass forms

For n ≥ 2, let Un (R) denote the group of upper triangular matrices with 1s on
the diagonal and real entries above the diagonal. Then every u ∈ Un (R) is of
the form
⎛ ⎞
1 u 1,2 u 1,3 · · · u 1,n
⎜ 1 u 2,3 · · · u 2,n ⎟
⎜ ⎟
⎜ .. .. ⎟ ,
u=⎜ . . ⎟ (5.2.1)
⎜ ⎟
⎝ 1 u n−1,n ⎠
1
with u i, j ∈ R for 1 ≤ i < j ≤ n. Similarly, we define Un (Z) with entries
u i, j ∈ Z for 1 ≤ i < j ≤ n.
If m = (m 1 , m 2 , . . . , m n−1 ) ∈ Zn−1 , then the function ψm :Un (R) → C×
defined by
ψm (u) = e2πi (m 1 u 1,2 +m 2 u 2,3 +···+m n−1 u n−1,n ) , (with u ∈ Un (R))
is a character of Un (R). This means that
ψm (u · v) = ψm (u)ψm (v) (5.2.2)
for all u, v ∈ Un (R). This can be quickly verified in the case n = 3 because
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
1 u 1,2 u 1,3 1 v1,2 v1,3 1 u 1,2 + v1,2 ∗
⎝0 1 u 2,3 ⎠ · ⎝ 0 1 v2,3 ⎠ = ⎝ 0 1 u 2,3 + v2,3 ⎠ ,
0 0 1 0 0 1 0 0 1
and it is easy to see that (5.2.2) holds in general.
For n ≥ 2, let φ be a Maass form for S L(n, Z) of type
ν = (ν1 , . . . , νn−1 ) ∈ Cn−1 . By analogy with the Fourier expansion tech-
niques introduced in Section 3.5, it is natural to introduce the function φ̃ m (z)
defined as follows:
 1  1 
φ̃ m (z) := ··· φ(u · z) ψm (u) du i, j . (5.2.3)
0 0 1≤i< j≤n

One might reasonably expect that φ̃ m (z) is a Fourier coefficient of φ and that
φ might be recoverable as a sum of such Fourier coefficients. Unfortunately,
the fact that Un (R) is a non–abelian group (for n > 2) complicates the issue
enormously, and it is necessary to go through various contortions in order to
 5.2 Whittaker functions associated to Maass forms 117

obtain a useful Fourier theory. We shall study this issue carefully in the next
section. For the moment, we focus on the integral (5.2.3).
Proposition 5.2.4 For n ≥ 2 and ν = (ν1 , ν2 , . . . , νn−1 ) ∈ Cn−1 , let φ be a
Maass form of type ν for S L(n, Z). Let m = (m 1 , . . . , m n−1 ) ∈ Zn−1 , and let
ψm be an additive character as in (5.2.2). Then the function φ̃ m (z) defined in
(5.2.3) satisfies the following conditions:
(1) φ̃ m (u · z) = ψm (u) · φ̃ m (z) (for all u ∈ Un (R)),
(2) Dφ̃ m = λ D φ̃ m , (for all D ∈ Dn ),
(3) |φ̃ m (z)|2 d ∗ z < ∞,
 √3 1
2 ,2

where  √3 , 1 denotes the Siegel set as in Definition 1.3.1, and d ∗ z is the left
2 2
invariant measure given in Proposition 1.5.3.
Remark Any smooth function: hn → C which satisfies conditions (1), (2),
(3) of Proposition 5.2.4 will be called a Whittaker function. A more formal
definition will be given in Section 5.4.
Proof First of all, the integral on the right-hand side of (5.2.3) is an integral
over Un (Z)\Un (R). Since both φ and ψm are invariant under Un (Z), the integral
is independent of the choice of fundamental domain for Un (Z)\Un (R).
Every z ∈ hn can be written in the form z = x · y, as in the beginning
of Section 5.1. In the integral (5.2.3), we make the change of variables
u → u · x −1 . It follows that
 1  1 
φ̃ m (z) = ··· φ(u · y) ψm (u · x −1 ) du i, j .
0 0 1≤i< j≤n

But (5.2.2) implies that ψm (u · x −1 ) = ψm (u) · ψm (x −1 ) = ψm (u) · ψm (x).

This proves that for u ∈ Un (R), φ̃(u · z) = ψm (u · x)φ̃(y) = ψm (u)φ̃(z). The
second part is an immediate consequence of the fact that φ is an eigenfunction
of every D ∈ Dn with eigenvalue λ D .
To prove (3), we use the Cauchy–Schwartz inequality and the fact that φ is
in L2 to deduce that
  1  1  
2 ∗
|φ̃ m (z)| d z < ··· |φ(u · z)|2 d ∗ z du i, j
0 0 1≤i< j≤n
 √3 1  √3 1
2 ,2 2 ,2

< |φ(z)|2 d ∗ z < ∞.
 √3 1
2 ,2

118 Maass forms and Whittaker functions for S L(n, Z)

5.3 Fourier expansions on SL(n, Z)\hn

The classical Fourier expansion theorem states that every smooth periodic func-
tion φ on Z\R has a Fourier expansion


φ(x) = φ̃ m (x), (5.3.1)
m∈Z

where
 1
φ̃ m (x) = φ(u + x)e−2πimu du.
0

We seek to generalize (5.3.1) to smooth automorphic functions on

S L(n, Z)\hn .

Theorem 5.3.2 For n ≥ 2, let Un denote the group of n × n upper triangular

matrices with 1s on the diagonal as in Section 5.2. Let φ be a Maass form for
S L(n, Z). Then for all z ∈ S L(n, Z)\hn




∞

∞   
γ
φ(z) = ··· φ̃ (m 1 ,...,m n−1 ) z ,
γ ∈ Un−1 (Z)\S L(n−1,Z) m 1 =0 m 2 =1 m n−1 =1
1

where the sum is independent of the choice of coset representatives γ and

 1  1
φ̃ (m 1 ,...,m n−1 ) (z) := ··· φ(u · z) e−2πi (m 1 u 1,2 +m 2 u 2,3 +···+m n−1 u n−1,n ) d ∗ u,
0 0
)
with u ∈ Un (R) given by (5.2.1) and d ∗ u = 1≤i< j≤n du i, j .

If φ satisfies conditions (1), (2), but does not satisfy condition (3) of
Definition 5.1.3, then the Fourier expansion takes the form



∞

∞

∞   
γ
φ(z) = ··· φ̃ (m 1 ,...,m n−1 ) z .
γ ∈Un−1 (Z)\S L(n−1,Z) m 1 =−∞ m 2 =0 m n−1 =0
1

The proof of Theorem 5.3.2 makes use of an elementary lemma in group

theory which we shall straightaway state and prove.

Lemma 5.3.3 Let C ⊆ B ⊆ A be groups. Let f : C\A → C be any function

+
such that the sum, f (γ ), converges absolutely. Then
γ ∈C\A




f (γ ) = f (δ  δ).
γ ∈C\A δ  ∈C\B δ∈B\A
 5.3 Fourier expansions on SL(n, Z)\hn 119

Proof It is clear that if δ ∈ B\A and δ  ∈ C\B then δ  δ ∈ C\A. On the other
hand, every γ ∈ C\A can be written in the form γ = ca for c ∈ C, a ∈ A. If
we now set δ  = c · 1 ∈ C\B and δ = 1 · a ∈ B\A, then we have expressed
γ = δ  δ.

Proof of Theorem 5.3.2 Since φ is automorphic for S L(n, Z), we have

⎛⎛ ⎞ ⎞
1 m1
⎜⎜ 1 ⎟ ⎟
m2
⎜⎜ ⎟ ⎟
⎜⎜ .. ⎟ ⎟
..
φ ⎜⎜ . ⎟ · z ⎟ = φ(z)
.
⎜⎜ ⎟ ⎟
⎝⎝ 1 m n−1 ⎠ ⎠
1

for all m 1 , m 2 , . . . , m n−1 ∈ Z. It then follows from classical one–dimensional

Fourier theory that
⎛⎛ ⎞ ⎞
1 v1
  ⎜⎜ 1 v2 ⎟ ⎟

1 1 ⎜⎜ ⎟ ⎟
⎜⎜ .. .. ⎟ ⎟
φ(z) = ··· φ ⎜⎜ . . ⎟ · z⎟
⎜⎜ ⎟ ⎟
m 1 ,...,m n−1 ∈ Z 0 0
⎝⎝ 1 vn−1 ⎠ ⎠
1
× e−2πi (m 1 v1 +···+m n−1 vn−1 ) dv1 · · · dvn−1 . (5.3.4)

Here, we are simply using the fact that the matrices

⎛ ⎞ ⎛ ⎞ ⎛ ⎞
1 1 1 0 1 0
⎜ 1 0⎟ ⎜ 1 1⎟ ⎜ 1 0⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜ .. .. ⎟ , ⎜ .. .. ⎟ , ⎜ .. .. ⎟
⎜ . .⎟ ⎜ . .⎟ ... ,⎜ . .⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎝ 1 0⎠ ⎝ 1 0⎠ ⎝ 1 1⎠
1 1 1

commute with each other and generate the abelian group of all matrices of the
form
⎧⎛ ⎞ ⎫
⎪ 1 m1  ⎪
⎪
⎪  ⎪
⎪
⎪
⎪ ⎜ 1 m ⎟  ⎪
⎪
⎨⎜ 2 ⎟  ⎬
⎜ . . ⎟ 
⎜ . . .
. ⎟  m 1 , . . . , m n−1 ∈ Z .
⎪
⎪ ⎜ ⎟  ⎪
⎪
⎪
⎪ ⎝ 1 m n−1 ⎠  ⎪
⎪
⎪
⎩ ⎪
⎭
1

120 Maass forms and Whittaker functions for S L(n, Z)

We now rewrite (5.3.4) with a more compact notation. For

⎛ ⎞
1 v1
⎜ 1 v2 ⎟
⎜ ⎟
.. ⎟, d ∗ v =  dv ,
n−1
⎜ ..
m = (m 1 , . . . , m n−1 ) ∈ Zn−1 , v = ⎜ . . ⎟ i
⎜ ⎟
⎝ 1 vn−1 ⎠ i=1

1
define
 1  1
φ̂ m (z) := ··· φ(vz)e−2πiv,m d ∗ v,
0 0
+n−1
where v, m = i=1 vi m i . Note the difference between φ̂ and φ̃, for
example, φ̃ involves integration with repect to all the variables u i, j with
1 ≤ i < j ≤ n − 1. With this notation, (5.3.4) becomes


φ(z) = φ̂ m (z). (5.3.5)
m∈Zn−1

The Fourier expansion (5.3.5) does not make use of the fact that φ is auto-
morphic for all S L(n, Z). To proceed further, we need the following lemma.

Lemma 5.3.6 Let n > 2. Fix an integer M = 0, and let γ ∈ S L(n − 1, Z).
Then
  
γ 0
φ̂ Men−1 γ (z) = φ̂ (0,...,0,M) ·z ,
0 1

where en−1 = (0, . . . , 0, 1) lies in Zn−1 .

Proof Let
⎛ ⎞
a1,1 a1,2 ··· a1,n−1
⎜ .. .. .. ⎟
⎜ ⎟
γ =⎜ . . . ⎟,
⎝ an−2,1 an−2,2 · · · an−2,n−1 ⎠
γ1 γ2 ··· γn−1
and
⎛ ⎞ ⎛ ⎞
1 v1 1 v1
⎜ 1 v2⎟ ⎜ 1 v2 ⎟
⎜ ⎟ ⎜ ⎟
⎜ .. ⎟
.. ⎜ .. .. ⎟
v=⎜ . ⎟,
. v = ⎜ . . ⎟.
⎜ ⎟ ⎜ ⎟
⎝ 1 vn−1 ⎠ ⎝ 1 
vn−1 ⎠
1 1
 5.3 Fourier expansions on SL(n, Z)\hn 121

Then we have the identity

   
γ 0  γ 0
·v =v · (5.3.7)
0 1 0 1
where

v1 = a1,1 v1 + · · · + a1,n−1 vn−1 ,

v2 = a2,1 v1 + · · · + a2,n−1 vn−1 ,
..
.

vn−2 = an−2,1 v1 + · · · + an−2,n−1 vn−1

vn−1 = γ1 v1 + · · · + γn−1 vn−1 .
  
γ 0
Now, since γ ∈ S L(n − 1, Z), we have φ(vz) = φ vz . It then fol-
0 1
lows from (5.3.7) and a simple change of variables that

φ̂ (Mγ1 ,...,Mγn−1 ) (z) = φ̂ Men−1 γ (z)

 1  1   
γ 0
= ··· φ vz e−2πi M (γ1 v1 +···+γn−1 vn−1 ) dv1 · · · dvn−1
0 0 0 1
 1  1    
γ 0
= ··· φ v z e−2πi M (γ1 v1 +···+γn−1 vn−1 ) dv1 · · · dvn−1
0 0 0 1
 1  1    
 γ 0 
= ··· φ v z e−2πi M·vn−1 dv1 · · · dvn−1 
0 0 0 1
  
γ 0
= φ̂ (0,...,0,M) ·z .
0 1

For n > 2, the group S L(n − 1, Z) acts on Zn−1 with two orbits:

{0}, Zn−1 − {0} = Z+ · en−1 · S L(n − 1, Z),

where Z+ denotes the positive integers. The second orbit above is a consequence
of the fact that every non-zero m ∈ Zn−1 can be expressed in the form

m = (m 1 , . . . , m n−1 ) = M · (γ1 , . . . , γn−1 )

 
∗
= M · en−1 · ,
γ1 γ 2 · · · γn−1
where M = gcd(m 1 , . . . , m n−1 ), M > 0, and m j = Mγ j (for j =
1, 2, . . . , n − 1). The stabilizer of en−1 in S L(n − 1, Z) (under right
122 Maass forms and Whittaker functions for S L(n, Z)

multiplication) is Pn−1 (Z) where

 
∗
Pn−1 (Z) = ,
0 0 ··· 0 1

and ∗ denotes arbitrary integer entries.

Remark 5.3.8 The above argument breaks down when n = 2. In this case,
the orbit consists of all integers M = 0.

It now follows from this discussion and Lemma 5.3.6 that we may rewrite
(5.3.5) as follows:

∞

φ(z) = φ̂ (0,...,0) (z) + φ̂ M·en−1 ·γ (z)
M=1 γ ∈Pn−1 (Z)\S L(n−1,Z)


∞
  
γ 0
= φ̂ (0,...,0) (z) + φ̂ (0,...,0,M) ·z .
M=1 γ ∈Pn−1 (Z)\S L(n−1,Z)
0 1
(5.3.9)

The fact that φ is a Maass form implies that φ̂ (0,...,0) (z) = 0. Replacing M by
m n−1 , and setting Pn−1 = Pn−1 (Z), S L n−1 = S L(n − 1, Z), we may, therefore,
rewrite (5.3.9) in the form

∞
     
1 1
γ
φ(z) = ··· φ u· ·z
m n−1 =1 γ ∈Pn−1 \S L n−1 0 0 1

× e−2πim n−1 ·u n−1,n d ∗ u, (5.3.10)

where
⎛ ⎞
1 u 1,n
⎜ 1 ⎟ u 2,n
⎜ ⎟ 
n−1
⎜ .. ⎟ ..
u=⎜ . ⎟, . d ∗u = du j,n .
⎜ ⎟
⎝ 1 u n−1,n ⎠
j=1

Lemma 5.3.11 The function

 1  1
φ̂ (0,...,0,M) (z) = ··· φ(v · z)e−2πi Mvn−1 d ∗ v
0 0
 5.3 Fourier expansions on SL(n, Z)\hn 123

where
⎛ ⎞
1 v1
⎜ 1 ⎟ v2
⎜ ⎟ 
n−1
⎜ .. ⎟ ..
v=⎜ . ⎟, . d ∗v = dvi ,
⎜ ⎟
⎝ 1 vn−1 ⎠
i=1

1
is invariant under left multiplication by matrices of the form
⎛ ⎞
m 1,1 m 1,2 · · · m 1,n−1 0
⎜ m · · · m 2,n−1 0⎟
⎜ 2,1 m 2,2 ⎟
⎜ . .. .. .. ⎟
⎜ . ⎟
m=⎜ ⎜
. . . . ⎟ ∈ S L(n, Z).
⎟
⎜ m n−2,1 m n−2,2 · · · m n−2,n−1 0 ⎟
⎜ ⎟
⎝ 0 0 ··· 1 0⎠
0 0 ··· 0 1

Proof We have
 1  1
φ̂ (0,...,0,M) (m · z) = ··· φ(v · m · z)e−2πi Mvn−1 d ∗ v
0 0
 1  1
= ··· φ(m · v  · z)e−2πi Mvn−1 d ∗ v
0 0
 1  1
= ··· φ(v  · z)e−2πi Mvn−1 d ∗ v,
0 0

where
⎛ ⎞
1 v1
⎜ 1 v2 ⎟
⎜ ⎟
 ⎜ .. .. ⎟
v =⎜ . . ⎟
⎜ ⎟
⎝ 
1 vn−1 ⎠
1
is chosen so that v · m = m · v  . A simple computation shows that we may take
v1 = m 1,1 v1 + m 1,2 v2 + · · · + m 1,n−1 vn−1
v2 = m 2,1 v1 + m 2,2 v2 + · · · + m 2,n−1 vn−1
..
.

vn−2 = m n−2,1 v1 + m n−2,2 v2 + · · · + m n−2,n−1 vn−1

vn−1 = vn−1 .
124 Maass forms and Whittaker functions for S L(n, Z)

Finally, if we make the change of variables v  → v, it follows from the above

discussion that
 1  1

φ̂ (0,...,0,M) (m · z) = ··· φ(v  · z)e−2πi Mvn−1 d ∗ v
0 1 0 1

= ··· φ(v  · z)e−2πi Mvn−1 d ∗ v 
0 0
= φ̂ (0,...,0,M) (z),

because the Jacobian of the transformation is 1.

We remark here that in the derivation of (5.3.10), we only used the left
invariance of φ(z) with respect to Pn−1 (Z). In view of Lemma 5.3.11, we may
then reiterate all previous arguments and obtain, instead of (5.3.10), the more
general form

∞
∞

 1  1
φ(z) = ···
m n−2 =1 m n−1 =1 γn−2 ∈Pn−2 \S L n−2 γn−1 ∈Pn−1 \S L n−1 0 0
⎛ ⎛ ⎞ ⎞
γn−2  
γn−1
× φ ⎝u · ⎝ 1 ⎠· · z⎠
1
1
× e−2πi [m n−2 u n−2,n−1 +m n−1 u n−1,n ] d ∗ u, (5.3.12)

where
⎛ ⎞ ⎛ ⎞
1 u 1,n 1 u 1,n−1 0
⎜ ⎟ ⎜ 0⎟
⎜ 1 ⎟ ⎜
u 2,n 1 u 2,n−1 ⎟
⎜ ⎟ ⎜
.. .. .. ⎟
⎜ .. ⎟ ⎜ .. ⎟
u=⎜
⎜
. ⎟·⎜
.
⎟ ⎜
. . .⎟
⎟
⎜ ⎟ ⎜ u n−2,n−1 0⎟
⎜ ⎟ ⎜ ⎟
⎝ 1 u n−1,n ⎠ ⎝ 1 0⎠
1 1

⎛ ⎞
1 u 1,n−1 u 1,n
⎜ ⎟
⎜ 1 u 2,n−1 u 2,n ⎟
⎜ .. .. ⎟
⎜ .. ⎟
=⎜
⎜
. . . ⎟,
⎟
⎜ u n−2,n−1 u n−2,n ⎟
⎜ ⎟
⎝ 1 u n−1,n ⎠
1

and
 
d ∗u = du i,n−1 · du j,n .
1≤i≤n−2 1≤ j≤n−1
 5.3 Fourier expansions on SL(n, Z)\hn 125

Note that by Remark 5.3.8, the sum over m n−2 in formula (5.3.12) will range
over all m n−2 = 0 when n = 3.

Lemma 5.3.13 For n ≥ 2, define the following subgroups of

S L n = S L(n, Z):
⎧⎛ ⎞⎫
⎧⎛ ⎞⎫ ⎪ ∗ ··· ∗ ∗ ∗ ⎪
⎪ ∗ ··· ∗ ∗ ⎪ ⎪
⎪ ⎪
⎪
⎪ ⎪
⎪ ⎪
⎪⎜ .. .. .. .. ⎟⎪
⎪
⎨⎜ ..
⎜
.. .. ⎟⎬ ⎨⎜ . · · · . . . ⎟⎪⎟⎬
P̃ n,1 = ⎜ . · · · . .⎟ ⎜
⎟ = Pn (Z), P̃ n,2 = ⎜ ∗ · · · ⎟
⎪ ∗ ∗ ∗ ⎟⎪,
⎪
⎪
⎝∗ ··· ∗ ∗ ⎠⎪⎪
⎪
⎪
⎪
⎪
⎜ ⎪
⎩ ⎭ ⎪
⎪
⎝0 ··· 0 1 ∗ ⎠⎪ ⎪
⎪
0 ··· 0 1 ⎩ ⎭
⎧⎛ 0 ··· 0 0 1
⎞⎫
⎪
⎪ ∗ ··· ∗ ∗ ∗ ⎪ ⎪
⎪
⎪ ⎜. ⎪ ⎧⎛ ⎞⎫
⎪
⎪ ⎜ .. · · · .. .. .. ⎟
⎟
⎪
⎪
⎪ ⎪ 1 ⎪
⎪
⎪ ⎪ ⎪ ⎪
⎨⎜⎜
. . . ⎟⎪
⎟ ⎬ ⎪
⎨ ⎜ 1 ∗⎟
⎪
⎬
P̃ n,3 = ⎜ ∗ ··· ∗ ∗ ∗ ⎟ , . . . , P̃ n,n = ⎜ ⎜
⎟
⎟ .
⎪ ⎜ ⎟ ..
⎪
⎪ ⎜ 0 ··· 1 ∗ ∗ ⎟⎪ ⎪
⎪
⎪
⎪
⎪
⎝ . ⎠⎪⎪
⎪
⎪
⎪ ⎜ ⎟⎪⎪ ⎩ ⎭
⎪
⎪ ⎝ 0 ··· 0 1 ∗ ⎠⎪ ⎪ 1
⎪
⎩ ⎪
⎭
0 ··· 0 0 1
Then for r = 1, 2, . . . , n − 1, we have
Pn−r \S L n−r ∼ \ P̃ n,r , 
= P̃ n,r +1
γ
Pn−r · γ → P̃ n,r +1 · (γ ∈ S L n−r ),
Ir
where Ir denotes the r × r identity matrix.

Proof We may write

⎧⎛ ⎞⎫
⎪
⎪ ∗ ∗ ··· ∗ ⎪ ⎪
⎪
⎪ ⎜ .. ⎟ ⎪
⎪
⎪
⎪ ⎜ SL .. .. ⎪
⎪
⎪ ⎜ . . ··· .⎟ ⎟⎪
⎪
⎪
⎪
⎪ ⎜
n−r
⎟ ⎪
⎪
⎪
⎨⎜ ∗ ∗ ··· ∗ ⎟⎪ ⎬
⎜ ⎟
P̃ n,r = ⎜ 1 ∗ ··· ∗⎟ ,
⎪
⎪ ⎜ .. ⎟ ⎪
⎪
⎪ ⎜ .. ⎟⎪⎪
⎪
⎪ ⎜ . . ⎪ ⎟ ⎪
⎪
⎪ ⎜ ⎟⎪⎪
⎪
⎪
⎪
⎝ 1 ∗ ⎠⎪ ⎪
⎪
⎪
⎩ ⎭
1
⎧⎛ ⎞⎫
⎪
⎪ ∗ ∗ ··· ∗ ⎪ ⎪
⎪⎜
⎪ .. ⎟ ⎪
⎪
⎪
⎪ ⎜ P .. .. ⎪
⎪
⎪ ⎜ n−r . . ··· . ⎟⎪
⎟ ⎪
⎪
⎪
⎪ ⎜ ⎪
⎪
⎨⎜ ∗ ∗ ··· ∗⎟ ⎟⎬
⎪
⎪
⎜ ⎟
P̃ n,r +1 = ⎜ 1 ∗ ··· ∗⎟ .
⎪
⎪ ⎜ .. ⎟ ⎪
⎪
⎪ ⎜ .. ⎟⎪⎪
⎪
⎪ ⎜ . . ⎟⎪ ⎪
⎪
⎪ ⎜ ⎟⎪⎪
⎪
⎪
⎪
⎝ 1 ∗ ⎠⎪ ⎪
⎪
⎪
⎩ ⎭
1
126 Maass forms and Whittaker functions for S L(n, Z)

The lemma follows after one notes that a set of coset representatives for the
quotient P̃ n,r +1 \ P̃ n,r is given by

(  
γ
.
γ ∈Pn−r \S L n−r
Ir

We now apply Lemma 5.3.13 (in the form Pn−1−r \S L n−1−r with r = 1) to
the inner sum

γn−2 ∈Pn−2 \S L n−2


in (5.3.12). We can replace γn−2 ∈ Pn−2 \S L n−2 by γn−2 ∈ P̃ n−1,2 \Pn−1 , where
⎛ ⎞
γn−2  

⎝ ⎠= γn−2
1 ,
1
1

and where P̃ n−1,2 is defined as in Lemma 5.3.13. If we then apply Lemma 5.3.3
to the sums



,

γn−2 ∈ P̃ n−1,2 \Pn−1 γn−1 ∈Pn−1 \S L n−1

it follows that

∞

∞
     
1 1
γ
φ(z) = ··· φ u· ·z
m n−2 =1 m n−1 =1 γ ∈ P̃ n−1,2 \S L n−1 0 0 1

× e−2πi [m n−2 u n−2,n−1 +m n−1 u n−1,n ] d ∗ u, (5.3.14)

where
⎧⎛ ⎞⎫
⎪ ∗ ∗ ··· ∗ ∗ ⎪
⎪
⎪ ⎪
⎪⎜ .. ..
⎪ . . ⎟⎪ ⎪
⎨⎜ . . · · · .. .. ⎟⎪⎬
⎜ ⎟
P̃ n−1,2 = ⎜ ∗ ∗ · · · ∗ ∗ ⎟ ⊂ S L(n − 1, Z),
⎪⎜
⎪ ⎟⎪⎪
⎪
⎪ ⎝ 1 ∗ ⎠⎪ ⎪
⎪
⎩ ⎪
⎭
1
 5.3 Fourier expansions on SL(n, Z)\hn 127

and
⎛ ⎞
1 u 1,n−1 u 1,n
⎜ 1 u 2,n−1 u 2,n ⎟
⎜ ⎟
⎜ .. .. .. ⎟
u=⎜
⎜
. . . ⎟,
⎟
⎜ u n−2,n−1 u n−2,n ⎟
⎝ 1 u ⎠
n−1,n
1
 
∗
d u= du i,n−1 · du j,n .
1≤i≤n−2 1≤ j≤n−1

All steps previously taken can be iterated. For example, after one more
iteration equation (5.3.14) becomes

∞
∞
∞
 1  1    
γ
φ(z) = ··· φ u· ·z
m =1 m =1 m =1 0 0 1
n−3 n−2 n−1 γ ∈ P̃ n−1,3 \S L n−1
−2πi [m n−3 u n−3,n−2 +m n−2 u n−2,n−1 +m n−1 u n−1,n ]
×e d ∗ u, (5.3.15)

where
⎧⎛ ⎞⎫
⎪
⎪ ∗ ∗ ··· ∗ ∗ ⎪ ⎪
⎪
⎪⎜. ⎪
⎪
⎪⎜ .. .. .. .. ⎟ ⎪
⎪
⎪
⎨⎜
⎪ . · · · . . ⎟⎪⎟ ⎪
⎪
⎜ ⎟⎬
P̃ n−1,3 = ⎜ ∗ ∗ · · · ∗ ∗ ⎟ ⊂ S L(n − 1, Z),
⎪⎜ ⎟
⎪
⎪⎜ 1 ∗ ∗ ⎟⎪ ⎪
⎪
⎪⎜ ⎟⎪⎪
⎪
⎪
⎪
⎝ 1 ∗ ⎠⎪ ⎪
⎪
⎪
⎩ ⎭
1

and
⎛ ⎞
1 u 1,n−2 u 1,n−1 u 1,n
⎜ ⎟
⎜ 1 u 2,n−2 u 2,n−1 u 2,n⎟
⎜ .. .. .. ⎟
⎜ .. ⎟
⎜ . . . . ⎟
u=⎜
⎜ 1 u n−3,n−2 u n−3,n−1
⎟
u n−3,n ⎟ ,
⎜ ⎟
⎜ 1 u n−2,n−1 u n−2,n ⎟
⎜ ⎟
⎝ 1 u n−1,n ⎠
1
  
d ∗u = du i,n−2 · du j,n−1 · du k,n .
1≤i≤n−3 1≤ j≤n−2 1≤k≤n−1

Theorem 5.3.2 follows from (5.3.15) after continuing this process inductively
for n − 2 steps, and taking into account Remark 5.3.8.

128 Maass forms and Whittaker functions for S L(n, Z)

5.4 Whittaker functions for SL(n, R)

For n ≥ 2, let ν = (ν1 , ν2 , . . . , νn−1 ) ∈ Cn−1 . We have repeatedly used the fact
that the function Iν : hn → C (see Section 2.4) given in (5.1.1), i.e.,
 n−1
n−1  b ν
Iν (z) = yi i, j j , (with z ∈ hn )
i=1 j=1

and

ij if i + j ≤ n,
bi, j =
(n − i)(n − j) if i + j ≥ n,
is an eigenfunction of every S L(n, R)–invariant differential operator in Dn .
These are not the only possible eigenfunctions, however. For example, we have
shown in Section 3.4 that the functions
√ √
y ν , y 1−ν , y K ν− 12 (2π|m|y)e2πimx , y Iν− 12 (2π|m|y)e2πimx ,
 2 
(with m ∈ Z, m = 0) are all eigenfunctions of  = −y 2 ∂∂x 2 + ∂∂y 2 with
2

eigenvalue ν(1 − ν). Of these four functions, only

√
y K ν− 12 (2π|m|y)e2πimx ,
has good growth properties (exponential decay as y → ∞) and appears in the
Fourier expansion of Maass forms. This is the multiplicity one theorem of
Section 3.4. We seek to generalize these concepts to the group S L(n, Z) with
n > 2.
For n ≥ 2, let Un (R) denote the group of upper triangular matrices with
1s on the diagonal. Fix ψ :Un (R) → C to be a character of Un (R) which, by
definition, satisfies the identity
ψ(u · v) = ψ(u)ψ(v)
for all u, v ∈ Un (R).
Definition 5.4.1 Let n ≥ 2. An S L(n, Z)–Whittaker function of type
ν = (ν1 , ν2 , . . . , νn−1 ) ∈ Cn−1 ,
associated to a character ψ of Un (R), is a smooth function W : hn → C which
satisfies the following conditions:
(1) W (uz) = ψ(u)W (z) (for all u ∈ Un (R), z ∈ hn ),
 (z) = λ D W (z)
(2) DW (for all D ∈ Dn , z ∈ hn ),
(3) |W (z)|2 d ∗ z < ∞,
 √3 1
2 ,2
 5.5 Jacquet’s Whittaker function 129

where λ D is defined by D Iν (z) = λ D Iν (z), the Siegel set  √3 , 1 is as in

2 2
Definition 1.3.1, and the left invariant quotient measure d ∗ z is given by
Proposition 1.5.3.
The primordial example of a Whittaker function for S L(n, Z) is the integral
 1  1 
··· φ(u · z) ψ(u) du i, j
0 0 1≤i< j≤n

given in Proposition 5.2.4. Here φ(z) is a Maass form for S L(n, Z). This example
shows that Whittaker functions occur naturally in the Fourier expansion of
Maass forms. The importance of Whittaker functions cannot be underestimated.
They are the cornerstone for the entire theory of L–functions.
We shall show in the next section that it is always possible to explic-
itly construct one non–trivial Whittaker function. Remarkably, this special
Whittaker function has good growth properties and is the only Whittaker func-
tion that appears in the Fourier expansion of Maass forms (multiplicity one
theorem).

5.5 Jacquet’s Whittaker function

Whittaker functions for higher rank groups were first studied by Jacquet (1967).
The theory was subsequently fully worked out for G L(3, R) in (Bump, 1984),
and then for arbitrary real reductive groups in (Wallach, 1988). Jacquet intro-
duced the following explicit construction. For n ≥ 2, fix
m = (m 1 , . . . , m n−1 ) ∈ Zn−1 , ν = (ν1 , ν2 , . . . , νn−1 ) ∈ Cn−1 ,
and let
⎛ ⎞
1 u 1,2 u 1,3 ··· u 1,n
⎜ 1 u 2,3 ··· u 2,n ⎟
⎜ ⎟
⎜ .. .. ⎟
u=⎜ . . ⎟ ∈ Un (R).
⎜ ⎟
⎝ 1 u n−1,n ⎠
1
In order to simplify later notation, it is very convenient to relabel the superdiag-
onal elements
u 1 = u n−1,n , u 2 = u n−2,n−1 , ..., u n−1 = u 1,2 .
Define ψm to be the character of Un (R) defined by

ψm (u) := e2πi [m 1 u 1 +m 2 u 2 +···+m n−1 u n−1 ] .

130 Maass forms and Whittaker functions for S L(n, Z)

Note that all characters of Un (R) are of this form.

For z = x y ∈ hn (as in the beginning of Section 5.1) and m i = 0,
(1 ≤ i ≤ n − 1), define

WJacquet (z; ν, ψm ) := Iν (wn · u · z) ψm (u) d ∗ u (5.5.1)
Un (R)

to be Jacquet’s Whittaker function. Here (!x" denotes the largest integer ≤ x)

⎛ !n/2" ⎞
⎜
⎜
(−1) ⎟
⎟
⎜ ⎟
⎜ 1 ⎟
⎜ ⎟
wn = ⎜
⎜
⎜
⎜ .
⎟
⎟
⎟
⎟
∈ S L(n, Z),
⎜
⎜
⎜
.. ⎟
⎟
⎟
⎝ ⎠
1
 ∞ ∞ 
∗
d u= ··· du i, j .
1≤i< j≤n
Un (R) −∞ −∞

Proposition 5.5.2 Let n ≥ 2. Assume that (νi ) > 1/n for i = 1, 2, . . . ,

n − 1 and m i = 0, (1 ≤ i ≤ n − 1). Then the integral on the right-hand side
of (5.5.1) converges absolutely and uniformly on compact subsets of hn and has
meromorphic continuation to all ν ∈ Cn−1 . The function WJacquet (z; ν, ψm ) is
an S L(n, Z)–Whittaker function of type ν and character ψm . Furthermore, we
have the identity
 
WJacquet (z; ν, ψm ) = cν,m · WJacquet M z; ν, ψ m1 , m2 ,..., mn−1
|m 1 | |m 2 | |m n−1 |

= cν,m · ψm (x) · WJacquet (M y; ν, ψ1,...,1 ),

where
⎛ cν,m = 0 (depends
⎞ only on ν, m) and M=
|m 1 m 2 ···m n−1 |

⎜ . .. ⎟
⎝ |m 1 m 2 | ⎠.
|m 1 |
1

+
n−1
)
n−1 bi, j ν j −i(n−i)
Remark The reader may verify that cν,m = |m i | j=1 . For example
i=1
when the dimension n = 2, we have c2,ν = |m 1 |ν1 −1 , whereas for “n = 3” the
coefficient is c3,ν = |m 1 |ν1 +2ν2 −2 |m 2 |2ν1 +ν2 −2 .

Proof We shall defer the proof of the convergence and the meromorphic con-
tinuation of the integral until later. At this point, we show that WJacquet (z; ν, ψm )
satisfies Definition 5.4.1 (1) and (2) of a Whittaker function. First of all, note
 5.5 Jacquet’s Whittaker function 131

that if
⎛ ⎞
1 a1,2 a1,3 ··· a1,n
⎜ 1 a2,3 ··· a2,n ⎟
⎜ ⎟
⎜ .. .. ⎟
a = ⎜ . . ⎟ ∈ Un (R),
⎜ ⎟
⎝ 1 an−1,n ⎠
1

then after changing variables,


WJacquet (az; ν, ψm ) = Iν (wn · u · a · z) ψm (u) d ∗ u
Un (R)

= Iν (wn · u · z) ψm (u · a −1 ) d ∗ u
Un (R)

= ψm (a) WJacquet (z; ν, ψm ).

Second, using the fact that every differential operator D ∈ Dn is invariant under
left multiplication by S L(n, R), it follows from the definition of λ D given in
Definition 5.4.1, that

D Iν (wn · u · z) = λ D Iν (wn · u · z).

Consequently,

D WJacquet (z; ν, ψm ) = D (Iν (wn · u · z)) ψm (u) d ∗ u
Un (R)

= λD Iν (wn · u · z) ψm (u) d ∗ u
Un (R)

= λ D · WJacquet (z; ν, ψm ).

We have thus proved Proposition 5.5.2 under the assumption that the integral
(5.5.1) converges absolutely and uniformly on compact subsets of hn to an L2
function on the Siegel set  √3 , 1 .
2 2
Next, we prove the identity

WJacquet (z; ν, ψm ) = cν,m · WJacquet M z; ν, ψ1 ,2 ,...,n−1 ,

with i = m i /|m i |, (i = 1, 2, . . . , n − 1). We have, after making the transfor-

mations

u 1 → |m 1 |u 1 , u 2 → |m 2 |u 2 , ..., u n−1 → |m n−1 |u n−1 ,

132 Maass forms and Whittaker functions for S L(n, Z)

that


WJacquet M z; ν, ψ1 ,2 ,...,n−1 = Iν (wn · u · M z)e−2πi [1 u 1 +···+n−1 u n−1 ] d ∗ u
Un (R)

n−1 
= |m i | Iν (wn · Mu · z)e−2πi [|m 1 |1 u 1 +···+|m n−1 |n−1 u n−1 ] d ∗ u
i=1 Un (R)


n−1 
= |m i | Iν (wn Mwn · wn uz)e−2πi [m 1 u 1 +···+m n−1 u n−1 ] d ∗ u
i=1 Un (R)

= cν,m · WJacquet (z; ν, ψm ),

for some constant cν,m ∈ C. Now,

WJacquet (M z; ν, ψ1 ,2 ,...,n−1 ) = ψm (x) · WJacquet (M y; ν, ψ1 ,2 ,...,n−1 ).

To complete the proof of the identity in Proposition 5.5.2 it remains to show

that

WJacquet (M y; ν, ψ1 ,2 ,... ,n−1 ) = WJacquet (M y; ν, ψ1,...,1 ).

Note that this identity holds because M y is a diagonal matrix with positive
entries. To prove it, consider (for j = 1, 2, . . . , n − 1) the (n − j + 1)th row:

(0, . . . , 0, 1, u j , u n− j+1,n− j+2 , ..., u n− j+1,n ),

of the matrix u. It is easy to see that for each 1 ≤ j ≤ n − 1, we can make the
transformation

u j →  j u j ,

by letting

u → δ j · u · δ j ,

where δ j is a diagonal matrix with 1s along the diagonal except at the

(n − j + 1)th row where there is an  j . Note that the other u ,k with
1 ≤ ≤ k − 2 ≤ n − 2 may also be transformed by  j , but, as we shall soon
see, this will not be relevant.
 5.5 Jacquet’s Whittaker function 133

If we make the above transformations in the integral for the Whittaker func-
tion, then that integral takes the form:


WJacquet (M y; ν, ψ1 ,...,n−1 ) = Iν (wn · u · M y)e−2πi [1 u 1 +···+n−1 u n−1 ] d ∗ u
Un (R)

= Iν (wn · δ j uδ j · M y)e−2πi [1 u 1 +···+u j +···+n−1 u n−1 ] d ∗ u
Un (R)

= Iν (δ j wn · u · M yδ j )e−2πi [1 u 1 +···+u j +···+n−1 u n−1 ] d ∗ u
Un (R)

= Iν (wn · u · M y)e−2πi [1 u 1 +···+u j +···+n−1 u n−1 ] d ∗ u
Un (R)

= WJacquet (M y; ν, ψ1 ,..., 1 ,...,n−1 ).

"#$%
jth position

One may do the above procedure for each j = 1, . . . , n − 1. In the end, we

prove the required identity.
The proof of the absolute convergence and meromorphic continuation of the
integral (5.5.1) is muchmore difficult.
 We shall prove it now for n = 2. In this
y x
case we may take z = and ν ∈ C. It follows that
0 1

∞      
0 −1 1 u y x
|WJacquet (z; ν, ψm )| ≤ IRe(ν) · · du
1 0 0 1 0 1
−∞

∞  Re(ν)
y
= du
(x + u)2 + y 2
−∞

∞
du
=y 1−Re(ν)
, (5.5.3)
(u 2 + 1)Re(ν)
−∞

which converges absolutely for Re(ν) > 1/2. For n = 2, the meromorphic
continuation of (5.5.1) is obtained by direct computation of the integral
134 Maass forms and Whittaker functions for S L(n, Z)

 
1 u
WJacquet (z; ν, ψm ). With the choice ψm = e2πimu , we have
0 1
 ∞  ν
y
WJacquet (z; ν, ψm ) = e−2πimu du
−∞ (x + u) + y
2 2
 ∞ −2πimuy
e
=e 2πimx 1−ν
y ν
du
−∞ (u + 1)
2

2|m|ν− 2 π ν √
1

= y K ν− 12 (2π|m|y) · e2πimx . (5.5.4)

(ν)
Since the Bessel function, K s (y), satisfies the functional equation
K s (y) = K −s (y), we see that the Whittaker function satisfies the functional
equation
∗
WJacquet (z; ν, ψm ) := |mπ|−ν (ν)WJacquet (z; ν, ψm )
∗
= WJacquet (z; 1 − ν, ψm ). (5.5.5)

The proof of the absolute convergence and meromorphic continuation of

(5.5.1) for the case n > 2, is presented in Section 5.8. We shall deduce it using
properties of norms of exterior products of vectors in Rn . The theory of exterior
powers of Rn is briefly reviewed in the next section.

5.6 The exterior power of a vector space

Basic references for this material are: (Bourbaki, 1998a, 2003), (Edelen, 1985),
(Brown, 1988).
For n = 1, 2, . . . , = 1, 2, . . . , let ⊗ (Rn ) denote the th tensor product
of the vector space Rn (considered as a vector space over R). The vector space
⊗ (Rn ) is generated by all elements of type

v1 ⊗ v2 ⊗ · · · ⊗ v ,

with vi ∈ Rn for i = 1, 2, . . . , .
We define

Λ (Rn ) = ⊗ (Rn )/a ,

where a denotes the vector subspace of ⊗ (Rn ) generated by all elements of

type

v1 ⊗ v2 ⊗ · · · ⊗ v ,
 5.6 The exterior power of a vector space 135

where vi = v j for some i = j. It is not hard to show (see (Bourbaki, 1998a,

2003)) that Λ (Rn ) can be realized as the vector space (over R) generated by
all elements of the form
v1 ∧ v2 ∧ · · · ∧ v
where the wedge product, ∧, satisfies the rules
v ∧ v = 0, v ∧ w = −w ∧ v,
(a1 v1 + a2 v2 ) ∧ w = a1 v1 ∧ w + a2 v2 ∧ w,
for all v, v1 , v2 , w ∈ Rn and a1 , a2 ∈ R.
Example 5.6.1 Consider the wedge product in R2 , where the canonical basis
for R2 is taken to be e1 = (1, 0), e2 = (0, 1). Then we have
(a1 e1 + a2 e2 ) ∧ (b1 e1 + b2 e2 ) = (a1 b2 − a2 b1 )e1 ∧ e2 .
There is a canonical inner product,  ,  : Rn × Rn → R, given by
v, w := v · t w, (5.6.2)
for all v, w ∈ Rn . It easily follows that we may extend this inner product to an
inner product  , ⊗ on ⊗ (Rn ) × ⊗ (Rn ) by defining


v, w⊗ := vi , wi , (5.6.3)
i=1

for all v = v1 ⊗ v2 ⊗ · · · ⊗ v , w = w1 ⊗ w2 ⊗ · · · ⊗ w ∈ ⊗ (Rn ). Note

that this agrees with the canonical inner product on ⊗ (Rn ) × ⊗ (Rn ) .
Our next goal is to define an inner product  ,  on Λ (Rn ) × Λ (Rn ).
Let e1 = (1, 0, . . . , 0), . . . , en = (0, . . . , 0, 1) denote the canonical basis for
Rn , and let


a= ai1 ,...,i ei1 ∧ · · · ∧ ei
1≤i 1 ,...,i ≤n

denote an arbitrary element of Λ (Rn ).

Consider the map φ : Λ (Rn ) → ⊗ (Rn ) given by
1


φ (a) := ai1 ,...,i Sign(σ ) · eσ (i1 ) ⊗ eσ (i2 ) ⊗ · · · ⊗ eσ (i ) ,
! 1≤i1 ,...,i ≤n σ ∈S
(5.6.4)
where S denotes the symmetric group of all permutations of {1, 2, . . . , } and
Sign(σ ) is plus or minus 1 according to whether σ ∈ S is an even or odd
permutation, i.e., it is a product of an even or odd number of transpositions.
136 Maass forms and Whittaker functions for S L(n, Z)

One easily checks that the map (5.6.4) is well defined and shows that Λ (Rn )
is isomorphic to a subspace of ⊗ (Rn ) generated by

ei1 ⊗ ei2 ⊗ · · · ⊗ ei ,

where 1 ≤ i 1 , i 2 , . . . , i ≤ n are distinct integers. We may then define

v, w := φ (v), φ (w)⊗ , (5.6.5)

for all v, w ∈ Λ (Rn ).

Now, the group S L(n, R) acts on the vector space Rn by right multiplication.
Thus, if v = (a1 , . . . , an ) ∈ Rn and g ∈ S L(n, R), then the action is given by
v · g, where · denotes the multiplication of a row vector by a matrix. This
action may be extended to an action ◦ of S L(n, R) on ⊗ (Rn ), by defining

v ◦ g := (v1 · g) ⊗ (v2 · g) ⊗ · · · ⊗ (v · g),

for all v = v1 ⊗ v2 ⊗ · · · ⊗ v ∈ ⊗ (Rn ) . In view of the isomorphism (5.6.4),

one may also define an action ◦ of S L(n, R) on Λ (Rn ) given by

v ◦ g := (v1 · g) ∧ (v2 · g) ∧ · · · ∧ (v · g),

for all v = v1 ∧ v2 ∧ · · · ∧ v ∈ Λ (Rn ) .

We now prove two lemmas which will allow us to construct, using norms on
Λ (Rn ), a function very similar to the function Iν (z) as defined in Section 2.4. In
fact, the sole purpose of this brief excursion into the theory of exterior powers
of a vector space is to ultimately realize Jacquet’s Whittaker function as an
integral of certain complex powers of norms on exterior product spaces.

Lemma 5.6.6 Let k ∈ O(n, R) with n ≥ 2. Then for ≥ 1, we have

v, w = v ◦ k, w ◦ k ,

&
||v|| := v, v = ||v ◦ k||,

for all v, w ∈ Λ (Rn ). Here || || denotes the canonical norm on Λ (Rn ).

Proof First note that for the inner product on Rn , we have

v · k, w · k = (v · k) · t (w · k) = v · k · t k · t w = v · t w = v, w,

for all v, w ∈ Rn . It immediately follows from (5.6.3) that

v ◦ k, w ◦ k⊗ (Rn ) = v, w⊗ (Rn ) ,

for all v, w ∈ ⊗ (Rn ). Finally, the invariance of the action by k on the inner
product can be extended to Λ (Rn ) by (5.6.5).

 5.6 The exterior power of a vector space 137

Lemma 5.6.7 Fix n ≥ 2. Let e1 = (1, 0, . . . , 0), . . . , en = (0, . . . , 0, 1)

denote the canonical basis for Rn . Then for all
⎛ ⎞
1 u 1,2 u 1,3 · · · u 1,n
⎜ 1 u 2,3 · · · u 2,n ⎟
⎜ ⎟
⎜ .. .. ⎟ ∈ S L(n, R),
u=⎜ . . ⎟
⎜ ⎟
⎝ 1 u n−1,n ⎠
1
and every 1 ≤ ≤ n − 1, we have
(en− ∧ · · · ∧ en−1 ∧ en ) ◦ u = en− ∧ · · · ∧ en−1 ∧ en .

Proof First of all, we have

(en− ∧ · · · ∧ en−1 ∧ en ) ◦ u = (en− · u) ∧ · · · ∧ (en−1 · u) ∧ (en · u).
Since ei · u = ei + linear combination of e j with j > i, we see that the extra
linear combination is killed in the wedge product.

Lemma 5.6.8 (Cauchy–Schwartz type inequalities) Let n ≥ 2. Then for all

v, w ∈ Λ (Rn ), we have
|v, w |2 ≤ v, v · w, w ,
||v ∧ w|| ≤ ||v|| · ||w|| .

Proof The classical Cauchy–Schwartz inequality

|v, w⊗ |2 ≤ v, v⊗ · w, w⊗

is well known on the tensor product space ⊗ (Rn ). It extends to Λ (Rn ) by the
identity (5.6.5).
To prove the second Cauchy–Schwartz type inequality consider


v= ai1 ,i2 ,...,i ei1 ⊗ ei2 ⊗ · · · ⊗ ei ∈  (Rn ),
i 1 ,i 2 ,...,i


w= b j1 , j2 ,..., j e j1 ⊗ e j2 ⊗ · · · ⊗ e j ∈  (Rn ).
j1 , j2 ,..., j

Note that we are thinking of v, w as also lying in ⊗ (Rn ). Then

||v||2 = |ai1 ,i2 ,...,i |2 , ||w||2 = |b j1 , j2 ,..., j |2 ,
i 1 ,i 2 ,...,i j1 , j2 ,..., j
138 Maass forms and Whittaker functions for S L(n, Z)

and


||v||2 ||w||2 = |ai1 ,i2 ,...,i |2 |b j1 , j2 ,..., j |2 .
i 1 ,i 2 ,...,i
j1 , j2 ,..., j

Now

v∧w

 
= ai1 ,i2 ,...,i b j1 , j2 ,..., j ei1 ⊗ ei2 ⊗ · · · ⊗ ei ∧ e j1 ⊗ e j2 ⊗ · · · ⊗ e j
i 1 ,i 2 ,...,i
j1 , j2 ,..., j

1



= 2
ai1 ,i2 ,...,i b j1 , j2 ,..., j Sign(σ ) Sign(σ  )
( !) i1 ,i2 ,...,i 
σ ∈S σ ∈S
j1 , j2 ,..., j
 
× eσ (i1 ) ⊗ · · · ⊗ eiσ ( ) ⊗ eσ  ( j1 ) ⊗ · · · ⊗ eσ  ( j ) .

The result now follows because ||u + u  ||2⊗ ≤ ||u||2⊗ + ||u  ||2⊗ for all
u, u  ∈ ⊗ (Rn ) .

5.7 Construction of the Iν function using wedge products

We now construct, using wedge products and norms on Λ (Rn ), the function
Iν (z) as defined in Section 5.4. Recall from Section 1.2 that every z ∈ hn can
be uniquely written in the form
⎛ ⎞ ⎛ ⎞
1 x1,2 x1,3 ··· x1,n y1 y2 · · · yn−1
⎜ 1 x2,3 ··· x2,n ⎟ ⎜ y1 y2 · · · yn−2 ⎟
⎜ ⎟ ⎜ ⎟
⎜ .. .. ⎟ ⎜ .. ⎟
z=⎜ . . ⎟·⎜ . ⎟.
⎜ ⎟ ⎜ ⎟
⎝ 1 xn−1,n ⎠ ⎝ y1 ⎠
1 1
(5.7.1)

Lemma 5.7.2 For n ≥ 2, let z be given by (5.7.1) and let || || denote the norm
on  (Rn ) as in Lemma 5.6.6. Then for ν = (ν1 , . . . , νn−1 ) ∈ Cn−1 , we have
the identity
 n−2
  +
n−1
  iν
Iν (z) := (en−i ∧ · · · ∧ en−1 ∧ en ) ◦ z −nνn−i−1 · |Det(z)| i=1 n−i .
i=0
 5.7 Construction of the Iν function using wedge products 139

Proof It follows from Lemma 5.6.6 that the function

 n−2
   +
n−1
 iν
Iν∗ (z) := (en−i ∧ · · · ∧ en−1 ∧ en ) ◦ z −nνn−i−1 · |Det(z)| i=1 n−i ,
i=0

is invariant under the transformation

z → z·k
for all k ∈ S O(n, R). Here, we have used the fact that Det(z · k) = Det(z) since
Det(k) = 1. It also follows from Lemma 5.6.7 that Iν∗ (z) = Iν∗ (y) with
⎛ ⎞
y1 y2 · · · yn−1
⎜ y1 y2 · · · yn−2 ⎟
⎜ ⎟
⎜ .. ⎟
y=⎜ . ⎟.
⎜ ⎟
⎝ y1 ⎠
1
One also easily checks that
Iν∗ (az) = Iν∗ (z)
for a ∈ R× . Thus Iν∗ (z) is well defined on hn . To show that Iν∗ (y) = Iν (y), note
that
en−i · y = y1 · · · yi , ··· , en−1 · y = y1 , en · y = 1.
Consequently,
  
(en−i ∧ · · · ∧ en−1 ∧ en ) ◦ y −nνn−i−1 = y1 · (y1 y2 ) · · · (y1 · · · yi ) −nνn−i−1

−nνn−i−1

i
= y i+1− .
=1

Furthermore,

+
n−1
+ i νn−i

n−1
i νn−i n−1 i=1
|Det(z)| i=1 = y n− ,
=1

from which the result follows by a brute force computation.

In order to demonstrate the power of the exterior algebra approach, we

explicitly compute the integral for the Jacquet Whittaker function on G L(3, R).
Example 5.7.3 ⎛
(G L(3,⎞
R) Whittaker
⎛
function)
⎞
Let n = 3, ν = (ν1 , ν2 )
⎜
⎜
1 u 2 u 3⎟
⎟
y y
⎜ 1 2
⎜
⎟
⎟
∈ C2 , u = ⎜⎜⎜⎜ 1 u 1 ⎟⎟⎟⎟, y = ⎜⎜⎜⎜ y1 ⎟⎟⎟, and for fixed m = (m 1 , m 2 ) ∈ Z2
⎟

⎝ ⎠ ⎝ ⎠
1 1
140 Maass forms and Whittaker functions for S L(n, Z)

define ψm (u) = e2πi(m 1 u 1 +m 2 u 2 ) . Then we have the explicit integral represen-

tation:
WJacquet (y; ν, ψm )
∞ ∞ ∞
3 4−3ν1 /2
= y1ν1 +2ν2 y22ν1 +ν2 y12 y22 + u 21 y22 + (u 1 u 2 − u 3 )2
−∞ −∞ −∞
3 4−3ν2 /2 −2πi(m 1 u 1 +m 2 u 2 )
× y12 y22 + u 22 y12 + u 23 e du 1 du 2 du 3 .

Proof In view of (5.5.1), it is enough to compute Iν (w3 uy). It follows from

Lemma 5.7.2 that
 ν2 +2ν1
Iν (w3 uy) = ||e3 w3 uy||−3ν2 · ||(e2 w3 uy) ∧ (e3 w3 uy)||−3ν1 · y12 y2 .
(5.7.4)

We compute
⎛ ⎞⎛ ⎞⎛ ⎞
−1 1 u2 u3 y1 y2
e3 w3 uy = (0, 0, 1) ⎝ 1 ⎠⎝ 1 u1 ⎠ ⎝ y1 ⎠
1 1 1
⎛ ⎞
y1 y2 u 2 y1 u3
= (1, 0, 0) ⎝ y1 u1 ⎠
1
= y1 y2 e1 + u 2 y1 e2 + u 3 e3 .

Hence
3 41
||e3 w3 uy|| = y12 y22 + u 22 y12 + u 23 2 .

Similarly,
⎛ ⎞⎛ ⎞⎛ ⎞
−1 1 u2 u3 y1 y2
e2 w3 uy = (0, 1, 0) ⎝ 1 ⎠⎝ 1 u1 ⎠ ⎝ y1 ⎠
1 1 1
⎛ ⎞
y1 y2 u 2 y1 u3
= (0, 1, 0) ⎝ y1 u1 ⎠
1
= y1 e2 + u 1 e3 ,

so that

(e2 w3 uy)∧ (e3 w3 uy) = y12 y2 e1 ∧ e2 + (u 1 u 2 y1 − u 3 y1 )e2 ∧ e3 + u 1 y1 y2 e1 ∧ e3 .

 5.8 Convergence of Jacquet’s Whittaker function 141

Consequently
3 41
||(e2 w3 uy) ∧ (e3 w3 uy)|| = y1 y12 y22 + (u 1 u 2 − u 3 )2 + u 21 y22 2 .

The result follows upon substituting the results of the above computations into
(5.7.4).

5.8 Convergence of Jacquet’s Whittaker function

Assume Re(νi ) > 1/n (for i = 1, 2, . . . , n − 1.) The absolute convergence
of the integral (5.5.1) has already been proved for the case n = 2 in (5.5.4). The
proof for all n ≥ 2 is based on induction on n and will now be given.
Recall the notation:
⎛ !n/2" ⎞
(−1)
⎜ 1 ⎟
wn = ⎜ ⎝
⎟ ∈ S L(n, Z),
⎠
···
1
⎛ ⎞ ⎛ ⎞
1 u 1,2 u 1,3 · · · u 1,n y1 y2 · · · yn−1
⎜ 1 u 2,3 · · · u 2,n ⎟ ⎜ y1 y2 · · · yn−2 ⎟
⎜ ⎟ ⎜ ⎟
⎜ .. .. ⎟ , ⎜ .. ⎟
u=⎜ . . ⎟ y=⎜ . ⎟.
⎜ ⎟ ⎜ ⎟
⎝ 1 u n−1,n ⎠ ⎝ y1 ⎠
1 1
It is enough to prove the absolute convergence of

WJacquet (y; ν, ψ) := Iν (wn · u · y) ψ(u) d ∗ u. (5.8.1)
Un (R)

Now, it follows from the Cauchy–Schwartz inequality, Lemma 5.6.8, and

Lemma 5.7.2, that


 
n−2 −n(νn−i−1 )
|Iν (wn uy)|  ||en · wn uy|| −nV  (en−i ∧ · · · ∧ en−1 ) ◦ wn uy  ,
y
i=1
(5.8.2)
where


n−2
V = (νn−i−1 ).
i=0

The  constant in (5.8.2) is independent of u and depends only on y. It appears

because Det(wn uy) = Det(y) does not depend on u, but only on y. In view of
142 Maass forms and Whittaker functions for S L(n, Z)

the identity
||en · wn uy|| = ||e1 · uy||
3 41
= (y1 · · · yn−1 )2 + (u 1,2 y1 · · · yn−2 )2 + · · · + (u 1,n−1 y1 )2 + (u 1,n )2 2 ,
it immediately follows from (5.8.2) that (5.8.1) converges absolutely for
Re(νi ) > 1/n (1 ≤ i ≤ n − 1) if the following two integrals converge
absolutely:
∞ ∞
3 4−nV /2 
n
··· (y1 · · · yn−1 )2 + (u 1,2 y1 · · · yn−2 )2 + · · · + (u 1,n )2 du 1,k ,
k=1
−∞ −∞
(5.8.3)

∞ ∞
n−2
 


−nνn−i−1
···  (en−i ∧ · · · ∧ en−1 ) ◦ wn uy  du i, j .
i=1 1<i< j≤n
−∞ −∞
(5.8.4)
Clearly, the first integral converges absolutely if Re(νi ) > 1/n, for
1 ≤ i ≤ n − 1. Furthermore, if 1 < i < n, then ei · wn = e for some = 1.
This immediately implies that
ei · wn uy = ei · wn u  y 
where
⎛ ⎞ ⎛ ⎞
1 0 0 ··· 0 1
⎜ 1 u 2,3 ··· u 2,n ⎟ ⎜ y1 y2 · · · yn−2 ⎟
⎜ ⎟ ⎜ ⎟
⎜ .. .. ⎟ ⎜ .. ⎟
u = ⎜ . . ⎟, y = ⎜ . ⎟.
⎜ ⎟ ⎜ ⎟
⎝ 1 u n−1,n ⎠ ⎝ y1 ⎠
1 1
It follows that the second integral (5.8.4) may be rewritten as
∞ ∞
n−2
 



··· (en−i ∧ · · · ∧ en−1 ) ◦ wn u  y   n−i−1
−nν
du i, j .
i=1 1<i< j≤n
−∞ −∞
(5.8.5)
The remarkable thing is that the integral in (5.8.5) can be interpreted as a Jacquet
Whittaker function for S L(n − 1, Z). This allows us to apply induction from
which the absolute convergence of (5.8.1) follows.
Recall the definition of wn given just after (5.5.1). To see that the integral
(5.8.5) is a Jacquet Whittaker function for S L(n − 1, Z), we use the matrix
 5.8 Convergence of Jacquet’s Whittaker function 143

identity:
⎛ ⎞ ⎛ ⎞
0 1 1 0
⎜ 0 1 ⎟ ⎜ !n/2"⎟
0 (−1) ⎟
⎜ ⎟ ⎜
⎜ .. .. ⎟ ⎜ . . ⎟
wn = ⎜ . . ⎟·⎜ .. .. ⎟.
⎜ ⎟ ⎜ ⎟
⎝ 0 1⎠ ⎝ 0 1 ⎠
1 0 0 1
Further,
⎛ ⎞
0 1
⎜ 0 1 ⎟
⎜ ⎟
⎜ .. .. ⎟
e · ⎜ . . ⎟ = e +1
⎜ ⎟
⎝ 0 1⎠
1 0
for 1 ≤ ≤ n − 1. It immediately follows from these remarks that (5.8.5) may
be rewritten as
∞ ∞
n−2
 



···     −nνn−i−1
(en+1−i ∧ · · · ∧ en ) ◦ w u y du i, j ,
n
i=1 1<i< j≤n
−∞ −∞
(5.8.6)
where
⎛ ⎞
1 0
⎜ 0 (−1)!n/2" ⎟  
⎜ ⎟
⎜ . . ⎟ 1
wn =⎜ .. .. ⎟= .
⎜ ⎟ wn−1
⎝ 0 1 ⎠
0 1
Furthermore, we may write
   
1 1
u = , y = ,
µ η
with µ ∈ Un−1 (R) and η a diagonal matrix in G L(n − 1, R). With these obser-
vations, one may deduce that
 
n−2 
(en+1−i ∧ · · · ∧ en ) ◦ w u  y  −nνn−i−1
n
i=1

n/(n−1)
 
(n−1)−2 
= (en−1−i ∧ · · · ∧ en−1 ) ◦ wn−1 µη−(n−1)νn−2−i .
i=0
144 Maass forms and Whittaker functions for S L(n, Z)

It follows that
∞ ∞
n−2
 


−nνn−i−1
··· (en+1−i ∧ · · · ∧ en ) ◦ w u y 
  
du i, j
n
i=1 1<i< j≤n
−∞ −∞
∞ ∞
  
···  I(nν/(n−1)),n−1 (wn−1 µη) dµ,
y
−∞ −∞

where Iν,n−1 denotes the Iν function (as defined in Section 5.4) for
G L(n − 1, R). By induction, we obtain the absolute convergence in the region
Re(ν) > 1/n.

5.9 Functional equations of Jacquet’s Whittaker function

In order to explicitly state the group of functional equations satisfied
by Jacquet’s Whittaker function, we need to introduce some prelimi-
nary notation. Fix an integer n ≥ 2, and let Wn denote the Weyl group
of S L(n, Z) consisting of all n × n matrices in S L(n, Z) which have
exactly one ±1 in each row and column. For each fixed w ∈ Wn , and
every ν − n1 = ν1 − n1 , . . . , ν n−1 − n1 ∈ Cn−1 , let us define ν  so that
ν  − n1 = ν1 − n1 , . . . , νn−1

− n1 ∈ Cn−1 satisfies
Iν− n1 (y) = Iν  − n1 (wy), (5.9.1)
for all
⎛ ⎞
y1 y2 · · · yn−1
⎜ y1 y2 · · · yn−2 ⎟
⎜ ⎟
⎜ .. ⎟
y=⎜ . ⎟.
⎜ ⎟
⎝ y1 ⎠
1
Definition 5.9.2 Fix an integer n ≥ 2, and let ψ be a character of Un (R). We
define
 
n−1  
∗ − 12 −v j,k 1
WJacquet (z; ν, ψ) = WJacquet (z; ν, ψ) · π  + v j,k ,
j=1 j≤k≤n−1
2

where


j−1
nνn−k+i − 1
v j,k = ,
i=0
2
and WJacquet (z; ν, ψ) denotes Jacquet’s Whittaker function (5.5.1).
 5.9 Functional equations of Jacquet’s Whittaker function 145

We may now state the functional equations of the Whittaker functions.

Theorem 5.9.3 Fix an integer n ≥ 2, and fix ψ = ψ1,1,...,1 so that

+
n−1
2πi u , +1
ψ(u) = e =1 ,

∗
for u ∈ Un (R). Then the Whittaker function WJacquet (z; ν, ψ) has a holomorphic
continuation to all ν ∈ C . For each w ∈ Wn , let ν, ν  satisfy (5.9.1). Then
n−1

we have the functional equation

∗ ∗
WJacquet (z; ν, ψ) = WJacquet (z; ν  , ψ).

∗ ∗
Remarks It is clear that WJacquet (z; ν, ψ) and WJacquet (z; ν  , ψ) are both
Whittaker functions of type ν and character ψ. It follows from Shalika’s multi-
plicity one theorem (Shalika, 1974) that the functional equation must hold up to
a constant depending on ν. The assumption that ψ(u) = ψ1,1,...,1 is not restric-
tive because Proposition 5.5.2 tells us that there is a simple identity relating
∗ ∗
WJacquet (z; ν, ψm ) and WJacquet (M z; ν, ψ1,1,...,1 ).
Before giving the proof of the functional equation, we will obtain explicit
versions of the functional equation ν → ν  given by (5.9.1). The Weyl group
Wn is generated by the simple reflections

⎛ ⎞
In−i−1
⎜ 0 −1 ⎟
σi = ⎜
⎝
⎟,
⎠ (i = 1, 2, . . . , n − 1),
1 0
Ii−1

where Ia denotes the a × a identity matrix. We adopt the convention that I0 is the
⎛ ⎞ ⎛ ⎞
In−2 0 −1
empty set so that σ1 = ⎝ 0 −1 ⎠ and σn−1 = ⎝ 1 0 ⎠.
1 0 In−2
Since σi (i = 1, . . . , n − 1) generate Wn , it is enough to give the functional
equation ν → ν  for the simple reflections w = σi . Fix an integer i with
1 ≤ i ≤ n − 1. In this case, ν  is defined by the equation

Iν− n1 (y) = Iν  − n1 (σi y) = Iν  − n1 σi yσi−1 = Iν  − n1 (y  ) (5.9.4)
146 Maass forms and Whittaker functions for S L(n, Z)

where

⎛ ⎞
y1 · · · yn−1
⎜ .. ⎟
⎜ . ⎟
⎜ ⎟
⎜ y1 · · · yi ⎟
⎜ ⎟
⎜ ⎟
y = ⎜ y1 · · · yi yi+1 ⎟
⎜ ⎟
⎜ .. ⎟
⎜ . ⎟
⎜ ⎟
⎝ y1 ⎠
1

is the diagonal matrix y with the (n − i)th and (n − i + 1)th rows interchanged.
It follows that if we put y  in Iwasawa form

⎛ ⎞
y1 · · · yn−1


⎜ .. ⎟
⎜ . ⎟
y = ⎜ ⎟,
⎝ y1 ⎠
1

then we must have

⎧
⎪
⎪ y if = i − 1, i, i + 1,
⎪
⎪
⎨y y if = i − 1,
+1
y  =
⎪ y −1
⎪ if = i,
⎪
⎪
⎩
y −1 y if = i + 1,

for 1 ≤ i ≤ n − 1 and 1 ≤ ≤ n − 1. Equation (5.9.4), together with the def-

inition of the Iν function given in Definition 5.4.1 imply (for 1 < i ≤ n − 1)
the following system of linear equations:

n−1  

n−1  
1 1
b , j νj − = b , j ν j − , (for = i),
j=1
n j=1
n
(5.9.5)


n−1  

n−1  
1 1
bi, j νj − = (bi−1, j − bi, j + bi+1, j ) · ν j − .
j=1
n j=1
n
 5.9 Functional equations of Jacquet’s Whittaker function 147

For each 1 ≤ i ≤ n − 1, the system of linear equations (5.9.5) has the

solution

 1
νn−i−1 =− + νn−i−1 + νn−i , (if i = n − 1),
n
 2
νn−i = − νn−i ,
n
(5.9.6)
 1
νn−i+1 =− + νn−i + νn−i+1 , (if i = 1),
n
ν  = ν ( = n − i − 1, n − i, n − i + 1).

Example 5.9.7 For n = 2, 3 the functional equations (5.9.6) take the explicit
form:

ν1 = 1 − ν1 , (n = 2),
2 1
ν2 = − ν2 , ν1 = − + ν1 + ν2 , (n = 3, i = 1),
3 3
2 1
ν1 = − ν1 , ν2 = − + ν1 + ν2 , (n = 3, i = 2).
3 3

It immediately follows from these computations that Theorem 5.9.3 can be

put in the following more explicit form.

Theorem 5.9.8 Fix an integer n ≥ 2 and fix an integer i with 1 ≤ i ≤ n − 1.

Let ν, ν  satisfy (5.9.6). Then we have the functional equations:

∗ ∗
WJacquet (z; ν, ψ) = WJacquet (z; ν  , ψ),
  
nνn−i
π − 2 νn−i 
n 
2
WJacquet (z; ν, ψ) = − n ν ·  nνn−i · WJacquet (z; ν  , ψ)
π 2 n−i  2

−(1− n2 νn−i )  1 − nνn−i
π
= ·  2
· WJacquet (z; ν  , ψ).
π − 2 νn−i
n
 nν2n−i
148 Maass forms and Whittaker functions for S L(n, Z)

Proof Fix a simple reflection σi with 1 ≤ i ≤ n − 1. The second functional

equation in Theorem 5.9.8 is equivalent to the first and can be obtained by
computing the affect of the transformations (5.9.6) on the v j,k given in Defini-
tion 5.9.2. It remains to prove the second functional equation in Theorem 5.9.8.
Recall that
⎛ !n/2" ⎞
(−1)
⎜ 1 ⎟
⎜ ⎟
wn = ⎜ . ⎟ ∈ S L(n, Z),
⎝ .. ⎠
1
where !x" denotes the smallest integer ≤ x. Define

wi := σi−1 wn .

The group N = Un (R) of upper triangular matrices with coefficients in R and

1s on the diagonal decomposes as

N = wi−1 Ni wi−1 · Ni (5.9.9)

where Ni is a one-dimensional subgroup with 1s on the diagonal, an arbitrary

real number at position {n − i, n − i + 1}, and zeros elsewhere, and Ni is the
subgroup of N with a zero at position {n − i, n − i + 1}.

Lemma 5.9.10 For 1 ≤ i ≤ n − 1, we have wi−1 Ni wi−1 = Nn−i .

Proof Note that if

⎛ ⎞
1
⎜ .. ⎟
⎜ . ⎟
⎜ ⎟
ni = ⎜
⎜ 1 a ⎟ ∈ Ni
⎟
⎜ .. ⎟
⎝ . ⎠
1
then
⎛ ⎞
1
⎜ .. ⎟
⎜ . ⎟
⎜ ⎟
σi n i σi−1 =⎜
⎜ −a 1 ⎟.
⎟
⎜ .. ⎟
⎝ . ⎠
1
Further, conjugating this by wn moves the −a to the position {i, i + 1}.

 5.9 Functional equations of Jacquet’s Whittaker function 149

It follows from (5.9.9) that we may factor every u ∈ Un (R) = N as

u = n i · n i with n i ∈ wi−1 Ni wi and n i ∈ Ni . For example, we may take
⎛ ⎞
1 0 0 0 0
⎜ 0 0 −1 0⎟
⎜ 0 ⎟
⎜ ⎟
σi = ⎜ 0 1 0 0 0⎟,
⎜ ⎟
⎝0 0 0 1 0⎠
0 0 0 0 1
⎧⎛ ⎞⎫ ⎧⎛ ⎞⎫
⎪
⎪ 1 0 0 0 0 ⎪ ⎪ ⎪
⎪ 1 0 0 0 0 ⎪ ⎪
⎪
⎪ ⎜0 1 ∗ ⎪ ⎪ ⎪
⎪
⎨⎜ 0 0⎟⎟⎪
⎪
⎬
⎪
⎪
⎨⎜
⎜0 1 0 0 0⎟⎟⎪
⎪
⎬
⎜ ⎟ ⎜ ⎟
Ni = ⎜ 0 0 1 0 0⎟ , wi−1 Ni wi = ⎜ 0 0 1 ∗ 0⎟
⎪
⎪ ⎜ ⎟⎪ ⎪ ⎜ ⎟⎪
⎪⎝ 0 0 0
⎪ 1 0 ⎠⎪⎪
⎪
⎪
⎪⎝ 0
⎪ 0 0 1 0 ⎠⎪⎪
⎪
⎪
⎩ 0 0 0 ⎪ ⎪ ⎪
0 1 ⎭ ⎩ 0 0 0 0 1 ⎭
⎧⎛ ⎞⎫
⎪
⎪ 1 ∗ ∗ ∗ ∗ ⎪ ⎪
⎪
⎪⎜ 0 1 ∗ ⎪
⎪
⎨⎜ ∗ ∗⎟⎟⎪
⎪
⎬
⎜ ⎟
Ni = ⎜ 0 0 1 0 ∗⎟ .
⎪⎜
⎪ ⎟⎪
⎪
⎪ ⎝0 0 0 1 ∗ ⎠⎪⎪
⎪
⎪
⎩ 0 0 0 ⎪
0 1 ⎭

Furthermore, by Lemma 5.9.10, we may express Jacquet’s Whittaker function

(5.5.1) in the form


WJacquet (z; ν, ψ) = Iν (wn · u · z) ψ(u) d ∗ u
U (R)
 0 1
n

= Iν (σi n i wi n i z)ψ̄(n i ) dn i ψ̄(n i ) dn i .

Ni Ni

The inner integral above (over the region Ni ) is a Whittaker function for the
group S L(2, R), and has a functional equation of type (5.5.5). This functional
equation is independent of the choice of wi n i z, and is precisely what is needed
to complete the proof of Theorem 5.9.8. We shall find the functional equation
by examining the case when wi n i z is the identity matrix.
We first compute Iν (σi n i ) for

⎛ ⎞ ⎛ ⎞
In−i−1 In−i−1
⎜ 0 −1 ⎟ ⎜ 1 u ⎟
σi = ⎜
⎝
⎟,
⎠ ni = ⎜
⎝
⎟.
⎠
1 0 0 1
Ii−1 , Ii−1 ,
150 Maass forms and Whittaker functions for S L(n, Z)

It follows that
⎛ ⎞
In−i−1
⎜ ⎟
(u 2 + 1)− 2
1 1
⎜ −u/(u 2 + 1) 2 ⎟
σi n i = ⎜ 1 ⎟·k
⎝ 0 (u 2 + 1) 2 ⎠
Ii−1 ,
for some orthogonal matrix k. If we then put σi n i in standard Iwasawa form,
we have
⎛ ⎞ ⎛a ···a ⎞
In−i−1 1 n−1
⎜ 1 −u/(u 2 + 1) ⎟ ⎜ .. ⎟
σi n i = ⎜ ⎟·⎜⎜ . ⎟
⎟
⎝ 0 1 ⎠ ⎝ ⎠
a1
Ii−1 , 1
with
⎧
⎪
⎪ 1 if = i − 1, i, i + 1
⎪
⎪
⎨(u 2 + 1) 12 if = i − 1
a = .
⎪
⎪(u 2 + 1)−1 if = i
⎪
⎪
⎩ 2 1
(u + 1) 2 if = i + 1.
Consequently
+
n−1
(bi−1, j −2bi, j +bi+1, j ) ν2j
Iν (σi n i ) = (u + 1)
2 j=1
.
The functional equation now follows from (5.5.5).

5.10 Degenerate Whittaker functions

Jacquet’s Whittaker function was constructed by integrating Iν (wn z) where wn
is the so-called long element of the Weyl group as in (5.5.1). Since the Iν
function is an eigenfunction of the invariant differential operators, its integral
inherits all those properties and gives us a Whittaker function.
It is natural to try this type of construction with other Weyl group elements
besides the long element wn . To get a feel for what is going on, let us try to do
this on G L(4). As an example, we shall consider the Weyl group element
⎛ ⎞
1 0 0 0
⎜0 0 0 1⎟
w=⎜ ⎝0 1 0 0⎠.
⎟

0 0 1 0
 5.10 Degenerate Whittaker functions 151

Let
⎛ ⎞ ⎛ ⎞
1 u 1,2 u 1,3 u 1,4 y1 y2 y3
⎜0 1 u 2,3 u 2,4 ⎟ ⎜ y1 y2 ⎟
u=⎜
⎝0
⎟, y=⎜ ⎟.
0 1 u 3,4 ⎠ ⎝ y1 ⎠
0 0 0 1 1

With a brute force computation, one sees that for ν = (ν1 , ν2 , ν3 ) ∈ C3 we have

 ν1 +2ν2 +ν3
y22 u 23,4 + y12
Iν (wuy) =   2
(u 2,4 − u 2,3 u 3,4 )2 + u 23,4 + y12 y22

  (ν1 +2ν2 +3ν3 )/2
y12 (u 2,4 − u 2,3 u 3,4 )2 + u 23,4 + y12 y22
×  2 2
u 3,4 + y12
  (3ν +2ν +ν )/2
× (u 2,4 − u 2,3 u 3,4 )2 + u 23,4 + y12 y22 y3 1 2 3 .

Clearly, the function does not involve the variables u 1,2 , u 1,3 , u 1,4 , so it is not
possible to integrate the function Iν (wuy) over the entire u space. We may only
consider some type of partial integral which does not involve all the u-variables.
We leave it to the reader to work out a general theory of degenerate Whittaker
functions and only briefly indicate how to define these objects. Let U = Un (R)
denote the group of upper triangular n × n matrices with real coefficients and
1s on the diagonal (upper triangular unipotent matrices). For each element w
in the Weyl group of S L(n, R) define

Uw := (w −1 · U · w) ∩ U, Ū w = (w−1 ·t U · w) ∩ U.

For example, if
⎛ ⎞
1 0 0 0
⎜0 0 0 1⎟
w=⎜
⎝0
⎟
1 0 0⎠
0 0 1 0
then
⎛ ⎞ ⎛ ⎞
1 ∗ ∗ ∗ 1
⎜ 1 ∗ ⎟ ⎜ 1 ∗⎟
Uw = ⎜
⎝
⎟,
⎠ Ū w = ⎜
⎝
⎟.
1 1 ∗⎠
1 1

We may think of Ū w as the group opposite or complementary to Uw in the upper

triangular unipotent matrices. These spaces have natural Lebesgue measures.
For example in the above situation we may write every element u ∈ Ū w in the
152 Maass forms and Whittaker functions for S L(n, Z)

⎛ ⎞
⎜
⎜
1 ⎟
⎟
⎜ ⎟
⎜ 1 u 2,4 ⎟⎟⎟ ∗
form u = ⎜
⎜
⎜
⎜
⎟, with u 2,4 , u 3,4 ∈ R. The natural measure, du , on
⎜
⎜
1 u 3,4 ⎟⎟⎟
⎝ ⎠
1
the space Ū w is du ∗ = du 2,4 du 3,4 .
Definition 5.10.1 For n ≥ 2, let z ∈ hn , ν = (ν1 , ν2 , . . . , νn−1 ) ∈ Cn−1 , and
let ψm be a character as in (5.5.1). Then the degenerate Whittaker function
associated to w is defined to be

Iν (wuz)ψm (u) d ∗ u,
Ū w

∗
where d u is the natural measure on Ū w .
Remark It may be shown that the degenerate Whittaker function can be
meromorphically continued and satisfies the same group of functional equations
as Jacquet’s Whittaker function as given in Theorem 5.9.8.
GL(n)pack functions The following GL(n)pack functions, described in the
appendix, relate to the material in this chapter:
FunctionalEquation IFun ModularGenerators
Wedge d Whittaker
WhittakerGamma WMatrix.
 6
Automorphic forms and L-functions for
S L(3, Z)

6.1 Whittaker functions and multiplicity one for SL(3, Z)

The generalized upper half-plane h3 was introduced in Example 1.2.4 and con-
sists of all matrices z = x · y with
⎛ ⎞ ⎛ ⎞
1 x1,2 x1,3 y1 y2 0 0
x = ⎝0 1 x2,3 ⎠ , y=⎝ 0 y1 0 ⎠ ,
0 0 1 0 0 1
where x1,2 , x1,3 , x2,3 ∈ R, y1 , y2 > 0. A basis for the ring D3 , of differen-
tial operators in ∂/∂ x1,2 , ∂/∂ x1,3 , ∂/∂ x2,3 , ∂/∂ y1 , ∂/∂ y2 which commute with
G L(3, R), can be computed by Proposition 2.3.3, and is given by (see also
(Bump, 1984)):
∂2 2 ∂
2
∂2  2 ∂2
1 = y12 + y − y1 y2 + y 2
x + y 2
∂ y12 2
∂ y22 ∂ y1 ∂ y2 1 1,2 2
∂ x1,3
2

(6.1.1)
∂2 2 ∂
2
∂2
+ y12 + y + 2y 2
x 1,2 ,
∂ x2,3
2 2
∂ x1,2
2 1
∂ x2,3 ∂ x1,3

∂3 ∂3 ∂3 ∂3
2 = −y12 y2 + y1 y2
2
− y 3 2
y
1 2 + y1 y2
2
∂ y12 ∂ y2 ∂ y1 ∂ y22 ∂ x1,3
2
∂ y1 ∂ x1,2
2
∂ y1
∂ 3  ∂3
− 2y12 y2 x1,2 + −x1,2 2
+ y22 y12 y2 2
∂ x2,3 ∂ x1,3 ∂ y2 ∂ x1,3 ∂ y2
∂3 ∂3 ∂3
− y12 y2 + 2y 2 2
y + 2y 2 2
y x 1,2
∂ x2,3
2
∂ y2 1 2
∂ x2,3 ∂ x1,2 ∂ x1,3 1 2
∂ x1,2 ∂ x1,3
2

∂2 ∂2 ∂2  2 ∂2
+ y12 2 − y22 2 + 2y12 x1,2 + x1,2 + y22 y12 2
∂ y1 ∂ y2 ∂ x2,3 ∂ x1,3 ∂ x1,3
∂2 2 ∂
2
+ y12 − y2 .
∂ x2,3
2
∂ x1,2
2

153
154 Automorphic forms and L-functions for S L(3, Z)

Let m = (m 1 , m 2 ) with m 1 , m 2 ∈ Z and let ν = (ν1 , ν2 ), with ν1 , ν2 ∈ C.

The Jacquet Whittaker function for S L(3, Z) was introduced in (5.5.1), and
takes the form
∞ ∞ ∞
WJacquet (z, ν, ψm ) = Iν (w3 · u · z)ψm (u) du 1,2 du 1,3 du 2,3 , (6.1.2)
−∞ −∞ −∞

where
⎛ ⎞ ⎛ ⎞
1 1 u 1,2 u 1,3
w3 = ⎝ −1 ⎠, u=⎝ 1 u 2,3 ⎠ ,
1 1

and

ψm (u) = e2πi (m 1 u 2,3 +m 2 u 1,2 ) , Iν (z) = y1ν1 +2ν2 y22ν1 +ν2 .

It was shown in Section 5.8, 5.9 that WJacquet (z, ν, ψ1,1 ) has meromorphic con-
tinuation to all ν1 , ν2 ∈ C and satisfies the functional equations:

∗ ∗
 
WJacquet (z, (ν1 , ν2 ), ψ1,1 ) = WJacquet z, ν1 + ν2 − 13 , 23 − ν2 , ψ1,1
∗
 
= WJacquet z, 23 − ν1 , ν1 + ν2 − 13 , , ψ1,1 ,

where
   
∗ 3ν1 3ν2
2 −3ν1 −3ν2
1
WJacquet (z, (ν1 , ν2 ), ψ1,1 ) =π  
2 2
 
3ν1 + 3ν2 − 1
× WJacquet (z, ν, ψ1,1 ).
2

Vinogradov and Takhtadzhyan (1982) and Stade (1990) have obtained the
following very explicit integral representation

∗
WJacquet (y, (ν1 , ν2 ), ψ1,1 )
∞  & 
1+(ν −ν )/2 1−(ν −ν )/2
= 4y1 1 2 y2 1 2 K 3ν1 +3ν2 −2 2π y2 1 + u −2
2
0
 &  3ν1 −3ν2 du
× K 3ν1 +3ν2 −2 2π y1 1 + u 2 u 2 . (6.1.3)
2 u
Using the above representation, or alternatively following Bump (1984), one
 6.1 Whittaker functions and multiplicity one for SL(3, Z) 155

may obtain the double Mellin transform pair

∞ ∞
dy1 dy2
W̃Jacquet (s, ν) := ∗
WJacquet (y, (ν1 , ν2 ), ψ1,1 ) y1s1 −1 y2s2 −1
y1 y2
0 0
π −s1 −s2
= · G(s1 , s2 ), (6.1.4)
4
∗
WJacquet (y, (ν1 , ν2 ), ψ1,1 )
 2+i∞
2+i∞ 
1
= G(s1 , s2 )(π y1 )1−s1 (π y2 )1−s2 ds1 ds2 ,
(4π 2 i)2
2−i∞ 2−i∞

where
 s1 +α  s1 +β  s1 +γ  s2 −α  s2 −β  s2 −γ
  2  2  2  2  2
G(s1 , s2 ) = 2
 , (6.1.5)
 s1 +s
2
2

and

α = −ν1 − 2ν2 + 1, β = −ν1 + ν2 , γ = 2ν1 + ν2 − 1.

The Gamma function,

 ∞
du
(s) = e−u u s ,
0 u
is uniquely characterized by its functional equation (s + 1) = s(s), growth
conditions, and the initial condition (1) = 1. This is the well-known Bohr–
Mollerup theorem (Conway, 1973). A simple proof of the Bohr–Mollerup
theorem can be obtained by assuming that if there exists another such function
F(s) then F(s)/(s) would have to be a periodic function. From the periodicity
and the growth conditions, one can conclude (Ahlfors, 1966) that F(s)/(s)
must be the constant function. Remarkably, this method of proof generalizes to
S L(3, Z) with periodic functions replaced by doubly periodic functions. The
following proof of multiplicity one was found by Diaconu and Goldfeld. It is
not clear if it can be generalized to S L(n, Z) with n > 3.

Theorem 6.1.6 (Multiplicity one) Fix ν = (ν1 , ν2 ) ∈ C2 . Let ν (z) be an

S L(3, Z) Whittaker function of type ν associated to a character ψ as in
Definition 5.4.1. Assume that ν (z) has sufficient decay in y1 , y2 so that
∞ ∞
dy1 dy2
y1σ1 y2σ2 |ν (y)|
y1 y2
0 0
156 Automorphic forms and L-functions for S L(3, Z)

converges for sufficiently large σ1 , σ2 . Then

ν (z) = c · WJacquet (z, ν, ψ),

for some constant c ∈ C.

Proof It is enough to prove the theorem for the case when ψ(x) = eπi (x2,3 +x1,2 )
since, in general, ν (z) = ν (y)ψ(x) for z = x y ∈ h3 . For s = (s1 , s2 ) ∈ C2 ,
consider the double Mellin transform
∞ ∞
dy1 dy2
˜ ν (s) =
 y1s1 y2s2 ν (y)
y1 y2
0 0

which is well defined for (s1 ), (s2 ) sufficiently large by the assumption in
our theorem. Define

Js (z) := ψ(x)y1s1 y2s2

for z = x y ∈ h3 .
Define the inner product, , , on L2 (U (Z)\h3 ) by

∞ ∞ 1 1 1
dy1 dy2
 f, g = f (z)g(z) d x1,2 d x1,3 d x2,3 ,
(y1 y2 )3
0 0 0 0 0

for all f, g ∈ L2 U (Z)\h3 . Taking f (z) = ν (z) and g(z) = Js̄ (z), it follows
that
∞ ∞
dy1 dy2
ν , Js̄  = ν (y) · y1s1 −2 y2s2 −2 ˜ ν (s ∗ )
=
y1 y2
0 0

where s ∗ = (s1∗ , s2∗ ) with s1∗ = s1 − 2, and s2∗ = s2 − 2.

Let D denote the polynomial ring over C of G L(3, R) invariant differen-
tial operators generated by 1 , 2 given in (6.1.1). Then since ν (z) is an
eigenfunction of every D ∈ D, we may write

Dν = λν (D)ν (6.1.7)

for some λν (D) ∈ R. Since D is a self-adjoint operator with respect to the above
inner product, it follows that

˜ ν (s ∗ ) = Dν , Js̄  = ν , D Js̄ .

λν (D) (6.1.8)
 6.1 Whittaker functions and multiplicity one for SL(3, Z) 157

Lemma 6.1.9 Let 1 , 2 ∈ D be given by (6.1.1). Let s = (s1 , s2 ) ∈ C2 . We

have


i Js (z) = c(s, σ, i )Js+σ (z), (i = 1, 2),
σ =(σ1 ,σ2 )∈S

where the sum ranges over the finite set S = {(0, 0), (0, 2), (2, 0)}, and
c(s, (0, 0), i ) = λs  (i ), (i = 1, 2),
c(s, (2, 0), 1 ) = c(s, (0, 2), 1 ) = −(2πi)2 ,
c(s, (2, 0), 2 ) = (2πi)2 (1 − s2 ),
c(s, (0, 2), 2 ) = (2πi)2 (s1 − 1),

where s  = 2s23−s1 , 2s13−s2 .
The proof of Lemma 6.1.9, first obtained by Friedberg and Goldfield
(1993), is given by a simple brute force computation which we omit. Note
that the map s → s  denotes the linear transformation of C2 such that
y1s1 y2s2 = Is  (z) with Iw (z) = y1w1 +2w2 y22w1 +w2 for all w = (w1 , w2 ) ∈ C2 .

It immediately follows from (6.1.8) and Lemma 6.1.9 that

˜ ν (s ∗ ) =
λν (i ) ˜ ν (s ∗ + σ ),
c(s, σ, i ) (6.1.10)
σ =(σ1 ,σ2 )∈S

for i = 1, 2, and where s ∗ = (s1 − 2, s2 − 2) as before. From (6.1.10), we

obtain
˜ ν (s ∗ ) = 4π 2 (
(λν (1 ) − λs  (1 )) ˜ ν (s ∗ + (2, 0)) +  ˜ ν (s ∗ + (0, 2)))
˜ ν (s ∗ ) = −4π 2 ((1 − s2 )
(λν (2 ) − λs  (2 )) ˜ ν (s ∗ + (2, 0))
˜ ν (s ∗ + (0, 2))).
+ (s1 − 1)
Consequently, 
˜ ν (s) must satisfy the shift equations
˜ ν (s) = A(s)
 ˜ ν (s + (2, 0)) = B(s)
˜ ν (s + (0, 2)),

for certain meromorphic functions A(s), B(s).

Stirling’s formula for the Gamma function (Whittaker and Watson, 1935)
tells us that
π π
|t|σ − 2 e− 2 |t|  |(σ + it)|  |t|σ − 2 e− 2 |t|
1 1
(6.1.11)
for σ, t ∈ R and |t| sufficiently large. If we combine (6.1.11) with the Mellin
transform (6.1.5), we obtain
π

|t1 |σ1 + 2 (α+β+γ −s2 −2) e− 2 |t1 |
1
W̃Jacquet ((s1 , s2 ), ν)  (6.1.12)
158 Automorphic forms and L-functions for S L(3, Z)

for s1 = σ1 + it1 , s2 = σ2 + it2 and with |t1 | → ∞ and s2 fixed. We also have
a similar estimate for s1 fixed and |t2 | → ∞.
Let us define the quotient function
˜ ν (s)
Fν (s) := .
W̃Jacquet (s, ν)
If we fix s2 and let (s1 ), (s2 ) be sufficiently large, then (6.1.12) implies that
π
Fν ((s1 , s2 ))  e 2 |t1 | (6.1.13)

as t1 → ∞. Similarly, for s1 fixed and (s1 ), (s2 ) sufficiently large, we have

π
Fν ((s1 , s2 ))  e 2 |t2 | (6.1.14)

as t2 → ∞.
Note that the shift equations imply that

Fν ((s1 , s2 )) = Fν ((s1 + 2, s2 )) = Fν ((s1 , s2 + 2)). (6.1.15)

Since Fν (s) is holomorphic for (s1 ), (s2 ) sufficiently large, it follows that
Fν (s) is entire. If we fix s2 and consider Fν ((s1 , s2 )) as a function of s1 , it is an
immediate consequence of (6.1.15) that Fν ((s1 , s2 )) will be periodic (of period
2) in s1 . Thus, for fixed s2 , Fν ((s1 , s2 )) will be a function of z 1 = eπis1 and will
have a Laurent expansion in the variable z 1 of the form


∞
Fν ((s1 , s2 )) = cn (s2 )z 1n ,
n=−∞

where the coefficients cn (s2 ) are entire and periodic of period 2. If we fix s2 ∈ C,

then
1

∞
|Fν ((s1 , s2 ))|2 ds1 = |cn (s2 )|2 e−2π nt1 ≥ |ck (s2 )|2 e−2π kt1
n=−∞
(s1 )=0

for every k = 0, ±1, ±2, . . .

Therefore, we have the Fourier expansion


∞

∞
Fν ((s1 , s2 )) = am,n eπims1 eπins2 .
m=−∞ n=−∞

The bound (6.1.13) implies that

1
|Fν ((s1 , s2 ))|2 ds1  eπ |t1 | ,
(s1 )=0
 6.2 Maass forms for SL(3, Z) 159

for |t1 | → ∞ and s2 fixed. This implies that ck (s2 ) = 0 for k = ±1, ±2, . . .
Similarly,
1
|Fν ((s1 , s2 ))|2 ds2 ≥ |a0, j |2 e−2π jt2
(s2 )=0

for j = 0, ±1, ±2, . . . We also have by (6.1.14) that

1
|Fν ((s1 , s2 ))|2 ds2  eπ|t2 | ,
(s2 )=0

for |t2 | → ∞ and s1 fixed. It follows that a0, j = 0 for j = ±1, ±2, . . . Thus
Fν ((s1 , s2 )) must be a constant. This completes the proof of Theorem 6.1.6.

6.2 Maass forms for SL(3, Z)

We want to study Maass forms for S L(3, Z). To this end, let us recall Theorem
5.3.2. If Un denotes the group of n × n upper triangular matrices with 1s on the
diagonal as in Section 5.2, and φ is a Maass form for S L(3, Z), then for all
z ∈ S L(3, Z)\h3


∞
∞   
γ
φ(z) = φ̃ (m 1 ,m 2 ) z ,
γ ∈U (Z)\S L(2,Z) m =1 m 2 =−∞
1
2 1
m 2 =0

where the sum is independent of the choice of coset representatives γ and

 1 1 1
φ̃ (m 1 ,m 2 ) (z) := φ(u · z) e−2πi(m 1 u 1 +m 2 u 2 ) d ∗ u,
0 0 0

with
⎛ ⎞ ⎛ ⎞
1 u 1,2 u 1,3 1 u2 u 1,3
u=⎝ 1 u 2,3 ⎠ = ⎝ 1 u 1 ⎠ ∈ U3 (R)
1 1
and d ∗ u = du 1 du 2 du 1,3 . Note that we have relabeled the super diagonal
elements u 1 = u 2,3 , u 2 = u 1,2 as in Proposition 5.5.2.
Now, we have shown that φ̃ (m 1 ,m 2 ) (z) is a Whittaker function. Further, φ̃ (m 1 ,m 2 )
will inherit the growth properties of the Maass form φ and will satisfy the
conditions of Theorem 6.1.6. The multiplicity one Theorem 6.1.6 tells us that
only the Jacquet Whittaker function (6.1.2) can occur in the Fourier expansion
of a Maass form for S L(3, Z), and that φ̃ (m 1 ,m 2 ) must be a constant multiple of the
160 Automorphic forms and L-functions for S L(3, Z)

Jacquet Whittaker function. It follows from Theorem 5.3.2 and Proposition 5.5.2
that if φ is a Maass form of type ν = (ν1 , ν2 ) ∈ C2 for S L(3, Z) then


∞

A(m 1 , m 2 )
φ(z) =
γ ∈U2 (Z)\S L(2,Z) m 1 =1 m 2 =0
|m 1 m 2 |
⎛⎛ ⎞ ⎞
|m 1 m 2 |  
×WJacquet ⎝⎝ m1 ⎠ γ z, ν, ψ1, |mm2 | ⎠ , (6.2.1)
1 2
1
⎛⎛ ⎞⎞
1 u2 u 1,3
where A(m 1 , m 2 ) ∈ C and ψ1 ,2 ⎝⎝ 1 u 1 ⎠⎠ = e2πi(1 u 1 +2 u 2 ) .
1
The particular normalization A(m 1 , m 2 )/|m 1 m 2 | is chosen so that later
formulae are as simple as possible.

Lemma 6.2.2 (Fourier coefficients are bounded) Let φ be a Maass form for
S L(3, Z) as in (6.2.1). Then for all integers m 1 ≥ 1, m 2 = 0,

A(m 1 , m 2 )
= O(1).
|m 1 m 2 |
⎛ ⎞⎛ ⎞
1 x2 x1,3 y1 y2
Proof Let z = ⎝ 1 x1 ⎠ ⎝ y1 ⎠ . A simple computation
1 1
shows that
⎛⎛ ⎞ ⎞
|m 1 m 2 |y1 y2
A(m 1 , m 2 )
· WJacquet ⎝⎝ m 1 y1 ⎠, ν, ψ1, m2 ⎠
|m 1 m 2 | |m | 2
1
1 1 1
= φ(z) e−2πi [m 1 x1 +m 2 x2 ] d x1 d x2 d x1,3 .
0 0 0

We choose y1 = |m 1 |−1 c1 , y2 = |m 2 |−2 c2 , where c1 , c2 are so chosen that

⎛⎛ ⎞ ⎞
c1 c2
WJacquet ⎝⎝ c1 ⎠ , ν, ψ1, m2 ⎠ = 0.
|m | 2
1

Since φ is bounded everywhere, this implies that

A(m 1 , m 2 ) = O |m 1 m 2 |) .

 6.3 The dual and symmetric Maass forms 161

6.3 The dual and symmetric Maass forms

Let φ(z) be a Maass form for S L(3, Z) as in (6.2.1). We shall now define,
φ̃(z), the dual Maass form associated to φ which plays an important role in
automorphic form theory.
Proposition 6.3.1 Let φ(z) be a Maass form of type (ν1 , ν2 ) ∈ C2 as in (6.2.1).
Then
⎛ ⎞
 t −1 1
φ̃(z) := φ w · (z ) · w , w=⎝ −1 ⎠,
1
is a Maass form of type (ν2 , ν1 ) for S L(3, Z). The Maass form φ̃ is called the
dual Maass form. If A(m 1 , m 2 ) is the (m 1 , m 2 )th Fourier coefficient of φ then
A(m 2 , m 1 ) is the corresponding Fourier coefficient of φ̃.
Proof First, for every γ ∈ S L(3, Z),
 
φ̃(γ z) = φ w · t ((γ z)−1 ) · w = φ γ  w · t (z −1 ) · w = φ̃(z)
since γ  = w · t (γ −1 ) · w ∈ S L(3, Z). Thus φ̃ satisfies the automorphic condi-
tion (1) of Definition 5.1.3 of a Maass form.
Next, note that if
⎛ ⎞ ⎛ ⎞
1 x2 x1,3 y1 y2
z=⎝ 1 x1 ⎠ · ⎝ y1 ⎠,
1 1
then
⎛ ⎞ ⎛ ⎞
1 x1 x1 x2 − x1,3 y1 y2
w · t (z −1 ) · w = ⎝ 1 x2 ⎠·⎝ y2 ⎠. (6.3.2)
1 1
It easily follows that
1 1 1 1
φ̃(z) d x2 d x1,3 = φ̃(z) d x1 d x1,3 = 0.
0 0 0 0

Thus φ̃ satisfies the cuspidality condition (3) of Definition 5.1.3.

Now

Iν1 ,ν2 (z) = y1ν1 +2ν2 y22ν1 +ν2 = Iν2 ,ν1 w · t (z −1 ) · w
since the involution z → w · t (z −1 ) · w interchanges y1 and y2 . It then follows
from (6.1.1), using the chain rule, that φ̃ is a Maass form of type (ν2 , ν1 ).
162 Automorphic forms and L-functions for S L(3, Z)

Finally, it follows from the identity (6.3.2) that if we integrate

1 1 1
φ̃(z) e−2πi [m 1 x1 +m 2 x2 ] d x1 d x2 d x1,3
0 0 0

to pick off the (m 1 , m 2 )th Fourier coefficient, then because x1 and x2 are
interchanged we will actually get A(m 2 , m 1 ).

In the S L(2, Z) theory, the notions of even and odd Maass forms
(see Section 3.9) played an important role. If a(n) is the nth Fourier coeffi-
cient of an S L(2, Z) Maass form then a(n) = ±a(−n) depending on whether
the Maass form is even or odd. We shall see that there is a quite different situa-
tion in the case of S L(3, Z) and that there are no odd Maass forms in this case.
The cognoscenti will recognize that there are no odd Maass forms on S L(3, Z)
because our definition of Maass form requires a trivial central character.
Consider a diagonal matrix δ of the form
⎛ ⎞
δ 1 δ2
δ := ⎝ δ1 ⎠
1
where δ1 , δ2 ∈ {+1, −1}. We define an operator Tδ which maps Maass forms
to Maass forms, and is given by
Tδ φ(z) := φ(δzδ).
Note that
⎛⎛ ⎞ ⎛ ⎞⎞
1 x2 x1,3 y1 y2
Tδ φ ⎝⎝ 1 x1 ⎠ · ⎝ y1 ⎠⎠
1 1
⎛⎛ ⎞ ⎛ ⎞⎞
1 x2 δ2 x1,3 δ1 δ2 y1 y2
= φ ⎝⎝ 1 x 1 δ1 ⎠ · ⎝ y1 ⎠⎠ . (6.3.3)
1 1
Clearly (Tδ )2 is the identity transformation, so the eigenvalues of Tδ can only
be ±1.
Definition 6.3.4 A Maass form φ of type ν = (ν1 , ν2 ) ∈ C2 for S L(3, Z) is
said to be symmetric if Tδ φ = ±φ for all Tδ as in (6.3.3).
We shall now show that every Maass form φ for S L(3, Z) is even, i.e.,
⎛ ⎞
δ1 δ2
Tδ φ = φ, for all Tδ = ⎝ δ1 ⎠ with δ1 , δ2 ∈ {+1, −1}. The reason is
1
 6.4 Hecke operators for SL(3, Z) 163

that a Maass form φ(z) is invariant under left multiplication by elements in

S L(3, Z), in particular by the elements:
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
−1 −1 1
⎝ −1 ⎠, ⎝ 1 ⎠, ⎝ −1 ⎠.
1 −1 −1
It is also invariant by the central element
⎛ ⎞
−1
⎝ −1 ⎠.
−1
Since these elements generate all the possible Tδ , this proves our assertion.
Proposition 6.3.5 Let φ be a Maass form of type ν = (ν1 , ν2 ) ∈ C2 for
S L(3, Z) with Fourier–Whittaker expansion



∞

A(m 1 , m 2 )
φ(z) =
γ ∈U2 (Z)\S L(2,Z) m 1 =1 m 2 =0
|m 1 m 2 |
⎛⎛ ⎞ ⎞
|m 1 m 2 |  
× WJacquet ⎝⎝ m1 ⎠ γ z, ν, ψ1, |mm2 | ⎠ ,
1 2
1
as in (6.2.1). Then for all m 1 ≥ 1 and m 2 = 0,
A(m 1 , m 2 ) = A(m 1 , −m 2 ).

⎛ ⎞
−1
Proof Let Tδ = ⎝ 1 ⎠. Then since δzδ transforms x2 → −x2 and
1
x1,3 → −x1,3 it easily follows that
 1 1 1
Tδ φ(z)e−2πim 1 x1 e−2πim 2 x2 d x1 d x2 d x1,3
0 0 0

picks off the A(m 1 , −m 2 ) coefficient of φ(z); and this equals A(m 1 , m 2 ) because
Tδ φ(z) = φ(z).

6.4 Hecke operators for SL(3, Z)

We recall the general definition of Hecke operators given in Definition 3.10.5.
Consider a group G that acts continuously on a topological space X . Let 
164 Automorphic forms and L-functions for S L(3, Z)

be a discrete subgroup of G. For every g in C G (), the commensurator of

 in G, (i.e., (g −1 g) ∩  has finite index in both  and g −1 g) we have a
decomposition of a double coset into disjoint right cosets of the form
(
g = αi . (6.4.1)
i

For each such g, the Hecke operator Tg : L2 (\X ) → L2 (\X ) is defined by

Tg f (x) = f (αi x),
i

where f ∈ L2 (\X ), x ∈ X, and αi are given by (6.4.1). The Hecke ring con-
sists of all formal sums


ck Tgk
k

with integer coefficients ck and gk in a semigroup  as in Definition 3.10.8.

Since two double cosets are either identical or totally disjoint, it follows that
unions of double cosets are associated to elements in the Hecke ring. Finally, we
recall Theorem 3.10.10 which states that the Hecke ring is commutative if there
exists an antiautomorphism g → g ∗ (i.e., (gh)∗ = h ∗ g ∗ ) for which  ∗ =  and
(g)∗ = g for every g ∈ .
We now specialize to the case where

G = G L(3, R),  = S L(3, Z), X = G L(3, R)/(O(3, R) · R× ) = h3 .

For every triple of positive integers m 0 , m 1 , m 2 , the matrix

⎛ ⎞
m0m1m2
⎝ m0m1 ⎠ ∈ C G (),
m0

the commensurator of  in G (defined in (3.10.2)). We define  to be the

semigroup generated by all such matrices. As in the case of S L(2, Z), we have
the antiautomorphism

g → t g, g ∈ ,

where t g denotes the transpose of the matrix g. It is again clear that the conditions
of Theorem 3.10.10 are satisfied so that the Hecke ring is commutative.
The following lemma is analogous to Lemma 3.12.1, which came up in the
S L(2, Z) situation.
 6.4 Hecke operators for SL(3, Z) 165

Lemma 6.4.2 Fix a positive integer n ≥ 1. Define the set

⎧⎛ ⎞  ⎫
⎨ a b1 c1  ⎬
 a,b,c ≥ 1
Sn := ⎝ 0 b c2 ⎠  .
⎩  abc = n
0 0 c  0 ≤ b1 < b, 0 ≤ c1 ,c2 < c ⎭

Then one has the disjoint partition

⎛ ⎞
( m0m1m2 (
⎝ m0m1 ⎠ = α. (6.4.3)
m 30 m 21 m 2 =n m0 α∈Sn

Proof First of all we claim the decomposition is disjoint. If not, there exists
⎛ ⎞
γ1,1 γ1,2 γ1,3
⎝ γ2,1 γ2,2 γ2,3 ⎠ ∈ 
γ3,1 γ3,2 γ3,3
such that
⎛ ⎞ ⎛ ⎞ ⎛  ⎞
γ1,1 γ1,2 γ1,3 a b1 c1 a b1 c1
⎝ γ2,1 γ2,2 γ2,3 ⎠ · ⎝ 0 b c2 ⎠ = ⎝ 0 b c2 ⎠ . (6.4.4)
γ3,1 γ3,2 γ3,3 0 0 c 0 0 c

This implies that γ2,1 = γ3,1 = γ3,2 = 0. Consequently, γ1,1 a = a  , γ2,2 b = b ,

and γ3,3 c = c . But γ1,1 γ2,2 γ3,3 = 1 and a, b, c, a  , b , c ≥ 1. It easily follows
that γ1,1 = γ2,2 = γ3,3 = 1. Note that the above shows that a  = a, b = b,
c = c. Therefore, (6.4.4) takes the form
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
1 γ1,2 γ1,3 a b1 c1 a b1 c1
⎝ 1 γ2,3 ⎠ · ⎝ 0 b c2 ⎠ = ⎝ 0 b c2 ⎠ .
1 0 0 c 0 0 c

Since 0 ≤ b1 , b1 < b and 0 ≤ c1 , c2 , c1 , c2 < c, one concludes that
γ1,2 = γ1,3 = γ2,3 = 0, and the decomposition is disjoint as claimed.
Now, by Theorem 3.11.2, every element on the right-hand side of (6.4.3) can
be put into Smith normal form, so must occur as an element on the left-hand
side of (6.4.3). Similarly, by Theorem 3.11.1, every element on the left-hand
side of (6.4.3) can be put into Hermite normal form, so must occur as an element
on the right-hand side of (6.4.3). This proves the equality of the two sides of
(6.4.3).

By analogy with the S L(2, Z) situation (see (3.12.3)), it follows that for
every integer n ≥ 1, we have a Hecke operator Tn acting on the space of square
166 Automorphic forms and L-functions for S L(3, Z)

integrable automorphic forms f (z) with z ∈ h3 . The action is given by the

formula
⎛⎛ ⎞ ⎞

a b1 c 1
1
Tn f (z) = f ⎝⎝ 0 b c2 ⎠ · z ⎠ . (6.4.5)
n
abc=n 0 0 c
0≤c1 ,c2 <c
0≤b1 <b

Note the normalizing factor of 1/n which was chosen to simplify later formulae.
Clearly, T1 is just the identity
 operator.

The C-vector space L2 \h3 has a natural inner product, denoted , , and
defined by

 f, g = f (z)g(z) d ∗ z,
\h3
⎛ ⎞ ⎛ ⎞
 1 x1,2 x1,3 y1 y2 0 0
for all f, g ∈ L2 \h3 , z = ⎝ 0 1 x2,3 ⎠ · ⎝ 0 y1 0 ⎠ ∈ h3 ,
0 0 1 0 0 1
and where
dy1 dy2
d ∗ z = d x1,2 d x1,3 d x2,3
(y1 y2 )3
denotes the left invariant measure given in Proposition 1.5.3.
In the case of S L(2, Z), we showed in Theorem 3.12.4 that the Hecke oper-
ators are self-adjoint with respect to the Petersson inner product. For S L(n, Z)
with n ≥ 3, it is no longer true that the Hecke operators are self-adjoint. What
happens is that the adjoint operator is again a Hecke operator and, therefore,
the Hecke operator commutes with its adjoint, which means that it is a normal
operator.
Theorem 6.4.6 (Hecke operators are normal operators) Consider the
Hecke operators Tn , (n = 1, 2, . . . ) defined in (6.4.5). Let Tn∗ be the adjoint
operator which satisfies
Tn f, g =  f, Tn∗ g
for all f, g ∈ L2 (\h3 ). Then Tn∗ is another Hecke operator which commutes
with Tn so that Tn is a normal operator. Explicitly, Tn∗ is associated to the
following union of double cosets:
⎛ 2 2 ⎞
( m0m1m2
⎝ m 20 m 1 m 2 ⎠ . (6.4.7)
m 30 m 21 m 2 =n 2
m0m1
 6.4 Hecke operators for SL(3, Z) 167

Proof It follows from (6.4.3), and also from the fact that transposition is an
antiautomorphism (as in the proof of Theorem 3.10.10), that

⎛ ⎞
( m0m1m2 ( (
⎝ m0m1 ⎠ = α = α. (6.4.8)
m 30 m 21 m 2 =n m0 α∈Sn α∈Sn

Since the action of the Hecke operator is independent of the choice of right
coset decomposition, we obtain


1
Tn f, g = f (αz) g(z) d ∗ z
n α∈Sn
\h3


1
= f (z) g(α −1 z) d ∗ z
n α∈Sn
\h3
⎛⎛ ⎞ ⎞

n
1
= f (z) g ⎝⎝ n ⎠α −1 z ⎠ d ∗ z, (6.4.9)
n α∈Sn
\h3 n

after making the change of variables z → α −1 z. Multiplying by the diagonal

⎛ ⎞
n
matrix ⎝ n ⎠ above does not change anything because g is well defined
n
on h3 . ⎛ ⎞
1
Now, it follows from (6.4.8) that for ω = ⎝ −1 ⎠, we have
1

⎛ ⎞−1
( ( m0m1m2
α −1 =  ·ω⎝ m0m1 ⎠ ω−1 · 
α∈Sn m 30 m 21 m 2 =n m0

⎛ ⎞
( m −1
0
= ·⎝ (m 0 m 1 )−1 ⎠ · .
m 30 m 21 m 2 =n −1
(m 0 m 1 m 2 )
(6.4.10)
168 Automorphic forms and L-functions for S L(3, Z)

Finally, if we multiply both sides of (6.4.10) by the diagonal matrix

⎛ ⎞
n
⎝ n ⎠, with n = m 30 m 21 m 2 , it follows that the adjoint Hecke operator
n
defined by (6.4.9) is, in fact, associated to the union of double cosets given in
(6.4.7). This completes the proof.

The Hecke operators commute with the differential operators 1 , 2 given

in (6.1.1) and they also commute with the operators Tδ given in (6.3.3). It fol-
lows by standard methods in functional analysis, that we may simultaneously
diagonalize the space L2 (S L(3, Z)\h3 ) by all these operators. We shall be inter-
ested in studying Maass forms which are eigenfunctions of the full Hecke ring
of all such operators. The following theorem is analogous to Theorem 3.12.8
which came up in the S L(2, Z) situation.

Theorem 6.4.11 (Multiplicativity of the Fourier coefficients) Consider

∞

A(m 1 , m 2 )
f (z) =
γ ∈U2 (Z)\S L(2,Z) m 1 =1 m 2 =0
|m 1 m 2 |
⎛⎛ ⎞ ⎞
|m 1 m 2 |  
× WJacquet ⎝⎝ m1 ⎠ γ z, ν, ψ1, |mm2 | ⎠ ,
1 2
1
a Maass form for S L(3, Z), as in (6.2.1). Assume that f is an eigenfunction of
the full Hecke ring. If A(1, 1) = 0, then f vanishes identically. Assume f = 0
and it is normalized so that A(1, 1) = 1. Then

Tn f = A(n, 1) · f, ∀ n = 1, 2, . . .

Furthermore, we have the following multiplicativity relations

A(m 1 m 1 , m 2 m 2 ) = A(m 1 , m 2 ) · A(m 1 , m 2 ), if (m 1 m 2 , m 1 m 2 ) = 1,

 
m 1 d0 m 2 d1
A(n, 1)A(m 1 , m 2 ) = A , ,
d1 d2
d0 d1 d2 =n
d1 |m 1
d2 |m 2

 
m 1 d2 m 2 d0
A(1, n)A(m 1 , m 2 ) = A , ,
d1 d2
d0 d1 d2 =n
d1 |m 1
d2 |m 2

m m2 
1
A(m 1 , 1)A(1, m 2 ) = A , .
d|(m 1 ,m 2 )
d d
 6.4 Hecke operators for SL(3, Z) 169

Proof Let z = x · y with

⎛ ⎞ ⎛ ⎞
1 x2 x1,3 y1 y2
x =⎝ 1 x1 ⎠ , y=⎝ y1 ⎠.
1 1

In view of Theorem 5.3.2 and (6.2.1), we may write (for m 1 ≥ 1, m 2 = 0)

1 1 1
f (z)e−2πi(m 1 x1 +m 2 x2 ) d x1,3 d x1 d x2
0 0 0
⎛⎛ ⎞ ⎞
|m 1 m 2 |y1 y2
A(m 1 , m 2 )
= · WJacquet ⎝⎝ m 1 y1 ⎠, ν, ψ1, m2 ⎠ .
|m 1 m 2 | |m | 2
1
(6.4.12)

If f is an eigenfunction of the Hecke operator Tn defined by (6.4.5), then we

have Tn f (z) = λn f (z) for some eigenvalue λn . We can compute λn directly
using a variation of (6.4.12). We begin by considering
n n n
1
Tn f (z)e−2πi(m 1 x1 +m 2 x2 ) d x1,3 d x1 d x2
n3
0 0 0
⎛⎛ ⎞ ⎞
|m 1 m 2 |y1 y2
A(m 1 , m 2 )
= λn · WJacquet ⎝⎝ m 1 y1 ⎠, ν, ψ1, m2 ⎠
|m 1 m 2 | |m | 2
1
⎛⎛ ⎞⎛ ⎞ ⎞
n n n a b1 c1 1 x2 x1,3
1


= 4 f ⎝⎝ 0 b c2 ⎠ ⎝ 0 1 x1 ⎠ · y ⎠
n abc=n 0≤c1 ,c2 <c
0≤b1 <b 0 0 0 0 0 c 0 0 1
× e−2πi(m 1 x1 +m 2 x2 ) d x1,3 d x1 d x2 . (6.4.13)

Next, if we let
⎛ ⎞⎛ ⎞ ⎛ ⎞⎛ ⎞
a b1 c1 1 x2 x1,3 1 α2 α1,3 a
⎝ 0 b c2 ⎠ ⎝ 0 1 x1 ⎠ = ⎝ 1 α1 ⎠ ⎝ b ⎠,
0 0 c 0 0 1 1 c

then we may solve for α1 , α2 , α1,3 , by considering

⎛ ⎞ ⎛ ⎞
a ax2 + b1 ax1,3 + b1 x1 + c1 a bα2 cα1,3
⎝ b bx1 + c2 ⎠ = ⎝ b cα1 ⎠ .
c c
170 Automorphic forms and L-functions for S L(3, Z)

It follows that the right-hand side of (6.4.13) can be expressed in the form
n n n
1


n 4 abc=n 0≤c1 ,c2 <c
0≤b1 <b 0 0 0
⎛⎛ ⎞ ⎞
ax2 + b1 ax1,3 + b1 x1 + c1
⎛ ⎞
⎜⎜1 b c ⎟ ay1 y2 ⎟
⎜⎜ ⎟ ⎟
× f⎜⎜
⎜⎜0 1
bx1 + c2 ⎟·⎝
⎟ by1 ⎠⎟
⎟
⎝⎝ c ⎠ c ⎠
0 0 1
× e−2πi(m 1 x1 +m 2 x2 ) d x1,3 d x2 d x1 ,

which after the elementary transformations

ax2 + b1 bx1 + c2 ax1,3 + b1 x1 + c1
x2 = , x1 = , 
x1,3 = ,
b c c
becomes
c2 b1 b1 x1 +c1
ab2 + a 2 c+ a 2 b+
  
1
cbc

c b c

n 4 abc=n a a b 0≤c1 ,c2 <c c2 b1 b1 x1 +c1

0≤b1 <b c b c
⎛⎛  ⎞ ⎛ ⎞⎞
1 x1,3 x2
ay1 y2
× f ⎝ ⎝ 0 x1 ⎠ · ⎝ 1 by1 ⎠⎠
0 1 0 c

m 1 c2 m 2 b1
  cx  bx 
2πi b + a
e−2πi m 1 b +m 2 a d x1,3 
d x2 d x1 .
1 2
×e

In view of the fact that the integrand above is periodic and does not change
under transformations of the form

x1 → x1 + ab2 , x2 → x2 + a 2 c, 

x1,3 
→ x1,3 + 1,

we immediately deduce that the above integral is the same as

⎛⎛ ⎞ ⎛ ⎞⎞
 ab2 a 2 c a 2 b 1 x2 x1,3 ay1 y2
1
c2 ⎝ ⎝ ⎠ ⎝ ⎠⎠
f 0 1 x1 · by1
n 4 abc=n a 2
0 0 0 0 0 1 c
⎛ ⎞
   
⎜
2πi
m 1 c2
+
m 2 b1
⎟ −2πi cx bx
m 1 b1 +m 2 a2
×⎝ e b a
⎠e d x1,3 d x2 d x1 . (6.4.14)
0≤c1 ,c2 <c
0≤b1 <b
 6.4 Hecke operators for SL(3, Z) 171

But the above integral vanishes unless b|m 1 c and a|m 2 b. Furthermore, in the
case that b|m 1 c and a|m 2 b, we have
 


m c m b
2πi 1b 2 + 2a 1 c2 b if b|m 1 , a|m 2 ,
e =
0≤c1 ,c2 <c 0 otherwise.
0≤b1 <b

Consequently, it follows that our triple integral (6.4.14) may be written in the
form:
1
c2
· ab2 · a 2 c · a 2 b · c2 b
n 4 abc=n a 2
b|m 1 , a|m 2
⎛⎛ ⎞ ⎛ ⎞⎞
1 1 1 1 x2 x1,3 ay1 y2
× f ⎝⎝ 0 1 x1 ⎠ · ⎝ by1 ⎠⎠
0 0 0 0 0 1 c
 
cx bx
−2πi m 1 b1 +m 2 a2
×e d x1,3 d x2 d x1 .
Here we have used the fact that if F : R → C is a periodic integrable function
satisfying F(x + 1) = F(x), then for any integer M ≥ 1, we have
 M  1
F(x) d x = M · F(x) d x.
0 0

Finally, the triple integral above can be evaluated with (6.4.12) and has the value
 ⎛⎛ ⎞ ⎞
|m 1 m 2 |y1 y2 c
A mb1 c , ma2 b
 m c m b  · WJacquet ⎝⎝ m 1 y1 c ⎠, ν, ψ1, m2 ⎠
 1 · 2  |m 2 |
b a c
 ⎛⎛ ⎞ ⎞
|m 1 m 2 |y1 y2
A mb1 c , ma2 b
=  m 1 c m 2 b  · WJacquet ⎝⎝ m 1 y1 ⎠, ν, ψ1, m2 ⎠ ,
 · a  |m 2 |
b 1
from which it follows from (6.4.13) and (6.4.14) that
⎛⎛ ⎞ ⎞
|m 1 m 2 |y1 y2
A(m 1 , m 2 )
λn · WJacquet ⎝⎝ m 1 y1 ⎠, ν, ψ1, m2 ⎠
|m 1 m 2 | |m | 2
1

1
c2 A mb1 c , ma2 b
= 4 · ab · a c · a b · c b ·  m 1 c m 2 b 
2 2 2 2
n a2  · a 
abc=n b
b|m 1 , a|m 2
⎛⎛ ⎞ ⎞
|m 1 m 2 |y1 y2
× WJacquet ⎝⎝ m 1 y1 ⎠, ν, ψ1, m2 ⎠ .
|m | 2
1
172 Automorphic forms and L-functions for S L(3, Z)

If we cancel the Whittaker functions on both sides of the above identity and
simplify the expressions, we obtain

 
m1c m2b
λn A(m 1 , m 2 ) = A , . (6.4.15)
abc=n
b a
b|m 1 , a|m 2

We now explore the consequences of the assumption that A(1, 1) = 0. It fol-

lows easily from (6.4.15) that A(n, 1) = 0 for all integers n, and then the left-
hand side of (6.4.15) vanishes for all n, m 1 as long as m 2 = 1. By choosing
m 2 = 1, m 1 = p, n = p one obtains A(1, p) = 0. Arguing inductively, we may
choose m 2 = 1, m 1 = p, n = p for = 1, 2, . . . from which one can conclude
that A( p , p) = 0 for all = 0, 1, 2, . . . One then obtains that the left-hand side
of (6.4.15) vanishes as long as m 1 = p. One can continue in the same manner
to show that A( pi , p j ) = 0 for all non-negative integers i, j. One may then
proceed to products of two primes, products of three primes, etc. to eventually
obtain that if A(1, 1) = 0 then all coefficients A(m, n) must vanish.
If f = 0 then we may assume it is normalized so that A(1, 1) = 1. If we now
choose m 1 = m 2 = 1, it immediately follows from (6.4.15) that λn = A(n, 1).
Substituting this into (6.4.15) proves the identity

 
m1c m2b
A(n, 1)A(m 1 , m 2 ) = A , .
abc=n
b a
b|m 1 , a|m 2

The rest of the proof of Theorem 6.4.11 follows easily.

6.5 The Godement–Jacquet L-function

Let



∞

A(m 1 , m 2 )
f (z) =
γ ∈U2 (Z)\S L(2,Z) m 1 =1 m 2 =0
|m 1 m 2 |
⎛⎛ ⎞ ⎞
|m 1 m 2 |  
⎝ ⎝ ⎠ γ
× WJacquet m1 z, ν, ψ1, |mm2 | ⎠ ,
1 2
1
be a non-zero Maass form for S L(3, Z), normalized so that A(1, 1) = 1, which
is a simultaneous eigenfunction of all the Hecke operators as in Theorem 6.4.11.
We want to build an L-function out of the Fourier coefficients of f . Lemma 6.2.2
tells us that we may form absolutely convergent Dirichlet series in a suitable
half-plane.
 6.5 The Godement–Jacquet L-function 173

The sum over γ ∈ U2 (Z)\S L(2, Z) in the Fourier expansion of f

creates seemingly insurmountable complications, and it is not possible to simply
set
x1,2 = x1,3 = x2,3 = 0,
and then take the double Mellin transform in y1 , y2 which would be the analogue
of what we did to create L-functions in the S L(2, Z) situation. The ingenious
construction of the L-functions and the proof of their functional equations was
first obtained by Godement and Jacquet (1972) and is based on Tate’s thesis
(Tate, 1950). The original construction of Godement and Jacquet did not use
Whittaker models. We follow here a different method of construction as in
(Jacquet and Piatetski-Shapiro and Shalika, 1979). The Godement–Jacquet L-
function is also commonly referred to as the standard L-function.
By Theorem 6.4.11, the Fourier coefficients, A(m 1 , m 2 ), of f must satisfy
the multiplicativity relations
A(m 1 m 1 , m 2 m 2 ) = A(m 1 , m 2 ) · A(m 1 , m 2 ), if (m 1 m 2 , m 1 m 2 ) = 1,

 
m 1 d2 m 2 d0
A(1, n)A(m 1 , m 2 ) = A , ,
d d d =n
d1 d2
0 1 2
d1 |m 1
d2 |m 2

m m2 
1
A(m 1 , 1)A(1, m 2 ) = A , .
d|(m 1 ,m 2 )
d d

It follows that

 
p pk
A( p, 1)A(1, p ) =k
A , = A(1, p k−1 ) + A( p, p k )
d| p
d d

 p k+1 d0

A(1, p)A(1, p ) =
k+1
A d2 , = A(1, p k+2 ) + A( p, p k ),
d0 d2 = p
d2
d2 | p k+1

with the understanding that A(1, p −1 ) = 0.

Therefore,
A( p, 1)A(1, p k ) − A(1, p)A(1, p k+1 ) A(1, p k−1 ) − A(1, p k+2 )
= .
p ks p ks
(6.5.1)
If we define

∞
A(1, p k )
φ p (s) := ,
k=0
p ks
174 Automorphic forms and L-functions for S L(3, Z)

then, after summing over k, equation (6.5.1) implies that

A( p, 1) · φ p (s) − A(1, p) · [φ p (s) p s − p s ]
= φ p (s) p −s − [φ p (s) p 2s − p 2s − A(1, p) p s ].
Multiplying through by p −2s in the above and solving for φ p (s) yields
 −1
φ p (s) = 1 − A(1, p) p −s + A( p, 1) p −2s − p −3s .
In a manner completely analogous to the situation of S L(2, Z), as in
Definition 3.13.3, it is natural to make the following definition.
Definition 6.5.2 Let s ∈ C with (s) > 2, and let f (z) be a Maass form for
S L(3, Z) as in Theorem 6.4.11. We define the Godement–Jacquet L-function
L f (s) (termed the L-function associated to f ) by the absolutely convergent
series

∞  −1
L f (s) = A(1, n)n −s = 1 − A(1, p) p −s + A( p, 1) p −2s − p −3s .
n=1 p

Remark It is clear that the L-function associated to the dual Maass form f˜
takes the form


∞  −1
L f˜ (s) = A(n, 1)n −s = 1 − A( p, 1) p −s + A(1, p) p −2s − p −3s .
n=1 p

By analogy with the G L(2) situation, we would like to construct the L-

function L f (s) as a Mellin transform of the Maass form f . Before taking
the Mellin transform, it is necessary to kill the sum over G L(2) in the Fourier–
Whittaker expansion (6.2.1). The procedure to do this uses an auxilliary integral
which requires some preliminary preparation.
Set
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
|m 2 m 1 | 1 x1,2 x1,3 y1 y2
M= ⎝ |m 1 | ⎠, x = ⎝ ⎠
1 x2,3 , y = ⎝ y1 ⎠.
1 1 1
A simple computation gives
⎛ ⎞
1 |m 2 |x1,2 |m 1 m 2 |x1,3
M·x =⎝ 1 |m 1 |x2,3 ⎠ · M.
1
To simplify the subsequent notation, it is very convenient to set
x1 := x2,3 , x2 := x1,2 ,
 6.5 The Godement–Jacquet L-function 175

so that the super diagonal elements of the matrix x are x1 , x2 . We also define

z 1 := x1 + i y1 , z 2 := x2 + i y2 .

It follows from Definition 5.4.1 (1), that that for any integers 1 , 2 , the Jacquet
Whittaker function satisfies

WJacquet (M z, ν, ψ1 ,2 )

⎛ ⎛ ⎞ ⎞
y1 y2
= e2πi [|m 1 |1 x1 +|m 2 |2 x2 ] · WJacquet ⎝ M ⎝ y1 ⎠, ν, ψ1 ,2 ⎠ .
1
(6.5.3)
 
a b
Further, for any S L(2, Z) matrix , we may put the G L(3) matrix,
c d
⎛ ⎞⎛ ⎞
a b y1 y2 y1 x2 x1,3
⎝c d ⎠⎝ y1 x1 ⎠ into Iwasawa form:
1 1
⎛ ⎞⎛ ⎞ ⎛      ⎞
a b y1 y2 y1 x2 x1,3 y1 y2 y1 x2 x1,3
⎝c d ⎠⎝ y1 x1 ⎠ ≡ ⎝ y1 x1 ⎠ (mod Z 3 O(3, R)),
1 1 1
(6.5.4)

where
az 2 + b
z 2 = x2 + i y2 = , y1 = |cz 2 + d|y1 ,
cz 2 + d
x1 = cx1,3 + d x1 , 
x1,3 = ax1,3 + bx1 .

It immediately follows from (6.5.3) and (6.5.4) that

⎛ ⎛ ⎞ ⎞
a b
⎝ ⎝
WJacquet M c d ⎠ z, ν, ψ1 ,2 ⎠
1
6  7
az +b
2πi |m 1 |1 (cx1,3 +d x1 ) + |m 2 |2  cz 2+d
=e 2

⎛ ⎛ ⎞ ⎞
y1 y2
|cz 2 +d|
⎜ ⎜ ⎟ ⎟
× WJacquet ⎝ M ⎝ y1 · |cz 2 + d| ⎠, ν, ψ1 ,2 ⎠ . (6.5.5)
1
176 Automorphic forms and L-functions for S L(3, Z)

Lemma 6.5.6 For all ν ∈ C2 , 1 , 2 ∈ {±1}, and any matrix

⎛ ⎞
y1 y2
y=⎝ y1 ⎠ with y1 , y2 > 0, we have
1
WJacquet (y, ν, ψ1 ,2 ) = WJacquet (y, ν, ψ1,1 ).

Proof Recall the definition

⎛⎛ ⎞⎛ ⎞ ⎞
∞ ∞ ∞ 1 1 u2 u 1,3
WJacquet (y, ν, ψ1 ,2 ) = Iν ⎝⎝ −1 ⎠⎝ 1 u1 ⎠ y⎠
−∞ −∞ −∞ 1 1
−2πi(1 u 1 +2 u 2 )
×e du 1 du 2 du 1,3
⎛ ' ⎞ν1 +2ν2
∞ ∞ ∞ y (u − u u )2 + u 2 + y 2 y 2
1 1,3 1 2 1 1 2
= ⎝  ⎠
u 21,3 + y12 u 22 + y22
−∞ −∞ −∞
⎛ '  ⎞2ν1 +ν2
y2 u 21,3 + y12 u 22 + y22
×⎝  ⎠
(u 1,3 − u 1 u 2 )2 + u 21 + y12 y22

× e−2πi(1 u 1 +2 u 2 ) du 1 du 2 du 1,3 .

To complete the proof of the lemma, we simply make the transformation
u 1 → 1 u 1 , u 2 → 2 u 2 , u 1,3 → 1 2 u 1,3 .

Finally, we obtain the following theorem, which is the basis for the
construction of the L-function L f (s) (given in Definition 6.5.2) as a Mellin
transform.
Theorem 6.5.7 Let f (z) be a Maass form of type ν for S L(3, Z) as in (6.2.1).
Then we have the representation
6  7


∞

A(m 1 , m 2 ) 2πi m 1 (cx1,3 +d x1 )+m 2  azcz 2+d
+b
f (z) = ·e 2

 m 1 =1 m 2 =0
|m 1 m 2 |
ab ∈ U2 (Z)\S L(2,Z)
cd
⎛⎛ ⎞⎛ y1 y2 ⎞ ⎞
|m 1 m 2 | |cz 2 +d|
⎜ ⎠⎜ ⎟ ⎟
× WJacquet ⎝⎝ m1 ⎝ y1 · |cz 2 + d| ⎠, ν, ψ1,1⎠.
1 1
 6.5 The Godement–Jacquet L-function 177

Proof The proof follows from Theorem 6.4.11, (6.5.5) and Lemma 6.5.6.

Corollary 6.5.8 Let f (z) be a Maass form of type ν for S L(3, Z) as in (6.2.1).
Then
⎛⎛ ⎞ ⎞
1 1 1 u3
f ⎝⎝ 1 u 1 ⎠ · z ⎠ e−2πiu 1 du 1 du 3
0 0 1
⎛⎛ ⎞ ⎞

A(1, m 2 ) |m 2 |y1 y2
= e2πi(x1 +m 2 x2 ) · WJacquet ⎝⎝ y1 ⎠, ν, ψ1,1 ⎠ .
m 2 =0
|m 2 |
1

Proof By Theorem 6.5.7, we see that

⎛⎛ ⎞ ⎞
1 1 1 u3
f ⎝⎝ 1 u 1 ⎠ · z ⎠ e−2πiu 1 du 1 du 3
0 0 1


∞

A(m 1 , m 2 )
=

 m 1 =1 m 2 =0
|m 1 m 2 |
ab ∈U2 (Z)\S L(2,Z)
cd
1 1 3 az +b
4
2πi m 1 (c(u 3 +x1,3 )+d(u 1 +x1 ))+m 2  cz 2+d
× e 2 e−2πiu 1 du 1 du 3
0 0
⎛⎛ ⎞⎛ y1 y2 ⎞ ⎞
|m 1 m 2 | |cz 2 +d|
⎜⎝ ⎠⎜ ⎟ ⎟
× WJacquet ⎝ m1 ⎝ y1 · |cz 2 + d| ⎠, ν, ψ1,1⎠.
1 1

The integrals

1 1
e 2πim 1 cu 3
du 3 , e2πi(m 1 d−1)u 1 du 1 ,
0 0

vanish unless c = 0 and m 1 d = 1, in which case they take the value 1. The
proof of Corollary 6.5.8 follows immediately from this.

Our next objective is to construct L-functions associated to a Maass form

and show that they satisfy functional equations. In view of the S L(3, Z) Hecke
theory we have shown that the natural definition of the L-function associated
178 Automorphic forms and L-functions for S L(3, Z)

to a Maass form f is given by Definition 6.5.2:

∞  −1
L f (s) = A(1, n)n −s = 1 − A(1, p) p −s + A( p, 1) p −2s − p −3s .
n=1 p

This L-function appears naturally in a Mellin transform applied to the integral

in Corollary 6.5.8. It, therefore, seems prudent to consider
⎛⎛ ⎞ ⎛ ⎞⎞
∞∞1 1 1 u3 y1 y2
dy1 dy2
f ⎝⎝ 1 u 1 ⎠ · ⎝ y1 ⎠⎠ e−2πiu 1 du 1 du 3 y1s1 y2s2 ,
y1 y2
0 0 0 0 1 1
*1*1
where the inner double integral 0 0 has the sole function of picking off the
Fourier coefficients A(m 1 , m 2 ) of f with m 1 = 1. This is the analogue of the
Mellin transform
 ∞  
y dy
f ys
0 1 y

which occurs in the S L(2, Z) theory in Section 3.13. The functional equa-
tion in  Z) case arises from the symmetry f (z) = f (ω · z), where
 the S L(2,
−1
ω= .
1
One is, thus, highly motivated to try to generalize this idea to S L(3, Z) by
considering symmetries f (z) = f (ω · z) where ω is in the Weyl group. Curi-
⎛ ⎞
−1
ously, the choice ω = ⎝ 1 ⎠ does not work in the S L(3, Z) situation,
1
⎛ ⎞
1
but fortunately the choice ω = ⎝ 1 ⎠ does.
1
We shall now prove Lemma 6.5.9 which contains the symmetry required
to obtain the functional equation of the Godement–Jacquet L-function. A new
feature which does not appear in the G L(2) theory is the unbalanced nature of
this symmetry. One side has a double integral, while the other side has a triple
integral!
For the following Lemma 6.5.9, we define for any function f : h3 → C, its
dual function
⎛ ⎞
 t −1 1
f˜(z) = f w (z )w , w=⎝ −1 ⎠,
1
 6.5 The Godement–Jacquet L-function 179

as in Section 6.3. We also introduce

⎛ ⎞ ⎛ ⎞
1 1
w1 = ⎝ 1 ⎠, w1 = ⎝ −1 ⎠,
1 −1

and
⎛ ⎞ ⎛ ⎞
1 1
z∗ = ⎝ −1 ⎠ · t (z −1 ) · ⎝ 1⎠.
1 −1
⎛ ⎞ ⎛ −1 ⎞
y1 y2 x3 y2
If z = ⎝ y1 x1 ⎠ , then z ∗ = y1−1 ⎝ x3 y2−1 y1 x1 ⎠ .
1 1

Lemma 6.5.9 Let f : h3 → C be such that f˜(z) has a Fourier–Whittaker

expansion of type (6.2.1). Then for any z ∈ h3 , we have the identity
⎛ ⎛ ⎞ ⎞
1 1 1 u3
f ⎝w1 · ⎝ 1 u 1 ⎠ · z ⎠ e−2πiu 1 du 1 du 3
0 0 1
⎛ ⎛ ⎞ ⎞
1 1 1 u 1 −u 3
= f˜ ⎝w1 · ⎝ 1 ⎠ · w t (z −1 )w ⎠ e−2πiu 1 du 1 du 3
0 0 1
⎛ ⎛ ⎞ ⎛ ⎞ ⎞
∞ 1 1 1 u3 1
= f˜ ⎝⎝ 1 u1 ⎠ ⎝ u 1 ⎠ · z ∗ ⎠ e−2πiu 1 du 1 du 3 du.
−∞ 0 0 1 1

Proof By assumption

f (z) = f˜ w · t (z −1 ) · w .

Consequently
⎛ ⎛ ⎞ ⎞ ⎛ ⎛ ⎞ ⎞
1 u3 1 u1 −u 3
⎝
f w1 ⎝ ⎠ ⎠ ⎝
1 u 1 · z = f w1
˜  ⎝
1 ⎠ · w · t (z −1 ) · w ⎠ .
1 1
(6.5.10)

Integrating both sides of (6.5.10) with respect to u 1 , u 3 gives the first identity
in Lemma 6.5.9.
180 Automorphic forms and L-functions for S L(3, Z)

It follows from (6.5.10) that

⎛ ⎛ ⎞ ⎞
1 u3
f ⎝w1 ⎝ 1 u1 ⎠ · z⎠
1
⎛ ⎛ ⎞ ⎞
1 u 1 −u 3
= f˜ ⎝w1 ⎝ 1 ⎠ w1 −1 · w1 · w · t (z −1 ) · ww1 −1 ⎠
1
⎛⎛ ⎞ ⎞
1
−1
= f˜ ⎝⎝ u 3 1 u 1 ⎠ · w1 w · t (z −1 ) · ww1 ⎠ . (6.5.11)
1

⎛ ⎞
1
Note that w1 w = ⎝ −1 ⎠. Recall that z ∗ = w1 w · t (z −1 ) · ww1 −1 . If
1
we integrate both sides of (6.5.11) then

⎛⎛ ⎞ ⎞
1 1 1 u3
f ⎝⎝ 1 u 1 ⎠ · z ⎠ e−2πiu 1 du 1 du 3
0 0 1
⎛ ⎛ ⎞⎛ ⎞ ⎞
1 1 1 1
= f˜ ⎝⎝ 1 u1 ⎠ ⎝ u3 1 ⎠ · z ∗ ⎠ e−2πiu 1 du 1 du 3 .
0 0 1 1

Finally, the proof may be completed by applying the following Lemma 6.5.12
to the integral above.

Lemma 6.5.12 Assume f : h3 → C has a Fourier–Whittaker expansion of

type (6.2.1). Then for any z ∈ h3 , we have the identity

⎛⎛ ⎞⎛⎞ ⎞
1 1 1 1
f ⎝⎝ 1 u1 ⎠ ⎝ u 1 ⎠ · z ⎠ e−2πiu 1 du 1 du
0 0 1 1
⎛⎛ ⎞⎛ ⎞ ⎞
∞ 1 1 1 u3 1
= f ⎝⎝ 1 u1 ⎠ ⎝ u 1 ⎠ · z ⎠ e−2πiu 1 du 1 du 3 du.
−∞ 0 0 1 1
 6.5 The Godement–Jacquet L-function 181

Proof For any integer m note that since f is automorphic for S L(3, Z), we
have
⎛⎛ ⎞⎛ ⎞ ⎞
1 1 1 u3 1
f ⎝⎝ 1 u1 ⎠ ⎝ m 1 ⎠ · z ⎠ e−2πiu 1 du 1 du 3
0 0 1 1
⎛ ⎛ ⎞ ⎛ ⎞ ⎞
1 1 1 1 u3
= f ⎝⎝ m 1 ⎠⎝ 1 u 1 − mu 3 ⎠ · z ⎠ e−2πiu 1 du 1 du 3
0 0 1 1
⎛⎛ ⎞ ⎞
 
1 1 1 u3
= f ⎝⎝ 1 u 1 ⎠ · z ⎠ e−2πiu 1 e−2πimu 3 du 1 du 3 . (6.5.13)
0 0 1

Furthermore, by Fourier theory

⎛⎛ ⎞ ⎞

1 1 1 u3
f (z) = f ⎝⎝ 1 u 1 ⎠ · z ⎠ e−2πim 1 u 1 e−2πim 3 u 3 du 1 du 3 ,
m 1 ,m 3 ∈Z 1
0 0

which implies that

⎛⎛ ⎞ ⎞
1 1
f ⎝⎝ 1 ξ1 ⎠ · z ⎠ e−2πiξ1 dξ1
0 1
⎛⎛ ⎞ ⎞

  
1 1 1 1 u3
= f ⎝⎝ 1 u 1 + ξ1 ⎠ · z ⎠
m 1 ,m 3 ∈Z 1
0 0 0

×e−2πim 1 u 1 e−2πim 3 u 3 e−2πiξ1 du 1 du 3 dξ1

⎛⎛ ⎞ ⎞

  ⎜⎜ 1
1 1 u3
⎟ ⎟
= f ⎝⎝ 1 u 1 ⎠ · z ⎠ e−2πiu 1 e−2πim 3 u 3 du 1 du 3 . (6.5.14)
m 3 ∈Z 1
0 0

Changing ξ1 to u 1 in (6.5.14) and combining with (6.5.13) we obtain

⎛⎛ ⎞ ⎞
1 1
f ⎝⎝ u 1 ⎠ · z ⎠ e−2πiu 1 du 1
1
0 1
⎛⎛ ⎞⎛ ⎞ ⎞

 
1 1 1 u3 1
= f ⎝ ⎝ 1 u1 ⎠ ⎝ m 1 ⎠ · z ⎠ e−2πiu 1 du 1 du 3 .
m∈Z
0 0 1 1
182 Automorphic forms and L-functions for S L(3, Z)

⎛ ⎞
1
Replacing z by ⎝ u 1 ⎠ · z and then integrating over u in the above identity
1
yields
⎛⎛ ⎞ ⎛ ⎞ ⎞
1 1 1 1
f ⎝⎝ 1 u1 ⎠ · ⎝ u 1 ⎠ z ⎠ e−2πiu 1 du 1 du
0 0 1 1
⎛⎛ ⎞ ⎛ ⎞ ⎞

  
1 1 1 1 u3 1
= f ⎝⎝ 1 u 1 ⎠ · ⎝ u + m 1 ⎠z ⎠ e−2πiu 1 du 1 du 3 du
m∈Z
0 0 0 1 1
⎛ ⎛ ⎞ ⎛ ⎞ ⎞
∞ 1 1 1 u3 1
= f ⎝⎝ 1 u1 ⎠ · ⎝ u 1 ⎠ z ⎠ e−2πiu 1 du 1 du 3 du.
−∞ 0 0 1 1

Theorem 6.5.15 Let f be a Maass form of type ν = (ν1 , ν2 ) for S L(3, Z)

with dual f˜ as in Proposition 6.3.1. Then L f (s)(respectively L f˜ (s)) (given in
Definition 6.5.2) have a holomorphic continuation to all s ∈ C and satisfy the
functional equation

G ν (s)L f (s) = G̃ ν (1 − s)L f˜ (1 − s),

where
     
−3s/2 s + 1 − 2ν1 − ν2 s + ν 1 − ν2 s − 1 + ν1 + 2ν2
G ν (s) = π    ,
2 2 2
     
−3s/2 s + 1 − ν1 − 2ν2 s − ν1 + ν2 s − 1 + 2ν1 + ν2
G̃ ν (s) = π    .
2 2 2

Proof An indirect proof of Theorem 6.5.15 was obtained by Bump (1984)

who showed that the functional equation of a Maass form must be the same as
that of an Eisenstein series which can be easily derived. We shall follow this
method later in this book for the case of S L(n, R) with n ≥ 3.
A direct proof is much more difficult. We present a proof of Hoffstein and
Murty (1989) which makes use of a double Mellin transform. This is different
than the proof of Jacquet and Piatetski-Shapiro and Shalika (1979) which uti-
⎛ ⎞
y1 y2
lizes a single Mellin transform. Accordingly, let us set z = ⎝ y1 ⎠.
1
 6.5 The Godement–Jacquet L-function 183

Since we are assuming that f is a Maass form, then we know that f (w1 z) =
f (z), and Lemma 6.5.9 takes the form
⎛⎛ ⎞⎞
1 1 y1 y2 u3
f ⎝⎝ y1 u 1 ⎠⎠ e−2πiu 1 du 1 du 3
0 0 1
⎛ ⎛ ⎞ ⎛ −1 ⎞⎞
∞ 1 1 1 u3 y2
= f˜ ⎝⎝ 1 u 1 ⎠ ⎝ uy2−1 y1 ⎠⎠ e−2πiu 1 du 1 du 3 du,
−∞ 0 0 1 1
(6.5.16)

The left-hand side of (6.5.16) can be evaluated by Corollary 6.5.8. We have

⎛⎛ ⎞⎞
1 1 y1 y2 u3
f ⎝⎝ y1 u 1 ⎠⎠ e−2πiu 1 du 1 du 3
0 0 1
⎛⎛ ⎞ ⎞

A(1, m 2 ) |m 2 |y1 y2
= · WJacquet ⎝ ⎝ y1 ⎠, (ν1 , ν2 ), ψ1,1 ⎠ .
m 2 =0
|m 2 |
1
(6.5.17)

In a similar manner, the right-hand side of (6.5.16) can be evaluated to give

⎛⎛ ⎞⎞
1 1 y1 y2 u3
f ⎝⎝ y1 u 1 ⎠⎠ e−2πiu 1 du 1 du 3
0 0 1
⎛⎛ ⎞ ⎞

A(m 1 , 1) 
∞ |m 1 |y2−1
⎜⎜ ⎟ ⎟
= WJacquet ⎝⎝ uy2−1 y1 ⎠ , (ν2 , ν1 ), ψ1,1 ⎠ du
m 1 =0
|m 1 |
−∞ 1
⎛⎛ ⎞ ⎞

A(m 1 , 1) 
∞ |m 1 |y2−1
= y2 · WJacquet ⎝⎝ u y1 ⎠, (ν2 , ν1 ), ψ1,1⎠ du.
m 1 =0
|m 1 |
−∞ 1
(6.5.18)

Note that the right-hand sides of (6.5.17) and (6.5.18) are identical. Then
the double Mellin transforms in y1 , y2 of the right-hand sides of (6.5.17) and
(6.5.18) must be the same. For (s1 ), (s2 ) sufficiently large, the double Mellin
184 Automorphic forms and L-functions for S L(3, Z)

transform of the right-hand side of (6.5.17) converges absolutely and is:

⎛⎛ ⎞⎞
∞ ∞ 1 1 y1 y2 u3
dy1 dy2
f ⎝⎝ y1 u 1 ⎠⎠ e−2πiu 1 y1s1 −1 y2s2 −1 du 1 du 3
y1 y2
0 0 0 0 1
⎛⎛ ⎞ ⎞
∞ ∞ y1 y2
= 2L f (s2 ) · WJacquet ⎝⎝ y1 ⎠, (ν1 , ν2 ), ψ1,1 ⎠
0 0 1
dy1 dy2
× y1s1 −1 y2s2 −1 . (6.5.19)
y1 y2

Similarly, for −(s1 ), −(s2 ) sufficiently large, the double Mellin transform
of the right-hand side of (6.5.18) converges absolutely and equals:
⎛⎛ ⎞⎞
∞ ∞ 1 1 y1 y2 u3
dy1 dy2
f ⎝⎝ y1 u 1 ⎠⎠ e−2πiu 1 y1s1 −1 y2s2 −1 du 1 du 3
y1 y2
0 0 0 0 1
⎛ ⎛⎛ −1 ⎞ ⎞ ⎞
∞ ∞ ∞ y2
= 2L f˜ (1 − s2 ) · ⎝ WJacquet ⎝⎝ u y1 ⎠, (ν2 , ν1 ), ψ1,1 ⎠ du ⎠
0 0 −∞ 1
dy1 dy2
× y1s1 −1 y2s2
y1 y2
⎛ ∞ ⎛⎛ ⎞ ⎞ ⎞
 
∞ ∞  y2
= 2L f˜ (1 − s2 ) · ⎝ WJacquet ⎝⎝ u y1 ⎠, (ν2 , ν1 ), ψ1,1 ⎠ du ⎠
0 0 −∞ 1
dy1 dy2
× y1s1 −1 y2−s2 . (6.5.20)
y1 y2

First, the holomorphic continuation of L f (s) follows by Riemann’s trick of

breaking the line of integration in the y2 variable into two pieces [0,1], [1,∞],
and then making the transformation y2 → 1/y2 in the first piece [0,1]. The
proof of the functional equation in Theorem 6.5.15 follows immediately from
the following lemma.

Lemma 6.5.21 The ratio of double Mellin transforms

⎛⎛ ⎞ ⎞
∞ ∞ y1 y2
WJacquet ⎝⎝ y1 ⎠, (ν1 , ν2 ), ψ1,1⎠ y1s1 −1 y2s2 −1 dy1 dy2
y1 y2
0 0 1
⎛ ⎛⎛ ⎞ ⎞ ⎞
∞ ∞ ∞ y2
⎝ WJacquet ⎝⎝ u y1 ⎠, (ν2 , ν1 ), ψ1,1⎠ du⎠ y1s1 −1 y2−s2 dy1 dy2
y1 y2
0 0 −∞ 1
 6.5 The Godement–Jacquet L-function 185

is precisely equal to
G ν (s2 )
.
G̃ ν (1 − s2 )

Proof The triple integral in the denominator of (6.5.21) can be evaluated as

⎛ ⎞
y2
follows. We first put the matrix ⎝ u y1 ⎠ into Iwasawa form to obtain
1
⎛ ' ' ⎞
⎛ ⎞ y y / u 2 + y 2 uy / u 2 + y 2
y2 ⎜ 1 2 1 2 1 ⎟
⎝ u y1 ⎠ = ⎜ ' ⎟
⎜ ⎟ (mod O(3, R) · R× ).
⎝ u 2 + y12 ⎠
1
1
To simplify notation, we also write W (z) instead of WJacquet (z, (ν2 , ν1 ), ψ1,1 ).
It follows,√ after successively √ making the transformations, u → u · y1 ,
y2 → y2 · u 2 + 1, y1 → y1 · 1/ u 2 + 1, that
⎛ ⎛⎛ ⎞⎞ ⎞
∞ ∞ ∞ y2
⎝ W ⎝⎝ u y1 ⎠⎠ du ⎠ y1s1 −1 y2−s2 dy1 dy2
y1 y2
0 0 −∞ 1
⎛ ⎛⎛ ' ' ⎞⎞ ⎞
  ⎜
∞ ∞ ∞ y y / u 2 + y2 uy / u 2 + y2
⎜⎜ 1 2 1 2 1 ⎟⎟ ⎟
⎜ ⎜⎜ ' ⎟⎟ ⎟
= ⎜ W ⎜⎜ ⎟⎟ du ⎟
⎝ ⎝⎝ u 2 + y12 ⎠⎠ ⎠
0 0 −∞
1
dy1 dy2
× y1s1 −1 y2−s2
y1 y2
⎛ ⎛⎛ √ √ ⎞⎞ ⎞
∞ ∞ ∞ y2 / u + 1 uy2 / u + 1
2 2
⎜ ⎜⎜ √ ⎟⎟ ⎟
= ⎝ W ⎝⎝ y1 u 2 + 1 ⎠⎠ du ⎠
0 0 −∞ 1
dy1 dy2
× y1s1 y2−s2
y1 y2
⎛ ⎛⎛ ⎞⎞ ⎞
∞ ∞ ∞ y2 uy2
⎝
s +s dy1 dy2
= W ⎝⎝ y1 ⎠⎠ (u 2 + 1) − 12 2
du ⎠ y1s1 y2−s2
y1 y2
0 0 −∞ 1
⎛⎛ ⎞⎞ ⎛ ⎞
∞ ∞ y1 y2 ∞
s +s
= W ⎝⎝ y1 ⎠⎠ ⎝ e2πiuy2 (u 2 + 1) − 12 2
du ⎠
0 0 1 −∞
dy1 dy2
× y1s1 −s2 y2−s2 .
y1 y2
186 Automorphic forms and L-functions for S L(3, Z)

It is well known that

∞
π s s− 12
e2πiuy2 (u 2 + 1)−s du = 2 y K s− 12 (2π y2 ).
(s) 2
−∞
Consequently, the denominator in Lemma 6.5.21 takes the form:
⎛ ⎛⎛ ⎞⎞ ⎞
∞ ∞ ∞ y2
⎝ W ⎝⎝ u y1 ⎠⎠ du ⎠ y1s1 −1 y2−s2 dy1 dy2
y1 y2
0 0 −∞ 1
⎛⎛ ⎞⎞
∞∞ y1 y2
2π (s1 +s2 )/2
= W ⎝⎝ y1 ⎠⎠
((s1 + s2 )/2)
0 0 1
s1 −s2 −1
dy1 dy2
× K s1 +s2 −1 (2π y2 ) y1s1 −s2 y2 2 . (6.5.22)
2 y1 y2
The double integral in (6.5.22) was first evaluated by Bump (1984) and is
equal to
     s1 +β  s1 +γ
24 π 2s1 −s2 −1  1−s22 +α  1−s22 +β  1−s22 +γ  s1 +α  2  2
 2
,
 s1 +s
2
2

(6.5.23)
where
α = −ν1 − 2ν2 + 1,
β = −ν1 + ν2 ,
γ = 2ν1 + ν2 − 1.
A number of years later Stade (1990) found another method to obtain (6.5.23).
Finally, we may complete the proof of Lemma 6.5.21 by evaluating the
numerator of the expression in Lemma 6.5.21 using (6.1.4) and then explicitly
computing the ratio of double Mellin transforms given in Lemma 6.5.21.

6.6 Bump’s double Dirichlet series

In (Bump, 1984) it was shown that if


∞

A(m 1 , m 2 )
f (z) =
γ ∈U2 (Z)\S L(2,Z) m 1 =1 m 2 =0
|m 1 m 2 |
⎛⎛ ⎞ ⎞
|m 1 m 2 |  
⎝ ⎝ ⎠ γ
× WJacquet m1 z, ν, ψ1, |mm2 |⎠ (6.6.1)
1 2
1
 6.6 Bump’s double Dirichlet series 187

is a non-zero Maass form for S L(3, Z), normalized so that A(1, 1) = 1, which is
a simultaneous eigenfunction of all the Hecke operators as in Theorem 6.4.11,
then the double Dirichlet series

∞
∞
A(m 1 , m 2 )
(6.6.2)
m 1 =1 m 2 =1
m s11 m s22

has a meromorphic continuation to all s1 , s2 ∈ C2 and satisfies certain functional

equations. This result is a consequence of the following proposition.

Proposition 6.6.3 Let f be a Maass form for S L(3, Z) as in (6.6.1). Then we

have the factorization

∞
∞
A(m 1 , m 2 ) L f˜ (s1 )L f (s2 )
s1 s2 = .
m 1 =1 m 2 =1
m1 m2 ζ (s1 + s2 )

Proof By Theorem 6.4.11, the Fourier coefficients of f satisfy

m m2 
1
A(m 1 , 1)A(1, m 2 ) = A , .
d|m 1 d|m 2
d d

It follows that

∞
∞
A(m 1 , 1) A(1, m 2 )
∞
∞

m m 
m −s −s2
1 2
s1 s2 = A , 1
1 m2
m 1 =1 m 2 =1
m 1 m 2 m 1 =1 m 2 =1 d|m 1 d|m 2
d d


∞

∞

∞
A(m 1 /d, m 2 /d)
=
d=1 m1 = 1 m 2 =1
m s11 m s22
m 1 ≡ 0 (mod d) m 2 ≡ 0 (mod d)


∞

∞

∞
A(m 1 , m 2 )
=
d=1 m  =1 m  =1
(m 1 d)s1 (m 2 d)s2
1 2

∞
∞
A(m 1 , m 2 )
= ζ (s1 + s2 ) · .
m 1 =1 m 2 =1
m s11 m s22

A direct proof of the meromorphic continuation and functional equation of

Bump’s double Dirichlet series (6.6.2) has been found by M. Thillainatesan.
By Proposition 6.6.3, this gives a new proof of the functional equation of the
Godement–Jacquet L-function L f (s). The new proof of the functional equation
of (6.6.2) is based on the following two propositions.
188 Automorphic forms and L-functions for S L(3, Z)

Proposition 6.6.4 Let f be a Maass form of type ν for S L(3, Z) as in (6.6.1).

For (s1 ), (s2 ) sufficiently large, define
⎛⎛ ⎞⎛ ⎞⎞
∞ ∞ 1 1 0 u3 y1 y2
(s1 , s2 ) = f ⎝⎝ 0 1 0 ⎠ ⎝ y1 ⎠⎠
0 0 0 0 0 1 1
dy1 dy2
× y1s1 −1 y2s2 −1 du 3 .
y1 y2
Then
⎛⎛ ⎞ ⎞
∞ ∞ y1 y2
4L f˜ (s1 )L f (s2 )
(s1 , s2 ) = WJacquet ⎝⎝ y1 ⎠, ν, ψ1,1 ⎠
ζ (s1 + s2 )
0 0 1
dy1 dy2
× y1s1 −1 y2s2 −1 .
y1 y2

Proposition 6.6.5 Let f be a Maass form of type ν for S L(3, Z) as in (6.6.1).

For (s1 ), −(s2 ) sufficiently large and (s1 + s2 ) > 12 , define
⎛⎛ ⎞⎛ ⎞⎞
∞ ∞ 1 1 0 u3 y1 y2
1 (s1 , s2 ) = f 1 ⎝⎝ 0 1 0 ⎠ ⎝ y1 ⎠⎠
0 0 0 0 0 1 1
dy dy
× y1s1 −1 y2s2 −1 du 3
1 2
,
y1 y2
⎛ ⎞
0 0 1
where f 1 (z) = f˜(w2 t (z −1 )) with w2 = ⎝ 1 0 0 ⎠ . Then
0 1 0

1 (s1 , s2 )
⎛⎛ ⎞ ⎞
∞ ∞ y1 y2
4L f˜ (s1 )L f˜ (1 − s2 )
= WJacquet ⎝⎝ y1 ⎠, ν, ψ1,1 ⎠
((s1 + s2 )/2)ζ (s1 + s2 )
0 0 1
(s −s2 −1)/2 dy1 dy2
× y1s1 −s2 y2 1 K s1 +s2 −1 (y2 ) .
2 y1 y2

Since f is automorphic, it is clear that f = f 1 and (s1 , s2 ) = 1 (s1 , s2 ).

The proof of the functional equation (in Theorem 6.5.15) of the Godement–
Jacquet L-function now follows from Lemma 6.5.21 and (6.5.22).
Proof of Proposition 6.6.4 We may substitute the Whittaker expansion in
Theorem 6.5.7 into the integral for (s1 , s2 ). The integral over u 3 forces c = 0
 6.6 Bump’s double Dirichlet series 189

and d = ±1 (see below), and kills the sum over S L(2, Z). We have
(s1 , s2 )


∞
A(m 1 , m 2 )
=


m 1 =1 m 2 =0
|m 1 m 2 |
ab ∈U2 (Z)\S L(2,Z)
cd
∞ ∞  1 6 
ai y +b
7 
2πi m 1 cu 3 +m 2  ci y 2+d
× e 2 du 3
0
⎛
0 ⎛0 ⎞ ⎞
y1 y2 m 1 |m 2 |
|ci y2 +d|
⎜⎜ ⎟ ⎟
× WJacquet ⎝⎝ y1 m 1 · |ci y2 + d| ⎠, ν, ψ1,1 ⎠
1
dy1 dy2
× y1s1 −1 y2s2 −1
y1 y2
⎛⎛ ⎞ ⎞

∞
∞∞ y1 y2 m 1 |m 2 |
A(m 1 , m 2 )
=2 WJacquet ⎝⎝ m 1 y1 ⎠, ν, ψ1,1 ⎠
m 1 =1 m 2 =0
|m 1 m 2 |
0 0 1
dy1 dy2
× y1s1 −1 y2s2 −1
y1 y2
⎛⎛ ⎞ ⎞

∞
∞ ∞ ∞ y1 y2
A(m 1 , m 2 ) ⎝ ⎝ ⎠, ν, ψ1,1 ⎠
=4 W Jacquet y1
m 1 =1 m 2 =1
m s11 |m 2 |s2
0 0 1
dy1 dy2
× y1s1 −1 y2s2 −1 .
y1 y2

Proof of Proposition 6.6.5 By definition

1 (s1 , s2 )
⎛⎛ ⎞⎛ ⎞⎞
∞ ∞ 1 1 0 u y1 y2
= f1 ⎝ ⎝ 0 1 0⎠⎝ y1 ⎠⎠ y1s1 −1 y2s2 −1 du dy1 dy2
y1 y2
0 0 0 0 0 1 1
⎛⎛ ⎞⎛ ⎞⎞
  
∞ ∞ 1 1 −u 0 y1 y2 y2−1
dy1 dy2
= f˜ ⎝⎝ 0 1 0⎠⎝ y2−1 ⎠⎠ y1s1 −1 y2s2 −1 du
y1 y2
0 0 0 0 0 1 1
⎛⎛ ⎞⎛ ⎞⎞
  
∞ ∞ 0 1 u 0 y1 y2
= f˜ ⎝⎝ 0 1 0⎠⎝ y1 ⎠⎠ y1s1 −s2 y2s1 −1 du dy1 dy2 ,
y1 y2
0 0 −1 0 0 1 1


190 Automorphic forms and L-functions for S L(3, Z)

where the last identity above is obtained after making the successive transforma-
tions y2 → y1−1 , y1 y2 → y2 , and u → −u. We now substitute the Whittaker
expansion of Theorem 6.5.7 into the above. In this case the sum over S L(2, Z)
is not killed. Remarkably, however, the integral can be significantly simplified
after several clever transformations. We have

A(m 2 , m 1 )   
∞ ∞ 0


∞
1 (s1 , s2 ) =


m 1 =1 m 2 =0
|m 1 m 2 |
ab ∈ U2 (Z)\S L(2,Z)
0 0 −1
cd
⎛⎛ ⎞ ⎞
y1 y2 m 1 |m 2 |
 
|c(u+i y2 )+d|
2πim 2 (
a(u+i y2 )+b
⎜⎜ ⎟ ⎟
×e c(u+i y2 )+d
· WJacquet ⎝⎝ y1 m 1 ·|c(u+i y2 )+d| ⎠, ν, ψ1,1 ⎠
1
dy1 dy2
× y1s1 −s2 y2s1 −1 du .
y1 y2

The term corresponding to c = 0 will be killed in the above integral

over u. So, we may assume c = 0. Note the identity: (az + b)/(cz + d) =
(a/c) − (1/c(cz + d)). Hence,
   
a(u + i y2 ) + b a 1
 = − .
c(u + i y2 ) + d c c(c(u + i y2 ) + d)

With this in mind, in the above integrals, make the transformations

cu + d → cu, d = cq + r (1 ≤ r ≤ |c|, (r, c) = 1).

Consequently

∞




|c|
2πim 2 r̄ A(m 2 , m 1 )
1 (s1 , s2 ) = e c

m 1 =1 m 2 =0 q∈Z c=0 r =1
|m 1 m 2 |
(r,c)=1
r
∞ ∞ 
q+ c
2πim 2 u
−
× e |c(u+i y2 )|2

0 0 q+ rc −1
⎛⎛ ⎞ ⎞
y1 y2 m 1 |m 2 |
|c(u+i y2 )|
⎜⎜ ⎟ ⎟
× WJacquet ⎝⎝ y1 m 1 ·|c(u+i y2 )| ⎠, ν, ψ1,1 ⎠
1
dy1 dy2
× y1s1 −s2 y2s1 −1 du . (6.6.6)
y1 y2
 6.6 Bump’s double Dirichlet series 191

Note that for fixed r, c,

r


q+ c
∞
= . (6.6.7)
q∈Z
q+ rc −1 −∞

Next, we combine (6.6.6) and (6.6.7) and then make the successive transfor-
mations
y1 u y2
y1 → , u → 2 , y2 → 2 .
|c(u + i y2 )| c c
We obtain

∞


|c| 2πim 2 r̄ ∞ ∞ ∞ 2πim u
e c A(m 2 , m 1 ) − 2
1 (s1 , s2 ) = e |u+i y2 |2

m 1 =1 m 2 =0 c=0 r =1
|c|s1 +s2 |m 1 m 2 |
(r,c)=1 0 0 −∞
⎛⎛ ⎞ ⎞
y1 y2 m 1 |m 2 |
|u+i y2 |2
⎜⎜ ⎟ ⎟ s −s s −1
× WJacquet ⎝⎝ y1 m 1 ⎠, ν, ψ1,1 ⎠ y11 2 y21
1
dy1 dy2
× |u + i y2 |s2 −s1 du . (6.6.8)
y1 y2
We now successively make the transformations
|m 2 | y1
u → u · y2 , y2 → y2 · , y1 →
u2 + 1 m1
in (6.6.8). It follows that

∞


|c| 2πim 2 r̄ ∞ ∞ ∞
e c A(m 2 , m 1 ) −
2πim 2 u
1 (s1 , s2 ) = +s
e y2 (u2 +1)
m 1 =1 m 2 =0 c=0 r =1
|c| 1 2 |m 1 m 2 |
s
(r,c)=1 0 0 −∞
⎛⎛ ⎞ ⎞
y1 m 1 |m 2 |
y2 (u 2 +1)
⎜⎜ ⎟ ⎟ s −s s
× WJacquet ⎝⎝ y1 m 1 ⎠, ν, ψ1,1 ⎠ y11 2 y22
1
dy1 dy2
× (u 2 + 1)(s2 −s1 )/2 du
y1 y2

∞


|c| 2πim 2 r̄
e c A(m 2 , m 1 )
= s1 +s2 1+s1 −s2
m 1 =1 m 2 =0 c=0 r =1
|c| m1 |m 2 |1−s2
(r,c)=1
⎛⎛ ⎞ ⎞
∞ ∞ ∞ m2
y1 y2−1
− 2πiu
y2 · |m 2 |
× e WJacquet ⎝⎝ y1 ⎠, ν, ψ1,1⎠
0 0 −∞ 1
y1s1 −s2 y2s2 dy1 dy2
× du . (6.6.9)
(u 2 + 1)(s1 +s2 )/2 y1 y2
192 Automorphic forms and L-functions for S L(3, Z)

Note that the Dirichlet series completely separates from the triple integral of the
Whittaker function in (6.6.9). To complete the proof, we need two lemmas. The
first lemma evaluates the Dirichlet series in (6.6.9) while the second evaluates
the triple integral of the Whittaker function.

Lemma 6.6.10 We have

∞


|c| 2πim 2 r̄
e c A(m 2 , m 1 ) 4L f˜ (s1 )L f˜ (1 − s2 )
+s −s
= .
m 1 =1 m 2 =0 c=0 r =1
|c| s 1 2 1+s
m1 1 2
|m 2 |1−s 2 ζ (s1 + s2 )
(r,c)=1

Lemma 6.6.11 We have

⎛⎛ ⎞ ⎞
∞ ∞ ∞ m2
y1 y2−1
− 2πiu
y2 · |m 2 |
e WJacquet ⎝⎝ y1 ⎠, ν, ψ1,1 ⎠
0 0 −∞ 1
y1s1 −s2 y2s2 dy1 dy2
× du
(u 2 + 1)(s1 +s2 )/2 y1 y2
⎛⎛ ⎞ ⎞
∞ ∞ y1 y2
= WJacquet ⎝⎝ y1 ⎠, ν, ψ1,1 ⎠
0 0 1
(s −s2 −1)/2 dy1 dy2
× y1s1 −s2 y2 1 K s1 +s2 −1 (y2 ) .
2 y1 y2

Proof To prove Lemma 6.6.10, we apply Proposition 3.1.7. Since every Maass
form is even, we obtain

∞


|c|
e(2πim 2 r̄ )/c A(m 2 , m 1 )
|c|s1 +s2 1+s1 −s2
m 1 =1 m 2 =0 c=0 r =1 m1 |m 2 |1−s2
(r,c)=1

4
∞
∞

A(m 2 , m 1 )
= 1 −s2
ζ (s1 + s2 ) m 1 =1 m 2 =1 d|m 2 d s 1 +s2 −1 m 1+s
1 m 1−s
2
2

4
∞
∞
∞
A(m 2 d, m 1 )
=
ζ (s1 + s2 ) m 1 =1 m 2 =1 d=1 m 1 1 −s2 m 1−s
1+s
2
2 s1
d
4
∞
∞
∞
A(m 2 d, m 1 )
=
ζ (s1 + s2 ) m 1 =1 m 2 =1 d=1 (m 1 m 2 )1−s2 (m 1 d)s1

4
∞
∞
A m 2 d/m 21 , m 1
=
ζ (s1 + s2 ) m 2 =1 d=1 m 1 |(d,m 2 ) m 1−s
2
2 s1
d
 6.6 Bump’s double Dirichlet series 193

4
∞
∞
A(d, 1) A(m 2 , 1)
= ·
ζ (s1 + s2 ) m 2 =1 d=1 d s1 m 1−s
2
2

4L f˜ (s1 )L f˜ (1 − s2 )
= .
ζ (s1 + s2 )
Finally, to prove Lemma 6.6.11, we use the identity
 ∞
−s1 −s2
e±2πiu/y2 (u 2 + 1) 2 du
−∞
s +s
2π 2
1 2
1−s1 −s2 
=  s1 +s2 y2 2 K s1 +s2 −1 2π y2−1 .
 2 2

The lemma immediately follows after applying the transformation y2 → y2−1 .

This completes the proof of Proposition 6.6.5.

Remark 6.6.12 Note that Bump’s double Dirichlet series is generalized to

G L(n) in (Bump and Friedberg, 1990). This is a Rankin–Selberg construction
involving two complex variables, one from an Eisenstein series, one of “Hecke”
type. For G L(3) it produces the Bump double Dirichlet series.
GL(n)pack functions The following GL(n)pack functions, described in the
appendix, relate to the material in this chapter:
ApplyCasimirOperator GetCasimirOperator
WeylGenerator Whittaker
WhittakerGamma. SpecialWeylGroup
 7
The Gelbart–Jacquet lift

7.1 Converse theorem for S L(3, Z)

It was shown in Theorem 6.5.15 that the Godement–Jacquet L-function asso-
ciated to a Maass form for S L(3, Z) is entire and satisfies a simple functional
equation. The development of a converse theorem for such L-functions (which
generalizes Theorem 3.15.3) is really due to the efforts of Piatetski-Shapiro over
several decades. In (Jacquet, Piatetski-Shapiro and Shalika, 1979) a Dirichlet
+∞
series A(n, 1)n −s (with A(1, 1) = 1) is considered. It is assumed that this
n=1
Dirchlet series is entire and bounded in vertical strips (EBV) and that the A(m, n)
satisfy the Hecke relations as given in Theorem 6.4.11. They proved that if all
the twists


∞
A(n, 1)χ (n)n −s , (χ a Dirichlet character)
n=1

satisfy a suitable functional equation (very similar to the functional equation

given in Theorem 6.5.15 with ν ∈ C2 ) then


∞

A(m 1 , m 2 )
f (z) =
γ ∈U2 (Z)\S L(2,Z) m 1 =1 m 2 =0
m 1 |m 2 |
⎛⎛ ⎞ ⎞
m 1 |m 2 |  
⎝ ⎝ ⎠ γ
× WJacquet m1 z, ν, ψ1, |mm2 | ⎠ ,
1 2
1
is, in fact, a Maass form for S L(3, Z). This is the converse thoerem for S L(3, Z),
and provides a generalization of the converse theorem for S L(2, Z) as given in
Section 3.15.
The idea of introducing twisted functional equations to prove a converse
theorem is due to Weil (1967). Cogdell and Piatetski-Shapiro have obtained

194
 7.1 Converse theorem for S L(3, Z) 195

a vast generalization of Weil’s original converse theorem. In (Cogdell and

Piatetski-Shapiro, 1994) it is shown that to prove an L-function is automor-
phic for G L(n), it is necessary to twist by G L(n − 1) while in (Cogdell and
Piatetski-Shapiro, 1999) this is improved to twists by G L(n − 2). In (Cogdell
and Piatetski-Shapiro, 2001) there is an improvement in a different direction
which involves restricting the ramification of the twisting. These papers have
proved of fundamental importance in establishing cases of Langlands conjec-
tures. A non-adelic version of the G L(3) converse theorem was obtained by
Miller and Schmid (2004). We shall present a new proof found recently by
Goldfeld and Thillainatesan.
Interlude on Dirichlet characters A Dirichlet character (mod q) is a
character of the cyclic group (Z/qZ)∗ . If χ is not trivial then χ (n) = 0 if
(n, q) > 1, and, otherwise is a φ(q)th root of unity. It is a periodic function
of n with period q. If q is the least period, then χ is said to be primitive. The
Dirichlet characters (mod q), both primitive and imprimitive, form a basis for
the functions on (Z/qZ)∗ . For Dirichlet characters χ (mod q) and (n, q) = 1,
we have the following well-known identity (see (Davenport, 1967))


q

q
τ (χ̄) · χ (n) = χ̄ ( )e2πi n/q , τ (χ ) = χ ( )e2πi /q , (7.1.1)
=1 =1

where τ (χ) is the Gauss sum. This allows us to represent χ (n) as a linear
combination of qth roots of unity. Note that in (7.1.1), the condition (n, q) = 1
can be dropped if χ is primitive. We may also represent each qth root of unity
as a linear combination of characters by the formula
1

χ̄ ( )τ (χ ) = e2πi /q . (7.1.2)
φ(q) χ (mod q)

Theorem 7.1.3 (S L(3, Z) Converse theorem) For integers m 1 , m 2 = 0, let

A(m 1 , m 2 ) ∈ C satisfy: A(m 1 , m 2 ) = A(−m 1 , m 2 ) = A(m 1 , −m 2 ), A(1, 1) = 1,
and also the multiplicativity relations given in Theorem 6.4.11. Assume that for
every primitive Dirichlet character χ (mod q), the Dirichlet series

∞
A(m, 1)χ (m)
∞
A(1, m)χ (m)
L χ (s) := , L̃ χ (s) := ,
m=1
ms m=1
ms

converge absolutely for (s) sufficiently large, and for fixed ν = (ν1 , ν2 ) ∈ C2 ,
satisfy the functional equation
τ (χ )2
q 2 s G ν (s + k)L χ (s) = i −k
3 3
√ · q 2 (1−s) G̃ ν (1 + k − s) L̃ χ̄ (1 − s),
τ (χ̄ ) q
196 The Gelbart–Jacquet lift

where k = 0 or 1 according as χ (−1) is +1 or −1, G ν , G̃ ν are prod-

ucts of three Gamma functions as in Theorem 6.5.15, and the functions
G ν (s)L(s), G̃ ν (s) L̃(s) are EBV. Then


∞

A(m 1 , m 2 )
γ ∈U2 (Z)\S L(2,Z) m 1 =1 m 2 =0
|m 1 m 2 |
⎛⎛ ⎞ ⎞
|m 1 m 2 |  
⎝ ⎝ ⎠ γ
× WJacquet m1 z, ν, ψ1, |mm2 | ⎠ ,
1 2
1
is a Maass form for S L(3, Z).

Proof of the converse theorem, Step I Define

∞
A(m 1 , m 2 )
f (z) :=
γ ∈U2 (Z)\S L(2,Z) m 1 =1 m 2 =0
|m 1 m 2 |
⎛⎛ ⎞ ⎞
|m 1 m 2 |  
γ
×WJacquet ⎝⎝ m1 ⎠ z, (ν1 , ν2 ), ψ1, |mm2 | ⎠, (7.1.4)
1 2
1
and


∞

A(m 2 , m 1 )
f˜(z) :=
γ ∈U2 (Z)\S L(2,Z) m 1 =1 m 2 =0
|m 1 m 2 |
⎛⎛ ⎞ ⎞
|m 1 m 2 |  
γ
×WJacquet ⎝⎝ m1 ⎠ z, (ν2 , ν1 ), ψ1, |mm2 | ⎠. (7.1.5)
1 2
1
In the following, let
⎛ ⎞ ⎛ ⎞
1 0 u3 0 0 1
u = ⎝ 0 1 u1 ⎠ , w2 = ⎝ 1 0 0⎠.
0 0 1 0 1 0
If f is automorphic then

f (z) = f˜ w2 · t z −1

for all z ∈ h3 and

1 1 1 1
−2πiku 1
f (Auz)e du 1 du 3 = f˜(w2 · t (Auz)−1 )e−2πiku 1 du 1 du 3
0 0 0 0
(7.1.6)
 7.1 Converse theorem for S L(3, Z) 197

⎛ ⎞
1 0 0
for all z ∈ h3 , k ∈ Z, A = ⎝ α 1 0 ⎠ with α ∈ Q. Our aim is to prove a
0 0 1
converse to (7.1.6).
Basic Lemma 7.1.7 Let f, f˜ be defined by (7.1.4), (7.1.5), respectively.
Assume
1 1 1 1
−2πiqu 1

f (Auz)e du 1 du 3 = f˜ w2 · t (Auz)−1 e−2πiqu 1 du 1 du 3
0 0 0 0
⎛ ⎞
1 0 0
for all z ∈ h3 , h, q ∈ Z, A = ⎝ h/q 1 0 ⎠, with q =  0. Then f is a Maass
0 0 1
form of type ν for S L(3, Z) and f˜ is its dual form.
Proof of Lemma 7.1.7 First of all, we claim that
⎛ ⎞
∗ ∗ ∗
f ( pz) = f (z), f˜( pz) = f˜(z), ∀ p ∈ P = ⎝ ∗ ∗ ∗ ⎠ ⊂ S L(3, Z).
0 0 1
(7.1.8)

This is due to the fact that P is generated by elements of the form

⎛ ⎞ ⎛ ⎞
a b 1 r
⎝c d ⎠, ⎝ 1 s ⎠ , (with a, b, c, d, r, s ∈ Z, ad − bc = 1).
1 1
⎛ ⎞
a b
We first check that (7.1.8) holds for p = ⎝ c d ⎠ . This follows easily
1
because



∞

A(m 1 , m 2 )
f ( pz) =
γ ∈U2 (Z)\S L(2,Z) m 1 =1 m 2 =0
|m 1 m 2 |
⎛⎛ ⎞ ⎞
|m 1 m 2 |  
⎝ ⎝ ⎠ γ
× WJacquet m1 · pz, ν, ψ1, |mm2 | ⎠
1 2
1
= f (z),

since the sum over γ is permuted by p. A similar argument holds for f˜.
198 The Gelbart–Jacquet lift

⎛ ⎞
1 r
It only remains to show that (7.1.8) holds for p = ⎝ 1 s ⎠ . To show
1
this, we just use the identity
⎛ ⎞⎛ ⎞ ⎛ ⎞
|m 1 m 2 | a b 1 r
⎝ m1 ⎠⎝c d ⎠·⎝ 1 s⎠
1 1 1
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
1 r |m 1 m 2 | a b
= ⎝ 1 s ⎠ · ⎝ m1 ⎠⎝c d ⎠,
1 1 1
where
r  = |m 1 m 2 |(ar + bs), s  = m 1 (cr + ds).
It follows that


∞

A(m 1 , m 2 )
f ( pz) =
γ ∈U2 (Z)\S L(2,Z) m 1 =1 m 2 =0
|m 1 m 2 |
⎛⎛ ⎞ ⎛ ⎞ ⎞
1 r |m 1 m 2 |  
γ
× WJacquet ⎝⎝ 1 s ⎠·⎝ m1 ⎠ z, ν, ψ1, |mm2 | ⎠
1 2
1 1
= f (z),
because
⎛⎛ ⎞ ⎞
1 r
WJacquet ⎝⎝ 1 s  ⎠ · z, ν, ψ ⎠ = WJacquet (z, ν, ψ)
1
for all z, ν, ψ, if r  , s  ∈ Z. Again, a similar argument applies for f˜.
Lemma 7.1.9 We have
 
f˜ w2 · t z −1 = f˜ w2 · t ( pz)−1
⎛ ⎞
1 r
for all p = ⎝ 1 s ⎠ , with r, s ∈ Z.
1
Proof By (7.1.8), we may, without loss of generality replace w2 by
⎛ ⎞
1 0 0
w2 = ⎝ 0 0 1 ⎠.


0 1 0
 7.1 Converse theorem for S L(3, Z) 199

We compute:
⎛ ⎞ ⎛ ⎞
1 1
−1 ⎠ · w2 −1
w2 · t p −1 · w2 = w2 · ⎝ 1 = ⎝ −r 1 −s ⎠ ∈ P.
−r −s 1 1

Thus,
   −1
f˜ w2 · t ( pz)−1 = f˜ w2 · t p −1 · t z −1 = f˜ w2 · t p −1 · w2 · w2 · t z −1
⎛⎛ ⎞ ⎞
1 
= f˜ ⎝⎝ −r 1 −s ⎠ · w2 · t z −1 ⎠ = f˜ w2 · t z −1 .
1

Here we have used the fact that f˜ is invariant under left multiplication by
elements in P.

⎛ ⎞
1 0 0
Now, if (7.1.6) holds for all z ∈ h3 and all h, q ∈ Z, A = ⎝ h/q 1 0 ⎠,
0 0 1
−1 
with q = 0, then on choosing z = A z , we must have
1 1
f (Au A−1 z  )e−2πiqu 1 du 1 du 3
0 0
1 1
= f˜(w2 · t (Au A−1 z  )−1 )e−2πiqu 1 du 1 du 3
0 0

for all z  ∈ h3 . Since

⎛ ⎞
1 0 u3
⎜ ⎟
Au A−1 = ⎝0 1 hu 3
q
+ u1 ⎠ ,
0 0 1

it easily follows after a change of variables that

1 1
3  4
f (uz) − f˜ w2 · t (uz)−1 e−2πiqu 1 e−2πi hu 3 du 1 du 3 = 0, (7.1.10)
0 0

for all z ∈ h3 , and all h, q ∈ Z, with q = 0.

We will next show that (7.1.10) holds for all h ∈ Z and q = 0. In fact, this
case holds without any assumptions on f .
200 The Gelbart–Jacquet lift

*1 3 4
Lemma 7.1.11 We have f (uz) − f˜(w2 · t (uz)−1 ) du 1 = 0, wher e
⎛ ⎞ 0
10 0
u = ⎝ 0 1 u 1 ⎠.
00 1

Proof It follows from Theorem 6.5.7, after integrating in u 1 , that

1
3 4
f (uz) − f˜(w2 · t (uz)−1 ) du 1
0
⎡ ⎛ ⎛ ⎞ ⎞


∞
0 1 0
⎣ A(m , m ) ⎝ ⎝ 0 ⎠ z⎠
1 2
= WJacquet M 1 0
m 1 =1 m 2 =0
m 1 |m 2 |
0 0 1
⎤
A(m 2 , m 1 ) 
− W̃Jacquet Mw2 · t z −1 ⎦ ,
m 1 |m 2 |

⎛ ⎞
m 1 |m 2 |
where M = ⎝ m1 ⎠ , and WJacquet is given as in Theorem 6.4.11.
1
But
⎛ ⎞
 1
W̃Jacquet (z) = WJacquet w3 · t z −1 · w3 −1
, w3 = ⎝ 1 ⎠.
1

The proof follows upon noting that


W̃Jacquet (Mw2 · t z −1 ) = WJacquet w3 · Mw2 · t z −1 · w3−1
⎛ ⎛ ⎞ ⎞
0 1 0
= WJacquet ⎝ M  ⎝ 1 0 0 ⎠ z ⎠ ,
0 0 1

where M  is the same as M, but with m 1 , m 2 interchanged.

It follows from Lemma 7.1.9, that the function ⎛ − f˜(w2 · t (z)−1 ) is

f (z) ⎞
1 ∗
invariant under left multiplication by the group ⎝ 1 ∗ ⎠ . It, therefore, has
1
a Fourier expansion in x1 , x3 . The identity (7.1.10) together with Lemma 7.1.11
 7.1 Converse theorem for S L(3, Z) 201

tells us that every Fourier coefficient vanishes. Consequently

 
f (z) = f˜ γ · t (z)−1 = f˜ t (γ z)−1

with γ = w2 , for all z ∈ h3 . Finally, we may complete the proof of Lemma 7.1.7.
Note that f (z) = f˜(t (γ z)−1 ) must hold for all γ ∈ S L(3, Z) because it will hold
with γ replaced by p1 γ p2 and p1 , p2 ∈ P. To see this  p1 ∈ P implies
 note that 
γ γ γ
that p1 is either of the form or of the form w2 with
1 1 1

γ , γ ∈ S L(2, Z). In the former case, the invariance due to multiplication by p1
on the left follows from (7.1.8), while in the latter case we have already proved
the invariance of w2 on the left, so we are reduced to the first case. The invariance
due to multiplication by p2 on the right follows by letting z → p2−1 · z. So, we
have shown that f is automorphic.

⎛ ⎞
1 0 0
Proof of the converse theorem, Step II Fix A = ⎝ h/q 1 0 ⎠ with
0 0 1
h, q ∈ Z, and q = 0. Let f, f˜ be defined as in (7.1.4) and (7.1.5) and set

1 1
F(z, h, q) = f (Auz)e−2πiqu 1 du 1 du 3 ,
0 0
1 1

F1 (z, h, q) = f˜ w2 · t (Auz)−1 e−2πiqu 1 du 1 du 3 ,
0 0

⎛ ⎞ ⎛ ⎞
u3 1 1
where we recall that u = ⎝ 1 u 1 ⎠ , w2 = ⎝ 1 ⎠ . If we can show
1 1
that F(z, h, q) = F1 (z, h, q) for all z, h, q, k then the basic Lemma 7.1.7 tells
us that f is a Maass form for S L(3, Z). It remains to show that the functional
equations for L χ (s) stated in Theorem 7.1.3 imply that F(z, h, q) = F1 (z, h, q)
for all z, h, q. We shall do this in two remaining steps. First we show that the
functional equations for L χ (s) imply that
   k     k 
∂ ∂  ∂ ∂ 
F(z, h, q)  = F1 (z, h, q)  .
∂ x1 ∂ x2 x1 =x2 =0
∂ x1 ∂ x2 x1 =x2 =0

In the final step 3 of the proof, we extend the above result and prove the basic
Lemma 7.1.7 in all cases.
202 The Gelbart–Jacquet lift

The key point is that the functional equations for L χ (s) imply that
   k ∞ ∞ 
∂ ∂ dy1 dy2 
F(z, h, q) y1s1 −1 y2s2 −1
∂ x1 ∂ x2 y1 y2 x1 =x2 =0
0 0
   k ∞ ∞ 
∂ ∂ dy1 dy2 
= F1 (z, h, q)y1s1 −1 y2s2 −1 ,
∂ x1 ∂ x2 y1 y2 x1 =x2 =0
0 0

and then by taking inverse Mellin transforms we obtain the desired identity.
The above idea is exemplified in the following two lemmas.

Lemma 7.1.12 Let q, h ∈ Z with q = 0. Define δ := (h, q), qδ := q/δ,

h δ := h/δ, and h̄ δ · h δ ≡ 1 (mod qδ ). Then for integers 0 ≤ , k, we have
   k ∞ ∞ 
∂ ∂ dy1 dy2 
F(z, h, q) y1s1 −1 y2s2 −1
∂ x1 ∂ x2 y1 y2 x1 =x2 =0
0 0

(2πiq)
A(δ, m 2 )  2πim 2 k

= s1 −2s2 +1 e2πim 2 h̄ δ /qδ
qδ δ s1 m 2 =0 |m|s2 qδ2
∞ ∞
dy1 dy2
× WJacquet (y, ν, ψ1,1 )y1s1 −1 y2s2 −1 .
y1 y2
0 0

Proof Note that

       
a b 1 0 a + bh/q b a b
· = = .
c d h/q 1 c + dh/q d c d

It follows from Theorem 6.5.7 that

∞ ∞

∞
∞ ∞ 1 1
dy1 dy2 A(m 1 , m 2 )
F(z, h, q) y1s1 −1 y2s2 −1 =
y1 y2 (c,d)=1 m 1 =1 m 2 =0
m 1 |m 2 |
0 0 0 0 0 0
     
a z +b
2πim 1 c+ dh (x3 +u 3 ) 2πim 1 d(x1 +u 1 ) −2πiqu 1 2πim 2  c z 2+d
×e q
e e 2 e
⎛⎛ ⎞ ⎞
m 1 |m 2 |y1 y2
|c z 2 +d|
⎜⎜ ⎟ ⎟
× WJacquet ⎝⎝ m 1 y1 · |c z 2 + d| ⎠ , ν, ψ1, |mm2 | ⎠
2
1
dy1 dy2
× y1s1 −1 y2s2 −1 du 1 du 3 .
y1 y2
 7.1 Converse theorem for S L(3, Z) 203

The integrals over u 1 , u 3 are zero unless c = −dh/q, m 1 d = q. Hence c = 0,

and d = q/m 1 . Since ad − bc = 1 it immediately follows that
a(q/m 1 ) + b(h/m 1 ) = 1. Consequently, m 1 = (q, h) := δ, and letting
qδ = q/δ, h δ = h/δ, h̄ δ = h −1 δ (mod qδ ), the above identity becomes:
 
∞ ∞
dy1 dy2
F(z, h, q)y1s1 −1 y2s2 −1
y1 y2
0 0

A(δ, m 2 ) 2πim2 h̄δ 2πim 2 x2

= e qδ e2πiq x1 e qδ
2

m 2 =0
δ · |m 2 |
⎛⎛ ⎞ ⎞
δ|m 2 |y1 y2
∞ ∞
⎜⎜ qδ ⎟ ⎟ s −1 s −1 dy1 dy2
× WJacquet ⎝⎝ δy1 · qδ ⎠, ν, ψ1, |mm2 | ⎠ y11 y22 .
2 y1 y2
0 0 1
The lemma follows after taking partial derivatives and then performing a simple
variable change.

Lemma 7.1.13 Let q, h ∈ Z with q = 0. Then for , k ∈ Z with ≥ 0,

k ∈ {0, 1}, we have
   ∞ ∞ 
∂ ∂ k 
s1 −1 s2 −1 dy1 dy2 
F1 (z, h, q) y1 y2
∂ x1 ∂ x2 y1 y2 x1 =x2 =0
0 0
+k
s1 +s2


 
(2πiq) π 2 A(m 2 , m 1 ) im 2 k
=  s1 +s2
q s1 +s2  2 2k+1−2s2
m 1 |q m 2 =0 m 1 |m 2 |k+1−s2 |m 2 |

1
q/m ∞ ∞
−hr +m 2 r̄
2πim 1
× e q WJacquet (y, (ν2 , ν1 ), ψ1,1 )
 r =1 0 0
r, mq =1
1

2k+s1 −s2 −1 dy1 dy2

× K s1 +s2 −1−2k (2π y2 )y12k+s1 −s2 y 2 .
2 y1 y2

Proof Note that

⎛ x2 (u 3 +x3 ) ⎞
1 −u 3 − x3 −u 1 − x1 + x22 +y22
⎜ ⎟
w2 · (t (Auz)−1 ) · w2−1 = ⎝ 1 −h/q − x2
x22 +y22 ⎠
1
⎛ ⎞
√ y1 y2
x22 +y22
⎜ ⎟
×⎝ y2
⎠ mod (O(3, R) · R× ),
x22 +y22
1
204 The Gelbart–Jacquet lift

and
∞ ∞
dy1 dy2
F1 (z, h, q) y1s1 −1 y2s2 −1
y1 y2
0 0
∞∞1 1
 dy1 dy2
= f˜ w2 · (t (Auz)−1 ) · w2−1 e−2πiqu 1 du 1 du 3 y1s1 −1 y2s2 −1 .
y1 y2
0 0 0 0

If we now make use of Theorem 6.5.7, in the above, we obtain

∞ ∞
dy1 dy2
F1 (z, h, q) y1s1 −1 y2s2 −1
y1 y2
0 0
0 1


∞
∞ ∞ 1 1
A(m 2 , m 1 ) x (u +x )
2πim 1 c −u 1 −x1 + 2 2 3 23
= e x2 +y2

(c,d)=1 m 1 =1 m 2 =0
m 1 |m 2 |
  0 0 0 0
x2 az 2 +b
−2πim 1 d q + x 2 +y 2
h
2πim 2  cz 2 +d
×e 2 2 e e−2πiqu 1 du 1
⎛⎛ ⎞ ⎞
√ m2 1 |m22 |y1 y2
⎜⎜ x2 +y2 · |cz 2 +d| ⎟ ⎟
× WJacquet ⎜
⎝⎝
⎜ m 1 y2 |cz 2 +d| ⎟ , (ν2 , ν1 ), ψ1,1 ⎟
⎠ ⎠
x22 +y22
1
dy1 dy2
× du 3 y1s1 −1 y2s2 −1 ,
y1 y2
where
'
z 2 = −u 3 − x3 + i y1 x22 + y22 .

The integral in u 1 above, namely

 1
e−2πim 1 cu 1 e−2πiqu 1 du 1
0

vanishes unless m 1 c = −q. Further, let d = c + r with ∈ Z and 1 ≤ r < c

where (r, c) = 1. If we make the change of variables u 3 → u 3 + + r m 1 /q,
then we may re-express the above identity in the form
∞ ∞
dy1 dy2
F1 (z, h, q) y1s1 −1 y2s2 −1
y1 y2
0 0


A(m 2 , m 1 )

1
q/m
2πim 1 r h 2πim 1 m 2 r̄
= e2πiq x1 e− q e− q

m 1 |q m 2 =0
m 1 |m 2 | ∈Z
 r =1
r, mq =1
1
 7.1 Converse theorem for S L(3, Z) 205

r m1
+ +1  
∞ ∞ q u 3 +x3 −1
−2πiq x2 2πim 2 
c2 z 2
× e x22 +y22
e
r m1
0 0 + q
⎛⎛ ⎞ ⎞
√m 12|m 2 |y1 y2

⎜⎜ x2 +y2 · |cz 2 | ⎟ ⎟
2

× WJacquet ⎜
⎝⎝
⎜ m 1 y2 |cz 2 | ⎟, (ν2 , ν1 ), ψ1,1 ⎟
⎠ ⎠
x22 +y22
1
dy1 dy2
× du 3 y1s1 −1 y2s2 −1 ,
y1 y2
which after summing over ∈ Z, and,'then successively making the transfor-
mations: u 3 → u 3 − x3 , u 3 → u 3 · y1 x22 + y22 , becomes

A(m 2 , m 1 )
1
q/m
2πim 1 r h 2πim 1 m 2 r̄
= e2πiq x1 e q e q
m 1 |m 2 |
m 1 |q m 2 =0 
r =1
r, mq =1
1

∞ ∞ ∞ √2
−2πiq x2 y1 u 3
√
2
2πim 1 m 2 u 3

x2 +y22 x22 +y22 (u 23 +1)

× e e q 2 y1

0 0 −∞
⎛⎛ m 21 |m 2 |y2
⎞ ⎞
 √
⎜⎜ q x2 +y22
2
u 23 +1
√ ⎟ ⎟
⎜⎜ ⎟ ⎟
× WJacquet ⎜⎜ qy1 y2
√ u 23 +1 ⎟, (ν2 , ν1 ), ψ1,1 ⎟
⎝⎝ x22 +y22 ⎠ ⎠
1
'
dy1 dy2
× du 3 y1s1 y2s2 −1 x22 + y22 .
y1 y2
If we now take partial derivatives with respect to x1 , x2 , set x1 = x2 = 0, and
make the substitutions
y1 m 2 |m 2 |y2
y1 → ' , y2 → '1 ,
q u 23 + 1 q u 23 + 1

it follows that
   ∞ ∞ 
∂ ∂ k dy1 dy2 

F1 (z, h, q) y1s1 −1 y2s2 −1 
∂ x1 ∂ x2 y1 y2 x1 =x2 =0
0 0

(2πiq)


1
q/m
A(m 2 , m 1 ) 2πim 1 r h 2πim 1 m 2 r̄
= e q e q
q s1 +s2 m 1 |q m 2 =0 m 1−2s 2
|m 2 |1−s2 
1
r =1
r, mq =1
1
206 The Gelbart–Jacquet lift

⎛⎛ ⎞ ⎞
∞ ∞  k y2−1
2πiqy1 u 3
× WJacquet ⎝⎝ y1 ⎠, (ν2 , ν1 ), ψ1,1 ⎠
m 21 |m 2 |y2
0 0 1
∞ 2πiu 3 m 2  −(s1 +s2 )/2 dy1 dy2
× e y1 y2 |m2 | u 23 + 1 du 3 y1s1 y2s2 .
y1 y2
−∞

Next, make the successive transformations: y2 → y2−1 , y2 → y2 · y1 . We

obtain
(2πiq)


1
q/m
A(m 2 , m 1 ) (−hr +m 2 r̄ )

+s 1−2s
e2πim 1 q
q s1 2
m 1 |q m 2 =0 m 1
2
|m 2 |1−s2  r =1
r, mq =1
1
⎛⎛ ⎞ ⎞
∞ ∞  k y1 y2
2πiqy12 y2
× WJacquet ⎝⎝ y1 ⎠, (ν2 , ν1 ), ψ1,1 ⎠
m 21 |m 2 |
0 0 1
⎛ ⎞
∞
−2πiu 3 y2 |m 2 |  2 −(s1 +s2 )/2 k dy1 dy2
m
×⎝ e 2 u3 +1 u 3 du 3 ⎠ y1s1 −s2 y2−s2 .
y1 y2
−∞

We may complete the proof of Lemma 7.1.13 by invoking the identities:

∞
πs
e2πiuy2 (u 2 + 1)−s du = 2
1
|y2 |s− 2 K s− 12 (2π |y2 |),
(s)
−∞
∞  
y2 πs
e2πiuy2 (u 2 + 1)−s u du = 2 i
1
· |y2 |s− 2 K s− 32 (2π |y2 |).
|y2 | (s)
−∞

Lemmas 7.1.12 and 7.1.13 involve additive twists of the Godement–Jacquet

L-function which have very complicated functional equations. We would like
to pass to twists by primitive Dirichlet characters which are much easier to deal
with. This is the motivation for the next lemma.
Lemma 7.1.14 Assume
   ∞ ∞ 
∂ ∂ k dy1 dy2 

F(z, h, q) y1s1 −1 y2s2 −1 
∂ x1 ∂ x2 y1 y2 x1 =x2 =0
0 0
   k ∞ ∞ 
∂ ∂ dy1 dy2 
= F1 (z, h, q) y1s1 −1 y2s2 −1 (7.1.15)
∂ x1 ∂ x2 y1 y2 x1 =x2 =0
0 0
 7.1 Converse theorem for S L(3, Z) 207

for all h, q, ∈ Z with q = 0, ≥ 0, and k ∈ {0, 1}. Define δ := (h, q) and

qδ := q/δ. Then
 
G ν (s2 ) τ (χ̄ )q 3s2 −1
A(δ, m 2 ) χ (m 2 ) 2πim 2 k
·
G̃ ν (1 − s2 + 2k) τ (χ )π (s1 +s+2)/2 m 2 =0 δ s1 |m 2 |s2 qδ2


A(m 2 , m 1 )χ̄ (−m 1 )  im 2 k
= (2πiq)k
m 1 |δ m 2 =0 m 1
2k+1−2s2
|m 2 |k+1−s2 |m 2 |

1
q/m
2πim 1 m 2 r̄
× χ̄ (r )e q , (7.1.16)
r =1
(r,(q/m 1 ))=1

for every character χ (mod qδ ), with G ν , G̃ ν as in Theorem 6.5.15. Further, if

(7.1.16) holds for every character χ , each k ∈ {0, 1}, and all s1 , s2 ∈ C, then
(7.1.15) holds for all h, q, ∈ Z, k ∈ {0, 1}, ≥ 0 and q = 0.

Note The identity (7.1.16) is a formal identity. It is understood that all series
are well defined by analytic continuation. We present (7.1.16) and its proof in
this form to simplify the exposition of the ideas.

Proof We have h = h δ · δ. Consequently if (7.1.15) holds, then

   k ∞ ∞

dy2 
qδ
∂ ∂ dy1
χ (h δ )F(z, h δ δ, q)y1s1 −1 y2s2 −1
∂ x1 ∂ x2 h δ =1
y1 y2 x1 =x2 =0
0 0
   k ∞ ∞

dy2 
qδ
∂ ∂ dy1
= χ (h δ )F1 (z, h δ δ, q)y1s1 −1 y2s2 −1 .
∂ x1 ∂ x2 h δ =1
y1 y2 x1 =x2 =0
0 0

The result follows after several routine computations using Lemmas 7.1.12,
7.1.13, and formula (7.1.1). For example, on the left-hand side we will have
from Lemma 7.1.12 the character sum


qδ

qδ
χ (h δ )e2πim 2 h̄ δ /qδ = χ̄ (h δ ) e2πim 2 h δ /qδ = τ (χ̄ ) · χ (m 2 ),
h δ =1 h δ =1

alternatively, on the right-hand side we will get from Lemma 7.1.13 (for
(r, qδ ) = 1), the character sum



qδ


χ(h δ )e−2πim 1 r ·h δ /qδ = τ (χ )χ̄ (−m 1r ) = τ (χ )χ̄ (−m 1r ),
m 1 |q h δ =1 m 1 |q m 1 |δ

since χ (m 1 ) = 0 if (m 1 , qδ ) > 1.
208 The Gelbart–Jacquet lift

The last statement in Lemma 7.1.14 is a consequence of formula (7.1.2)

which says that every qth root of unity can be expressed as a linear combination
of Dirichlet characters (mod q).

Let us explore Lemma 7.1.14 in the special case that k = 0, χ is an even

primitive character (mod q) and (h, q) = 1 so that δ = 1. In this case (using
(7.1.1), Lemma 6.5.21 and (6.5.22)), the functional equation in Lemma 7.1.14
takes the simple form

3 τ (χ )2 3
q 2 s G̃ ν (s)L χ (s) = √ · q 2 (1−s) G̃ ν (1 − s) L̃ χ̄ (1 − s),
τ (χ̄ ) q

where

∞
A(n, 1)χ (n)
∞
A(1, n)χ̄ (n)
L χ (s) = , L̃ χ̄ (s) = ,
n=1
ns n=1
ns

and where G ν (s), G̃ ν (s) are products of Gamma functions as defined in Theorem
6.5.15. Note that if we take k = 1 in the above situation, then the identity (7.1.16)
holds automatically because each side will simply vanish since the sum over
m 2 with positive and negative terms will cancel out. We may also consider odd
primitive characters χ . In this case, the identity (7.1.16) automatically holds if
k = 0 and we get a functional equation

3 τ (χ )2 3
q 2 s G̃ ν (s + 1)L χ (s) = i √ · q 2 (1−s) G̃ ν (2 − s) L̃ χ̄ (1 − s),
τ (χ̄ ) q

when k = 1. The two functional equations may be combined into a single

functional equation
3 τ (χ )2 3
q 2 s G̃ ν (s + k)L χ (s) = i k √ · q 2 (1−s) G̃ ν (1 + k − s) L̃ χ̄ (1 − s),
τ (χ̄ ) q
(7.1.17)

where k = 0 if χ (−1) = 1 and k = 1 if χ (−1) = −1.

These are the simplest instances of the type of twisted functional equations
needed to obtain the converse theorem. Once one knows these functional equa-
tions, then all the other required functional equations follow by the usual tech-
niques (modified for the G L(3) situation) for constructing functional equations
with imprimitive characters. We shall not pursue this further as the technical
complications become extremely messy.
Proof of the converse theorem, Step III We have proved that if L χ (s) sat-
isfies the functional equation (7.1.17) for all primitive Dirichlet characters
 7.1 Converse theorem for S L(3, Z) 209

χ then
 
∂ ∂k  ∂ ∂k 

F(z, h, q) =
F (z, h, q)
k 1
, (7.1.18)
∂ x1 ∂ x2
k
x1 =x2 =0 ∂ x1 ∂ x2 x1 =x2 =0

for all = 0, 1, 2, 3, . . . , and k = 0, 1. Note that the proofs of Lemmas 7.1.12,

7.1.13 imply that F(z) := F(z, h, q), F1 (z) := F1 (z, h, q) are independent of
x3 and are functions of x1 , x2 , y1 , y2 only. Consequently, if
⎛ ⎞⎛ ⎞
1 x2 x3 y1 y2
z=⎝ 1 x1 ⎠⎝ y1 ⎠,
1 1
then the function F(z) − F1 (z) does not depend on x3 .
Since F(z) − F1 (z) is a real analytic function, it has a power series expansion
of the form

∞
∞
j
F(z) − F1 (z) = ci, j (y1 , y2 )x1i x2 .
i=0 j=0

It immediately follows from (7.1.18) that

c ,k (y1 , y2 ) = 0, (for = 0, 1, 2, 3, . . . , k = 0, 1). (7.1.19)
Now, F, F1 are eigenfunctions of the invariant differential operators on
G L(3, R). In particular, they are eigenfunctions of 1 given in (6.1.1). Recall
that
∂2 ∂2 ∂2  ∂2
1 = y12 2 + y22 2 − y1 y2 + y12 x22 + y22
∂ y1 ∂ y2 ∂ y1 ∂ y2 ∂ x32
∂ 2
∂ 2
∂2
+ y12 2 + y22 2 + 2y12 x2 .
∂ x1 ∂ x2 ∂ x1 ∂ x3
We calculate the action of 1 on F − F1 and use this to prove F(z) − F1 (z) ≡ 0
by showing that ci, j = 0 for all i, j ≥ 0.
 j j ∂ ci, j
2
2 i j ∂ ci, j
2
i j ∂ ci, j
2
1 x1i x2 ci, j = y12 x1i x2 + y x x − y1 y2 x x
∂ y12 2 1 2
∂ y22 1 2
∂ y1 ∂ y2
i−2 j i j−2
+ y1 i(i − 1)x1 x2 ci, j + y2 j( j − 1)x1 x2 ci, j ,
2 2

  

∞
∞
∂ 2 ci, j 2 ∂ ci, j ∂ 2 ci, j
2
j j
1 x1i x2 ci, j = x1i x2 y12 + y − y y
1 2
i=0, j=0 i=0, j=0 ∂ y12 2
∂ y22 ∂ y1 ∂ y2

∞
j
+ y12 i(i − 1)x1i−2 x2 ci, j
i=2, j=0


∞
j−2
+ y22 j( j − 1)x1i x2 ci, j ,
i=0, j=2
210 The Gelbart–Jacquet lift

  


∞

∞
∂ 2 ∂ ci, j ∂ 2 ci, j
2 2
j j ci, j
1 x1i x2 ci, j = x1i x2 y12 + y2 − y1 y2
i=0, j=0 i=0, j=0 ∂ y12 ∂ y22 ∂ y1 ∂ y2

∞
j
+ y12 (i + 2)(i + 1)x1i x2 ci+2, j
i=0, j=0

∞
j
+ y22 ( j + 2)( j + 1)x1i x2 ci, j+2 .
i=0, j=2

For fixed (i, j) we have

∂ 2 ci, j ∂ 2 ci, j ∂ 2 ci, j

y12 + y22 − y1 y2 + y12 (i + 2)(i + 1)ci+2, j
∂ y12
∂ y22 ∂ y1 ∂ y2
+ y22 ( j + 2)( j + 1)ci, j+2 − γ1 ci, j = 0. (7.1.20)

Now assume for fixed j, we have ci, j = 0, this is certainly true for all i
provided j = 0 or 1. Then from (7.1.20), it follows that ci, j+2 = 0. Further,
from (7.1.19) and induction on j, we see that ci, j = 0 for all i and j. This
completes the proof of the converse Theorem 7.1.3.

7.2 Rankin–Selberg convolution for GL(2)

Let

&
f (z) = a(n) 2π y · K ν f − 12 (2π |n|y) · e2πinx , (7.2.1)
n=0

&
g(z) = b(n) 2π y · K νg − 12 (2π |n|y) · e2πinx , (7.2.2)
n=0

be Maass forms of type ν f , νg , respectively, for S L(2, Z) as in Proposition 3.5.1.

Both Rankin (1939) and Selberg (1940) independently found the meromorphic
continuation and functional equation of the convolution L-function

∞
a(n)b(n)
L f ×g (s) = ζ (2s) . (7.2.3)
n=1
ns

They introduced, for the first time, the bold idea that L f ×g (s) can be constructed
explicitly by taking an inner product of f · ḡ with an Eisenstein series. This
beautiful construction has turned out to be extraordinarily important and has had
many ramifications totally unforeseen by the original discoverers. The Rankin–
Selberg convolution (for the case of Maass forms on S L(2, Z)) and its proof
are given in the following theorem.
 7.2 Rankin–Selberg convolution for GL(2) 211

Theorem 7.2.4 (Rankin–Selberg convolution) Let f (z), g(z) be Maass

forms of type ν f , νg , for S L(2, Z) as in (7.2.1), (7.2.2). Then L f ×g (s), defined
in (7.2.3), has a meromorphic continuation to all s ∈ C with at most a simple
pole at s = 1. Further, we have the functional equation
 f ×g (s) = π −2s G ν f ,νg (s)L f ×g (s) =  f ×g (1 − s).
where G ν f ,νg (s) is the product of four Gamma factors
       
s +1 − ν f − νg s +ν f − νg s − ν f + νg s − 1+ν f + νg
    .
2 2 2 2
If
  
α p −1

α p −1
L f (s) = 1− s 1− s
p p p
and
  
β p −1

β p −1
L g (s) = 1− s 1− s ,
p p p
then
 αpβp
−1 
α p β p
−1 
α p β p
−1 
α p β p
−1
L f ×g (s) = 1− 1− 1− 1− .
p ps ps ps ps

Proof Let E(z, s) denote the Eisenstein series given in Definition 3.1.2. We
compute, for (s) sufficiently large,   the inner product
d xd y
ζ (2s) f ḡ, E(∗, s̄) = ζ (2s) f (z) g(z) · E(z, s̄)
y2
S L(2,Z)\h2

ζ (2s)
d xd y
= f (z) g(z) Is (γ z)
2 γ ∈ \S L(2,Z) y2
∞
 S L(2,Z)\h2
ζ (2s) d xd y
= f (z) g(z) y s
2 y2
∞ \h2
∞ 1
ζ (2s) d xd y
= f (z) g(z) y s
2 y2
0
0
= πζ (2s) a(n)b(n)
n=0
∞
dy
× K ν f − 12 (2π|n|y)K νg − 12 (2π |n|y) y s
y
0
∞
dy
= (2π) 1−s
L f ×g (s) · K ν f − 12 (y)K νg − 12 (y) y s ,
y
0
212 The Gelbart–Jacquet lift

where
∞
dy
K ν f − 12 (y)K νg − 12 (y) y s
y
0        
s+1−ν f −νg s+ν f −νg s−ν f +νg s−1+ν f +νg
 2  2  2  2
=2 s−3
.
(s)
This computation gives the meromorphic continuation of L f ×g (s). By
Theorem 3.1.10, the Eisenstein series E(z, s) has a simple pole at s = 1 with
residue 3/π. It follows that L f ×g (s) has a simple pole at s = 1 if and only if
 f, g = 0. Finally, the functional equation for  f ×g is a consequence of the
functional equation for the Eisenstein series
E ∗ (z, s) = π −s (s)ζ (2s)E(z, s) = E ∗ (z, 1 − s)
given in Theorem 3.1.10.
Finally, it remains to prove the Euler product representation for L f ×g (s). We
shall actually prove a more general result (see (Bump, 1987)). Let u, v : Z −→C
be functions. Let z be a complex variable. Assume that

∞
u(n)z n = (1 − αz)−1 (1 − α  z)−1 ,
n=0


∞
v(n)z n = (1 − βz)−1 (1 − β  z)−1 ,
n=0

and for |z| sufficiently small, the above series converge absolutely. We will
show that

∞
1 − αα  ββ  z 2
u(n)v(n)z n = . (7.2.5)
n=0
(1 − αβz)(1 − α  βz)(1 − αβ  z)(1 − α  β  z)
In fact, let us now show that (7.2.5) implies the Euler product representation of
L f ×g in Theorem 7.2.4. Recalling (7.2.3), it follows that
∞ 
+ n n ) p −ns
 n=0 a( p )b( p
L f ×g (s) = . (7.2.6)
p (1 − p −2s )

One may now easily check that our result follows from (7.2.6) and (7.2.5) if we
choose
u(n) = a( p n ), v(n) = b( p n ), z = p −s .
To prove (7.2.5), let us define


∞

∞
U (z) := u(n)z n , V (z) := v(n)z n ,
n=0 n=0
 7.3 Statement and proof of the Gelbart–Jacquet lift 213

and consider the integral


1  dz 1
U (z · z 1 ) V z 1−1 ,
2πi C z1
where C is the circle |z 1 | = . Here,  > 0, is sufficiently small so that the poles
of U (zz 1 ) are outside C and the poles of V (z 1−1 ) are inside. It follows that

∞
1  dz 1
U (z · z 1 ) V z 1−1 = u(n)v(n)z n .
2πi C z1 n=0

Thus, to prove (7.2.5), it is sufficient to show that


1  −1  −1 dz 1
(1 − αzz 1 )−1 (1 − α  zz 1 )−1 1 − βz 1−1 1 − β  z 1−1
2πi C z1
1 − αα  ββ  z 2
= .
(1 − αβz)(1 − α  βz)(1 − αβ  z)(1 − α  β  z)
This is easily proved, however, because the left-hand side is just equal to the
sum of the residues at the poles: z 1 = β and z 1 = β  , and summing these gives
exactly the right-hand side of the above formula.

This completes the proof of Theorem 7.2.4.

7.3 Statement and proof of the Gelbart–Jacquet lift

The functional equation of a Maass form of type (ν1 , ν2 ) for S L(3, Z) is given
in Theorem 6.5.15 and involves a product of three Gamma factors:
     
s + 1 − 2ν1 − ν2 s + ν 1 − ν2 s − 1 + ν1 + 2ν2
   .
2 2 2
If the Maass form is self dual then ν1 = ν2 = ν, say, and the Gamma factors
take the simpler form:
     
s + 1 − 3ν s s − 1 + 3ν
   .
2 2 2
In this case, if L ν (s) denotes the L-function associated to the S L(3, Z) Maass
form, then its functional equation will be
     
s + 1 − 3ν s s − 1 + 3ν
ν (s) := π −3s/2    L ν (s) = ν (1 − s).
2 2 2
(7.3.1)
This represents the simplest type of functional equation we may have on
S L(3, Z).
An intriguing question that arises in the mind of researchers first encoun-
tering this field is whether there might be any candidate L-functions occurring
214 The Gelbart–Jacquet lift

somewhere in the G L(2) theory which have the functional equation (7.3.1). If
one could prove that all the twists by Dirichlet characters of such a candidate
L-function satisfy the functional equations and other conditions given in the
converse Theorem 7.1.3, then the candidate L-function would have to be asso-
ciated to a Maass form on S L(3, Z). The fact that this phenomenon occurs is
the substance of the Gelbart–Jacquet lift. In fact, if f is a Maass form of type
ν f for S L(2, Z), then by Theorem 7.2.4, and the functional equation
s   
−s/2 −(1−s)/2 1−s
π  ζ (s) = π  ζ (1 − s)
2 2
of the Riemann zeta function, we see that
     
s + 1 − 2ν f s s − 1 + 2ν f L f × f (s)
π −3s/2   
2 2 2 ζ (s)
     
2 − s − 2ν 1 − s −s + 2ν f L f × f (1 − s)
= π − 2 (1−s) 
3 f
  ,
2 2 2 ζ (1 − s)
which exactly matches the functional equation of a self-dual Maass form on
S L(3, Z) of type (2ν f /3, 2ν f /3). We now state and prove the Gelbart–Jacquet
lift from S L(2, Z) to S L(3, Z).
Theorem 7.3.2 (Gelbart–Jacquet lift) Let f be a Maass form of type ν f for
S L(2, Z) with Fourier expansion (7.2.1). Assume that f is an eigenfunction
of all the Hecke operators. Let L f × f (s) be the convolution L-function as in
(7.2.3). Then
L f × f (s)
ζ (s)
is the Godement–Jacquet L-function of a self-dual Maass form of type
(2ν f /3, 2ν f /3) for S L(3, Z).
Proof Let us define, for (s) sufficiently large,
L f × f (s)
∞
L(s) := = A(n, 1)n −s . (7.3.3)
ζ (s) n=1
We need to show that if we define A(n, 1) = A(1, n) for all n = 1, 2, 3, . . . ,
then for (s) sufficiently large,
 −1
L(s) = 1 − A( p, 1) p −s + A(1, p) p −2s − p −3s , (7.3.4)
p

and, in addition, we also must show that for every primitive Dirichlet character
χ , the twisted L-function


∞
L χ (s) := A(n, 1)χ (n)n −s
n=1
 7.3 Statement and proof of the Gelbart–Jacquet lift 215

satisfies the EBV condition and functional equation given in the converse
Theorem 7.1.3. The proof will be accomplished in three independent
steps.

Proof of the Gelbart–Jacquet lift (Euler Product), Step I Consider the L-

function, L f (s), associated to the Maass form f . Since f is an eigenfunction
of all the Hecke operators we know by (3.13.2) that, for (s) sufficiently large,
L f (s) has a degree two Euler product
  
α p −1

α p −1
L f (s) = 1− s 1− s
p p p

where α p · α p = 1 for every rational prime p.

Lemma 7.3.5 For (s) sufficiently large, the L-function, L(s), (defined in
(7.3.3)) has a degree three Euler product

−1  
−1
 α 2p α p α  p −1 α  2p
L(s) = 1− s 1− 1− s .
p p ps p

Proof This follows immediately from the Euler product representation in

)
Theorem 7.2.4 because the term p (1 − (α p α p / p s ))−1 just corresponds to
ζ (s) since α · α  = 1.

Note that Lemma 7.3.5 immediately implies (7.3.4).

Proof of the Gelbart–Jacquet lift (Functional Equations), Step II In order to

show that the candidate L-function
L f × f (s)
∞
L(s) := = A(n, 1)n −s ,
ζ (s) n=1

originally defined in (7.3.3) is actually the Godement–Jacquet L-function of

a Maass form for S L(3, Z) it is necessary to show that for every primitive
Dirichlet character χ , the twisted L-function


∞
L χ (s) = A(n, 1)χ (n)n −s ,
n=1

satisfies the functional equation specified in the converse Theorem 7.1.3. Such
functional equations, in some special cases of holomorphic modular forms,
were first considered by Rankin (1939). This method was later generalized in
(Li, 1975), (Atkin and Li, 1978), (Li, 1979). Also, Manin and Pančiškin (1977)
obtained the required functional equations for the case of χ (mod q) where q is
216 The Gelbart–Jacquet lift

a prime power. An adelic version of the Rankin–Selberg method for G L(2) was
given in (Jacquet, 1972), but the requisite functional equations are not given
in explicit form. We briefly sketch the method for obtaining such functional
equations.

Interlude on automorphic forms for 0 (N ) It is necessary, at this point,

to introduce automorphic forms for the congruence subgroup
   
a b 
0 (N ) := 
∈ S L(2, Z)  c ≡ 0 (mod N )
c d

with N = 1, 2, 3, . . . A function F : h2 −→ C is said to be automorphic for

0 (N ), with character ψ, if it satisfies the automorphic condition F(γ z) =
ψ(γ )F(z) for all γ ∈ 0 (N ), z ∈ h2 , and for some character ψ of the group
0 (N ). Note that if χ is a Dirichlet character (mod N ), then
 
a b
ψ := χ(d)
c d
will always
 be a character
 of 0 (N ). To simplify notation, we will sometimes
a b
write χ to denote χ (d). The cusps of 0 (N ), denoted by Gothic
c d
letters a, b, c, . . . , are defined to be elements of Q ∪ {i∞}. Two cusps a, b,
are termed equivalent if there exists γ ∈ 0 (N ) such that b = γ a. The stability
group of a cusp a is an infinite cyclic group, γa, generated by some γa ∈ 0 (N )
which is defined by
 
γa := γ ∈ 0 (N )  γ a = a .

Then there exists a scaling matrix σa ∈ S L(2, Q) (which is determined up to

right multiplication by a translation) by the conditions
 
1 1
σai∞ = a, σa−1 γaσa = .
0 1
If F(z) is automorphic for 0 (N ) and a is any cusp, then F(σa z) is invariant
under the translation z → z + 1. It, therefore, has a Fourier expansion


F(σa z) = Aa(n, y)e2πinx ,
n∈Z

with Aa(n, y) ∈ C for all n ∈ Z, y > 0. We say F is a cusp form if Aa(m, y) = 0

for all cusps a, every m ≤ 0, and all y > 0. If, in addition, F is an eigenfunction
of the Laplacian −y 2 ((∂ 2 /∂ 2 x) + (∂ 2 /∂ 2 y)), then F is said to be a Maass form.
If F does not come from a form of lower level then we say F is a newform.
More precisely, if F is automorphic for 0 (N ) then for r = 2, 3, 4, . . . , F(r z)
 7.3 Statement and proof of the Gelbart–Jacquet lift 217

will be automorphic for 0 (r N ) because of the identity

     
r 0 a b a br r 0
= .
0 1 c d c/r d 0 1
In this case F(r z) is called an old form coming from F which lives on level N
which is a lower level than r N .
The key ideas for obtaining meromorphic continuation and functional equa-
tions of the twisted L-functions L χ (s) are based on the following two lemmas.
Lemma 7.3.6 Let

&
F(z) = A(n) 2π y K ν− 12 (2π|n|y) e2πinx
n=0

be a Maass form of type ν for 0 (N ) with trivial character. Let χ be a primitive

character (mod q). Then

&
Fχ (z) = A(n)χ (n) 2π y K ν− 12 (2π |n|y) e2πinx
n=0

is a Maass form for 0 (N q 2 ) with character χ 2 . If N = 1, then

 
−1 τ (χ )
Fχ 2 = Fχ̄ (z).
q z τ (χ̄ )

Proof Since χ is primitive, the Gauss sum τ (χ̄ ), given by (7.1.1), cannot be
zero. It also immediately follows from (7.1.1) that

q  
−1
Fχ (z) = τ (χ̄ ) χ̄ ( ) · F z + .
=1
q
 
a b
Assume that γ = ∈ 0 (N q 2 ). We have the following matrix identity:
c d
⎛ ⎞
   a + c b − (ad − 1) d − cdq 2  1 d 2 /q 
2 2

1 /q a b
=⎝ ⎠
q q
.
0 1 c d d − cd 0 1
2
c q

It follows that


q  
d 2
Fχ (γ z) = τ (χ̄ )−1 χ̄ ( ) · F z +
=1
q

q  
d 2
= τ (χ̄ )−1 χ̄ ( d 2 )χ (d)2 · F z +
=1
q
= χ (d)2 Fχ (z).

This shows that Fχ is automorphic for 0 (N q 2 ) with character χ 2 .

218 The Gelbart–Jacquet lift

For the last part, suppose that ( , q) = 1. Then there exist integers r, s such
that rq − s = 1. Further, we have the matrix identity
      
1 /q 0 −1 0 −q q −s 1 s/q
= .
0 1 q2 0 q 0 − r 0 1
     
0 −q 0 −1
Note that F z =F z = F(z) for all z. Setting s = ¯
q 0 1 0
where ¯ ≡ 1 (mod q), it follows from the prior matrix identity that
 
q    
−1 1 /q 0 −1
τ (χ̄ )Fχ 2 = χ̄ ( )F z
q z =1
0 1 q2 0

q   
1 /q
¯
= χ̄ ( )F z
=1
0 1

q   
1 /q
= χ ( )F z
=1
0 1
= τ (χ )Fχ̄ (z).

Lemma 7.3.7 Let χ be an even primitive Dirichlet character (mod q). The
Eisenstein series
1
ys
E(z, s, χ) = χ (d)
2 (c,d)=1 |cz + d|2s
c ≡ 0 (mod q 2 )

is an automorphic form for 0 (q 2 ) with character χ . For any fixed z ∈ h2 , the

function

5 s
∗ q2
E (z, s, χ) := L(2s, χ)(s)E(z, s, χ)
π

is entire in s and satisfies the functional equation

 
τ (χ ) −1
E ∗ (z, s, χ ) = √ · E ∗ 3 , 1 − s, χ̄ .
q q z

Proof It is easy to see that E(z, s, χ) is automorphic when it is written in the

form:
1

E(z, s, χ ) = χ (γ ) (γ z)s .
2 γ ∈ \ (q 2 )
∞ 0
 7.3 Statement and proof of the Gelbart–Jacquet lift 219

+
Let w = u + iv with v > 0. The function |m + w|−2s is periodic in u
m∈Z
and one may easily derive the Fourier expansion

(2s − 1)
|m + w|−2s = 2π |2v|1−2s
m∈Z
(s)2
2π s
1 −s s− 1
+ |v| 2 |m| 2 K s− 12 (2π |m|v)e2πimu .
(s) m=0

It now follows from the definition of the Eisenstein series and the above Fourier
expansion (after making the substitution d = mcq 2 + r ) that

L(2s, χ)E(z, s, χ )


∞

ys
= y s L(2s, χ ) + χ (d) ·
c=1 d∈Z
|cq 2 z + d|2s
 s

∞

 −2s
2
y
cq

= y s L(2s, χ ) + |c|−2s χ (r ) m + z + r 
q4  cq 2 
c=1 r =1 m∈Z
 s

∞

cq 2
y
= y s L(2s, χ ) + |c|−2s χ (r )
q4 c=1 r =1
0
(2s − 1) 2π s
 y  12 −s
× 2π |2y| 1−2s
+  
(s)2 (s) m=0 m
 1
2πim x+ r 2
× K s− 12 (2π|m|y)e cq

√
2π s y
1
= y s L(2s, χ ) + |m|s− 2 K s− 12 (2π |m|y)e2πimx
q 4s (s) m=0

∞

2
cq
2πimr
−2s
× |c| χ (r )e cq 2 . (7.3.8)
c=1 r =1

Next, we show that




2
cq
2πimr/cq 2 τ (χ )χ̄ ( ) · qc if m = cq,
χ (r )e = (7.3.9)
r =1 0 otherwise.

In fact, this follows easily because every r in the above sum is of the form
r = r1 + tq with 1 ≤ r1 ≤ q and 0 ≤ t < cq. Hence, the sum takes the form

q qc−1 2πir1 m 2πimt
χ (r1 )e cq 2 e cq

r1 =1 t=0
220 The Gelbart–Jacquet lift

and the inner sum over t above is zero unless m = cq for some ∈ Z. Com-
bining (7.3.8) and (7.3.9), we obtain
L(2s,χ )E(z, s,χ )
√
2π s y
σ2s−1 (m, χ̄ )
= y L(2s, χ ) + τ (χ )
s
K s− 12 (2π |m|qy)e2πimqx ,
3s− 12 1
q (s) m=0 |m|s− 2
(7.3.10)
where

 m s
 
σs (m, χ ) = χ (m/ ) ·   .
|m

≥1

Next, we compute
 the Fourier
 expansion around z = 0. We make use of the
0 −1
fact that wq = is a normalizer for 0 (q 2 ). This means that
q2 0

wq 0 (q 2 )wq−1 = 0 (q 2 ),
which can be easily seen from the calculation
     
0 −1 a b 0 q −2 d −c
= ∈ 0 (q 2 ).
q2 0 cq 2 d −1 0 −bq 2 a
Thus, for every γ ∈ ∞ \0 (q 2 ), there exists a unique γ  ∈ ∞ \0 (q 2 ) such
that γ wq = wq γ  . It follows, as in the computation for (7.3.8), that
 

−1 1
L(2s,χ)E 2 , s, χ = L(2s, χ) χ (γ ) (γ wq z)s
q z 2 γ ∈ \ (q 2 ) ∞ 0

1

= L(2s, χ) χ̄ (γ ) (wq γ z)s
2 γ ∈∞ \0 (q 2 )

 s
L(2s, χ ) −1
= χ̄ (d)  2
2   q · (az + b)/(cz + d)
a b
∈ ∞ \0 (q 2 )
c d
a 
ys
∞

χ (a) ys
∞
χ (a)

 r −2s
= = z + m + 
q 2s a=1 b∈Z |az + b|2s q 2s a=1 a 2s m∈Z r =1 a
2π 21−2s y 1−s (2s − 1)
= L(2s − 1, χ )
q 2s (s)2
√
2π s y
1
+ 2s σ2s−1 (m, χ )K s− 12 (2π|m|y)e2πimx .
q (s) m=0 |m|s− 12

The functional equation follows after comparing the above with (7.3.10).

 7.3 Statement and proof of the Gelbart–Jacquet lift 221

Finally, we return to the problem of determining the functional equation of

L f × f (s, χ) L f × fχ̄ (s)
L χ (s) = =
L(s, χ ) L(s, χ)
where f is a Maass form for S L(2, Z). By Lemma 7.3.6, we know that f χ is a
Maass form for 0 (q 2 ) with character χ 2 . It follows, as in the proof of Theorem
7.2.4, that we may construct L f × fχ̄ (s) as a Rankin–Selberg convolution

 f, f χ̄ · E(∗, s, χ 2 ),

proving that L f × fχ̄ (s) has a meromorphic continuation to all s ∈ C. If χ 2 is a

primitive character, then we have the functional equation of Lemma 7.3.7. One
may then easily show that L χ (s) has exactly the functional equation given in
the converse Theorem 7.1.3. This method will also work if χ 2 is not primitive,
and we leave the details to the reader.

Proof of the Gelbart–Jacquet lift (EBV condition), Step III In Step II we have
shown how to obtain the meromorphic continuation and functional equation for
the twisted L-function

∞
L f × f (s, χ)
L χ (s) = A(n, 1)χ (n)n −s = .
n=1
L(s, χ )
It does not follow from these methods that L χ (s) is entire, however. The problem
is that we do not know that L(s, χ ) divides L f × f (s, χ), and, for all we know,
L χ (s) could have poles at all the zeros of L(s, χ ).
With a brilliant idea, Shimura (1975) was able to show that L(s, χ ) always
divides L f × f (s, χ), so that L χ (s) is entire for all Dirichlet characters χ . We
now explain Shimura’s idea.
It follows from the methods used to prove Lemma 7.3.5 that


∞

∞
L(s, χ) χ (n)a(n 2 )n −s = χ (n)a(n)2 n −s , (7.3.11)
n=1 n=1

where a(n) is the nth Fourier coefficient of f as in (7.2.1). Here, notice the
difference from a(n 2 ) to a(n)2 . Shimura showed that


∞
χ (n)a(n 2 )n −s
n=1

is entire by considering the Rankin–Selberg convolution of f with a theta

function
1
∞
χ̄ (n)n δ e2πin z ,
2
θχ (z) :=
2 n=−∞
222 The Gelbart–Jacquet lift

where δ = 0 or 1 according as χ (−1) = 1 or −1. It is known (see (Shimura,

1973)) that the theta function is an automorphic form for 0 (4q 2 ) where χ is
a primitive Dirichlet character (mod q). The automorphic relation takes the
explicit form
 
−1 δ  c  2δ+1
(cz + d)δ+ 2 θχ (z),
1
θχ (γ z) = χ̄ (d) 
d d d
   
a b
∀γ = ∈ 0 (4q ) ,
2
c d

where dc is the Kronecker symbol and

1 if d ≡ 1 (mod 4)
d =
i if d ≡ 3 (mod 4).

Note that θχ is automorphic of half-integral weight. The multiplier (factor of

automorphy)
 
−1 δ  c  2δ+1
ω(γ ) :=  ,
d d d
√
assures that we are always on the same branch of the square root cz + d
function. In order to proceed with the Rankin–Selberg convolution, we need
to construct an appropriate Eisenstein series which transforms with the same
half-integral weight multiplier system. For example, we may construct


ω(γ ) χ̄ (d)(cz + d)−δ− 2 (γ z)s .
1
Ẽ(z, s, χ) =
γ ∈∞ \0 (4q 2 )
a b
γ= c d

It is known that Ẽ(z, s, χ ) has analytic continuation in s ∈ C with at most a

simple pole at s = 1 − ((2δ + 1)/4). The Rankin–Selberg unfolding method
gives
 ∞ 1
d xd y d xd y
f (z) θχ (z) Ẽ(z, s, χ ) = f (z) θχ (z) y s
y2 y2
0 (4q 2 )\h2 0 0

√

∞ ∞
a(n 2 )n δ χ (n) 1 dy
= 2π 1 K ν f − 12 (y) y s− 2 .
n=1 (2πn 2 )s− 2 y
0

It now follows from this and (7.3.11) that

L f × f (s, χ)
L f (s, χ)
 7.4 Rankin–Selberg convolution for G L(3) 223

has at most a simple pole at s = 34 − 2δ . Actually, such a pole can only occur if
χ is the trivial character, and in this case, we know from the Rankin–Selberg
convolution in Step II that it cannot occur. The residue of the half-integral
1
weight Eisenstein series is, up to a constant factor, y 2 θ(z), where


∞
2
θ(z) = e2πin z

n=−∞

denotes the classical theta function. As a consequence, one obtains the inter-
esting result that

1 d xd y
f (z) |θ(z)|2 y 2 = 0.
y2
0 (4q 2 )\h2

Finally, from the growth properties of Ẽ(z, s, χ ), one may obtain the EBV
(entire and bounded in vertical strips of fixed width) conditions needed for the
completion of the proof of the Gelbart–Jacquet lift.

7.4 Rankin–Selberg convolution for G L(3)

The Rankin–Selberg convolution has been extended to automorphic forms on
G L(n) by Jacquet, Piatetski-Shapiro and Shalika (1983). A sketch of the method
in the classical setting had been given in (Jacquet, 1981). We now work out,
following (Friedberg, 1987b), the convolution for the special case of the group
S L(3, Z).
It is necessary to introduce the maximal parabolic Eisenstein series for
S L(3, Z). Let
⎛ ⎞
∗ ∗ ∗
ˆ = ⎝ ∗ ∗ ∗ ⎠ ∈ S L(3, Z).
0 0 1

Lemma 7.4.1 The cosets of \S ˆ L(3, Z) are in one-to-one correspondence

ˆ → last row of γ .
with the relatively prime triples of integers via the map: γ

Proof One easily verifies that the map is well defined and injective. It is
surjective because every relatively prime triple of integers can be completed to
a matrix in S L(3, Z).

Lemma 7.4.2 Let z ∈ h3 have the representation

⎛ ⎞⎛ ⎞
1 x2 x3 y1 y2
z=⎝ 1 x1 ⎠⎝ y1 ⎠.
1 1
224 The Gelbart–Jacquet lift

⎛ ⎞
∗ ∗ ∗
Let γ = ⎝ ∗ ∗ ∗ ⎠ be a representative for the coset \S
ˆ L(3, Z). Then
a b c
Det(z)
Det(γ z) =  32 ,
y12 |az 2 + b|2 + (ax3 + bx1 + c)2
where z 2 = x2 + i y2 , and Det : h3 → R+ denotes the determinant of a matrix,
where the matrix is in Iwasawa canonical form as in Proposition 1.2.6.
Proof Brute force computation.

We now introduce the Eisenstein series

Det(γ z)s
E(z, s) := .
2
γ ∈\S
ˆ L(3,Z)

Lemmas 7.4.1 and 7.4.2 then imply that the series above converges absolutely
for (s) > 1, and has the explicit representation
 2 s
1
y1 y2
E(z, s) =    . (7.4.3)
2 a,b,c ∈ Z y 2 az 2 + b2 + (ax3 + bx1 + c)2 3s/2
1
(a,b,c)=1

The most important properties of E(z, s) are given in the next proposition.
Proposition 7.4.4 The Eisenstein series E(z, s) has a meromorphic continu-
ation to all s ∈ C and satisfies the functional equation
   
−3s/2 3s −(3−3s)/2 3 − 3s
π  ζ (3s)E(z, s) = π  ζ (3 − 3s) Ẽ(z, 1 − s),
2 2
where Ẽ(z, 1 − s) = E(w t z −1 w, s) is the dual Eisenstein series, and w is the
long element of the Weyl group, as in Proposition 6.3.1. Furthermore,
 
∗ −3s/2 3s
E (z, s) := π  ζ (3s)E(z, s),
2
is holomorphic except for simple poles at s = 0, 1, with residues −2/3, 2/3,
respectively.
Proof The results will follow from the Fourier–Whittaker expansion in exactly
the same way in which they were obtained for Eisenstein series on S L(2, Z)
in Theorems 3.1.8 and 3.1.10. We break the sum for E ∗ (z, s), given in (7.4.3),
into two pieces corresponding to a = 0 and a = 0. It follows that E ∗ (z, s)
= E 1∗ (z, s) + E 2∗ (z, s) where
   2 s
3s
y1 y2
∗ −3s/2
E 1 (z, s) = π   2 ,
2 b,c ∈ Z2 −(0,0) y b2 + (bx1 + c)2 3s/2
1
 7.4 Rankin–Selberg convolution for G L(3) 225

and


∞

∞ ∞ ∞
 s
E 2∗ (z, s) =2 y12 y2
a=1 j, =−∞
−∞ −∞
−3s/2
(3s/2) · e−2πi( ju+ v)
π
×  2 3s/2 dudv. (7.4.5)
y 2 az 2 + u  + (ax3 + ux1 + v)2
1
 1 s
But E 1∗ (z, s) is y1 y2 times an S L(2, Z) Eisenstein series with Fourier expan-
2

sion given by Theorem 3.1.8 from which it follows that

  
 1 s 3s/2 3s
E 1∗ (z, s) = 2 y 2 y2 y1 π −3s/2  ζ (3s)
2
 
1−(3s/2) (1−3s)/2 3s − 1
+ y1 π  ζ (3s − 1)
2

1

+ 2y12
|m 1 |(3s−1)/2
σ1−3s (|m 1 |)K 2
3s−1 (2π |m|y1 )e 2πim 1 x1
.
m 1 =0
(7.4.6)
Now E ∗ (z, s) has a Fourier–Whittaker expansion of type




  
γ
E ∗ (z, s) = φ0,m 2 (z) + φm 1 ,m 2 z ,
m 2 ∈Z γ ∈ ∞ \S L(2,Z) m 1 =0 m 2 ∈Z
1
(7.4.7)
where
⎛⎛ ⎞ ⎞
1 1 1 1 ξ2 ξ3
φm 1 ,m 2 (z) = E ∗ ⎝⎝ 1 ξ1 ⎠z ⎠e−2πi (m 1 ξ1 +m 2 ξ2 ) dξ1 ξ2 dξ3 .
0 0 0 1
We immediately observe that (7.4.6) only contributes to the Fourier expan-
sion (7.4.7) when m 2 = 0. On the other hand, after some change of variables
in (7.4.5), the contribution of E 2∗ (z, s) to the Fourier coefficient φm 1 ,m 2 is given
by
  1 1 1

∞

∞
−3s/2 3s  s
2π  · y12 y2
2 a=1 j, =−∞
0 0 0

× e2πi ( a(x3 +ξ3 +ξ2 x1 )+( j− (x1 +ξ1 ))a(x2 +ξ2 )−m 1 ξ1 −m 2 ξ2 )
∞ ∞ 2πi (− v+ay2 u( (x1 +ξ1 )− j))
e
× ay2  2 2 3s/2 dudv dξ1 dξ2 dξ3 .
−∞ −∞
y1 y2 a 2 (u 2 + 1) + v 2
226 The Gelbart–Jacquet lift

But this vanishes unless = 0, and, in this case, m 1 = 0 also. Further, if

m 2 = 0, we get a contribution only when a j = m 2 , and when m 2 = 0, we
must have j = 0. Consequently everything simplifies to
 
3s  2 s 2πim 2 x2
δm 1 ,0 · 2π −3s/2  y1 y2 e
2

∞ ∞
e−2πim 2 y2 u
× ay2  2 2 3s/2 dudv,
a|m 2
−∞ −∞
y1 y2 a 2 (u 2 + 1) + v 2

when m 2 = 0, and to
 
−3s/23s  2 s
δm 1 ,0 · 2π  y1 y2
2


∞ ∞ ∞
 2 2 2 2 −3s/2
× ay2 y1 y2 a (u + 1) + v 2 dudv,
a=1
−∞ −∞

when m 2 = 0. It follows that the Fourier coefficient φm 1 ,m 2 (z) is given by

⎧  s 1
⎪
⎪
1 3s−1
4 y 2 y2 y12 |m 1 | 2 σ1−3s (|m 1 |)K 3s−1 (2π|m 1 |y1 )e2πim 1 x1 ,
⎪
⎪
⎪
⎪
2

⎪
⎪ if m 1 = 0, m 2 = 0,
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎨ 1−s 1−( 2s )
|m 2 | 2 −1 σ2−3s (|m 2 |)K 3s2 −1 (2π|m 2 |y2 )e2πim 2 x2 ,
3s
4y1 y2
⎪
⎪ if m 1 = 0, m 2 = 0,
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪  
⎪
⎪
−3s 1−3s
2y12s y2s π 2  3s2 ζ (3s) + 2y11−s y2s π 2  3s−1 ζ (3s − 1)
⎪
⎪
⎪
⎩ 1−s 2−2s 1− 3s2
 3s 2
+ 2y1 y2 π  2 − 1 ζ (3s − 2), if m 1 = m 2 = 0.
The meromorphic continuation and functional equation of E(z, s) are an
immediate consequence of the above explicit computation of the Fourier
coefficients given in (7.4.7).

The Rankin–Selberg convolution for S L(3, Z) is very similar to the convo-

lution for S L(2, Z) given in Theorem 7.2.4. The idea is to take the inner product
of two Maass forms with the Eisenstein series (7.4.3). Let


∞
   
A(m 1 , m 2 ) γ
f (z) = WJacquet M z, ν, ψ1, |m2 | ,
m

γ ∈U (Z)\S L(2,Z) m 1 =1 m 2 =0
|m 1 m 2 | 1 2
2

(7.4.8)



∞
B(m 1 , m 2 )    
γ
g(z) = WJacquet M z, ν  , ψ1, |mm2 | ,
γ ∈U2 (Z)\S L(2,Z) m 1 =1 m 2 =0
|m 1 m 2 | 1 2
 7.4 Rankin–Selberg convolution for G L(3) 227

be Maass forms of type ν, ν  , respectively, for S L(3, Z). Here M

⎛ ⎞
m 1 |m 2 |
=⎝ m 1 ⎠.
1

Theorem 7.4.9 (Rankin–Selberg convolution) Let f, g be Maass forms for

S L(3, Z) of types ν, ν  , as in (7.4.8). Then

∞
∞
A(m 1 , m 2 ) B(m 1 , m 2 )
L f ×g (s) := ζ (3s)
m 1 =1 m 2 =1 m 2s
1 m2
s

has a meromorphic continuation to all s ∈ C with at most a simple pole at

s = 1. The residue is proportional to  f, g, the Petersson inner product of f
with g.
Furthermore, if
∞ ∞
 s dy1 dy2
G ν,ν  (s) = WJacquet (y, ν, ψ1,1 ) · W̄ Jacquet (y, ν  , ψ1,1 ) y12 y2 ,
y13 y23
0 0

then L f ×g satisfies the functional equation

 
−3s/2 3s
 f ×g (s) = π  G ν,ν  (s)L f ×g (s) =  f˜×G̃ (1 − s),
2

where f˜, g̃ denote the dual Maass forms as in Proposition 6.3.1.

Proof We compute the Rankin–Selberg inner product of f · ḡ with the Eisen-

stein series E(z, s) given in (7.4.3). It follows that

 f · ḡ, E(∗, s̄) = f (z)g(z) · E(z, s̄) d ∗ z,
S L(3,Z)\h3

with d ∗ z, the invariant measure as given in Proposition 1.5.3. If we now apply

the usual unfolding trick, we obtain

1  s
 f · ḡ, E(∗, s̄) = f (z)g(z) · y12 y2 d ∗ z. (7.4.10)
2
\h
ˆ 3

Now
⎛ ⎞
∗ ∗ ∗
⎝
= ∗
ˆ ∗ ∗ ⎠ ⊂ S L(3, Z)
0 0 1
228 The Gelbart–Jacquet lift

is generated by matrices of type

⎛ ⎞ ⎛ ⎞
a b 0 1 0 e
⎝ c d 0 ⎠, ⎝0 1 f⎠
0 0 1 0 0 1

 a,b, c, d, e, f ∈ Z, ad − bc = 1. Consequently, if we define U2 (Z) =

with
1 ∗
⊂ S L(2, Z),
1
(  
γ ˆ 3 ∼
· \h = U3 (Z)\h3 , (7.4.11)
γ ∈ U (Z)\S L(2,Z)
1
2

where
⎛ ⎞
1 ∗ ∗
U3 (Z) = ⎝ 1 ∗ ⎠ ⊂ S L(3, Z).
1

This result was also obtained in Lemma 5.3.13. Since the Fourier expansion
of f (7.4.10) is given by a sum over S L(2, Z), we may unfold further using
(7.4.11) to obtain

 f · ḡ, E(∗, s̄)


∞

1 A(m 1 , m 2 )
= WJacquet (M y, ν, ψ1,1 )e2πi(m 1 x1 +m 2 x2 )
2 m 1 =1 m 2 =0
|m 1 m 2 |
U3 (Z)\h3


∞

B(m 1 , m 2 )
×
  m  =1 m  =0
|m 1 m 2 |
a b 1 2
∈U2 (Z)\S L(2,Z)
c d
6  7
az +b
−2πi m 1 (cx3 +d x1 )−m 2  cz 2+d
×e 2

⎛ ⎛ ⎞ ⎞
y1 y2
|cz 2 +d|
⎜ ⎜ ⎟  ⎟
× W̄ Jacquet ⎝ M  ⎝ y1 · |cz 2 + d| ⎠, ν , ψ1,1 ⎠
1
 s
× y12 y2 d ∗ z,

after applying Theorem 6.5.7 to g. Further,

 1 1 1 ∞ ∞
= .
U3 (Z)\h3 x1 =0 x2 =0 x3 =0 y1 =0 y2 =0
 7.4 Rankin–Selberg convolution for G L(3) 229

It then follows by the standard techniques introduced in Section 6.5 that

ζ (3s) f · ḡ, E(∗, s̄) = L f ×g (s) G ν,ν  (s).
The theorem is now a consequence of the functional equation of the Eisenstein
series given in Proposition 7.4.4.

We will end this section with an explicit computation of the Euler product
for L f ×g assuming that f, g are both eigenfunctions of the Hecke operators and
L f (s), L g (s) have Euler products as given in Definition 6.5.2.
Proposition 7.4.12 (Euler product) Let f, g be Maass forms for S L(3, Z)
with Fourier coefficients A(m 1 , m 2 ), B(m 1 , m 2 ), respectively, as in (7.4.8). Let

∞
∞
A(m 1 , m 2 ) B(m 1 , m 2 )
L f ×g (s) := ζ (3s)
m 1 =1 m 2 =1 m 2s
1 m2
s

be the Rankin–Selberg convolution. Assume L f , L g have Euler products:

 3    3  
αi, p −1 β j, p −1
L f (s) = 1− s , L g (s) = 1− s ,
p i=1 p p j=1 p

then

−1

3 
3
αi, p β j, p
L f ×g (s) = 1 − .
p i=1 j=1 ps

Proof The proof is based on a special case of Cauchy’s identity (Weyl, 1939),
(McDonald, 1979), (Bump, to appear), which takes the form:

3 
3

∞

∞
(1 − αi β j x)−1 = Sk1 ,k2 (α1 , α2 , α3 )Sk1 ,k2 (β1 , β2 , β3 )x k1 +2k2
i=1 j=1 k1 =0 k2 =0

× (1 − α1 α2 α3 β1 β2 β3 x 3 )−1 , (7.4.13)
where
 k1 +k2 +2 
 x1
 k +1 x2k1 +k2 +2 x3k1 +k2 +2 
 x 1 x2k1 +1 x3k1 +1 
 1
 1 1 1 
Sk1 ,k2 (x1 , x2 , x3 ) :=  2 
 x1 x22 x32 
 
 x1 x2 x3 
 
1 1 1

is the Schur polynomial given by a ratio of determinants. We shall defer the

proof of Cauchy’s identity for the moment. In the meantime, we will show
230 The Gelbart–Jacquet lift

that
 
A p k1 , p k2 = Sk1 ,k2 (α1, p , α2, p , α3, p ), B p k1 , p k2 = Sk1 ,k2 (β1, p , β2, p , β3, p ),
(7.4.14)

from which the proposition follows after choosing αi = αi, p , β j = β¯i, p

for (1 ≤ i, j ≤ 3) and x = p −s in (7.4.13).
To prove (7.4.14), we follow (Bump, 1984). We require the identity:


∞

∞
Sk1 ,k2 (α1 , α2 , α3 ) p −k1 s p −k2 s (7.4.15)
k1 =0 k2 =0

3 
= (1 − p −2s ) (1 − αi p −s )−1 (1 − αi α j p −s )−1 .
i=1 1≤i< j≤3

This is proved by noting that the ratio of determinants in the definition of the
Schur polynomial can be computed explicitly to yield

Sk1 ,k2 (α1 , α2 , α3 )

  
α k1 +k2 +2 α2k1 +1 − α3k1 +1 + α2k1 +k2 +2 α3k1 +1 − α1k1 +1 + α3k1 +k2 +2 α1k1 +1 − α2k1 +1
= 1 .
(α2 − α1 )(α3 − α2 )(α1 − α3 )

If we then multiply both sides of the above formula by p −(k1 +k2 )s and sum over
k1 , k2 , then the identity (7.4.15) follows after some algebraic manipulations.
Now, by Section 6.5, we have

1− A(1, p) p −s + A( p, 1) p −2s − p −3s = (1 − α1 p −s )(1 − α2 p −s )(1 − α3 p −s ),

1 − A( p, 1) p −s + A(1, p) p −2s − p −3s = (1 − α2 α3 p −s )(1 − α3 α1 p −s )
× (1 − α1 α2 p −s ).

It follows that


∞

∞

A p k1 , p k2 p −(k1 +k2 )s
k1 =0 k2 =0


∞

∞

= (1 − p −2s ) A p k1 −k , p k2 −k p −(k1 +k2 )s
k1 =0 k2 =0 k≤min(k1 ,k2 )


∞

∞
 
= (1 − p −2s ) A p k1 , 1 A 1, p k2 p −(k1 +k2 )s
k1 =0 k2 =0
−2s
 −1
= (1 − p ) 1 − A( p, 1) p −s + A(1, p) p −2s − p −3s
 −1
× 1 − A(1, p) p −s + A( p, 1) p −2s − p −3s
 7.4 Rankin–Selberg convolution for G L(3) 231

−1 −1 −1
= (1 − p −2s )(1 − α1 p −s ) (1 − α2 p −s ) (1 − α3 p −s )
−1 −1 −1
× (1 − α2 α3 p −s ) (1 − α3 α1 p −s ) (1 − α1 α2 p −s )


∞

∞
= Sk1 ,k2 (α1 , α2 , α3 ) p −(k1 +k2 )s .
k1 =0 k2 =0

Finally, (7.4.14) is obtained by comparing coefficients on both sides.

It remains to prove Cauchy’s identity (7.4.13). Note that Schur’s polynomial
is a ratio of determinants. The matrix in the denominator is a Vandermonde
matrix, which is a matrix whose columns all have the form (1, x j , x 2j , . . . , x n−1
j )
for some x j and some n > 2. Now, if two of the x j s are equal, then two columns
of the matrix are the same and the determinant is zero. This means, that as a
polynomial, the determinant must have (xi − x j ) as a factor for all 1 ≤ i < j
≤ n. Consequently, the determinant of the Vandermonde matrix must have

(xi − x j ) (7.4.16)
1≤i< j≤n

as a factor. By comparing degrees one easily sees that the determinant of the Van-
dermonde matrix is the product (7.4.16) up to a constant factor. To show that the
constant factor is 1, one may consider the main diagonal term x1n−1 x2n−2 · · · xn1 xn0 ,
which is exactly what one gets by taking the first positive terms in each factor
of the product (7.4.16). This proves that
 n−1 
 x1 x2n−1 · · · xnn−1 
 n−2 
 x1 x2n−2 · · · xnn−2 

 .. .. ..  =  (x − x ).
 . . ··· .  i j (7.4.17)

 x ···  1≤i< j≤n
 1 x2 xn 
 1 1 ··· 1 
Next, we consider Cauchy’s determinant which is defined to be the deter-
minant of the n × n matrix whose i, jth entry is 1/(1 − xi y j ) where xi , y j
(1 ≤ i, j ≤ n) are variables and n ≥ 2. We write this matrix as
 
1
.
1 − xi y j 1≤i, j≤n

Lemma 7.4.18 (Cauchy’s determinant) Let n ≥ 2 and xi , yi ∈ C (1 ≤ i

≤ n). Then
) )

   (xi − x j ) (yi − y j )
1 1≤i< j≤n 1≤i< j≤n
Det = .
1 − xi y j 1≤i, j≤n )n )n
(1 − xi y j )
i=1 j=1
232 The Gelbart–Jacquet lift

Proof We know the determinant is a rational function. The determinant will

be zero if any two of the xi or any two of the y j are equal since, in this case, either
two of the rows or columns of the matrix would be the same. Consequently,
) )
the numerator of the determinant must have factors (xi − x j ) (yi − y j ).
Clearly, the denominator of the determinant must have (1 − xi y j ) as a factor
for every 1 ≤ i, j ≤ n. This suggests that the determinant should be a constant
multiple of
) )
(xi − x j ) (yi − y j )
1≤i< j≤n 1≤i< j≤n
. (7.4.19)
)
n )
n
(1 − xi y j )
i=1 j=1

In fact, the numerator of (7.4.19) has degree n(n − 1) while the denominator
has degree n 2 which is also the case for the determinant because all of its terms
are products of n factors, each having one term in the denominator and none
in the numerator with a net increase of n in the denominator. To show that the
constant multiple must be 1, let xi = −yi−1 for i = 1, 2, . . . , n. In this case,
(7.4.19) becomes
)
(yi − y j )2
1≤i< j≤n
2−n · ) .
(yi + y j )2
1≤i< j≤n

We may now fix y2 , y3 , . . . , yn and let y1 → ∞. The above expression collapses

to
)
(yi − y j )2
2≤i< j≤n
2−n · ) .
(yi + y j )2
2≤i< j≤n

Next, fix y3 , . . . , yn and let y2 → ∞. Continue in this manner. The expression

(7.4.19) turns into 2−n . On the other hand, all upper triangular terms of the
matrix become 0 after taking the above limits, while the diagonal terms are all
1
2
. So the value of Cauchy’s determinant must also be 2−n . This completes the
proof of Lemma 7.4.18.

We shall conclude this section by proving a more general version of Cauchy’s

identity (7.4.13). In order to do this, however, it is necessary to introduce a
more general Schur polynomial. Let n ≥ 2 and k = (k1 , k2 , . . . , kn−1 ) be a set
 7.4 Rankin–Selberg convolution for G L(3) 233

of n − 1 non-negative integers. We define

 k1 +k2 +···+kn−1 +n−1 
x k +k +···+kn−1 +n−1 
· · · xn1 2
 1 
 x k1 +k2 +···+kn−2 +n−2 · · · x k1 +k2 +···+kn−2 +n−2 
 1 n 
 
 .. .. .. 
 . . . 
 
 
 
 x1k1 +1
··· k1 +1
xn 
 
 1 ··· 1 
Sk (x1 , x2 , . . . , xn ) =  n−1  ,
 x1 x2n−1 · · · xnn−1 
 n−2 
 x1 x2n−2 · · · xnn−2 

 .. .. .. 
 . . ··· . 

 x ··· xn 
 1 x2
 1 1 ··· 1 

where the denominator is the Vandermonde determinant which has the value
given by (7.4.17).

Proposition 7.4.20 (Cauchy’s identity) Let n ≥ 2 and αi , βi ∈ C for every

i = 1, . . . , n. For x ∈ C and |x| sufficiently small, we have the identity

n 
n
(1 − αi β j x)−1
i=1 j=1


∞
∞
Sk (α1 , . . . , αn ) Sk (β1 , . . . , βn ) · x k1 +2k2 +···+(n−1)kn−1
= ··· ,
k1 =0 kn−1 =0
1 − αβx n
k=(k1 ,...,kn−1 )

)
n )
n
where α = αi and β = βi .
i=1 i=1

Proof We follow (Macdonald, 1979).

    
1 
Det = Det 1 + xi y j + xi y j + · · · 1≤i, j≤n
2 2
1 − xi y j 1≤i, j≤n

∞

∞  
 j j
= ··· Det xi y j 1≤i, j≤n
1 = 0 n = 0


∞

= ··· a ((x1 , . . . , xn ))y1 1 · · · yn n ,
1 = 0 n = 0
= ( 1 ,..., n )
234 The Gelbart–Jacquet lift

where


a ((x1 , . . . , xn )) = (σ )σ x1 1 · · · xn n ,
σ ∈Sn

in which Sn is the group of permutations of {1, 2, . . . , n}, and (σ ) is the sign
of the permutation σ .
It is clear that the polynomial a (x) := a ((x1 , . . . , xn )) satisfies
σ (a (x)) = (σ )a (x)
for any σ ∈ Sn , and, therefore, vanishes unless 1 , . . . , n are all distinct. So,
we may assume 1 > 2 > · · · > n ≥ 0. In view of the skew symmetry, we
may write
( 1 , . . . , n ) = (λ1 + n − 1, λ2 + n − 2, λ3 + n − 3, . . . , λn ).
It follows that
 
 λ j +n− j
a (x) = aλ+δ = Det xi 1≤i, j≤n
,

where λ = (λ1 , . . . , λn ) and δ = (n − 1, n − 2, . . . , 1, 0).

If the above computations are combined with Lemma 7.4.18 (Cauchy’s deter-
minant) we obtain the identity

n 
n
aλ+δ (x)aλ+δ (y)
(1 − xi y j )−1 = ) ) , (7.4.21)
i=1 j=1 λ
(xi − x j ) (yi − y j )
1≤i< j≤n 1≤i< j≤n

where the sum goes over all λ = (λ1 , . . . , λn ) ∈ Zn with λ1 ≥ λ2 ≥ · · · ≥ λn

≥ 0. To complete the proof, note that if we choose
λ = (k1 + · · · + kn , k1 + · · · + kn−1 , . . . , k1 ),
then
aλ+δ (x)
) = Sk2 ,k3 ,...,xn (x1 , x2 , . . . , xn ).
(xi − x j )
1≤i< j≤n

Cauchy’s identity (Proposition 7.4.20) immediately follows from this after

renumbering (k2 , k3 , . . . , kn ) to (k1 , k2 , . . . , kn−1 ), and replacing xi , y j with
αi x, β j x, respectively in the identity (7.4.21).

GL(n)pack functions: The following GL(n)pack functions, described in the

appendix, relate to the material in this chapter:
HeckeMultiplicativeSplit SchurPolynomial.
 8
Bounds for L-functions and Siegel zeros

8.1 The Selberg class

We have investigated carefully the theory of automorphic functions for the
groups S L(2, Z) and S L(3, Z) as well as some of their subgroups. In these and
all other known examples of Dirichlet series with arithmetic significance there
has appeared certain expectations: a type of Riemann hypothesis will hold
if the Dirichlet series has a functional equation and an Euler product or a
converse theorem will hold allowing one to show that the Dirichlet series is, in
fact, associated to an automorphic function on an arithmetic group. Following
(Langlands, 1970), if
 n  
α p,i −1
L(s) = 1− s
p i=1 p

is the L-function associated to a Maass form on G L(n), then one may consider
    −1
L(s, ∨k ) := 1 − α p,i1 α p,i2 · · · α p,ik p −s ,
p 1≤i 1 ≤i 2 ≤···≤i k ≤n

the symmetric kth power L-function which is conjectured to be a Maass form

on G L(M) where


M = M(k, n) = 1.
1≤i 1 ≤i 2 ≤···≤i k ≤n

We have shown this conjecture to hold in the case n = 2, k = 2 in Chapter 7. In

this case, it is the Gelbart–Jacquet lift from G L(2) to G L(3). More recently, Kim
and Shahidi (2002) proved this conjecture in the case of the symmetric cube lift
from G L(2) to G L(4), i.e., when n = 2, k = 3, and Kim (2003) obtained the
symmetric fourth power lift from G L(2) to G L(5). See also (Henniart, 2002).
Selberg (see (Selberg, 1991)) axiomatized certain expected properties of
Dirichlet series and L-functions and introduced the Selberg class S, consisting

235
236 Bounds for L-functions and Siegel zeros

of all formal Dirichlet series (L-functions)

∞
am
L(s) = s
, (a1 = 1, am ∈ C for m = 2, 3, 4, . . . )
m=1
m

which satisfiy the following axioms.

Axiom 8.1.1 (Analyticity) The function (s − 1) L(s) is an entire function of

finite order for some non-negative integer .

Axiom 8.1.2 (Ramanujan hypothesis) For fixed  > 0, we have |am |  m  ,

for all m = 1, 2, 3, . . . , where the constant implied by the  symbol depends
at most on .

Axiom 8.1.3 (Functional equation) Define (s) := As G(s)L(s), with A > 0,

)n
where G(s) = (λ j s + µ j ) with λ j > 0, (µ j ) ≥ 0. Then we have the
j=1
functional equation

(s) =  · (1 − s̄), (|| = 1).

Axiom 8.1.4 (Euler product) We may express log L(s) by the Dirichlet series

∞
bm
log L(s) =
m=2
ms

where bm = 0 unless m is a positive power of a rational prime and |bm |  m θ

for all m = 2, 3, 4, . . . with some fixed θ < 12 .

Selberg made a number of interesting conjectures concerning Dirichlet series

in S. He also introduced the notion of primitive elements in S which cannot be
factored into the product of two or more non-trivial members of S. In (Murty,
1994), it is shown that Selberg’s conjectures imply Artin’s conjecture on the
holomorphy of the L-series attached to finite-dimensional complex representa-
tions of Gal(Q̄/Q), and that the θ of Axiom 8.1.4 behaves like 0 on average.
+
Here are Selberg’s conjectures. Sums of type p refer to sums over rational
primes.

Conjecture 8.1.5 (Regularity of distribution) Associated to each L ∈ S,

there is an integer n L ≥ 0, such that as x → ∞,

|a p |2
= n L log log x + O(1).
p≤x p
 8.1 The Selberg class 237

Conjecture 8.1.6 (Orthonormality) If L , L  ∈ S are distinct and primitive,

then n L = n L  = 1 and

a p a p
= O(1).
p≤x p

Conjecture 8.1.7 (G L(1) twists) Let χ be a primitive Dirichlet character and

+
∞ +
∞
for L(s) = am /m s ∈ S define L χ (s) = am χ (m)/m s to be the twisted
m=1 m=1
Dirichlet series. Then up to a finite Euler product, L χ is also in S.

Conjecture 8.1.8 (Riemann hypothesis) The Dirichlet series L(s) ∈ S have

all their non-real zeros on the critical line (s) = 12 . If L(β) = 0 with β ∈ R,
then β = 12 or β ≤ 0.

See (Kowalski, 2003), (Ramakrishnan and Wang, 2003) for further discus-
sions of the following very interesting conjecture.

Conjecture 8.1.9 For any L(s) ∈ S, if it has a pole of order k at s = 1, then

ζ (s)k |L(s), i.e., L(s) = ζ (s)k L 1 (s), with L 1 (s) ∈ S.

It follows from (Conrey and Ghosh, 1993) that Conjectures 8.1.5 and 8.1.6
imply Conjecture 8.1.9.
In order to classify the Dirichlet series L(s) in the Selberg class, it is conve-
nient to introduce the degree d L of L ∈ S as

n
dL = 2 λj.
j=1

A fundamental conjecture asserts that the degree is always an integer. It has

been shown by Richert (1957) that there are no elements in S with degree
d satisfying 0 < d < 1. Another proof was also found later by Conrey and
Ghosh (1993) who also showed that the only elements of degree zero in S
are the constant functions. Kaczorowski and Perelli (1999) determined the
structure of the Selberg class for degree 1, showing that it contains only the
Riemann zeta function and shifts of Dirichlet L-functions. Soundararajan (2004)
found a simpler proof of this result. In another paper, Kaczorowski and Perelli
(2002), showed that there are no elements of the Selberg class with degree
1 < d < 5/3. It seems likely that the only elements in the Selberg class with
degree d ≤ 3 are L-functions associated to automorphic functions on G L(2),
G L(3).
238 Bounds for L-functions and Siegel zeros

8.2 Convexity bounds for the Selberg class

The maximum principle (see (Ahlfors, 1966)) states that if f (z) is analytic on
a closed bounded set E, then the maximum of | f (z)| is taken on the boundary
of E. A variation of the maximum principle is given in the following version
of the Phragmén–Lindelöf theorem (Phragmén and Lindelöf, 1908) which has
important applications to the theory of Dirichlet series.

Theorem 8.2.1 (Phragmén–Lindelöf) Fix real numbers σ1 < σ2 . Let φ(s) be

holomorphic in the strip {s | σ1 ≤ (s) ≤ σ2 }. Suppose that φ(s) satisfies the
bound
α
|φ(σ + it)| ≤ Ce|t| , (for some α, C > 0)

in this strip. Assume further that

|φ(σ1 + it)| ≤ B (1 + |t|) M1 , |φ(σ2 + it)| ≤ B (1 + |t|) M2 ,

for fixed constants B > 0, M1 , M2 ≥ 0, and for all t ∈ R. Then

M1 (σ2 − σ ) + M2 (σ − σ1 )
|φ(σ + it)| ≤ B (1 + |t|) M(σ ) , with M(σ ) = ,
σ2 − σ1
for all σ1 ≤ σ ≤ σ2 and t ∈ R.

Proof We first consider the simpler case that M1 = M2 = 0. Fix an inte-

ger m > α such that m ≡ 2 (mod 4). Then for (s) = t → +∞, we have
s/|s| → i, so that (s/|s|)m → −1. Consequently, there exists T0 > 0 such
that
 m 
s 1
 <−
|s| 2
for t > T0 .
Now fix  > 0, and define ,m (s) = es φ(s). Since
m

 m 
s 
(s m ) =  ·  · |s|m < − · |s|m
|s| 2
it follows that for T > T0 , sufficiently large, that

+|s|α
|,m (s)| ≤ e− 2 |s|
m
· |φ(s)| ≤ |φ(s)| ≤ B,

for (s) = σ1 , (s) = σ2 , (s) = t ≥ T . Consequently, |,m (s)| ≤ B on

any line segment LT (with T ≥ T ) where
 
LT = s  (s) = T, σ1 ≤ (s) ≤ σ2 .
 8.2 Convexity bounds for the Selberg class 239

Therefore,

|φ(s)| ≤ e|s| B
m
(8.2.2)

on LT . A similar argument can be given for the case t < 0 from which it follows
that (8.2.2) also holds on any such L−T . By the maximum principle, we can
now assert that (8.2.2) holds for any  > 0 on the entire strip with vertical
sides consisting of the lines (s) = σ1 , (s) = σ2 . Letting  → 0 in (8.2.2)
establishes that

|φ(s)| ≤ B

everywhere inside this strip which proves the theorem in the particular case
considered.
In the general case, let
α
u(s) = e M(s) log((−is) ) ,

where the logarithm has its principal value. Then the function u(s) is holomor-
phic for σ1 ≤ (s) ≤ σ2 , and (s) = t ≥ 1. If we write M(s) = M(σ ) + ibt,
then

(M(s)α log(−is)) = ((M(σ )+ibt) · α · log(t − iσ )) = α M(σ ) log t +O(1).

Hence |u(s)| = |t| M(σ ) eO(1) , and, therefore, the function (s) = φ(s)/u(s)
satisfies the same conditions as φ(s) did in the first part. Thus, (s) is bounded
in the strip and the theorem follows.

The Phragmén–Lindelöf Theorem 8.2.1 allows one to obtain growth prop-

erties of holomorphic L-functions in the Selberg class. The key idea for doing
this is that we know L(s) is bounded in its region of absolute convergence
(s) > 1 + , say. By the functional equation of Axiom 8.1.3 we may obtain a
polynomial bound (in |s|) for the growth on the line −. The Phragmén–Lindelöf
theorem then gives us the growth of L(s) in the critical strip 0 ≤ (s) ≤ 1.
Growth properties obtained in this manner are called convexity bounds. We
now establish such convexity bounds for the important subfamily of the
Selberg class S where λ1 = λ2 = · · · = λn = λ and the µ j ( j = 1, 2, . . . , n)
occur in conjugate pairs. Recall that these quantities occur in the Gamma factors
of the functional equation of Axiom 8.1.3.
Restricting to this particular subfamily is not necessary. The reader may,
with a modicum of effort, obtain convexity bounds in more general situations
at the expense of more complicated notation. However, this subfamily includes
all the examples treated in this book, so we will not consider other cases.
240 Bounds for L-functions and Siegel zeros

Theorem 8.2.3 (Convexity bound) Let L(s) ∈ S have a pole of order at

s = 1 and satisfy the functional equation

n
(s) = As (λs + µ j )L(s) =  · (1 − s̄), (|| = 1),
j=1

where λ > 0 and µ j = γ j + iκ j (with γ j > 0, κ j ∈ R, j = 1, 2, . . . , n) occur

in conjugate pairs. Then for every  > 0, there exists an effectively computable
constant C > 0, such that
 
1−σ +
s−1  n
λ
|L(s)| ≤ C A · (1 + |λt + κ j |) ,
s+1 j=1

for all s = σ + it with 0 ≤ σ ≤ 1 and t ∈ R.

Proof By Axiom 8.1.2 we know that for every fixed  > 0, the L-function
L(s) converges absolutely in the region (s) > 1 + , and, hence, is bounded
in this region. Let s = − + it. By the functional equation
A1+2 |G(1 +  − it)|L(1 +  + it) |G(1 +  − it)|
|L(− + it)| =  A1+2 .
|G(− + it)| |G(− + it)|
In order to proceed further, we need to estimate the right-hand side of the above
equation. We make use of Stirling’s asymptotic formula
 it   
√ σ − 12 − π2t |t| 1
(σ + it) = 2π (it) e 1+O , (8.2.4)
e |t|
which is valid for fixed σ and |t| sufficiently large. It follows that
)
n
|(λ(1 +  − it)+µ j )|
n λ(1+2)
|G(1+ − it)| j=1 
= n  (1 + |λt + κ j |) .
|G(− + it)| )
|(λ(− + it)+µ j )| j=1
j=1

We have thus established that L(s) is bounded for (s) = 1 +  and is bounded
by

λ(1+2)

n
A 1+2
(1 + |λt + κ j |)
j=1

for (s) = −. Before applying the Phragmén–Lindelöf Theorem 8.2.1, we

first multiply L(s) by ((s − 1)/(s + 1)) to remove the multiple pole. The result
then immediately follows from Theorem 8.2.1.

A key ingredient in the proof of the convexity bound (Theorem 8.2.3) is

Axiom 8.1.2 which states that the mth Dirichlet coefficient of an L-function in
 8.3 Approximate functional equations 241

S satisfies the Ramanujan bound of O (m  ). The  here reappears as the  (not

necessarily the same ) in the convexity bound. Molteni (2002), using an idea
of Iwaniec (1990) (see also (Murty, 1994)) showed that Axiom 8.1.2 will auto-
matically hold on average (which is all that is really needed for applications to
convexity bounds) if certain symmetric square L-functions satisfy the expected
growth properties.

8.3 Approximate functional equations

It is possible, in many cases, by the use of approximate functional equations, to
obtain slight improvements on the convexity bound obtained in Theorem 8.2.3,
i.e., one may replace the term


n
λ
A · (1 + |λt + κ j |)
j=1

by a power of the logarithm. Such considerations become important, for exam-

ple, when estimating L-functions on the line (s) = 1. Improvements of this
type, based on the method of approximate functional equations, have a long
history (see (Titchmarsh, 1986)).
In (Chandrasekharan and Narasimhan, 1962), the method of approximate
functional equations is applied for the first time in a very general setting corre-
sponding to the Selberg class. Their main interest, however, was in deducing the
average order of the Dirichlet coefficients of an L-function rather than in obtain-
ing bounds for the L-function in the critical strip. They made the surprising dis-
covery that better error terms are available in the case where the Dirichlet series
has positive coefficients. Lavrik (1966) obtained an explicit approximate func-
tional equation for a very wide class of L-functions. See (Iwaniec and Kowalski,
2004) for a detailed exposition of more recent developments along these lines.
Following Lavrik (1966), (see also (Ivić, 1995), (Harcos, 2002), (Iwaniec
and Kowalski, 2004)), we now derive a very general form of the approximate
functional equation for L ∈ S. We assume that L(s) satisfies a functional equa-
tion of the type
n
(s) = As (λs + µ j )L(s) =  · (1 − s̄), (|| = 1), (8.3.1)
j=1

as given in Theorem 8.2.3. Define

n
qw := A · (3 + |λw + µ j |)λ . (8.3.2)
j=1
242 Bounds for L-functions and Siegel zeros

+
∞
Theorem 8.3.3 (Approximate functional equation) Let L(s) = am /m s ∈ S,
m=1
be entire and satisfy the functional equation (8.3.1). Define qw as in (8.3.2).
Then there exists a smooth function F : (0, ∞) → C such that for every w ∈ C
with 0 ≤ (w) ≤ 1, we have

∞  
∞  
am m am m
L(w) = F + λw F̄ ,
m=1
mw qw m=1
m 1−w q1−w

where λw = A1−w G(1 − w)/Aw G(w).

The function F and its partial derivatives F (k) , (k = 1, 2, . . . ) satisfy, for
any σ > 0, the following uniform growth estimates at 0 and ∞:

1 + Oσ (x σ )
F(x) = −σ
F (k) (x) = Oσ (x −σ ).
Oσ (x ),

The implied Oσ –constants depend only on σ, k, n.

Remarks The approximate functional equation allows one to effectively

compute and obtain bounds for special values of the L-function. It effectively
reduces the computation to a short sum of  max(|qw |, |q1−w |) terms. One
may also easily obtain a version of the approximate functional equation for
L-functions with a pole at s = 1.

Proof Let h(s) be a holomorphic function satisfying

h(s) = h(−s) = h(s̄), h(0) = 1, (8.3.4)

and which is bounded in the vertical strip −2 < (s) < 2. For every w ∈ C,
and x > 0, we define

2+i∞
1 G(s + w) ds
Hw (x) := h(s)x −s , (8.3.5)
2πi G(w) s
2−i∞

where

n
G(s) = (λs + µ j ),
j=1

as in Theorem 8.2.3. We assume that h has sufficient decay properties so that

the integral in (8.3.5) converges absolutely.
As a first step in the proof of Theorem 8.3.3, we first derive an approximate
functional equation in terms of the function Hw . For any small  > 0, and
 8.3 Approximate functional equations 243

w ∈ C with 0 ≤ (w) ≤ 1, consider the integral


1++i∞
1 As+w G(s + w)L(s + w) ds
I L (w) := h(s) .
2πi Aw G(w) s
1+−i∞

If we shift the line of integration to the left we pick up a residue of the pole of
the integrand at s = 0. It follows, after applying the functional equation (8.3.1),
and then transforming s → −s, that


−1−+i∞
1  A1−s−w G(1 − s − w)L(1 − s̄ − w̄) ds
I L (w) = L(w) + w
h(s)
2πi A G(w) s
−1−−i∞

A1−w G(1 − w)
= L(w) −
2πi Aw G(w)

1++i∞
 As+1−w G(s + 1 − w) L̃(s + 1 − w) ds
× h(s) ,
A1−w G(1 − w) s
1+−i∞

or equivalently

A1−w G(1 − w)
L(w) = I L (w) +  · I L̃ (1 − w), (8.3.6)
Aw G(w)

+
∞
where L̃(s) = am /m s denotes the dual L-function. In (8.3.6) we may substi-
m=1
tute the Dirichlet series for L(s) and L̃(s) and integrate term by term. It follows
that

∞
am m 
∞
am m 
L(w) = Hw + λ w H1−w . (8.3.7)
m=1
mw A m=1
m 1−w A

To pass from (8.3.7) to the approximate functional equation in Theorem 8.3.3

it is necessary to choose a suitable test function h so that Hw takes the form
of the function F in Theorem 8.3.3. To this end, we analyze Hw further using
Stirling’s approximation for the Gamma function. Let w = u + iv, s = σ + it.
It follows from (8.2.4) that
 
 (w + s)  u+σ − 12
   |w + s| 1 π
· e 2 (|w|−|w+s|)
 (w) 
|w|u− 2
π
 (|w| + 3)σ e 2 |s| .
244 Bounds for L-functions and Siegel zeros

Consequently
)
n
  |(λ(w + s) + µ j )|
 G(w + s) 
  j=1 π
 qσw e 2 nλ|s| .
 G(w)  = )
n (8.3.8)
|(λw + µ j )|
j=1

Next, following (8.3.5), we define


2+i∞
1 G(s + w) ds
F(x) := Hw (qw · x) = h(s)(qw x)−s . (8.3.9)
2πi G(w) s
2−i∞

In (8.3.9), under the assumption that h(s) has sufficient decay, we may shift
the line of integration either to the left (picking up the residue 1 at the pole at
s = 0) or we may shift to the right. After shifting to an arbitrary line R(s) = σ
and differentiating k times with respect to x, we obtain from (8.3.9) that
 k
d (−1)k
F(x) = δσ,k +
dx 2πi
σ+i∞
G(s + w) −s−k ds
× h(s)s(s + 1) · · · (s + k − 1)q−s
w x ,
G(w) s
σ −i∞
(8.3.10)

1 σ < 0, k = 0,
where δσ,k =
0 otherwise.
It now follows from (8.3.8) and (8.3.10) that
⎛ σ +i∞ ⎞
 k 
d π
F(x) = δσ,k + Oσ,k ⎝ e 2 n|s| (1 + |s|)k · |h(s)| · x −σ |ds|⎠ .
dx
σ −i∞

To complete the proof of Theorem 8.3.3 it is enough to choose a test function h

with sufficient decay properties so that the above integral converges absolutely.
As an example, one may choose h(s) = (cos(π s/2))−2n .

As an example of the approximate functional equation, consider a Dirichlet

L-function L(s, χ ) associated to a primitive Dirichlet character (mod q). Then
the functional equation (see (Davenport, 1967)) takes the form
 −(s+a)/2  
π s+a τ (χ )
(s, χ) =  L(s, χ ) = a√ · (1 − s, χ̄ ),
q 2 i q

0 if χ (−1) = 1,
where τ (χ ) is the Gauss sum, as in (7.1.1), and a =
1 if χ (−1) = −1.
 8.4 Siegel zeros in the Selberg class 245

In this case, qw as defined in (8.3.2), satisfies

&
|qw |  |q| · (1 + |w|).

It immediately follows from the approximate functional equation given in

Theorem 8.3.3 that

|L(1, χ)|  log |q|. (8.3.11)

8.4 Siegel zeros in the Selberg class

In the Selberg class S determined by Axioms 8.1.1–8.1.4, let us fix the integer
n ≥ 1, and we assume that we have a subfamily Sn , all of whose elements have
exactly n Gamma factors which are all of the same form. We shall now define
Siegel zeros for this subfamily. Note, however, that by Selberg’s Conjecture
8.1.8, we do not expect such Siegel zeros to exist.

Definition 8.4.1 (Siegel zero) Fix a constant c > 0 and an integer n ≥ 1. Let
L(s) ∈ Sn satisfy the functional equation given in Axiom 8.1.3. Assume that
L(β) = 0 for some real β satisfying
c
1− ≤ β ≤ 1,
log(λA + 1)
where λ = max (λi + |µi |). Then β is termed a Siegel zero for L(s) relative
1≤i≤n
to c.

Interlude on the history of Siegel zeros Let D < √0 denote the fundamental
discriminant of an imaginary quadratic field k = Q( D). Then D ≡ 1 (mod 4)
and square-free, or of the form D = 4m with m ≡ 2 or 3 (mod 4) and square-
free. Define
 
group of non-zero fractional ideals ba
h(D) = #
group of principal ideals (α), α ∈ k ×
to be the cardinality of the ideal class group of k. Gauss (1801) showed (using the
language of binary quadratic forms) that h(D) is always finite. He conjectured
that

h(D) → ∞ as D → −∞,

which was first proved by Heilbronn (1934).

The Disquisitiones also contains a number of tables of binary quadratic forms
(actually only even discriminants were considered) with small class numbers.
246 Bounds for L-functions and Siegel zeros

Gauss made the remarkable conjecture that his tables were complete. In modern
parlance, we may rewrite Gauss’ tables in the form (see also (Goldfeld, 2004)):

h(D) 1 2 3 4 5
# of fields 9 18 16 54 25 . (8.4.2)
largest |D| 163 427 907 1555 2683

The case of class number 1 is particularly interesting, because in this case, it

can be shown that the imaginary quadratic field has the unique factorization
property – that every integer in the field can
√ be uniquely factored into primes.
Note that unique factorization fails in Q( −5) because the integer 6 can be
factored in two distinct ways, i.e.,
√ √
6 = 3 · 2, 6 = (1 + −5) · (1 − −5),
√ √
and each of 2, 3, 1 − −5, 1 + −5, cannot be further factored, so are primes.
For the case of class number 1, Gauss’ conjecture takes the explicit form that
there are only nine discriminants

−d = −3, −4, −7, −8, −11, −19, −43, −67, −163

√
where the imaginary quadratic field Q( −d) has class number 1.
The problem of finding an algorithm which would enable one to effectively
determine all imaginary quadratic fields with a given class number h is now
known as the Gauss class number problem, and it is in connection with this prob-
lem that Siegel zeros first arose. If such an effective algorithm did not exist, then
a Dirichlet series associated to an imaginary quadratic field with small class
number and large discriminant would have to have a Siegel zero. More con-
cretely, it had been shown that if Gauss’ tables (8.4.2) were not complete then a
Siegel zero would have to exist and the Riemann hypothesis would be violated!
This problem has a long and colorful history, (see (Goldfeld, 1985)), the first
important milestones were obtained by Heegner (1952), Stark (1967, 1972),
and Baker (1971) whose work led to the solution of the class number 1 and 2
problems. Finally Goldfeld, Gross, and Zagier (see (Goldfeld, 1976, 1985),
(Gross and Zagier, 1986), (Iwaniec and Kowalski, 2004)) solved the Gauss
class number problem completely. In (Watkins, 2004), the range of the com-
plete (unconditional) solution for Gauss’ class number prolem was extended to
determining all imaginary quadratic fields with h(d) ≤ 100. This was achieved
after several months of computer computation.
Why are Siegel zeros so intimately related to the class numbers of imaginary
quadratic fields? Although it may not seem so at first, the answer to this question
is the substance of the following lemma.
 8.4 Siegel zeros in the Selberg class 247

Lemma 8.4.3 Let L ∈ S have non-negative Dirichlet coefficients. Assume

that L(s) has a simple pole at s = 1 with residue R, and that L(s) satisfies a
growth condition on the line (s) = 12 of the form
  
 
 L 1 + it  ≤ M(1 + |t|) B (8.4.4)
 2 

for some M ≥ 1, B ≥ 0, and all t ∈ R. If L(s) has no real zeros in the range
1 − (1/log M) < s < 1, then there exists an effective constant c(B) > 0 such
that

R −1 ≤ c(B) log M.

Proof Let r > B be a fixed integer. We shall make use of the well-known
integral transform
 2+i∞   r
1 xs 1
1 − x1 , x > 1,
ds = r !
2πi 2−i∞ s(s + 1) · · · (s + r ) 0, 0 < x ≤ 1,

which is proved by either shifting the line of integration to the left (if x > 1)
or to the right (if 0 < x ≤ 1), and then computing the sum of the residues with
Cauchy’s theorem. Now,


∞
a(m)
L(s) =
m=1
ms

with a(1) = 1, a(m) ≥ 0, for m = 2, 3, . . . It follows that for all x ≥ 2, and

any 12 < β < 1,
 2+i∞
1 L(s + β)x s
1 ds. (8.4.5)
2πi 2−i∞ s(s + 1) · · · (s + r )
Here, as throughout this proof, the constant implied by the –symbol is effec-
tive and depends at most on B. Now (8.4.4) implies that L(s) has polynomial
growth on the line (s) = 12 . Further, L(s) is bounded by L(3) on the line
(s) = 3. By a convexity argument, one obtains that L(σ + it) = O(|t| B ) for
all 12 ≤ σ ≤ 3, t ≥ 1. It follows that one may shift the line of integration of the
integral on the right-hand side of (8.4.5) to the line (s) = 12 − β < 0, picking
up residues at s = 1 − β, 0. Thus the integral becomes
Rx 1−β L(β) 
+ O M x 2 −β .
1
+
(1 − β)(2 − β) · · · (r + 1 − β) r!
If we now choose x = M C , for a sufficiently large constant C, it follows from
248 Bounds for L-functions and Siegel zeros

(8.4.5) and the above residue computations that

R M C(1−β)
1 + L(β). (8.4.6)
1−β
The key point is that for real s > 1, the function L(s) is positive. On the other
hand, for 1 − (1/log M) < s < 1, the function L(s) becomes negative since we
have crossed the pole at s = 1 where there is a sign change and we have assumed
that L(s) has no zeros in the interval. One may then choose 1 − β = 1/log M
so that L(β) ≤ 0. Then (8.4.6) implies that R −1  log M.

Let us now use Lemma 8.4.3 to relate the Siegel zero with the Gauss class
number problem for imaginary quadratic fields. Let D <√0 be the fundamen-
tal discriminant of an imaginary quadratic field k = Q( D). The relation is
through the zeta function


∞
a(m)
ζk (s) = ζ (s)L χ (s) = ,
m=1
ms

where χ (m) = (D/m) is the Kronecker symbol (primitive quadratic Dirichlet

character of conductor D) where a(1) = 1, and


a(m) = χ (d) ≥ 0 (8.4.7)
d|m

for all m = 1, 2, 3, . . . It is a classical theorem of Dirichlet (see (Davenport,

1967) that ζk (s) has a simple pole at s = 1 with residue
π h(D)
R= 1 (8.4.8)
|D| 2
provided D < −4. Further, by the convexity bound given in Theorem 8.2.3,
ζk (s) satisfies the growth condition
  
 
ζk 1 + it   |D| 12 + · (1 + |t|)1+ . (8.4.9)
 2 

Actually, much stronger bounds are known, but we do not need them here. It
follows from (8.4.7), (8.4.8), (8.4.9) that ζk (s) satisfies the conditions required
in Lemma 8.4.3. It then follows from Lemma 8.4.3 that ζk (s) has a Siegel zero
if the class number h(D) is so small that
1
|D| 2
h(D)  . (8.4.10)
log |D|
The above result was first published by Landau (1918), but Landau attributes
this result to a lecture given by Hecke.
 8.5 Siegel’s theorem 249

8.5 Siegel’s theorem

It was shown√in (8.4.10) that if the class number h(D) of an imaginary quadratic
field k = Q( D) is too small, then the zeta function ζk (s) of k must have a Siegel
zero. In fact, since ζk (s) factors as ζ (s)L(s, χ), and the Riemann zeta function
ζ (s) does not vanish for 0 < s < 1, it follows that the Dirichlet L-function
L(s, χ) must have a Siegel zero. This result goes back to Landau (1918).
Fifteen years later there was further, rather surprising, progress.
r Deuring (1933) proved that if the classical Riemann hypothesis is false then
h(D) ≥ 2 for −D sufficiently large.
r Mordell (1934) showed if the Riemann hypothesis is false, then h(D) → ∞
as −D → ∞.
r Heilbronn (1934) proved that the falsity of the generalized Riemann hypoth-
esis for Dirichlet L-functions implies that h(D) → ∞ as −D → ∞.

When combined with the Landau–Hecke result (8.4.10) this gave an uncondi-
tional proof of Gauss’ conjecture that the class number of an imaginary quadratic
field goes to infinity with the discriminant. The surprising aspect of this chain of
theorems is that first one assumes the Riemann hypothesis to establish a result
and then one assumes that the Riemann hypothesis is false to obtain the exact
same result! This is now called the Deuring–Heilbronn phenomenon, but has
the defect of being totally ineffective. Siegel (1935) practically squeezed the
last drop out of the Deuring–Heilbronn phenomenon. He proved the following
theorem, which is the main subject matter of this section.
√
Theorem 8.5.1 (Siegel’s theorem) Let Q( D) be an imaginary quadratic
field with fundamental discriminant D < 0 and class number h(D). Then
for every  > 0, there exists a constant c > 0 (which cannot be effectively
computed) such that

h(D) > c |D| 2 − .

1

Remarks Landau (1935) proved Theorem 8.5.1 with  = 18 (also not effec-
tive). Siegel’s theorem, with any  > 0, appeared in the same volume of Acta
Arithmetica, but did not reference Landau’s result at all!

Proof of Siegel’s theorem We follow Goldfeld (1974). In view of Dirichlet’s

theorem (8.4.8), it is enough to prove that for every fixed  > 0 and for all real
primitive quadratic Dirichlet characters χ (mod D), that

L(1, χ) > c |D|− .

250 Bounds for L-functions and Siegel zeros

Note that since h(D) is a positive rational integer, it follows from (8.4.8) that

L(1, χ )  |D|− 2 .
1

Let χ  be a real primitive quadratic Dirichlet character (mod D  ) for some

other fundamental discriminant D  . Consider the zeta function

Z (s) = ζ (s)L(s, χ )L(s, χ  )L(s, χ χ  ),

and let R = L(1, χ)L(1, χ  )L(1, χ χ  ) be the residue at s = 1. In view of the

Euler product
  
1 −1
 
χ ( p) −1
 
χ  ( p) −1

χ ( p)χ  ( p) −1
Z (s) = 1− s 1− s 1− s 1− ,
p p p p ps

and the fact that χ , χ  can only take values among {−1, 0, +1}, one readily
establishes that Z (s) is a Dirichlet series whose first coefficient is 1 and whose
other coefficients are non-negative.

Lemma 8.5.2 For every  > 0, there exists χ  (mod D  ) and β ∈ R satisfying
1 −  < β < 1 such that Z (β) ≤ 0 independent of what χ (mod D) may be.

Proof If there are no zeros in [1 − , 1] for any L(s, χ ), then Z (β) < 0 if
1 −  < β < 1 since ζ (β) < 0 and all L-functions L(s, χ) will be positive in
the interval. Here we use the fact that L(s, χ ) is positive for (s) > 1 and can
only change sign in the interval [1 − , 1] if the L-function vanishes in the
interval.
On the other hand, if such real zeros do exist, let β be such a zero, with χ 
the corresponding character. Then Z (β) = 0 independent of χ .

Next, fix β as in Lemma 8.5.2. It follows, as in the proof of Lemma 8.4.3

that for x ≥ 1,

2+i∞
1 xs
1 Z (s + β) · ds
2πi s(s + 1)(s + 2)(s + 3)(s + 4)
2−i∞
 
R · x 1−β Z (β) |D D  |1+ x −β
= + +O ,
(1 − β)(2 − β)(3 − β)(4 − β)(5 − β) 4! 1−β
after shifting the line of integration to (s) = −β and using convexity bounds of
Theorem 8.2.3 for the growth of Z (s). But Lemma 8.5.2 tells us that Z (β) ≤ 0.
Therefore,
x 1−β
1 R·
1−β
 8.6 The Siegel zero lemma 251

if |D D  |2+  x, since R  1/|D D  |. Consequently, since

R  L(1, χ ) (log |D D  |) (log |D|),
by (8.3.11), we get
|D|−(2+)(1−β)
L(1, χ)  ,
log |D|
where the implied constant in the –sign depends only on χ  , and, therefore,
only on . This proves Siegel’s theorem if (2 + )(1 − β) < /2 and |D| is
sufficiently large.

Tatuzawa (1951) went a step beyond Siegel and proved that Siegel’s
Theorem 8.5.1 must hold with an effectively computable constant c > 0 for
all D < 0, except for at most one exceptional discriminant D. This shows that
the family of real Dirichlet L-functions can have at most one L-function with a
Siegel zero. It can be shown (see (Davenport, 1967)) that complex Dirichlet
L-functions cannot have Siegel zeros. Thus, the entire family of G L(1)
L-functions can have at most one exceptional L-function with a Siegel zero.
The exceptional L-function must correspond to a real primitive Dirichlet char-
acter associated to a quadratic field.

8.6 The Siegel zero lemma

The following lemma plays a crucial role in all the recent work on the non-
existence of Siegel zeros. It first appeared in (Goldfeld, Hoffstein and Lieman,
1994). While we have defined Siegel zeros relative to a constant c > 0, we shall
suppress the constant in the following discussion because it can be easily com-
puted and it is not really important to the flow of ideas.
Lemma 8.6.1 (Siegel zero lemma) Let L(s) ∈ S have non-negative Dirich-
let coefficients and satisfy the functional equation in Axiom 8.1.3. Assume
L  (s)/L(s) is negative for s real and > 1, and that L(s) has a pole of order
m at s = 1. Assume further that (s) = s m (1 − s)m As G(s)L(s) = (1 − s̄),
)n
where G(s) = (λ j s + µ j ), as given in Axiom 8.1.3, is an entire function
j=1
of order 1. Then L(s) has at most m Siegel zeros.
Proof Since (s) is an entire function of order 1 it has (see (Davenport,
1967)) a Hadamard factorization and can be represented in the form
  s

(s) = ea+bs 1− es/ρ , (8.6.2)
(ρ)=0
ρ
252 Bounds for L-functions and Siegel zeros

for certain constants a, b. If (s) > 1, we may take logarithmic derivatives in

(8.6.2) to obtain
m m G  (s) L  (s)
 1 1

+ + log A + + =b+ + .
s s−1 G(s) L(s) ρ s−ρ ρ

By the functional equation

 1 1

 1 1

b+ + = −b − + ,
ρ s−ρ ρ ρ 1−s−ρ ρ

so that

1
b=− .
ρ ρ

It follows that
m m G  (s) L  (s)
1
+ + log A + + = .
s s−1 G(s) L(s) ρ s −ρ

Now, by assumption, L  (s)/L(s) is negative for s ∈ R, s > 1. Also, if we pair

+
conjugate roots, then every term of (s − ρ)−1 is positive, so there exists an
absolute effective constant c0 > 0, such that


r
1 m
≤ + c0 log M,
j=1
s − βj s−1

where r denotes the number of real zeros β j of L(s) in the interval

[1 − (c/log M), 1]. Here the the constant c0 can be computed from the integral
representation
   ∞  −t z
1 1 1 t t e
log (z) = z − log z − z + log(2π)+ − + t dt,
2 2 0 2 z e − 1 t
for the Gamma function. If the constant c is chosen small enough, compared to
the constant c0 , then a contradiction is obtained whenever r ≥ m + 1.

8.7 Non-existence of Siegel zeros for

Gelbart–Jacquet lifts
The existence of Siegel zeros for classical Dirichlet L-functions has been shown
to be equivalent to the existence of primitive quadratic Dirchlet characters χ
with the property that L(1, χ) takes on too small a value. This follows from
 8.7 Non-existence of Siegel zeros for Gelbart–Jacquet lifts 253

Lemma 8.4.3 when we choose ζ (s)L(s, χ) as our Dirichlet series, which has
all the required properties of Lemma 8.4.3 and has a simple pole at s = 1 with
residue L(1, χ). A real breakthrough was achieved in (Hoffstein and Lockhart,
1994) when they took the study of Siegel zeros outside the classical domain of
Dirichlet L-functions and considered, for the first time, the question of whether
such zeros could exist for L-functions associated to automorphic forms on
G L(3) occurring as symmetric square lifts (Gelbart–Jacquet lifts) from G L(2).
It now became possible to obtain lower bounds of special values of L-functions
in the same way as the classical methods (using the Deuring–Heilbronn phe-
nomenon) gave lower bounds for L(1, χ) with χ a real primitive Dirichlet
character. This established a powerful new tool in modern analytic number
theory.
In (Hoffstein and Lockhart, 1994), it was shown that if f is a Maass form
for S L(2, Z) which is an eigenfunction of the Hecke operators, and F is its
symmetric square lift (Gelbart–Jacquet lift) to S L(3, Z), then the lifted L-
function,
L f × f (s)
L F (s) = , (8.7.1)
ζ (s)

given in Theorem 7.3.2 cannot have a Siegel zero. Actually, they proved a
more general result valid for congruence subgroups of S L(2, Z) and also con-
sidered Gelbart–Jacquet lifts of both holomorphic modular forms and non-
holomorphic Maass forms. Their proof was based on a generalization of Siegel’s
Theorem 8.5.1 and, thereby, was not effective. In the appendix of their paper
Goldfeld, Hoffstein and Lieman (1994) obtained an effective version of their
theorem which was based on the Siegel-zero Lemma 8.6.1. This effective proof
is the subject matter of this section.
The key idea in the proof of the non-existence of Siegel zeros for L-functions
of type (8.7.1) is the construction of an auxiliary L-function which has non-
negative Dirichlet coefficients and a multiple pole at s = 1, and satisfies the
requirements of Lemma 8.6.1. Accordingly, we introduce

Z (s) := ζ (s)L F (s)2 L F×F (s). (8.7.2)

+
∞ +
∞
Here, if L F (s) = c(m)m −s , then L F×F (s) = ζ (3s) |c(m)|2 m −s as we
m=1 m=1
recall from Proposition 7.4.12. In terms of Euler products, if

  
α p −1

α p −1
L f (s) = 1− s 1− s ,
p p p
254 Bounds for L-functions and Siegel zeros

with α p · α p = 1, and

−1  −1
−1
 α 2p 1 α p 2
L F (s) = 1− 1− s 1− ,
p p ps p ps

then

−1
−1  −1
−1
 α 4p α 2p 1 α 2p
L F×F (s) = 1− 1− 1− s 1−
p ps ps p ps
 −1
−1  −1
1 α p 2 1
× 1− s 1− 1− s
p ps p

−1
−1
α p 2 α p 4
× 1− 1− . (8.7.3)
ps ps

Recall that the symmetric square lift of F is given by

−1
−1  −1
 α 4p α 2p 1
L F (s, ∨ ) =
2
1− 1− 1− s
p ps ps p
 −1
−1
−1
1 α p 2 α p 4
× 1− s 1− 1− .
p ps ps

It follows that

L F×F (s) = L F (s)L F (s, ∨2 ).

This actually factors further since

L F (s, ∨2 ) = ζ (s)L f (s, ∨4 ).

Finally, we obtain

Z (s) = ζ (s)L F (s)3 L F (s, ∨2 ). (8.7.4)

+
∞
Lemma 8.7.5 Let Z (s) = a(m)m −s be given by (8.7.2). Then a(1) = 1
m=1
and a(m) ≥ 0 for m = 2, 3, 4, . . .
 8.7 Non-existence of Siegel zeros for Gelbart–Jacquet lifts 255

Proof The fact that a(1) = 1 is an immediate consequence of the Euler prod-
uct for Z (s). Now, the Euler product for Z (s) takes the form

−1
 α p1 α p 2
1−
p ps

where the product is taken over all sixteen possible pairs (1 , 2 ), and where
1 , 2 , independently run through the values 2, 0, 0, 2. If one takes logarithms,
it follows that the pth term in the expansion of log Z (s) is


∞
(α 2 + α −2 + 2)(α  2 + α  −2 + 2)
.
=1
p s

Since α 2 , α  2 , are either non-negative real numbers or lie on the unit circle, it
follows that the above series has non-negative terms. Consequently, so does the
series for Z (s).

Theorem 8.7.6 Let f be a Maass form for S L(2, Z) which is an eigenform

for the Hecke operators. Let F be its symmetric square lift to S L(3, Z). Then
L F (s), given by (8.7.1), has no Siegel zero.

Proof We make use of (8.7.4). It follows from the Euler product that for s ∈ R,
s > 1 that Z  (s)/Z (s) < 0. In (Bump and Ginzburg, 1992) it is shown that when
f is not a lift from G L(1), then L F (s, ∨2 ) has a simple pole at s = 1, and any
zero of L F (s) will be a zero of Z (s) with order at least 3. Consequently, if
L(F, s) has a Siegel zero, then Z (s) = ζ (s)L F (s)3 L F (s, ∨2 ) will have 3 Siegel
zeros. Since Z (s) has a pole of order 2, we obtain a contradiction from the
Siegel zero Lemma 8.6.1.

In the classical case, the non-existence of Siegel zeros implies a lower bound
for the class number of an imaginary quadratic field. One may ask what takes
the place of class numbers in the G L(3) setting. The answer to this question
is given in Lemma 8.4.3 which says that we will obtain a lower bound for the
residue (at s = 1) of the relevant L-function. We state an important and useful
corollary to Theorem 8.7.6.

Corollary 8.7.7 Let f be a Maass form for S L(2, Z) of type ν. Then the
Petersson inner product of f with itself satisfies
1
 f, f   .
log(1 + |ν|)
256 Bounds for L-functions and Siegel zeros

Proof Let F be the symmetric square lift of f . Then

L f × f (s) = ζ (s)L F (s),
and L f × f (s) has no Siegel zero and satisfies the conditions of Lemma 8.4.3.
This gives a lower bound for L F (1) of the type stated in Corollary 8.7.7. But,
we know that L F (1) = c ·  f, f  (for some constant c) by Theorem 7.2.4. This
proves the corollary.

Remark Corollary 8.7.7 can be generalized to the case of holomorphic mod-

ular forms and Maass forms for congruence subgroups of S L(2, Z). It can also
be further generalized to L-functions in the Selberg class satisfying Langlands’
conjecture on the automorphicity of certain symmetric power L-functions.

8.8 Non-existence of Siegel zeros on G L(n)

We have already shown at the end of Section 8.5, that there is at most one
classical Dirichlet L-function with a Siegel zero. The exceptional L-function, if
it exists, will be associated to a real primitive Dirichlet character attached to a
quadratic field. This is the situation for L-functions on G L(1), and it has been
known for a long time (Davenport, 1967).
With the breakthrough of Hoffstein and Lockhart (1994) it became possible,
for the first time, to show the rarity of Siegel zeros of L-functions associated
to Maass forms on G L(n) with n ≥ 2. We give a brief account of the known
results and also specify the particular L-function used in conjunction with the
Siegel zero Lemma 8.6.1 to obtain these results.
In accordance with Theorem 7.2.4 and Proposition 7.4.12, it is natural to
formalize a more general version of the Rankin–Selberg convolution in terms
of Euler products. Let
 n    n  
αi, p −1 βi, p −1
L f (s) = 1− s , L g (s) = 1− s ,
p i=1 p p i=1 p
be two L-functions in the Selberg class S. We then define the Rankin–Selberg
L-function, L f ×g by the new Euler product
 n  n  −1
αi, p β¯ j, p
L f ×g (s) = 1− . (8.8.1)
p i=1 j=1 ps

In (Hoffstein and Ramakrishnan, 1995) it is shown that there are no

Siegel zeros for L-functions associated to cusp forms (holomorphic or non-
holomorphic Maass forms) on G L(2). Their proof works over any number
field. The idea of the proof is as follows. Let f be a cusp form on G L(2) where
 8.8 Non-existence of Siegel zeros on G L(n) 257

L f (s) has a Siegel zero. Let L f (s, ∨2 ) denote the symmetric square lift as in
Theorem 7.3.2, which is associated to a cusp form F on G L(3).
Now, construct

ζ (s) · L (s, ∨2 ) · L f (s),

"#$% " f #$ % " #$ %
G L(1) G L(3) G L(2)

which corresponds to an automorphic form G on G L(6). Let L G×G (s), as in

(8.8.1), denote the L-function of the Rankin–Selberg convolution of G with
itself which will have non-negative Dirichlet coefficients. Then we have the
identity

L G×G (s) = ζ (s)L F×F (s)L f × f (s)L F (s)2 L f (s)4 L f (s, ∨3 ),

which can be verified by comparing Euler products. It follows from results of

Bump and Ginzberg (1992), Bump, Ginzberg and Hoffstein (1996) and Shahidi
(1989) that L G×G (s) has a triple pole at s = 1. If L f (s) had a Siegel zero then
L G×G (s) would have to have four Siegel zeros which contradicts the Siegel zero
Lemma 8.6.1.
We now consider the case of a non-self dual Maass form f on G L(n) for
n ≥ 3. We follow (Hoffstein and Ramakrishnan 1995). Define

Z (s) = ζ (s)L f (s)L f˜ (s),

where f˜ denotes the Maass form dual to f . Let D(s) denote the Rankin–Selberg
convolution of Z (s) with itself. Then D(s) will have a pole of order 3, but it
will have two copies of L f (s) and an additional two copies of L f˜ (s) as factors.
So if L f (s) has a Siegel zero then D(s) will have four Siegel zeros and a pole
of order 3 at s = 1 which again contradicts the Siegel zero Lemma 8.6.1. This
establishes that L-functions associated to non-self dual Maass forms on G L(n),
for n ≥ 3, cannot have Siegel zeros. Note that this situation is analogous to the
way one proves that complex Dirichlet L-functions cannot have Siegel zeros.
Finally, we consider the case of self dual Maass forms on G L(n) with n ≥ 3.
We again follow Hoffstein and Ramakrishnan (1995) who proved that Siegel
zeros cannot exist if one assumes Langlands’ conjectures. Let f denote a self
dual Maass form on G L(n). Assume there exists some g = f where g is not
an Eisenstein series such that

L f × f (s) = L g (s) · D(s)

for some other Dirichlet series D(s). Construct

Z (s) = ζ (s)L g (s)L f (s),

258 Bounds for L-functions and Siegel zeros

and take the Rankin–Selberg convolution of Z with itself. Then this Rankin–
Selberg convolution will have non-negative coefficients, a pole of order 3 at
s = 1, and it will have L f (s)4 as a factor assuming everything else is analytic. If
L f (s) had a Siegel zero, then this would contradict the Siegel zero Lemma 8.6.1.
For the case of G L(3), Hoffstein and Ramakrishnan (1995) proved there are
no Siegel zeros subject to a certain analyticity hypothesis. This hypothesis was
subsequently proved in (Banks, 1997) which establishes that there are no Siegel
zeros on G L(3) except for the obvious cases.
 9
The Godement–Jacquet L-function

9.1 Maass forms for S L(n, Z)

We briefly review Maass forms which were introduced in Section 5.1. For n ≥ 2,
the generalized upper half plane hn (see Definition 1.2.3) consists of all n × n
matrices of the form z = x · y where
⎛ ⎞ ⎛ ⎞
1 x1,2 x1,3 · · · x1,n y1 y2 · · · yn−1
⎜ 1 x2,3 · · · x2,n ⎟ ⎜ y1 y2 · · · yn−2 ⎟
⎜ ⎟ ⎜ ⎟
⎜ .. . ⎟ ⎜ .. ⎟
x =⎜ . .. ⎟, y = ⎜ . ⎟,
⎜ ⎟ ⎜ ⎟
⎝ 1 xn−1,n ⎠ ⎝ y1 ⎠
1 1

with xi, j ∈ R for 1 ≤ i < j ≤ n and yi > 0 for 1 ≤ i ≤ n − 1.

Remark 9.1.1 It is particularly convenient to relabel the super diag-

onal elements of the unipotent matrix x so that x1,2 = xn−1 , x2,3 =
xn−2 , . . . , xn−1,n = x1 , i.e.,
⎛ ⎞
1 xn−1 x1,3 ··· x1,n
⎜ 1 xn−2 ··· x2,n ⎟
⎜ ⎟
⎜ .. .. .. ⎟ .
x =⎜ . . . ⎟
⎜ ⎟
⎝ 1 x1 ⎠
1

Henceforth, we will adhere to this notation.

Let Un denote the group of n × n upper triangular matrices with 1s on the

diagonal as in Section 5.2. Consider a Maass form φ(z) with z ∈ S L(n, Z)\hn

259
260 The Godement–Jacquet L-function

as defined in Definition 5.1.3. Then φ has a Fourier expansion of the form:

∞

∞
  
γ
φ(z) = ··· φ̃ (m 1 ,...,m n−1 ) z ,
γ ∈Un−1 (Z)\S L(n−1,Z) m 1 =1 m n−2 =1 m n−1 =0
1

where the sum is independent of the choice of coset representatives γ ,

 1  1 
φ̃ (m 1 ,...,m n−1 ) (z) := ··· φ(u · z) e−2πi m 1 u 1 +m 2 u 2 +···+m n−1 u n−1
d ∗ u,
0 0

with u ∈ Un (R) given by (5.2.1) and


d ∗ u = du 1 · · · du n−1 du i, j .
1≤i< j+1≤n

Note the change of notation in the Fourier expansion above to conform with
Remark 9.1.1.
Now, we have shown that φ̃ (m 1 ,...,m n−1 ) (z) is a Whittaker function. Further,
φ̃ (m 1 ,...,m n−1 ) (z) will inherit the growth properties of the Maass form φ and will
satisfy the conditions of the multiplicity one theorem of Shalika (1974) which
states that only the Jacquet Whittaker function (5.5.1) can occur in the Fourier
expansion of a Maass form for S L(n, Z) and that φ̃ (m 1 ,...,m n−1 ) must be a constant
multiple of the Jacquet Whittaker function. It follows from Theorem 5.3.2 and
Proposition 5.5.2 that if φ is a Maass form of type ν = (ν1 , . . . , νn−1 ) ∈ Cn−1
for S L(n, Z) then

∞

∞
A(m 1 , . . . , m n−1 )
φ(z) = ···
γ ∈Un−1 (Z)\S L(n−1,Z) m 1 =1 m n−2 =1 m n−1 =0
)
n−1
|m k |k(n−k)/2
k=1
   
γ
× WJacquet M · z, ν, ψ1,...,1, m n−1 , (9.1.2)
1 |m n−1 |

where
⎛ ⎞
m 1 · · · m n−2 · |m n−1 |
⎜ .. ⎟
⎜ . ⎟
⎜ ⎟
M =⎜ m1m2 ⎟, A(m 1 , . . . , m n−1 ) ∈ C,
⎜ ⎟
⎝ m1 ⎠
1
 9.2 The dual and symmetric Maass forms 261

and
⎛⎛ ⎞⎞
1 u n−1
⎜⎜ 1 u n−2 ∗ ⎟⎟ 
⎜⎜ ⎟⎟
⎜⎜ .. .. ⎟⎟ 2πi u 1 +···+u n−2 +u n−1
ψ1,...,1, ⎜⎜ . . ⎟⎟ = e .
⎜⎜ ⎟⎟
⎝⎝ 1 u1 ⎠⎠
1
A(m 1 ,... ,m n−1 )
The particular normalization )
n−1 is chosen so that later formulae are
|m k |k(n−k)/2
k=1
as simple as possible.
Lemma 9.1.3 (Fourier coefficients are bounded) Let φ be a Maass form for
S L(n, Z) as in (9.1.2). Then for all non-zero integers m 1 , . . . , m n−1 ,
A(m 1 , . . . , m n−1 )
= O(1).
)
n−1
|m k |k(n−k)/2
k=1

Proof Let z ∈ hn . By the Fourier expansion, we know that

A(m 1 , . . . , m n−1 )  
· W Jacquet M y, ν, ψ m n−1
)
n−1 1,...,1, |m n−1 |
|m k |k(n−k)/2
k=1
1 1 3 4
= ··· φ(z) e−2πi m 1 x1 +···+m n−1 xn−1
d ∗ x,
0 0

where

d ∗ x = d x1 · · · d xn−1 d xi, j .
1≤i< j+1≤n

Choosing y1 = |m 1 |−1 c1 , y2 = |m 2 |−1 c2 , . . . , yn−1 = |m n−1 |−1 cn−1 , for suit-

able c1 , . . . , cn−1 , and noting that φ is bounded implies that
A(m 1 , . . . , m n−1 )
= O(1).
)
n−1
|m k |k(n−k)/2
k=1

9.2 The dual and symmetric Maass forms

Let φ(z) be a Maass form for S L(n, Z) as in (9.1.2). We shall now define,
φ̃(z), the dual Maass form associated to φ which plays an important role in
automorphic form theory.
262 The Godement–Jacquet L-function

Proposition 9.2.1 Let φ(z) be a Maass form of type (ν1 , . . . , νn−1 ) ∈ Cn−1 as
in (9.1.2). Then (for !x" the largest integer ≤ x)
⎛ ⎞
(−1)[n/2]
⎜ 1 ⎟
⎜ ⎟
φ̃(z) := φ(w · t (z −1 ) · w), w = ⎜ . ⎟
⎝ .. ⎠
1

is a Maass form of type (νn−1 , . . . , ν1 ) for S L(n, Z). The Maass form φ̃ is
called the dual Maass form. If A(m 1 , . . . , m n−1 ) is the (m 1 , . . . , m n−1 )-Fourier
coefficient of φ then A(m n−1 , . . . , m 1 ) is the corresponding Fourier coefficient
of φ̃.

Proof First, for every γ ∈ S L(n, Z),

 
φ̃(γ z) = φ w · t ((γ z)−1 ) · w = φ γ  w · t (z −1 ) · w = φ̃(z)

since γ  = w · t (γ −1 ) · w ∈ S L(n, Z). Thus φ̃ satisfies the automorphic condi-

tion (1) of Definition 5.1.3 of a Maass form.
Next, note that if
⎛ ⎞⎛ ⎞
1 xn−1 x1,3 ··· x1,n y1 y2 · · · yn−1
⎜ 1 xn−2 ··· x2,n ⎟ ⎜ y1 y2 · · · yn−2 ⎟
⎜ ⎟⎜ ⎟
⎜ .. .. .. ⎟ ⎜ .. ⎟
z=⎜ . . . ⎟ ⎜ . ⎟,
⎜ ⎟⎜ ⎟
⎝ 1 x1 ⎠ ⎝ y1 ⎠
1 1

then
⎛ ⎞⎛ ⎞
1 δx1 y1 y2 · · · yn−1
⎜ 1 −x2 ∗ ⎟ ⎜ y2 y3 · · · yn−2 ⎟
⎜ ⎟⎜ ⎟
−1 −1 ⎜ . . ⎟ ⎜ . ⎟
w · (z ) · w =⎜
t .. .. ⎟⎜ .. ⎟
⎜ ⎟⎜ ⎟
⎝ 1 −xn−1 ⎠ ⎝ yn−1 ⎠
1 1
(9.2.2)
!n/2"+1
where δ = (−1) . One may then show that

φ̃(z) du = 0,
(S L(n,Z)∩U )\U
 9.2 The dual and symmetric Maass forms 263

for every upper triangular subgroup U of the form

⎧⎛ ⎞⎫
⎪
⎪ Ir1 ⎪
⎪
⎪⎜
⎨ ⎟⎪⎬
⎜ Ir 2
∗ ⎟
U= ⎜ .. ⎟ ,
⎪
⎪ ⎝ . ⎠⎪⎪
⎪
⎩ ⎪
⎭
Irb

with r1 + r2 + · · · rb = n. Here Ir denotes the r × r identity matrix. Thus φ̃

satisfies the cuspidality condition (3) of Definition 5.1.3.
Now

Iν1 ,...,νn−1 (z) = Iνn−1 ,...,ν1 (w · t (z −1 ) · w−1 )

since the involution z → w · t (z −1 ) · w−1 interchanges y j and yn− j for j =

1, 2, . . . , n − 1. This shows that φ̃ is a Maass form of type (νn−1 , . . . , ν1 ).
Finally, if we integrate
1 1 3 4
··· φ̃(z) e−2πi m 1 x1 +···+m n−1 xn−1
d∗x
0 0

to pick off the (m 1 , . . . , m n−1 )-Fourier coefficient, then because x j and xn− j
are interchanged (for j = 1, . . . , n − 1) we will actually get A(m n−1 , . . . , m 1 ).

In the S L(2, Z) theory, the notions of even and odd Maass forms (see
Section 3.9) played an important role. If a(n) is the nth Fourier coefficient
of an S L(2, Z) Maass form then a(n) = ±a(−n) depending on whether the
Maass form is even or odd. We shall see that a similar phenomenon holds in the
case of S L(n, Z) when n is even. On the other hand, if n is odd then all Maass
forms are actually even (see Section 6.3 for the example of n = 3).
Consider a diagonal matrix δ of the form
⎛ ⎞
δ1 · · · δn−1
⎜ .. ⎟
⎜ . ⎟
⎜ ⎟
δ := ⎜ δ1 δ2 ⎟
⎜ ⎟
⎝ δ1 ⎠
1

where δ j ∈ {+1, −1} for j = 1, . . . , n − 1. We define an operator Tδ which

maps Maass forms to Maass forms, and is given by

Tδ φ(z) := φ(δzδ) = φ(δz),

264 The Godement–Jacquet L-function

since φ, as a function on hn , is right-invariant under multiplication by O(n, R).

Note that
⎛⎛ ⎞⎛ ⎞⎞
1 xn−1 y1 y2 · · · yn−1
⎜⎜ 1 xn−2 ∗ ⎟ ⎜ y1 y2 · · · yn−2 ⎟⎟
⎜⎜ ⎟⎜ ⎟⎟
⎜⎜ .. .. ⎟⎜ .. ⎟⎟
Tδ φ ⎜⎜ . . ⎟⎜ . ⎟⎟
⎜⎜ ⎟⎜ ⎟⎟
⎝⎝ 1 x1 ⎠ ⎝ y1 ⎠⎠
1 1
(9.2.3)
⎛⎛ ⎞⎛ ⎞⎞
1 δn−1 xn−1 y1 y2 · · · yn−1
⎜⎜ 1 δn−2 xn−2 ∗⎟ ⎜ y1 y2 · · · yn−2 ⎟⎟
⎜⎜ ⎟⎜ ⎟⎟
⎜⎜ .. .. ⎟⎜ .. ⎟⎟
= φ ⎜⎜ . . ⎟⎜ . ⎟⎟ .
⎜⎜ ⎟⎜ ⎟⎟
⎝⎝ 1 δ1 x1 ⎠ ⎝ y1 ⎠⎠
1 1

Clearly (Tδ )2 is the identity transformation, so the eigenvalues of Tδ can only

be ±1.

Definition 9.2.4 A Maass form φ of type ν = (ν1 , . . . , νn−1 ) ∈ Cn−1 for

S L(n, Z) is said to be symmetric if Tδ φ = ±φ for all Tδ as in (9.2.3).

Remark Note that every Maass form is a linear combination of symmetric

1s.

Proposition 9.2.5 Assume n ≥ 2 is an odd integer. Then every Maass form φ

for S L(n, Z) is even, i.e., Tδ φ = φ, for all Tδ of the form (9.2.3). Furthermore,
if A(m 1 , . . . , m n−1 ) denotes the Fourier coefficient in the expansion (9.1.2)
then

A(m 1 , . . . , m n−2 , m n−1 ) = A(m 1 , . . . , m n−2 , −m n−1 ),

for all m j ≥ 1 (with j = 1, 2, . . . , n − 1).

Proof Any Maass form φ is invariant under left multiplication by elements

in S L(n, Z) of the form
⎛ ⎞
1
⎜ 2 ⎟
⎜ ⎟
⎜ .. ⎟
⎝ . ⎠
n
 9.2 The dual and symmetric Maass forms 265

)
n
with  j ∈ {+1, −1}, for j = 1, 2, . . . , n and  j = 1. It is also invariant by
j=1
the central element
⎛ ⎞
−1
⎜ −1 ⎟
⎜ ⎟
⎜ .. ⎟
⎝ . ⎠
−1
which has determinant −1. Since these elements generate all possible Tδ , this
proves Tδ φ = φ for all Tδ .
Next, let
⎛ ⎞
−1
⎜ 1 ⎟
⎜ ⎟
Tδ0 = ⎜ .. ⎟.
⎝ . ⎠
1
Then since δ0 zδ0 transforms xn−1 → −xn−1 and fixes x j with 1 ≤ j ≤ n − 2,
it follows that the integral
1 1 3 4
··· Tδ0 φ(z) e−2πi m 1 x1 +···+m n−1 xn−1
d∗x
0 0

picks off the A(m 1 , . . . , m n−2 , −m n−1 ) coefficient of φ, and this must be the
same as A(m 1 , . . . , m n−2 , m n−1 ) because Tδ0 φ = φ.

We next show that the theory of symmetric Maass forms for S L(n, Z) (with
n even) is very similar to the S L(2, Z) theory. Basically, there are only two
types of such Maass forms, even and odd Maass forms.
Proposition 9.2.6 Assume n ≥ 2 is an even integer. Let φ be a symmetric
Maass form for S L(n, Z) with Fourier coefficients A(m 1 , . . . , m n−1 ) as in the
expansion (9.1.2). Fix
⎛ ⎞
−1
⎜ 1 ⎟
⎜ ⎟
Tδ0 = ⎜ .. ⎟.
⎝ . ⎠
1
Then for all m j ≥ 1 (with j = 1, 2, . . . , n − 1), we have
A(m 1 , . . . , m n−2 , m n−1 ) = ±A(m 1 , . . . , m n−2 , −m n−1 ),
according as Tδ0 φ = ±φ. The Maass form φ is said to be even or odd
accordingly.
266 The Godement–Jacquet L-function

Proof The diagonal elements with ±1 entries are generated by such elements
with determinant 1 and the additional special element Tδ0 . Since φ is invariant
under left multiplication by S L(n, Z) it follows that φ is symmetric if and only
if Tδ φ = ±φ.
Now δ0 zδ0 transforms xn−1 → −xn−1 and fixes x j with 1 ≤ j ≤ n − 2, it
follows that the integral
1 1 3 4
··· Tδ0 φ(z) e−2πi m 1 x1 +···+m n−1 xn−1
d∗x
0 0

picks off the A(m 1 , . . . , m n−2 , −m n−1 ) coefficient of φ, and this must be the
same as ±A(m 1 , . . . , m n−2 , m n−1 ) because Tδ0 φ = ±φ.

9.3 Hecke operators for SL(n, Z)

We recall the general definition of Hecke operators given in Definition 3.10.5.
Consider a group G that acts continuously on a topological space X . Let 
be a discrete subgroup of G. For every g in C G (), the commensurator of
 in G, (i.e., (g −1 g) ∩  has finite index in both  and g −1 g) we have a
decomposition of a double coset into disjoint right cosets of the form
(
g = αi . (9.3.1)
i

For each such g, the Hecke operator Tg : L2 (\X ) → L2 (\X ) is defined by

Tg f (x) = f (αi x),
i

where f ∈ L (\X ), x ∈ X, and αi are given by (9.3.1). The Hecke ring consists
2

of all formal sums

ck Tgk
k

with integer coefficients ck and gk in a semigroup  as in Definition 3.10.8.

Since two double cosets are either identical or totally disjoint, it follows that
unions of double cosets are associated to elements in the Hecke ring. Finally, we
recall Theorem 3.10.10 which states that the Hecke ring is commutative if there
exists an antiautomorphism g → g ∗ (i.e., (gh)∗ = h ∗ g ∗ ) for which  ∗ =  and
(g)∗ = g for every g ∈ .
We now consider, for n ≥ 2, the general case
G = G L(n, R),  = S L(n, Z), X = G L(n, R)/(O(n, R) · R× ) = hn .
 9.3 Hecke operators for SL(n, Z) 267

For every n-tuple of positive integers (m 0 , m 1 , . . . , m n−1 ), the matrix

⎛ ⎞
m 0 · · · m n−1
⎜ .. ⎟
⎜ . ⎟
⎜ ⎟ ∈ C G (),
⎝ m0m1 ⎠
m0

the commensurator of  in G (defined in (3.10.2)). We define  to be the

semigroup generated by all such matrices. As in the case of S L(2, Z), we have
the antiautomorphism

g → t g, g ∈ ,

where tg denotes the transpose of the matrix g. It is again clear that the conditions
of Theorem 3.10.10 are satisfied so that the Hecke ring is commutative.
The following lemma is analogous to Lemma 3.12.1, which came up in the
S L(2, Z) situation.

Lemma 9.3.2 Fix a positive integer N ≥ 1. Define the set

⎧⎛ ⎞  ⎫
⎪ c1 c1,2 ··· c1,n  ⎪
⎪
⎪  c ≥ 1 ( =1,2,...,n) ⎪
⎪
⎨⎜ c2 ··· c2,n ⎟  ⎬
⎜ ⎟  )n
S N := ⎜ .. .. ⎟  c =N .
⎪
⎪ ⎝ . . ⎠  =1 ⎪
⎪
⎪
⎩  0≤ci, <c (1≤i< ≤n) ⎪
⎭
cn 

Then one has the disjoint partition

⎛ ⎞
m 0 · · · m n−1
( ⎜ .. ⎟ (
⎜ . ⎟
⎜ ⎟ = α. (9.3.3)
⎝ m0m1 ⎠ α∈S
1 ···m n−1 =N
m n0 m n−1 N

m0

Proof First of all we claim the decomposition is disjoint. If not, there exists
⎛ ⎞
γ1,1 γ1,2 ··· γ1,n
⎜ γ2,1 γ2,2 ··· γ2,n ⎟
⎜ ⎟
⎜ . .. .. ⎟ ∈ 
⎝ .. . . ⎠
γn,1 γn,2 ··· γn,n
268 The Godement–Jacquet L-function

such that
⎛ ⎞ ⎛ ⎞ ⎛    ⎞
γ1,1 γ1,2 · · · γ1,n c1 c1,2 · · · c1,n c1 c1,2 · · · c1,n
⎜ γ2,1 γ2,2 · · · γ2,n ⎟ ⎜ c2 · · · c2,n ⎟ ⎜  ⎟
c2 · · · c2,n
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜ . .. .. ⎟ · ⎜ .. .. ⎟ = ⎜ .. .. ⎟ .
⎝ .. . . ⎠ ⎝ . . ⎠ ⎝ . . ⎠
γn,1 γn,2 · · · γn,n cn cn
(9.3.4)

This implies that γi, j = 0 for 1 ≤ j < i ≤ n. Consequently, γ , · c = c  ,

)
n
for 1 ≤ ≤ n. But γ , = 1 and c , c  ≥ 1 (1 ≤ ≤ n). It easily follows
=1
that

γ1,1 = γ2,2 = · · · = γn,n = 1.

Note that the above shows that c = c  (1 ≤ ≤ n). Therefore, (9.3.4) takes
the form
⎛ ⎞ ⎛ ⎞ ⎛   ⎞
1 γ1,2 · · · γ1,n c1 c1,2 · · · c1,n c1 c1,2 · · · c1,n
⎜ ⎟
1 · · · γ2,n ⎟ ⎜ ⎜ ⎟
c2 · · · c2,n ⎟ ⎜ ⎜ c2  ⎟
· · · c2,n
⎜ ⎟
⎜ . .. ⎟ · ⎜ . .. ⎟ = ⎜ .. .. ⎟ .
⎝ . . . ⎠ ⎝ . . . ⎠ ⎝ . . ⎠
1 cn cn

Since 0 ≤ ci, , ci, < c for 1 ≤ i < ≤ n, one concludes that γi, = 0 for
1 ≤ i < ≤ n, and the decomposition is disjoint as claimed.
Now, by Theorem 3.11.2, every element on the right-hand side of (9.3.3) can
be put into Smith normal form, so must occur as an element on the left-hand
side of (9.3.3). Similarly, by Theorem 3.11.1, every element on the left-hand
side of (9.3.3) can be put into Hermite normal form, so must occur as an element
on the right-hand side of (9.3.3). This proves the equality of the two sides of
(9.3.3).

By analogy with the S L(2, Z) situation (see (3.12.3)), it follows that for
every integer N ≥ 1, we have a Hecke operator TN acting on the space of
square integrable automorphic forms f (z) with z ∈ hn . The action is given by
the formula
⎛⎛ ⎞ ⎞
c1 c1,2 · · · c1,n

⎜⎜ c2 · · · c2,n ⎟ ⎟
1 ⎜⎜ ⎟ ⎟
TN f (z) = n−1/2 f ⎜⎜ . . ⎟ · z ⎟ (9.3.5)
N )
n ⎝⎝ .. .. ⎠ ⎠
c =N
=1 cn
0≤ci, <c (1≤i< ≤n)
 9.3 Hecke operators for SL(n, Z) 269

Note the normalizing factor of 1/N (n−1)/2 which was chosen to simplify later
formulae. Clearly, T1 is just the identity operator.
The C-vector space L2 (\hn ) has a natural inner product, denoted , , and
defined by

 f, g = f (z)g(z) d ∗ z,
\hn

for all f, g ∈ L2 (\hn ), where d ∗ z denotes the left invariant measure given in
Proposition 1.5.3.
In the case of S L(2, Z), we showed in Theorem 3.12.4 that the Hecke oper-
ators are self-adjoint with respect to the Petersson inner product. For S L(n, Z)
with n ≥ 3, it is no longer true that the Hecke operators are self-adjoint. What
happens is that the adjoint operator is again a Hecke operator and, therefore,
the Hecke operator commutes with its adjoint which means that it is a normal
operator.

Theorem 9.3.6 (Hecke operators are normal operators) Consider the Hecke
operators TN , (N = 1, 2, . . . ) defined in (9.3.5). Let TN∗ be the adjoint operator
which satisfies
 
TN f, g = f, TN∗ g

for all f, g ∈ L2 (\hn ). Then TN∗ is another Hecke operator which commutes
with TN so that TN is a normal operator. Explicitly, TN∗ is associated to the
following union of double cosets:
⎛ ⎞
N · m −1
0
( ⎜ N · (m 0 m 1 )−1 ⎟
⎜ ⎟
⎜ . ⎟ .
⎝ . . ⎠
1 ···m n−1 =N
m n0 m n−1
−1
N · (m 0 · · · m n−1 )
(9.3.7)

Proof It follows from (9.3.3), and also from the fact that transposition is an
antiautomorphism (as in the proof of Theorem 3.10.10), that
⎛ ⎞
m 0 · · · m n−1
( ⎜ .. ⎟ ( (
⎜ . ⎟
⎜ ⎟= α = α. (9.3.8)
⎝ m0m1 ⎠ α∈S α∈S
n n−1
m0 m1 ···m n−1 =N N N

m0
270 The Godement–Jacquet L-function

Since the action of the Hecke operator is independent of the choice of right
coset decomposition, we obtain


1
TN f, g = (n−1)/2 f (αz) g(z) d ∗ z
N α∈S N
\hn


1
= (n−1)/2 f (z) g(α −1 z) d ∗ z (9.3.9)
N α∈S N
\hn
⎛⎛ ⎞ ⎞
N

⎜⎜ .. ⎟ ⎟
1 ⎜⎜ . ⎟ −1 ⎟ ∗
= f (z) g ⎜⎜ ⎟α · z ⎟ d z,
N (n−1)/2 α∈S N
⎝⎝ N ⎠ ⎠
\hn
N

after making the change of variables z → α −1 z. Multiplying by the diagonal

matrix (with N s on the diagonal) above does not change anything because g is
well defined on hn .
Now, it follows from (9.3.8) that for
⎛ ⎞
1
⎜ −1 ⎟
⎜ ⎟
⎜ 1 ⎟
ω=⎜ ⎟,
⎜ . ⎟
⎝ .. ⎠
1
we have
⎛ ⎞−1
m 0 · · · m n−1
( ⎜ .. ⎟
⎜ . ⎟
 ·ω⎜ ⎟ · ω−1 
⎝ m0m1 ⎠
1 ···m n−1 =N
m n0 m n−1
m0
⎛ ⎞
m −1
0
( ⎜ (m 0 m 1 )−1 ⎟
⎜ ⎟
= ·⎜ .. ⎟ · .
⎝ . ⎠
m n0 m 1n−1 ···m n−1 =N
(m 0 · · · m n−1 )−1
(9.3.10)

Finally,
⎛ if ⎞
we multiply both sides of (9.3.10) by the diagonal matrix
N
⎜ .. ⎟
⎠, with N = m n0 m 1 · · · m n−1 , it follows that the adjoint Hecke
n−1
⎝ .
N
 9.3 Hecke operators for SL(n, Z) 271

operator defined by (9.3.9) is, in fact, associated to the union of double cosets
given in (9.3.7). This completes the proof.

The Hecke operators commute with the G L(n, R)-invariant differential oper-
ators, and they also commute with the operators Tδ given in (9.2.3). It follows
by standard methods in functional analysis, that we may simultaneously diago-
nalize the space L2 (S L(n, Z)\hn ) by all these operators. We shall be interested
in studying Maass forms which are eigenfunctions of the full Hecke ring of all
such operators. The following theorem is analogous to Theorem 3.12.8 which
came up in the S L(2, Z) situation.
Theorem 9.3.11 (Multiplicativity of the Fourier coefficients) Consider



∞

∞
A(m 1 , . . . , m n−1 )
φ(z) = ···
γ ∈Un−1 (Z)\S L(n−1,Z) m 1 =1 m n−2 =1 m n−1 =0
)
n−1
|m k |k(n−k)/2
k=1
⎛⎛ ⎞ ⎞
m 1 · · · |m n−1 |
⎜⎜ .. ⎟  ⎟
⎜⎜ . ⎟ γ ⎟
× WJacquet ⎜⎜ ⎟· z, ν, ψ1,...,1, mn−1 ⎟,
⎝⎝ m1 ⎠ 1 |m n−1 |⎠

1
a Maass form for S L(n, Z), as in (9.1.2). Assume that φ is an eigenfunction
of the full Hecke ring. If A(1, . . . , 1) = 0, then φ vanishes identically. Assume
φ = 0 and it is normalized so that A(1, . . . , 1) = 1. Then
Tm φ = A(m, 1, . . . , 1) · φ, ∀ m = 1, 2, . . .
Furthermore, we have the following multiplicativity relations
 
A m 1 m 1 , . . . , m n−1 m n−1 = A(m 1 , . . . , m n−1 ) · A m 1 , . . . , m n−1 ,
if (m 1 · · · m n−1 , m 1 · · · m n−1 ) = 1, and

 
m 1 cn m 2 c1 m n−1 cn−2
A(m, 1, . . . , 1)A(m 1 , . . . , m n−1 ) = A , ,...,
)
n c1 c2 cn−1
c =m
=1
c1 |m 1 ,c2 |m 2 ,...,cn−1 |m n−1

Addendum Let φ̃ denote the Maass form dual to φ. As in Proposition 9.2.1,

the (m 1 , . . . , m n−1 ) Fourier coefficient of φ̃ is A(m n−1 , . . . , m 1 ). If φ is an
eigenform of the full Hecke ring then
A(m n−1 , . . . , m 1 ) = A(m 1 , . . . , m n−1 ),
i.e., it is the complex conjugate of the Fourier coefficient of φ.
272 The Godement–Jacquet L-function

Proof Let z = x · y with

⎛ ⎞ ⎛ ⎞
1 x1,2 x1,3 ··· x1,n y1 y2 · · · yn−1
⎜ 1 x2,3 ··· x2,n ⎟ ⎜ y1 y2 · · · yn−2 ⎟
⎜ ⎟ ⎜ ⎟
⎜ .. .. .. ⎟ ⎜ .. ⎟
x =⎜ . . . ⎟, y=⎜ . ⎟.
⎜ ⎟ ⎜ ⎟
⎝ 1 xn−1,n ⎠ ⎝ y1 ⎠
1 1

In view of Theorem 5.3.2, Remark 9.1.1, and formula (9.1.2), we may write
for, m 1 , . . . , m n−1 ≥ 1,

1 1
··· φ(z) e−2πi (m 1 xn−1,n +m 2 xn−2,n−1 +···+m n−1 x1,2 ) d ∗ x
0 0
⎛⎛ ⎞ ⎞
m 1 y1 · · · m n−1 yn−1
A(m 1 , . . . , m n−1 ) ⎜⎜ .. ⎟ ⎟
⎜⎜ . ⎟ ⎟
= n−1 · WJacquet ⎜⎜ ⎟, ν, ψ1,...,1⎟,
) k(n−k)/2 ⎝⎝ m 1 y1 ⎠ ⎠
mk
k=1 1
(9.3.12)

)
where d ∗ x = d xi, j . While the notation adopted in Remark 9.1.1 is
1≤i< j≤n−1
useful in most cases, it makes the Hecke operator computations extremely
gruesome, so we temporarily return to our earlier notation for x.
If φ is an eigenfunction of the Hecke operator Tm defined by (9.3.5), then we
have Tm f (z) = λm f (z) for some eigenvalue λm . We can compute λm directly
using a variation of (9.3.12). We begin by considering for m = 1, 2, . . . ,

⎛⎛ ⎞ ⎞
m 1 y1 · · · m n−1 yn−1
A(m 1 , . . . , m n−1 ) ⎜⎜ .. ⎟ ⎟
⎜⎜ . ⎟ ⎟
λm n−1 · WJacquet ⎜⎜ ⎟, ν, ψ1,...,1 ⎟
) k(n−k)/2 ⎝⎝ m 1 y1 ⎠ ⎠
mk
k=1 1
m m
1
= ··· Tm φ(z) e−2πi (m 1 xn−1,n +m 2 xn−2,n−1 +···+m n−1 x1,2 ) d ∗ x
m n(n−1)/2
0 0
 9.3 Hecke operators for SL(n, Z) 273

m m
1
= ···
m (n+1)(n−1)/2 )
n
c =m 0 0
=1
0≤ci, <c (1≤i< ≤n)
⎛ ⎛ ⎞ ⎞
⎛ ⎞ 1 x1,2 x1,3 ··· x1,n
⎜ c1 c1,2 · · · c1,n ⎜ ⎟ ⎟
⎜⎜ 1 x2,3 ··· x2,n
⎜⎜ c2 · · · c2,n ⎟
⎟⎜
⎜
..
⎟ ⎟
⎟ ⎟
× φ ⎜⎜ ⎟ ⎜ .. .. ⎟ · y⎟
⎜⎝ .. .
.. ⎠ ⎜ . . .⎟ ⎟
⎝ . ⎝ 1 xn−1,n ⎠ ⎠
cn
1
× e−2πi (m 1 xn−1,n +m 2 xn−2,n−1 +···+m n−1 x1,2 ) d ∗ x. (9.3.13)

Next, if we let

⎛ ⎞
⎛ ⎞ 1 x1,2 x1,3 ··· x1,n
c1 c1,2 · · · c1,n ⎜ ⎟
⎜ 1 x2,3 ··· x2,n
⎜ c2 · · · c2,n ⎟
⎟⎜
⎜
..
⎟
⎟
⎜ .. ..
.. .. ⎟ ⎜ . . . ⎟
⎝ . . ⎠⎜⎝
⎟
1 xn−1,n ⎠
cn
1
⎛    ⎞
1 x1,2 x1,3 ··· x1,n ⎛ ⎞
⎜ 1 
x2,3 ··· 
x2,n ⎟ c1
⎜ ⎟⎜ c2 ⎟
⎜ .. .. .. ⎟⎜ ⎟
=⎜ . . . ⎟⎜ .. ⎟,
⎜ ⎟⎝ . ⎠
⎝ 1 
xn−1,n ⎠
cn
1

then we may solve for xi, j (1 ≤ i < j ≤ n), and obtain

1

j
xi, j = ci,k xk, j , (9.3.14)
c j k=i

with the understanding that ci,i = ci and xi,i = 1 for i = 1, 2, . . . , n. Note the
special case:

 ci xi,i+1 + ci,i+1
xi,i+1 = (i = 1, 2, . . . , n − 1), (9.3.15)
ci+1

of (9.3.14).
274 The Godement–Jacquet L-function

With the computations (9.3.14), (9.3.15), the right-hand side of the identity
(9.3.13) becomes

ci m +
j ci,k xk, j
cj + cj

 
k=i+1
1
m (n+1)(n−1)/2 )
n 1≤i< j≤n
c =m +
j ci,k xk, j
=1 cj
0≤ci, <c (1≤i< ≤n) k=i+1

⎛⎛    ⎞ ⎞
1 x1,2 x1,3 ··· x1,n ⎛ ⎞
⎜⎜   ⎟ c1 ⎟
⎜⎜ 1 x2,3 ··· x2,n ⎟⎜ ⎟ ⎟
⎜⎜ .. ⎟⎜ c2 ⎟ ⎟
× φ ⎜⎜ .. .. ⎟⎜ ⎟ · y⎟
⎜⎜ . . . ⎟⎝ .. ⎠ ⎟
⎝⎝  ⎠ . ⎠
1 xn−1,n
cn
1
+ cr,r +1 m n−r
n−1 +
n−1
cr +1 m n−r 
2πi −2πi xr,r +1 cj
×e ·e · d xi, j .
cr cr
r =1 r =1
ci

In view of the periodicity of the above integrand, we may deduce that it takes
the form:

ci m

 c j
1
m (n+1)(n−1)/2 )
n 1≤i< j≤n
c =m 0
=1
0≤ci, <c (1≤i< ≤n)
⎛⎛    ⎞ ⎞
1 x1,2 x1,3 ··· x1,n ⎛ ⎞
⎜⎜   ⎟ c1 ⎟
⎜⎜ 1 x2,3 ··· x2,n ⎟⎜ ⎟ ⎟
⎜⎜ .. ⎟⎜ c2 ⎟ ⎟
× φ ⎜⎜ .. .. ⎟⎜ ⎟ · y⎟
⎜⎜ . . . ⎟⎝ .. ⎠ ⎟
⎝⎝  ⎠ . ⎠
1 xn−1,n
cn
1
+ cr,r +1 m n−r
n−1 +
n−1
cr +1 m n−r 
2πi −2πi xr,r +1 cj
×e ·e · d xi, j .
cr cr
r =1 r =1
ci

But the above integral vanishes unless cr | cr +1 m n−r for all r = 1, 2, . . . ,

n − 1. Furthermore, in this case we have

⎧)
n

2πi
+ cr,r +1 m n−r
n−1
⎨ ct−1 if cr | m n−r (1 ≤ r < n),
t
=
cr
e r =1 t=2
0≤ci, <c (1≤i< ≤n)
⎩
0 otherwise.
 9.3 Hecke operators for SL(n, Z) 275

Consequently, it follows that our integral (9.3.13) may be written in the form:
ci m


n  c j
1
ctt−1
m (n+1)(n−1)/2 )
n t=2 1≤i< j≤n
c =m 0
=1
⎛⎛    ⎞ ⎞
1 x1,2 x1,3 ··· x1,n ⎛ ⎞
⎜⎜   ⎟ c1 ⎟
⎜⎜ 1 x2,3 ··· x2,n ⎟⎜ ⎟ ⎟
⎜⎜ .. ⎟⎜ c2 ⎟ ⎟
× φ ⎜⎜ .. .. ⎟⎜ ⎟ · y⎟
⎜⎜ . . . ⎟⎝ .. ⎠ ⎟
⎝⎝  ⎠ . ⎠
1 xn−1,n
cn
1
+
n−1
cr +1 m n−r 
−2πi xr,r +1 cj
×e · d xi, j .
cr
r =1
ci
Further, if F : R → C is a periodic
*M function satisfying
*1 F(x + 1) = F(x) and
M is a positive integer, then 0 F(x) d x = M · 0 F(x) d x. Consequently, the
integral above is equal to


n  1
1
ctt−1
m (n−1)/2 )
n t=2 1≤i< j≤n
c =m 0
=1
⎛⎛    ⎞ ⎞
1 x1,2 x1,3 ··· x1,n ⎛ ⎞
⎜⎜   ⎟ c1 ⎟
⎜⎜ 1 x2,3 ··· x2,n ⎟⎜ ⎟ ⎟
⎜⎜ .. ⎟⎜ c2 ⎟ ⎟
× φ ⎜⎜ .. .. ⎟⎜ ⎟ · y⎟
⎜⎜ . . . ⎟⎝ .. ⎠ ⎟
⎝⎝  ⎠ . ⎠
1 xn−1,n
cn
1
+
n−1
cr +1 m n−r 
−2πi xr,r +1
×e d xi, j .
cr
r =1 (9.3.16)
) *1
Finally, the multiple integral, · · · , above can be evaluated with
1≤i< j≤n 0
(9.3.12) and has the value
 
A mcn−1
1 cn
, mc2n−2
cn−1
, . . . , m n−1
c1
c2

)
n−1 k(n−k)/2
cn+1−k m k
cn−k
k=1
⎛⎛ ⎞ ⎞
m 1 y1 · · · m n−1 yn−1 cn
⎜⎜ .. ⎟ ⎟
⎜⎜ . ⎟ ⎟
× WJacquet ⎜⎜ ⎟, ν, ψ1,...,1 ⎟ ,
⎝⎝ m 1 y1 cn ⎠ ⎠
cn
276 The Godement–Jacquet L-function

from which it follows from (9.3.13) and (9.3.16) that

⎛⎛ ⎞ ⎞
m 1 y1 · · · m n−1 yn−1
A(m 1 , . . . , m n−1 ) ⎜ ⎜ .. ⎟ ⎟
⎜⎜ . ⎟ ⎟
λn n−1 · WJacquet ⎜⎜ ⎟, ν, ψ1,...,1 ⎟
) k(n−k)/2 ⎝⎝ m 1 y1 ⎠ ⎠
mk
k=1 1
 
m 1 cn m 2 cn−1
1

n A cn−1
, cn−2 , . . . , m n−1
c1
c2

= ctt−1 ·
m (n−1)/2 )
n t=2
)
n−1
cn+1−k m k
k(n−k)/2
c =m cn−k
=1 k=1
⎛⎛ ⎞ ⎞
m 1 y1 · · · m n−1 yn−1 cn
⎜⎜ .. ⎟ ⎟
⎜⎜ . ⎟ ⎟
× WJacquet ⎜⎜ ⎟, ν, ψ1,...,1 ⎟ .
⎝⎝ m 1 y1 cn ⎠ ⎠
cn

Note that we may cancel the Whittaker functions on both sides of the above
identity because the Whittaker functions are invariant under multiplication by
scalar matrices. Further, one easily checks the identity

)
n
ctt−1
t=2
k(n−k)/2 = (c1 · c2 · · · cn ) = m (n−1)/2 .
(n−1)/2
)
n−1
cn+1−k
cn−k
k=1

It immediately follows that

 
m 1 cn m 2 cn−1 m n−1 c2
λm A(m 1 , . . . , m n−1 ) = A , ,..., .
)
n cn−1 cn−2 c1
c =m
=1
cn−1 |m 1 , cn−2 |m 2 ,..., c1 |m n−1
(9.3.17)

We now explore the consequences of the assumption that A(1, . . . , 1) = 0.

It follows easily from (9.3.17) that A(k, 1, . . . , 1) = 0 for all integers k, and
then the left-hand side of (9.3.17) vanishes for all m, m 1 as long as m 2 = m 3
= · · · = m n−1 = 1. By choosing m 2 = · · · = m n−1 = 1, m 1 = p, m = p
one obtains A(1, p, 1, . . . , 1) = 0. Arguing inductively, we may choose m 2 =
· · · = m n−1 = 1, m 1 = p, m = p for = 1, 2, . . . from which one can con-
clude that A( p , p, 1, . . . , 1) = 0 for all = 0, 1, 2, . . . One can continue
to show that A( pi1 , pi2 , 1, . . . , 1) = 0 for all non-negative integers i 1 , i 2 , and
 9.4 The Godement–Jacquet L-function 277

proceeding inductively one may show that


A pi1 , pi2 , . . . , pin−1 = 0

for all non-negative integers i 1 , . . . , i n−1 . One may then proceed to products
of two primes, products of three primes, etc. to eventually obtain that if
A(1, . . . , 1) = 0 then all coefficients A(m 1 , . . . , m n−1 ) must vanish.
If f = 0 then we may assume it is normalized so that A(1, . . . , 1) = 1.
If we now choose m 1 = m 2 = · · · = m n−1 = 1, it immediately follows from
(9.3.17) that λm = A(m, 1, . . . , 1). Substituting this into (9.3.17), and changing
indices (on the c j s), proves the identity

A(m, 1, . . . , 1)A(m 1 , . . . , m n−1 )

 
m 1 cn m 2 c1 m n−1 cn−2
= A , ,..., .
)n c1 c2 cn−1
c =m
=1
c1 |m 1 , c2 |m 2 ,..., cn−1 |m n−1

The rest of the proof of Theorem 9.3.11 follows easily.

To prove the addendum, consider the identity T φ, φ = φ, T ∗ φ, T is a
Hecke operator and T ∗ is the adjoint operator, and   denotes the Petersson
inner product. The addendum follows from Theorem 9.3.6.

9.4 The Godement–Jacquet L-function

Let



∞

∞
A(m 1 , . . . , m n−1 )
f (z) = ···
γ ∈Un−1 (Z)\S L(n−1,Z) m 1 =1 m n−2 =1 m n−1 =0
)
n−1
|m k |k(n−k)/2
k=1
⎛⎛ ⎞ ⎞
m 1 · · · m n−2 · |m n−1 |
⎜⎜ .. ⎟  ⎟
⎜⎜ . ⎟ ⎟
⎜⎜ ⎟ γ ⎟
× WJacquet ⎜⎜ m1m2 ⎟· z, ν, ψ1,...,1, mn−1 ⎟,
⎜⎜ ⎟ 1 |m n−1 |⎟
⎝⎝ m1 ⎠ ⎠
1

be a non-zero Maass form for S L(n, Z), normalized so that A(1, . . . , 1) =

1, which is a simultaneous eigenfunction of all the Hecke operators as in
Theorem 9.3.11. We want to build an L-function out of the Fourier coefficients
278 The Godement–Jacquet L-function

of f . Lemma 9.1.3 tells us that we may form absolutely convergent Dirichlet

series in a suitable half-plane.
The sum over γ ∈ Un−1 (Z)\S L(n − 1, Z) in the Fourier expansion of f
creates seemingly insurmountable complications, and it is not possible to simply
set

xi, j = 0 (1 ≤ i < j ≤ n),

and then take the n − 1-fold Mellin transform in y1 , y2 , . . . , yn−1 which

would be the exact analogue of what we did to create L-functions in the
S L(2, Z) situation. The ingenious construction of the L-functions and the
proof of their functional equations was first obtained by Godement and Jacquet
(1972).
By Theorem 9.3.11, the Fourier coefficients, A(m 1 , . . . , m n−1 ), of f satisfy
the multiplicativity relations

A(m, 1, . . . , 1)A(m 1 , . . . , m n−1 )

 
m 1 cn m 2 c1 m n−1 cn−2
= A , ,..., .
)n c1 c2 cn−1
c =m
=1
c1 |m 1 , c2 |m 2 ,..., cn−1 |m n−1

It follows that for all k = 1, 2, . . .

A( p k , 1, . . . , 1)A( p, 1, . . . , 1) = A( p k+1 , 1, . . . , 1) + A( p k−1 , p, 1, . . . , 1),

A( p k , 1, . . . , 1)A(1, p, 1, . . . , 1) = A( p k , p, 1, . . . , 1)
+ A( p k−1 , 1, p, 1, . . . , 1),
A( p k , 1, . . . , 1)A(1, 1, p, 1, . . . , 1) = A( p k , 1, p, 1, . . . , 1)
+ A( p k−1 , 1, 1, p, 1, . . . , 1),
..
.
A( p k , 1, . . . , 1)A(1, . . . , 1, p) = A( p k , 1, . . . , 1, p) + A( p k−1 , 1, . . . , 1),

with the understanding that A(1, . . . , p − j , . . . , 1) = 0 for any j ≥ 1.

Therefore,

1
n−2
· (−1)r A( p k−r , 1, . . . , 1)A( 1, . . . , p , . . . , 1)
p ks r =0 " #$ %
position r +1
A( p k+1
, 1, . . . , 1) + (−1) n−2
A( p k−n+1
, 1, . . . , 1)
= . (9.4.1)
p ks
 9.4 The Godement–Jacquet L-function 279

If we define

∞
A( p k , 1, . . . , 1)
φ p (s) := ,
k=0
p ks

then, after summing over k, equation (9.4.1) implies that

⎡ ⎤

n−2
φ p (s) · ⎣ (−1)r A( 1, . . . , p , . . . , 1) p −r s ⎦
" #$ %
r =0
position r +1

= φ p (s) p s − p s + (−1)n−2 φ p (s) p (−n+1)s .

Solving for φ p (s) yields


φ p (s) = 1 − A( p, . . . , 1) p −s + A(1, p, . . . , 1) p −2s −
−1
· · · + (−1)n−1 A(1, . . . , p) p (−n+1)s + (−1)n p −ns . (9.4.2)

In a manner completely analogous to the situation of S L(2, Z), as in

Definition 3.13.3, it is natural make the following definition.

Definition 9.4.3 Let s ∈ C with (s) > (n + 1)/2, and let f (z) be a Maass
form for S L(n, Z), with n ≥ 2, which is an eigenfunction of all the Hecke
operators as in Theorem 9.3.11. We define the Godement–Jacquet L-function
L f (s) (termed the L-function associated to f ) by the absolutely convergent
series


∞ 
L f (s) = A(m, 1, . . . , 1)m −s = φ p (s),
m=1 p

with φ p (s) given by (9.4.2).

Remark It is clear that the L-function associated to the dual Maass form f˜
takes the form


∞
L f˜ (s) = A(1, . . . , 1, m)m −s .
m=1

By analogy with the G L(2) situation, we would like to construct the L-

function L f (s) as a Mellin transform of the Maass form f . However, before
taking the Mellin transform, it is necessary to kill the sum over S L(n − 1) in the
Fourier Whittaker expansion (9.1.2). The procedure to do this uses an auxiliary
integral which requires some preliminary preparation.
280 The Godement–Jacquet L-function

Set
⎛ ⎞
m 1 · · · m n−2 · |m n−1 |
⎜ .. ⎟
⎜ . ⎟
⎜ ⎟
M =⎜ m1m2 ⎟,
⎜ ⎟
⎝ m1 ⎠
1
and z = x y with
⎛ ⎞
1 xn−1 x1,3 ··· x1,n
⎜ 1 xn−2 ··· x2,n ⎟
⎜ ⎟
⎜ .. .. .. ⎟ ,
x =⎜ . . . ⎟
⎜ ⎟
⎝ 1 x1 ⎠
1
⎛ ⎞
y1 y2 · · · yn−1
⎜ y1 y2 · · · yn−2 ⎟
⎜ ⎟
⎜ .. ⎟
y=⎜ . ⎟.
⎜ ⎟
⎝ y1 ⎠
1
A simple computation gives
⎛ ⎞
1 |m n−1 |xn−1
⎜ ∗ ⎟
⎜ ⎟
⎜ ⎟
⎜ .. .. ⎟
M·x =⎜ ⎜
. . ⎟ · M.
⎟
⎜ 1 m 2 x2 ⎟
⎜ ⎟
⎝ 1 m 1 x1 ⎠
1
It follows from Definition 5.4.1 (2), that for any integers 1 , . . . , n−1 , the
Jacquet Whittaker function satisfies

WJacquet (M z, ν, ψ1 ,...,n−1 ) = e2πi [m 1 1 x1 +···+m n−2 n−2 xn−2 +|m n−1 |n−1 xn−1 ]
× WJacquet (M · y, ν, ψ1 ,...,n−1 ).

Further, for any S L(n − 1, Z) matrix

⎛ ⎞
a1,1 ··· a1,n−1
⎜ a2,1 ··· a2,n−1 ⎟
⎜ ⎟
γ =⎜ . .. ⎟, (9.4.4)
⎝ .. . ⎠
an−1,1 · · · an−1,n−1
 9.4 The Godement–Jacquet L-function 281

 
γ
we may put the G L(n, R) matrix, · z, into Iwasawa form:
1
⎛ γ γ γ ⎞
1 xn−1 x1,3 · · · x1,n
⎜ γ γ ⎟
  ⎜ 1 xn−2 · · · x2,n ⎟
γ ⎜ ⎟
·z ≡⎜⎜
..
.
..
.
.. ⎟
⎟
1 ⎜ .
⎝ γ ⎟
1 x1 ⎠
1
⎛ γ γ γ ⎞
y1 y2 · · · yn−1
⎜ γ γ γ ⎟
⎜ y1 y2 · · · yn−2 ⎟
⎜ ⎟
×⎜⎜ . ⎟ (mod Z n O(n, R)),
.. ⎟
⎜ ⎟
⎝ y
γ ⎠
1
1
(9.4.5)
γ
where x1 = an−1,1 x1,n + an−1,2 x2,n + · · · + an−1,n−1 x1 .
It immediately follows from Proposition 5.5.2, (9.4.4) and (9.4.5) that we
may write
   
γ
WJacquet M · · z, ν, ψ1 ,...,n−1
1
3 γ γ
4
= e2πi m 1 1 (an−1,1 x1,n +an−1,2 x2,n +···+an−1,n−1 x1 )+m 2 2 x2 +···+|m n−1 |n−1 xn−1
⎛ ⎛ γ γ ⎞ ⎞
y1 · · · yn−1
⎜ ⎜ .. ⎟ ⎟
⎜ ⎜ . ⎟ ⎟
× WJacquet ⎜ M · ⎜ ⎟ , ν, ψ 1,...,1 ⎟ . (9.4.6)
⎝ ⎝ y
γ ⎠ ⎠
1
1

Finally, we obtain the following theorem which is the basis for the construc-
tion of the L-function L f (s) (given in Definition 9.4.3) as a Mellin transform.

Theorem 9.4.7 Let f (z) be a Maass form of type ν for S L(n, Z) as in (9.1.2).
Then we have the representation,



∞

∞
A(m 1 , . . . , m n−1 )
f (z) = ···
γ ∈Un−1 (Z)\S L(n−1,Z) m 1 =1 m n−2 =1 m n−1 =0
)
n−1
|m k |k(n−k)/2
k=1
γ γ
2πim 1 (an−1,1 x1,n +···+an−1,n−1 x1 ) 2πi (m 2 x2 +···+m n−1 xn−1 )
×e e
× WJacquet (M · y γ , ν, ψ1,...,1 ),
282 The Godement–Jacquet L-function

 
γ
where γ is given by (9.4.4) and, x γ , y γ , are defined by: z ≡ x γ · yγ ,
1
as in (9.4.5).

Proof The proof follows from the Fourier expansion (9.1.2) and the identity
(9.4.6).

Note that Theorem 9.4.7 is a direct generalization of Theorem 6.5.7. The

explicit realization of the Fourier expansion of the Maass form f given in
Theorem 9.4.7 is of fundamental importance. It is the basis for the construc-
tion of the Godement–Jacquet L-function as a Mellin transform of a certain
projection operator acting on the Maass form f .

Corollary 9.4.8 Let f (z) be a Maass form of type ν for S L(n, Z) as in (9.1.2).
Then
⎛⎛ ⎞ ⎞
1 0 · · · 0 u 1,n
1 1 ⎜ ⎜⎜
⎜ ..
.
.. .
. ..
.. ⎟ ⎟
. ⎟ ⎟ 
n−2
⎜⎜ ⎟ ⎟ −2πiu 1
··· f ⎜⎜ 1 0 u ⎟ · z ⎟ e du 1 du j,n
⎜⎜ n−2,n ⎟ ⎟
0 0 ⎝ ⎝ 1 u1 ⎠ ⎠ j=1

1



∞

∞

= ···
γ ∈Un−2 (Z)\S L(n−2,Z) m 2 =1 m n−2 =1 m n−1 =0
A(1, m 2 , . . . , m n−1 ) 2πi (x1 +m 2 x2γ +···+m n−1 xn−1
γ
)
× n−1 e
)
|m k |k(n−k)/2
k=2
⎛⎛ γ γ ⎞ ⎞
m 2 · · · |m n−1 |y1 y2 · · · yn−1
⎜⎜ .. ⎟ ⎟
⎜⎜ . ⎟ ⎟
⎜⎜ ⎟ ⎟
× WJacquet ⎜⎜ m
γ
2 y1 y2
⎟, ν, ψ1,...,1 ⎟ .
⎜⎜ ⎟ ⎟
⎝⎝ y1 ⎠ ⎠
1

Proof We may use Theorem 9.4.7, to compute the integral:

⎛⎛ ⎞ ⎞
1 ··· u 1,n
1 1 ⎜ ⎜
⎜⎜
..
.
.. ⎟ ⎟
. ⎟ ⎟ 
n−2
⎜⎜ ⎟ ⎟ −2πiu 1
··· f ⎜⎜ 1 u ⎟ · z ⎟ e du 1 du j,n .
⎜⎜ n−2,n ⎟ ⎟
0 0 ⎝⎝ 1 u1 ⎠ ⎠ j=1

1
 9.4 The Godement–Jacquet L-function 283

The key point is that the integral

1 1  
n−2
2πim 1 an−1,1 u 1,n +···+an−1,n−2 u n−2,n +an−1,n−1 u 1
··· e e−2πiu 1 du 1 du j,n
j=1
0 0

vanishes unless m 1 an−1,n−1 = 1 (which implies m 1 = 1, an−1,n−1 = 1),

and

an−1,1 = 0, an−1,2 = 0, ··· , an−1,n−2 = 0.

The proof of Corollary 9.4.8 follows from this after noting that

⎛ ⎞
a1,1 ··· a1,n−2 a1,n−1
⎜ a2,1 ··· a2,n−2 a2,n−1 ⎟
⎜ ⎟
⎜ .. .. .. .. ⎟
γ =⎜ . . . . ⎟ ∈ Un−1 (Z)\S L(n − 1, Z)
⎜ ⎟
⎝ an−2,1 an−2,n−2 an−2,n−1 ⎠
0 ··· 0 1

forces a1,n−1 = a2,n−1 = · · · = an−2,n−1 = 0, so that γ takes the form

⎛ ⎞
a1,1 ··· a1,n−2 0
⎜ a2,1 ··· a2,n−2 0⎟
⎜ ⎟
⎜ .. .. .. .. ⎟ ∈ U (Z)\S L(n − 2, Z).
γ =⎜ . . . .⎟ n−2
⎜ ⎟
⎝ an−2,1 an−2,n−2 0 ⎠
0 ··· 0 1

Theorem 9.4.9 Let f (z) be a Maass form of type ν for S L(n, Z) as in (9.1.2).
Then for

⎛ ⎞
1 0 u 1,3 ··· u 1,n
⎜ 1 u n−2 ··· u 2,n ⎟
⎜ ⎟ 
n−2 
⎜ .. .. .. ⎟ ,
û = ⎜ . . . ⎟ d ∗ û = uj du i, j ,
⎜ ⎟
⎝ 1 u1 ⎠
j=1 2≤i< j−1≤n−1

1
284 The Godement–Jacquet L-function

we have
1 1
··· f (ûz)e−2πi (u 1 +···+u n−2 ) d ∗ û
0 0

A(1, . . . , 1, m)
= e2πimxn−1 e2πi(x1 +···+xn−2 )
m=0
|m| n−1/2

⎛⎛ ⎞ ⎞
|m|
⎜⎜ 1 ⎟ ⎟
⎜⎜ ⎟ ⎟
× WJacquet ⎜⎜ .. ⎟ · y; ν, ψ1,...,1 ⎟ .
⎝⎝ . ⎠ ⎠
1

GL(n)pack functions The following GL(n)pack functions, described in the

appendix, relate to the material in this chapter:
HeckeCoefficientSum HeckeOperator HeckePowerSum
HeckeEigenvalue HeckeMultiplicativeSplit.
 10
Langlands Eisenstein series

The modern theory of Eisenstein series began with Maass (1949), who formally
defined and studied the series (see Section 3.1)
1
ys
E(z, s) = ,
2 (c,d)=1 |cz + d|2s

taking the viewpoint that it is an eigenfunction of the Laplacian. In (Roelcke,

1956), Eisenstein series for more general discrete groups commensurable with
S L(2, Z) were investigated. It was in (Selberg, 1956, 1963) that the spec-
tral theory and the meromorphic continuation of Eisenstein series was fully
worked out for G L(2). The Selberg spectral decomposition given in Section
3.16 underscores the supreme importance of Eisenstein series in number theory.
In (Selberg, 1960) (see also (Hejhal, 1983)) an extremely ingenious analytic
method is introduced for obtaining the meromorphic continuation of Eisenstein
series for higher rank groups, but it was not clear if the method would work
for Langlands Eisenstein series twisted by Maass forms defined on lower rank
groups.
The completion of this program in, perhaps, the most general context was
attained by (Langlands, 1966, 1976). There were two main parts to Langlands
theory.
r The meromorphic continuation of Eisenstein series.
r The complete spectral decomposition of arithmetic quotients \G where G
is a reductive group and  is an arithmetic group.

An excellent summary of Langlands theory of Eisenstein series was given in

(Arthur, 1979). In the intervening years, two books have been published giving
expositions of Langlands theory of Eisenstein series: (Osborne and Warner,
1981) and (Moeglin and Waldspurger, 1995) (see also (Jacquet, 1997)). A few

285
286 Langlands Eisenstein series

years ago, a new proof of the meromorphic continuation of Langlands Eisenstein

series has been attained by Bernstein (2002).
Another important direction in the theory of Eisenstein series was also ini-
tiated by Langlands who had the idea of studying automorphic L-functions by
investigating the constant term (see (Langlands, 1971)) in the Fourier expan-
sion of Eisenstein series. This theme was further developed in (Shahidi, 1981,
1988, 1990a, b, 1992), and is now called the Langlands–Shahidi method. This
method has had a number of striking successes, one of the first being (Moeglin
and Waldspurger, 1989) that the completed Rankin–Selberg L-function for
G L(n) × G L(m) is holomorphic in the region 0 < (s) < 1. More recently,
Gelbart and Shahidi (1988), Shahidi (1985, 1990a,b), Cogdell, Kim, Piatetski-
Shapiro and Shahidi (2001), and Kim and Shahidi (2000) have led to many
new examples of entire L-functions including the symmetric cube and fourth
power lifts of G L(2) Maass forms. The remarkable fact is that these particular
symmetric power L-functions occur in the constant term of Eisenstein series
which are associated to exceptional Lie groups! Unfortunately, since there are
only a few exceptional Lie groups this puts a severe constraint on what one can
expect to get by this method.
In this chapter we shall give an elementary exposition of Langlands Eisen-
stein series for the group S L(n, Z). We only discuss meromorphic continuation
that can be obtained from Fourier–Whittaker expansions or the Poisson sum-
mation formula. These methods work quite well for Eisenstein series that are
not twisted by Maass forms of lower rank. We present a short elementary intro-
duction to the Langlands–Shahidi method with an application to non-vanishing
of L-functions on the line (s) = 1. Finally, in Section 10.13, we give a simple
proof (due to M. Thillainatesan) of the Langlands spectral decomposition for
G L(3, R).

10.1 Parabolic subgroups

We shall define the standard parabolic subgroups for G L(n, R) (with n ≥ 2)
in an explicit manner, avoiding the more general abstract theory. Briefly,
the standard parabolic subgroups are certain subgroups containing the stan-
dard Borel subgroup B which consists of all upper triangular non-singular
matrices.
Each standard parabolic subgroup of G L(n, R) is associated to a partition
n = n 1 + n 2 + · · · + nr ,
where 1 ≤ n 1 , n 2 , . . . , n r < n are integers.
 10.1 Parabolic subgroups 287

Definition 10.1.1 The standard parabolic subgroup associated to the partition

n = n 1 + n 2 + · · · + n r is denoted Pn 1 ,...,nr , and is defined to be the group of
all matrices of the form
⎛ ⎞
mn 1 ∗ ··· ∗
⎜ 0 mn 2 · · · ∗ ⎟
⎜ ⎟
⎜ 0 0 ··· ∗ ⎟
⎜ ⎟,
⎜ . .. .. .. ⎟
⎝ .. . . . ⎠
0 0 · · · mnr
where mni ∈ G L(n i , R) for 1 ≤ i ≤ r. The integer r is termed the rank of the
parabolic subgroup Pn 1 ,...,nr .

Example 10.1.2 (Parabolic subgroups of G L(3, R)) There are three stan-
dard parabolic subgroups of G L(3, R) corresponding to the three partitions:

3 = 1 + 1 + 1, 3 = 1 + 2, 3 = 2 + 1.

Explicitly, we have
⎧⎛ ⎞⎫ ⎧⎛ ⎞⎫
⎨ ∗ ∗ ∗ ⎬ ⎨ ∗ ∗ ∗ ⎬
P1,1,1 = ⎝ 0 ∗ ∗ ⎠ , P1,2 = ⎝0 ∗ ∗⎠ ,
⎩ ⎭ ⎩ ⎭
0 0 ∗ 0 ∗ ∗
⎧⎛ ⎞⎫
⎨ ∗ ∗ ∗ ⎬
P2,1 = ⎝ ∗ ∗ ∗ ⎠ .
⎩ ⎭
0 0 ∗
Example 10.1.3 (Parabolic subgroups of G L(4, R)) There are seven stan-
dard parabolic subgroups of G L(4, R) corresponding to the seven partitions:

4 = 1 + 1 + 1 + 1, 4 = 1 + 1 + 2, 4 = 1 + 2 + 1,
4 = 2 + 1 + 1, 4 = 1 + 3, 4 = 2 + 2, 4 = 3 + 1.

Explicitly, we have
⎧⎛ ⎞⎫ ⎧⎛ ⎞⎫
⎪
⎪ ∗ ∗ ∗ ∗ ⎪⎪ ⎪
⎪ ∗ ∗ ∗
∗ ⎪⎪
⎨⎜ ⎬ ⎨⎜ ⎬
0 ∗ ∗ ∗⎟ 0 ∗ ∗⎟
∗
P1,1,1,1 = ⎜ ⎝
⎟ , P1,1,2 = ⎜ ⎟ ,
⎪
⎪ 0 0 ∗ ∗ ⎠⎪
⎪ ⎪
⎪
⎝ 0 0 ∗ ⎠⎪
∗ ⎪
⎩ ⎭ ⎩ ⎭
0 0 0 ∗ 0 0 ∗∗
⎧⎛ ⎞⎫ ⎧⎛ ⎞⎫
⎪
⎪ ∗ ∗ ∗ ∗ ⎪⎪ ⎪
⎪ ∗ ∗ ∗ ∗ ⎪⎪
⎨⎜ ⎬ ⎨⎜ ⎬
0 ∗ ∗ ∗⎟ ∗ ∗ ∗ ∗⎟
P1,2,1 = ⎜ ⎝
⎟ , P2,1,1 = ⎜ ⎟ ,
⎪
⎪ 0 ∗ ∗ ∗ ⎠⎪
⎪ ⎪
⎪
⎝ 0 0 ∗ ∗ ⎠⎪
⎪
⎩ ⎭ ⎩ ⎭
0 0 0 ∗ 0 0 0 ∗
288 Langlands Eisenstein series

⎧⎛ ⎞⎫ ⎧⎛ ⎞⎫
⎪
⎪ ∗ ∗ ∗ ∗ ⎪⎪ ⎪
⎪ ∗ ∗ ∗ ∗ ⎪ ⎪
⎨⎜ ⎬ ⎨⎜ ⎬
0 ∗ ∗ ∗⎟ ∗ ∗ ∗ ∗⎟
P1,3 = ⎜⎝0 ∗
⎟ , P3,1 = ⎝⎜ ⎟ ,
⎪
⎪ ∗ ∗ ⎠⎪
⎪ ⎪
⎪ ∗ ∗ ∗ ∗ ⎠⎪⎪
⎩ ⎭ ⎩ ⎭
0 ∗ ∗ ∗ 0 0 0 ∗
⎧⎛ ⎞⎫
⎪
⎪ ∗ ∗ ∗ ∗ ⎪⎪
⎨⎜ ⎬
∗ ∗ ∗ ∗⎟
P2,2 = ⎜⎝0 0
⎟ .
⎪
⎪ ∗ ∗ ⎠⎪
⎪
⎩ ⎭
0 0 ∗ ∗
Definition 10.1.4 (Associate parabolics) Fix integers n ≥ 2, 1 < r ≤ n.
Two standard parabolic subgroups Pn 1 ,...,nr , Pn  1 ,...,Pn r of G L(n, R), corre-
sponding to the partitions,

n = n 1 + · · · + n r = n 1 + · · · + n r ,

are said to be associate (Pn 1 ,...,nr ∼ Pn 1 ,...,nr ) if the set of integers {n 1 , . . . , n r }

is a permutation of the set of integers {n 1 , . . . , n r }.

Definition 10.1.5 (Weyl group of associate parabolics) Fix n ≥ 2 and

1 < r ≤ n. Let P = Pn 1 ,...,nr , P  = Pn 1 ,...,nr be two associate parabolic sub-
groups of G L(n, R) corresponding to the partitions,

n = n 1 + · · · + n r = n 1 + · · · + n r .

The Weyl group, denoted (P, P  ), consists of all σ ∈ Sr (permutation group

on r symbols) such that n i = n σ (i) for all i = 1, 2, . . . , r. We shall also let (P)
denote (P, P).

10.2 Langlands decomposition of parabolic subgroups

Let n ≥ 2, and fix a partition n = n 1 + n 2 + · · · + n r with 1 ≤ n 1 , n 2 , . . . , n r
< n. The parabolic subgroup (see Definition 10.1.1)
⎧⎛ ⎞⎫
⎪ mn 1 ∗ ··· ∗ ⎪
⎪
⎪ ⎪
⎪ ⎜ ⎟⎪⎪
⎨⎜ 0 mn 2 · · · ∗ ⎟⎪
⎪ ⎬
⎜ 0 0 ··· ∗ ⎟ ⎟
Pn 1 ,...,nr = ⎜ (10.2.1)
⎪
⎪ ⎜ . . . . ⎟⎪⎪
⎪⎝ .
⎪ . . . . . ⎠⎪
. ⎪
⎪
⎩
. ⎪
⎭
0 0 · · · mnr
of G L(n, R) can be factored as

Pn 1 ,...,nr = Nn 1 ,...,nr · Mn 1 ,...,nr (10.2.2)

 10.2 Langlands decomposition of parabolic subgroups 289

where
⎧⎛ ⎞⎫
⎪ In 1 ∗ ··· ∗ ⎪
⎪
⎪ ⎪
⎪⎜ 0
⎪ ··· ∗ ⎟⎪⎪
⎨⎜ In 2 ⎟⎪⎬
⎜ ··· ∗ ⎟
Nn 1 ,...,nr = ⎜ 0 0 ⎟ (Ik = k × k Identity matrix),
⎪⎜ .
⎪ ⎟⎪
⎪
⎪ ⎝ .. .. ..
.
.. ⎠⎪⎪
⎪
⎪
⎩
. . ⎪
⎭
0 0 ··· In r
is the unipotent radical and
⎧⎛ ⎞⎫
⎪ mn 1 0 ··· 0 ⎪
⎪
⎪ ⎪
⎪⎜ 0 mn 2
⎪ ··· ⎟⎪
⎪
⎨⎜ 0 ⎟⎪
⎬
⎜ ··· ⎟
Mn 1 ,...,nr = ⎜ 0 0 0 ⎟ (mk ∈ G L(k, R)),
⎪
⎪ ⎜ . ⎟⎪
⎪
⎪ ⎝ .. .. ..
.
.. ⎠⎪
⎪
⎪
⎪
⎩
. . ⎪
⎭
0 0 · · · mnr
is the so-called Levi component. The Levi component further decomposes into
the direct product

Mn 1 ,...,nr = An 1 ,...,nr · Mn 1 ,...,nr (10.2.3)

where An 1 ,...,nr is the connected center of Mn 1 ,...,nr :

⎧⎛ ⎞⎫
⎪ t1 · In 1 0 ··· 0 ⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎜ t2 · I n 2 · · · 0 ⎟ ⎪
⎨⎜ 0 ⎟⎪⎬
⎜ 0 0 ··· 0 ⎟ ⎟
An 1 ,...,nr = ⎜ (ti ∈ R, ti > 0),
⎪
⎪⎜ .. ⎟ ⎪
⎪
⎪
⎪ ⎝ ... .. ..
. . ⎠⎪ ⎪
⎪
⎩
. ⎪
⎭
0 0 · · · tr · Inr
and
⎧⎛  ⎞⎫
⎪ mn 0 ··· 0 ⎪
⎪
⎪⎜ 0 1 ⎪
⎪
⎪
⎪ mn 2 ··· ⎟⎪
⎨⎜ 0 ⎟⎪⎬  
⎜ ··· ⎟
Mn 1 ,...,nr = ⎜ 0 0 0 ⎟ det mni = ±1, i = 1, . . . , r .
⎪
⎪ ⎜ . ⎟⎪
⎪
⎪ ⎝ .. .. ..
.
.. ⎠⎪⎪
⎪
⎪
⎩
. . ⎪
⎭
0 0 · · · mnr

Definition 10.2.4 (Langlands decomposition) The Langlands decomposition

of a parabolic subgroup of the form (10.2.1) gives the factorizations (10.2.2),
(10.2.3).

Example 10.2.5 (Langlands decomposition on G L(3, R)) We explicitly

write down the Langlands decomposition for the three parabolic subgroups
290 Langlands Eisenstein series

of G L(3, R).
⎧⎛ ⎞⎫
⎨ ∗ ∗ ∗ ⎬
P1,1,1 = ⎝ 0 ∗ ∗⎠
⎩ ⎭
0 0 ∗
⎧⎛ ⎞ ⎛ ⎞ ⎛ ⎞⎫
⎨ 1 ∗ ∗ t1 0 0 ±1 0 0 ⎬
= ⎝0 1 ∗⎠ · ⎝ 0 t2 0⎠·⎝ 0 ±1 0⎠ ,
⎩ ⎭
0 0 1 0 0 t3 0 0 ±1
(t1 , t2 , t3 > 0)
⎧⎛ ⎞⎫
⎨ ∗ ∗ ∗ ⎬
P1,2 = ⎝0 ∗ ∗⎠
⎩ ⎭
0 ∗ ∗
⎧⎛ ⎞ ⎛ ⎞ ⎛ ⎞⎫
⎨ 1 ∗ ∗ t1 0 0 ±1 0 0 ⎬
= ⎝0 1 0⎠ · ⎝ 0 t2 0⎠·⎝ 0 a b⎠ ,
⎩ ⎭
0 0 1 0 0 t2 0 c d
(t1 , t2 > 0, ad − bc = ±1)
⎧⎛ ⎞⎫
⎨ ∗ ∗ ∗ ⎬
P2,1 = ⎝∗ ∗ ∗⎠
⎩ ⎭
0 0 ∗
⎧⎛ ⎞ ⎛ ⎞ ⎛ ⎞⎫
⎨ 1 0 ∗ t1 0 0 a b 0 ⎬
= ⎝ ⎠
0 1 ∗ · 0⎝ t1 0⎠·⎝c d 0⎠ ,
⎩ ⎭
0 0 1 0 0 t2 0 0 ±1
(t1 , t2 > 0, ad − bc = ±1).
Example 10.2.6 (Langlands decomposition for P1,3,2 ) Finally, we shall give
an example of the Langlands decomposition for the parabolic subgroup P1,3,2
of G L(6, R). We have
⎧⎛ ⎞⎫
⎪
⎪ ∗ ∗ ∗ ∗ ∗ ∗ ⎪ ⎪
⎪
⎪⎜ ⎟⎪⎪
⎪
⎪ ⎜ ∗ ∗ ∗ ∗ ∗ ⎟ ⎪
⎪
⎪
⎨⎜ ⎟⎪⎬
⎜ ∗ ∗ ∗ ∗ ∗⎟
P1,3,2 = ⎜ ⎟
⎪
⎪ ⎜ ∗ ∗ ∗ ∗ ∗ ⎟⎪
⎪⎜
⎪ ⎟⎪⎪
⎪
⎪ ⎝ ∗ ∗ ⎠⎪⎪
⎪
⎪
⎩ ⎪
⎭
∗ ∗
⎧⎛ ⎞⎫ ⎧⎛ ⎞⎫
⎪
⎪ 1 ∗ ∗ ∗ ∗ ∗ ⎪ ⎪ ⎪
⎪ ∗ ⎪
⎪
⎪
⎪ ⎜ ⎪
⎟⎪ ⎪
⎪ ⎜ ⎟⎪⎪
⎪
⎪ ⎜ 1 0 0 ∗ ∗ ⎪
⎟⎪ ⎪
⎪ ⎜ ∗ ∗ ∗ ⎟⎪⎪
⎪
⎨⎜ ⎪
⎟⎬ ⎨⎜⎪ ⎟⎪⎬
⎜ 0 1 0 ∗ ∗⎟ ⎜ ∗ ∗ ∗ ⎟
= ⎜ ⎟ · ⎜ ⎟ ,
⎪
⎪ ⎜ 0 0 1 ∗ ∗ ⎟⎪ ⎪ ⎜ ∗ ∗ ∗ ⎟⎪
⎪⎜
⎪ ⎟⎪
⎪
⎪
⎪
⎪
⎪ ⎜ ⎟⎪⎪
⎪
⎪
⎪ ⎝ 1 0 ⎪
⎠⎪ ⎪⎝⎪ ∗ ∗ ⎠⎪⎪
⎪
⎩ ⎪
⎭ ⎩ ⎪ ⎪
⎭
0 1 ∗ ∗
unipotent radical Levi component
 10.2 Langlands decomposition of parabolic subgroups 291

and the further decomposition of the Levi component

⎧⎛ ⎞⎫
⎪
⎪ ∗ ⎪
⎪
⎪
⎪ ⎜ ⎟⎪⎪
⎪
⎪ ⎜ ∗ ∗ ∗ ⎟ ⎪
⎪
⎪⎜
⎨ ⎟⎪⎬
⎜ ∗ ∗ ∗ ⎟
⎜ ⎟
⎪
⎪ ⎜ ∗ ∗ ∗ ⎟⎪
⎪
⎪ ⎜ ⎟⎪⎪
⎪
⎪ ⎝ ∗ ∗ ⎠⎪⎪
⎪
⎪
⎩ ⎪
⎭
∗ ∗
Levi component
⎧⎛ ⎞⎫ ⎧⎛ ⎞⎫
⎪
⎪ t1 ⎪
⎪ ⎪
⎪ ∗ ⎪
⎪
⎪
⎪ ⎜ ⎟⎪⎪ ⎪
⎪ ⎜ ⎟⎪
⎪
⎪
⎪ ⎜ t2 0 0 ⎟⎪⎪ ⎪
⎪⎜ ∗ ∗ ∗ ⎟⎪
⎪
⎪
⎨⎜ ⎟⎪⎬ ⎪
⎨⎜ ⎟⎪
⎬
⎜ 0 t2 0 ⎟ ⎜ ∗ ∗ ∗ ⎟
= ⎜ ⎟ · ⎜ ⎟ ,
⎪⎜
⎪ 0 0 t2 ⎟⎪⎪ ⎪
⎪ ⎜ ∗ ∗ ∗ ⎟⎪
⎪
⎪ ⎜ ⎟⎪ ⎪ ⎜ ⎟⎪
⎪
⎪⎝
⎪
⎪ t3 0 ⎠⎪ ⎪ ⎪
⎪
⎪
⎪⎝
⎪
⎪ ∗ ∗ ⎠⎪
⎪
⎪
⎪
⎩ ⎭ ⎩ ⎭
0 t3 ∗ ∗
A1,3,2 
M1,3,2


where t1 , t2 , t3 > 0 and the block matrices in M1,3,2 have determinant ±1.
The parabolic subgroups of G L(n, R) can be characterized as stabilizers of
flags on Rn . A flag of Rn is a sequence of subspaces:

φ ⊂ V1 ⊂ V2 · · · ⊂ Vr = Rn

where ⊂ denotes a proper subset, and φ is the empty set. The action of G L(n, R)
on a flag (V1 , . . . , Vr ) is defined in the canonical way. That is if g ∈ G L(n, R)
then the action is given by g(V1 , . . . , Vr ) = (gV1 , . . . , gVr ), where the action
of g on an element (a1 , . . . , an ) ∈ Rn is given by matrix multiplication
g · t(a1 , . . . , an ). The standard complete flag is φ ⊂ V1 ⊂ · · · ⊂ Vn = Rn
where Vi = Re1 ⊕ · · · ⊕ Rei and ei is the vector (of length n) with a 1 in
the ith position and zeros elsewhere. The stabilizer of a subflag (Vd1 , . . . , Vdr )
(with 0 < d1 < · · · < dr = n) of the standard flag has the form
⎛ ⎞
m11 m12 m13 · · ·
⎜ 0 m22 m23 · · · ⎟
⎜ ⎟
⎜ 0 0 m33 · · · ⎟ , (10.2.7)
⎝ ⎠
.. .. .. ..
. . . .

where mii is a square matrix of size di − di−1 for i = 1, 2, . . . , r, where, by

convention, d0 = 0.
If a parabolic subgroup P ≤ G L(n, R) is the stabilizer of a flag (V1 , . . . , Vr ),
then for every g ∈ P, we must have gVi = Vi for i = 1, . . . , r. It follows that
g induces an automorphism gi of Vi /Vi−1 for every 1 ≤ i ≤ r. Here, we set
292 Langlands Eisenstein series

V0 to be the empty set. In the case that P takes the form (10.2.7), we have
gi = mii . The unipotent radical N P of P is the subgroup of all g ∈ P so that
gi is the identity on Vi /Vi−1 for every i. For P of the form (10.2.7), mii must
be the identity matrix for every i. In a similar manner we may define the Levi
component using flags. For each 1 ≤ i ≤ r choose a complementary subspace
X i ⊂ Rn so that Vi = Vi−1 ⊕ X i . The Levi component M P of P is defined to
be the subgroup of P consisting of all g ∈ P which stablize each X i . In the
case that P is of the form (10.2.7), the Levi component requires that mi j = 0
for 1 ≤ i < j ≤ n.

10.3 Bruhat decomposition

Let Sn denote the symmetric group of all permutations of n symbols. We have
a homomorphism of Sn into G L(n, R) whose image is the Weyl group Wn
consisting of all n × n matrices which have exactly one 1 in each row and
column, and zeros elsewhere.

Example 10.3.1 (Weyl group for G L(3, R)) The Weyl group W3 is the group
of six elements:
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
1 0 0 1 0 0 0 1 0
⎝0 1 0⎠, ⎝0 0 1⎠, ⎝1 0 0⎠,
0 0 1 0 1 0 0 0 1
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
0 1 0 0 0 1 0 0 1
⎝0 0 1⎠, ⎝1 0 0⎠, ⎝0 1 0⎠.
1 0 0 0 1 0 1 0 0
Recall that the standard Borel subgroup Bn of G L(n, R) is the group of invertible
upper triangular matrices.

Proposition 10.3.2 (Bruhat decomposition) For n ≥ 2, we have

G L(n, R) = Bn Wn Bn .

Proof Let
⎛ ⎞
g11 g12 ··· g1n
⎜ g21 g22 ··· g2n ⎟
⎜ ⎟
g=⎜ . .. .. ⎟ ∈ G L(n, R).
⎝ .. . ··· . ⎠
gn1 gn2 ··· gnn
 10.3 Bruhat decomposition 293

Let gn denote the first non-zero entry in the bottom row of g. Then by right
multiplication by some b1 ∈ Bn , we can change this entry to 1 and make the
rest of the bottom row 0. The matrix gb1 now takes the form
⎛    ⎞
g11 ··· g1 ··· g1n
⎜ g21
··· 
g2 ··· 
g2n ⎟
⎜ ⎟
⎜ .. . . ⎟
gb1 = ⎜ . ··· .
. ··· . ⎟
. .
⎜ ⎟
⎝ g  
· · · gn−1, · · · gn−1,n ⎠
n−1,1
0 ··· 1 ··· 0
If we now multiply gb1 on the left by some matrix
⎛     ⎞
b11 b12 b13 · · · b1n
⎜ 0 b22  
b23  ⎟
· · · b2n
⎜ ⎟
⎜   ⎟
· · · b3n
b1 = ⎜ 0 0 b33 ⎟ ∈ Bn ,
⎜ . .. .. .. .. ⎟
⎝ .. . . . . ⎠
0 0 0 ··· 1
it is easy to see that we may choose the bi, j (1 ≤ i < j ≤ n) so that b1 gb1 takes
the form
⎛   ⎞
g11 ··· 0 ··· g1n
⎜ g21
··· 0 ··· 
g2n ⎟
⎜ ⎟
 ⎜ .. . . ⎟
b1 gb1 = ⎜ . ··· . ···. . ⎟
. .
⎜ ⎟
⎝ g ··· 0 ··· g  ⎠
n−1,1 n−1,n
0 ··· 1 ··· 0
We call this clearing the (n)th row and ( )th column.
We next consider the first non-zero entry in the (n − 1)st row of b1 gb1 .

Suppose it is gn−1, 1
.We may again multiply b1 gb1 on the right by some element
b2 ∈ Bn so that we change this entry to 1 and make all other entries in the
(n − 1)st row 0. By left multiplication by some b2 we can make all the other
entries in the ( 1 )st column 0 which results in clearing the (n − 1)st row and
( 1 )st column.
Continuing in this manner, we obtain a set of n matrices

b1 gb1 , b2 b1 gb1 b2 , b3 b2 b1 gb1 b2 b3 , ... , bn · · · b3 b2 b1 gb1 b2 b3 · · · bn

where the last entry must lie in Wn .

We now seek a more explicit realization of the Bruhat decomposition. We

follow (Friedberg, 1987a,c). It is necessary to introduce some more notation.
Fix an integer n ≥ 2. For every λ = ( 1 , . . . , k ) ∈ Zk (with 1 ≤ k ≤ n) and
294 Langlands Eisenstein series

g ∈ G L(n, R), define Mλ (g) to be the k × k minor of g formed from the bottom
k rows and the columns 1 , . . . , k , indexed by the elements of λ. We may
express Mλ (g) using wedge products of e1 , e2 , . . . , en , where ei denotes the
column vector of length n with a 1 at position i and zeros elsewhere. We
have
 
Mλ (g)e1 ∧ · · · ∧ en = e1 ∧ · · · ∧ en−k ∧ g · e 1 ∧ · · · ∧ g · e k . (10.3.3)

We shall also define

⎛ ⎞
1 ∗
⎜ .. ⎟
Un = ⎝ . ⎠ ⊂ Bn , (10.3.4)
0 1

to be the subgroup of upper triangular unipotent (1s on the diagonal) matrices.

Definition 10.3.5 Let w ∈ Wn . We define ω ∈ Sn to be the permutation of the

set {1, 2, . . . , n} associated to w, and defined by wei = eω−1 (i) for all 1 ≤ i ≤ n.

Proposition 10.3.6 (Explicit Bruhat decomposition) Every g ∈ G L(n, R)

(n ≥ 2) has a Bruhat decompostion g = u 1 cwu 2 with u 1 , u 2 ∈ Un , w ∈ Wn ,
and
⎛ ⎞
/cn−1
⎜ cn−1 /cn−2 ⎟
⎜ ⎟
⎜ .. ⎟
c=⎜ . ⎟,
⎜ ⎟
⎝ c2 /c1 ⎠
c1
 = det(w)det(g), ci =
 0 (1 ≤ i < n).

Furthermore, for each 1 ≤ i ≤ n − 1,

 
|ci | =  M(ω(n),ω(n−1),...,ω(n−i+1)) (g),

with ω as in Definition 10.3.5, and M(ω(n),ω(n−1),...,ω(n−i+1)) (g) defined by

(10.3.3).

Proof It easily follows from Proposition 10.3.2 that every g ∈ G L(n, R) has a
Bruhat decompostion g = u 1 cwu 2 with u 1 , u 2 ∈ Un , w ∈ Wn , and c a diagonal
matrix. To determine c, we utilize (10.3.3) and write

Mλ (g)e1 ∧ · · · ∧ en = e1 ∧ · · · ∧ en−i ∧ geω(n) ∧ · · · ∧ geω(n−i+1) ,

 10.4 Minimal, maximal, and general parabolic Eisenstein series 295

with λ = (ω(n), ω(n − 1), . . . , ω(n − i + 1)). Since uei − ei ∈ Span(e1 , . . . ,

ei−1 ) for any u ∈ Un , and ek ∧ ek = 0 for any 1 ≤ k ≤ n, it follows that

Mλ (g)e1 ∧ · · · ∧ en = e1 ∧ · · · ∧ en−i ∧ cwu 2 · eω(n) ∧ · · · ∧ cwu 2 · eω(n−i+1) .

(10.3.7)

We now write u 2 = (µi, j ) (where µi, j denotes the i, j entry of the matrix
u 2 ). Note that for any 1 ≤ ≤ n,



u 2 e = u r, er .
r =1

It follows from this and (10.3.7) that




ω(n)
Mλ (g)e1 ∧ · · · ∧ en = e1 ∧ · · · ∧ en−i ∧ cw u r1 ,ω(n) er1
r1 =1





ω(n−1)

ω(n−i+1)
∧ cw u r2 ,ω(n−1) er2 ∧ · · · ∧ cw u ri ,ω(n−i+1) eri
r2 =1 ri =1




ω(n)
= e1 ∧ · · · ∧ en−i ∧ c u r1 ,ω(n) eω−1 (r1 )
r1 =1





ω(n−1)

ω(n−i+1)
∧ c u r2 ,ω(n−1) eω−1 (r2 ) ∧ · · · ∧ c u ri ,ω(n−i+1) eω−1 (ri )
r2 =1 ri =1

= e1 ∧ · · · ∧ en−i ∧ cen ∧ cen−1 ∧ · · · ∧ cen−i+1

= ci e1 ∧ e2 ∧ · · · ∧ en−i ∧ en ∧ · · · ∧ en−i+1 .

10.4 Minimal, maximal, and general parabolic

Eisenstein series
The minimal (smallest) standard parabolic subgroup for G L(n, R) (n ≥ 2) is
⎛ ⎞
∗ ∗ ··· ∗
⎜ ∗ ··· ∗⎟
⎜ ⎟
P1, 1, . . . , 1 = ⎜ .. . ⎟.
" #$ % ⎝ . .. ⎠
n ones
∗

Set  = S L(n, Z) and Pmin = P1,1,...,1 ∩ .

296 Langlands Eisenstein series

Let z ∈ hn take the form

⎛ ⎞ ⎛ ⎞
1 x1,2 x1,3 · · · x1,n y1 y2 · · · yn−1
⎜ 1 x2,3 · · · ⎟ ⎜ y1 y2 · · · yn−2
x2,n ⎟
⎜ ⎟ ⎜ ⎟
⎜ .. ..⎟ ⎜ .. ⎟
z=⎜ . .⎟·⎜ . ⎟.
⎜ ⎟ ⎜ ⎟
⎝ 1 xn−1,n ⎠ ⎝ y1 ⎠
1 1
Then the function (see also (5.1.1))
 n−1
n−1  b sj
Is (z) = yi i, j
i=1 j=1

with

ij if i + j ≤ n,
bi, j =
(n − i)(n − j) if i + j ≥ n,

and s ∈ Cn−1 is invariant under transformations of the form

z → p · z

with p ∈ Pmin . It follows that the sum

E Pmin (z, s) := Is (γ z) (10.4.1)
γ ∈Pmin \

is well defined provided it converges absolutely.

Definition 10.4.2 The series (10.4.1) is called the minimal parabolic Eisen-
stein series for .

Proposition 10.4.3 The minimal parabolic Eisenstein series (10.4.1) con-

verges absolutely and uniformly on compact subsets of hn to a  invariant
function provided s = (s1 , . . . , sn−1 ) and Re(si ) is sufficiently large for every
i = 1, 2, . . . , n − 1.

Proof The fact that E Pmin (z, s) is invariant under  is easy to prove because
the function is formed as a sum over a coset of the group . For the absolute
convergence, we follow Godement (see (Borel, 1966)). It is enough to show
that for every point z 0 ∈ \hn and some (non-zero volume) compact subset C z0
of \hn (with z 0 ∈ C z0 ), that the integral

 
 E P (z, s)d ∗ z
min

C z0
 10.4 Minimal, maximal, and general parabolic Eisenstein series 297

converges. Here d ∗ z is the invariant measure given in Theorem 1.6.1. Without

loss of generality, we may assume the si to be real. It follows that it is enough
to show that the integral


∗
Is (γ z) d z = Is (z) d ∗ z
γ ∈Pmin \
C z0 (Pmin \)·C z0

converges. Now, it follows from Proposition 1.3.2 that there will be only finitely
many γ ∈ Pmin \ such that γ z 0 ∈  √3 , 1 . By a continuity argument, one may
2 2
deduce, for sufficiently small C z0 , that there are only finitely many γ ∈ Pmin \
such that γ z ∈  √3 , 1 for all z ∈ C z0 . We immediately deduce that there exists
√ 2 2

some a ≥ 3/2 such that

γ z ∈ a, 12
for all γ ∈ Pmin \, z ∈ C z0 . It follows that
  1  1 a   n−1
a n−1  
Is (z) d ∗ z ≤
b sj
··· ··· yi i, j d xi, j
0 0 0 0 i=1 j=1 1≤i< j≤n
(Pmin \)·C z0


n−1
× yk−k(n−k)−1 dyk ,
k=1

where the integrals from 0 to 1 are integrals for the variables

xi, j (1 ≤ i < j ≤ n) and the integrals from 0 to a are integrals with respect
to the variables y1 , y2 , . . . , yn−1 . It is clear that the latter integral converges
absolutely if the si are all sufficiently large.

The two largest (maximal) parabolic subgroups of  = S L(n, Z) are

⎛ ⎞ ⎛ ⎞
∗ ∗ ··· ∗ ∗ ··· ∗ ∗
⎜0 ∗ ··· ∗⎟ ⎜ .. . .⎟
⎜ ⎟ ⎜ · · · .. .. ⎟
P1,n−1 = ⎜ . . .. ⎟ , Pn−1,1 = ⎜ . ⎟.
⎝ . . ··· . ⎠
. . ⎝∗ ··· ∗ ∗⎠
0 ∗ ··· ∗ 0 ··· 0 ∗
In a manner similar to the way we defined the minimal parabolic Eisenstein
series (10.4.1), we would like to define maximal parabolic Eisenstein series, and
more generally, Eisenstein series associated to any standard parabolic subgroup
P of G L(n, R). Since the function Is (z) is not usually invariant under transla-
tions z → γ z (with γ ∈ P ∩ ), we cannot simply sum over all left translations
γ ∈ P ∩ \ as we did in (10.4.1). Nevertheless, we will show that for proper
choice of s we can make Is (z) invariant under left multiplication z → γ z with
γ ∈ P ∩ .
298 Langlands Eisenstein series

The construction of such a P-invariant function Is (z) requires some prelim-

inary preparation. We shall make use of the Langlands decomposition
⎧⎛ ⎞ ⎛ ⎞⎫
⎪ In 1 ∗ · · · ∗ mn 1 0 ··· 0 ⎪
⎪
⎪ ⎪
⎪ ⎜ ⎟ ⎜ ⎟⎪⎪
⎨⎜ 0 In 2 · · · ∗ ⎟ ⎜ 0 mn 2 · · · 0 ⎟⎪
⎪ ⎬
⎜ 0 ⎟
0 ··· ∗ ⎟ · ⎜ 0 ⎜ 0 ··· 0 ⎟ ⎟
P = NM = ⎜
⎪
⎪ ⎜ . .. .. .. ⎟ ⎜ . .. .. .. ⎟ ⎪
⎪
⎪
⎪ ⎝ .. . . ⎠ ⎝ .. . . ⎠⎪⎪
⎪
⎩
. . ⎪
⎭
0 0 · · · In r 0 0 · · · mnr
(10.4.4)

where mk ∈ G L(k, R).

Definition 10.4.5 Let Pn 1 ,...,nr be a parabolic subgroup of G L(n, R), asso-

ciated to the partition: n = n 1 + · · · + n r , with a Langlands decomposition
+
r
(10.4.4). Let s = (s1 , . . . , sr ) ∈ Cr satisfy n i si = 0. Let K = O(n 1 , R) ×
i=1
O(n 2 , R) × · · · × O(n r , R) be the direct product of orthogonal groups. We
define the function, Is (∗, Pn 1 ,...,nr ), which maps,

Pn 1 ,...,nr /(K · R× ) → C,

by the formula

r 
 
Is (g, Pn 1 ,...,nr ) = Det mn (g) si ,
i
i=1

for all
⎛ ⎞ ⎛ ⎞
In 1 ∗ ··· ∗ mn 1 (g) 0 ··· 0
⎜ 0 In 2 ··· ∗ ⎟ ⎜ 0 mn 2 (g) · · · 0 ⎟
⎜ ⎟ ⎜ ⎟
⎜ ··· ∗ ⎟ ⎜ 0 ··· 0 ⎟
g=⎜ 0 0 ⎟·⎜ 0 ⎟ ∈ Pn 1 ,...,nr .
⎜ . .. .. .. ⎟ ⎜ . .. .. .. ⎟
⎝ .. . . . ⎠ ⎝ .. . . . ⎠
0 0 ··· In r 0 0 · · · mnr (g)

Remarks Here mni is the whole group G L(n i , R) while mni (g) is a particular
+
r
element in this group. The condition n i si = 0 insures that
i=1

Is (δg, Pn 1 ,...,nr ) = Is (g, Pn 1 ,...,nr )

for any matrix δ of the form δ = t · In with t ∈ R× , so that Is (∗, Pn 1 ,...,nr ) is well
defined on Pn 1 ,...,nr /R× . It is also clear that Is (gk, Pn 1 ,...,nr ) = Is (g, Pn 1 ,...,nr ) for
k ∈ K since the determinant of an orthogonal matrix has absolute value 1.
 10.4 Minimal, maximal, and general parabolic Eisenstein series 299

It follows that Is is well defined on the generalized upper half-plane hn . This

is because we may bring each matrix m ni (g) ∈ G L(n i , R) (i = 1, . . . , r ) into
diagonal form by right multiplication by an orthogonal matrix (see Proposition
1.2.6, the Iwasawa decomposition); and, then, with an additional multiplication
by t In with t ∈ R× we may bring the entire matrix into Iwasawa form.
In particular, if
⎛ ⎞ ⎛ ⎞
1 x1,2 x1,3 ··· x1,n y1 y2 · · · yn−1
⎜ 1 x2,3 ··· x2,n ⎟ ⎜ y1 y2 · · · yn−2 ⎟
⎜ ⎟ ⎜ ⎟
⎜ .. .. ⎟ ⎜ .. ⎟
z=⎜ . . ⎟·⎜ . ⎟ ∈ hn ,
⎜ ⎟ ⎜ ⎟
⎝ 1 xn−1,n ⎠ ⎝ y1 ⎠
1 1

then z ∈ Pn 1 ,...,nr , since the standard Borel subgroup lies in every parabolic
subgroup Pn 1 ,...,nr . It follows that

s1
s2

n 1
n−n
Is (z, Pn 1 ,...,nr ) = Y j1 · Y j2
j1 =n−n 1 +1 j2 =n−n 1 −n 2 +1

s3
sr

n−n 1 −n 2 
nr
× Y j3 ··· Y jr .
j3 =n−n 1 −n 2 −n 3 +1 jr =1

where we have defined Y1 , Y2 , . . . , Yn by

⎛ ⎞
⎛ ⎞ y1 y2 · · · yn−1
Yn ⎜ y1 y2 · · · yn−2 ⎟
⎜ Yn−1 ⎟ ⎜ ⎟
⎜ ⎟ ⎜ .. ⎟
⎜ .. ⎟=⎜ . ⎟.
⎝ . ⎠ ⎜ ⎟
⎝ y1 ⎠
Y1
1

One easily checks that Is (z, Pn 1 ,...,nr ) is precisely the standard function Is  (z)
for suitable choice of s  depending on s.

Lemma 10.4.6 Let Pn 1 ,...,nr be a parabolic subgroup of G L(n, R) with a

+
r
Langlands decomposition (10.4.4). Let s ∈ Cr where n i si = 0. The function
i=1
Is (∗, Pn 1 ,...,nr ), as in Definition 10.4.5, satisfies
 
Is γ z, Pn 1 ,...,nr = Is z, Pn 1 ,...,nr

for all γ ∈ Pn 1 ,...,nr ∩ S L(n, Z) and all z ∈ hn .

300 Langlands Eisenstein series

Proof If γ ∈ Pn 1 ,...,nr ∩ S L(n, Z), then γ has a Langlands decomposition

⎛ ⎞ ⎛ ⎞
In 1 ∗ ··· ∗ mn 1 (γ ) 0 ··· 0
⎜ 0 In 2 ··· ∗ ⎟ ⎜ 0 mn 2 (γ ) · · · 0 ⎟
⎜ ⎟ ⎜ ⎟
⎜ ··· ∗ ⎟ ⎜ 0 ··· 0 ⎟
γ =⎜ 0 0 ⎟·⎜ 0 ⎟
⎜ . .. .. .. ⎟ ⎜ . .. .. .. ⎟
⎝ .. . . . ⎠ ⎝ .. . . . ⎠
0 0 ··· In r 0 0 · · · mnr (γ )

where Det(mni (γ )) = ±1 for all i = 1, . . . , r. The lemma immediately

follows.

Definition 10.4.7 Let P = Pn 1 ,...,nr be a parabolic subgroup of G L(n, R)

+r
with a Langlands decomposition (10.4.4). Let s ∈ Cr where n i si = 0. We
i=1
define the Eisenstein series associated to P, denoted E P (z, s), by the infinite
series


E P (z, s) := Is (γ z, P),
γ ∈(P∩)\

where Is (z, P) is given in Definition 10.4.5.

The absolute convergence of the Eisenstein series E P (z, s) for s in a suitable

range follows from Proposition 10.4.3. This is because the function Is (z, P) is
actually a special case of the I-function given in (5.1.1) and the set P ∩ \
has fewer elements than the set Pmin \ since Pmin ⊂ P ∩ .

Example 10.4.8 Maximal parabolic Eisenstein series for S L(3, Z) Let

P1,2 , P2,1 denote the two maximal parabolic subgroups of G L(3, R) given
in Example 10.1.2. Then for s = (s1 , s2 ) ∈ C2 satisfying s1 + 2s2 = 0
and
⎛ ⎞ ⎛ ⎞
1 x1,2 x1,3 y1 y2
z=⎝ 1 x2,3 ⎠ · ⎝ y1 ⎠ ∈ h3 ,
1 1

we have
 1 s1
Is (z, P1,2 ) = (y1 y2 )s1 · y1s2 = y12 y2 .

Similarly, for s = (s1 , s2 ) ∈ C2 satisfying 2s1 + s2 = 0, we have

 s 1
Is (z, P2,1 ) = y12 y2 .
 10.5 Eisenstein series twisted by Maass forms 301

10.5 Eisenstein series twisted by Maass forms

Let P = Pn 1 ,...,nr be a parabolic subgroup of G L(n, R) which has a Langlands
decomposition P = N M as in (10.4.4).
We define a function:

m P : P → M,

by the formulae
⎛ ⎞
mn 1 (g) 0 ··· 0
⎜ 0 mn 2 (g) · · · 0 ⎟
⎜ ⎟
⎜ ··· 0 ⎟
m P (g) = ⎜ 0 0 ⎟,
⎜ . .. .. .. ⎟
⎝ .. . . . ⎠
0 0 · · · mnr (g)

for all
⎛ ⎞ ⎛ ⎞
In 1 ∗ ··· ∗ mn 1 (g) 0 ··· 0
⎜ 0 In 2 ··· ∗ ⎟ ⎜ 0 mn 2 (g) ··· 0 ⎟
⎜ ⎟ ⎜ ⎟
⎜ ··· ∗ ⎟ ⎜ 0 ··· ⎟
g =⎜ 0 0 ⎟·⎜ 0 0 ⎟ ∈ P.
⎜ . .. .. .. ⎟ ⎜ . .. .. .. ⎟
⎝ .. . . . ⎠ ⎝ .. . . . ⎠
0 0 ··· In r 0 0 · · · mnr (g)
(10.5.1)

Here, each mni (g) ∈ G L(n i , R) (i = 1, . . . , r ) as in (10.4.4). Note that every

z ∈ hn is also an element of P so that m P (z) is well defined.
Let φ be a Maass form for the group M. Then φ is really a set of r Maass forms
φ1 , φ2 , . . . , φr where each φi is a Maass form for G L(n i , R) (i = 1, 2, . . . , r ).
For g ∈ P, of the form (10.5.1), we define

r

φ (m P (g)) = φi mni (g) . (10.5.2)
i=1

We may now define the Langlands Eisenstein series twisted by Maass forms of
lower rank.

Definition 10.5.3 Let P = Pn 1 ,...,nr be a standard parabolic subgroup of

G L(n, R) with Langlands decomposition: P = NM, as in (10.4.4). Let
s = (s1 , . . . , sr ) ∈ Cr satisfy


r
n i si = 0,
i=1
302 Langlands Eisenstein series

and put  = S L(n, Z). Then for φ a cusp form on M and z ∈ hn , we define the
Eisenstein series E P (z, s, φ) by the infinite series


E P (z, s, φ) = φ (m P (γ z)) · Is (γ z, P),
γ ∈P∩\

where Is (z, P) is given by Definition 10.4.5 and φ is given by (10.5.2).

Example 10.5.4 (S L(3, Z)-Eisenstein series twisted by Maass forms) There
are two classes of Eisenstein series twisted by Maass forms in the case of
S L(3, Z). Let φ be a Maass form for S L(2, Z) with Fourier expansion
⎛⎛ ⎞⎞
y x
√
φ ⎝⎝ 1 ⎠⎠ = a(n) y K r (2π|n|y)e2πinx , (x ∈ R, y > 0),
1 n=0

say, as in Proposition 3.5.1.Note that, in order to conform to the notation (10.5.2),

it is necessary to consider φ as being defined on 3 × 3 matrices. Let s = (s1 , s2 )
with 2s1 + s2 = 0, and
⎛ ⎞ ⎛ ⎞
1 x1,2 x1,3 y1 y2
z=⎝ 1 x2,3 ⎠ · ⎝ y1 ⎠ ∈ h3 .
1 1
The first class of such Eisenstein series is associated with the parabolic subgroup
⎛ ⎞
∗ ∗ ∗
P2,1 = ⎝ ∗ ∗ ∗ ⎠ ,
0 0 ∗
and consists of series of the form:


E P2,1 (z, s, φ) = Is (γ z, P2,1 )φ m P2,1 (γ z) . (10.5.5)
γ ∈(P2,1 ∩S L(3,Z))\S L(3,Z)

We seek a more explicit version of (10.5.5). By Example 10.4.8, we have

 s 1
Is (z, P2,1 ) = y12 y2 = Det(z)s1 .
It is also easy to see that
⎛ ⎞⎛ ⎞
y2 x1,2 0 y1 0 0
⎝
m P2,1 (z) = 0 1 0 ⎠ ⎝ 0 y1 0⎠.
0 0 1 0 0 1
Then we have the explicit representation:


E P2,1 (z, s, φ) = Det(γ z)s1 · φ m P2,1 (γ z) . (10.5.6)
γ ∈(P2,1 ∩S L(3,Z))\S L(3,Z)
 10.6 Fourier expansion of minimal parabolic Eisenstein series 303

10.6 Fourier expansion of minimal parabolic

Eisenstein series
Let n ≥ 2 and consider E Pmin (z, s), the minimal parabolic Eisenstein series
defined in (10.4.1). We shall first compute the Fourier coefficients of E Pmin (z, s)
as in Theorem 5.3.2.Let m = (m 1 , m 2 , . . . , m n−1 ) ∈ Zn−1 . Then the mth Fourier
coefficient is

Em (z, s) = E Pmin (u · z, s) ψm (u) d ∗ u,
U (Z)\U (R)

where
ψm (u) = e2πi(m 1 u 1,2 +···+m n−1 u n−1,n ) ,
and U denotes the group of upper triangular matrices with 1s on the diagonal.
The computation of this Fourier coefficient is based on the explicit Bruhat
decomposition given in Proposition 10.3.6:
(
G L(n, R) = Gw,
w∈W
where W denotes the Weyl group and
G w = U DwU = U w DU,
with U the group of upper triangular matrices with 1s on the diagonal (as
above), and where D denotes the multiplicative group of diagonal matrices
with non-zero determinant. Consider Pmin (Z) = Pmin ∩ S L(n, Z).
The minimal parabolic Eisenstein series E Pmin (z, s) is constructed as a sum
over the left quotient space Pmin (Z)\S L(n, Z). By the Bruhat decomposition,
we may realize this left quotient space as a union:
( <
Pmin (Z)\S L(n, Z) = Pmin (Z) (S L(n, Z) ∩ G w ).
w∈W

It is natural then, for each w ∈ W , to study the left coset space

<
Pmin (Z) (S L(n, Z) ∩ G w ). (10.6.1)
We would also like to take a further quotient of (10.6.1) on the right. In Lemma
10.6.3 we show that

w = w −1 · t Pmin (Z) · w ∩ Pmin (Z)
acts properly on the right on (10.6.1), so that
< 
Pmin (Z) (S L(n, Z) ∩ G w ) w (10.6.2)
is a well defined double coset space.
304 Langlands Eisenstein series


Lemma 10.6.3 The group w = w−1 · t Pmin (Z) · w ∩ Pmin (Z) acts properly
on the right on the left coset space Pmin (Z)\(S L(n, Z) ∩ G w ).
Proof For each w ∈ W , we introduce two additional spaces.
Uw = (w −1 · U · w) ∩ U
(10.6.4)
Ūw = (w −1 · t U · w) ∩ U.
To get a feel for these spaces, consider the example of G L(3, R) where we have
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
1 ∗ ∗ 1 1
Uw1 = ⎝ 1 ∗ ⎠ ∩ U, Ūw1 = ⎝ ∗ 1 ⎠ ∩ U, w1 = ⎝ 1 ⎠,
1 ∗ ∗ 1 1
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
1 ∗ ∗ 1 1
Uw2 = ⎝ 1 ⎠ ∩ U, Ūw2 = ⎝ ∗ 1 ∗ ⎠ ∩ U, w2 = ⎝ 1⎠,
∗ 1 ∗ 1 1
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
1 ∗ 1 ∗ 1
Uw3 = ⎝ ∗ 1 ∗ ⎠ ∩ U, Ūw3 = ⎝ 1 ⎠ ∩ U, w3 = ⎝ 1 ⎠,
1 ∗ ∗ 1 1
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
1 1 ∗ ∗ 1
Uw4 = ⎝ ∗ 1 ∗ ⎠ ∩ U, Ūw4 = ⎝ 1 ⎠ ∩ U, w4 = ⎝ 1⎠,
∗ 1 ∗ 1 1
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
1 ∗ 1 ∗ 1
Uw5 = ⎝ 1 ⎠ ∩ U, Ūw5 = ⎝ ∗ 1 ∗ ⎠ ∩ U, w5 = ⎝ 1 ⎠,
∗ ∗ 1 1 1
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
1 1 ∗ ∗ 1
Uw6 = ⎝ ∗ 1 ⎠ ∩ U, Ūw6 = ⎝ 1 ∗ ⎠ ∩ U, w6 = ⎝ 1 ⎠.
∗ ∗ 1 1 1
⎛ ⎞ ⎛ ⎞
1 ∗ 1 ∗
For example, Uw5 = ⎝ 1 ⎠ and Ūw5 = ⎝ 1 ∗ ⎠ . We may think of
1 1
Ū w as the space opposite to Uw in U . It is clear that for each w ∈ W , we have
U = Uw · Ū w = Ū w · Uw . (10.6.5)
Returning to the proof of our lemma, it is plain that for every w ∈ W , the group
w acts on Pmin (Z)\(S L(n, Z) ∩ G w ) in the sense that right multiplication by
w maps left cosets to left cosets. We only need to show that this action is
proper, i.e., only the identity element acts trivially. To show this, suppose that

γ ∈ w = w−1 · t Pmin (Z) · w ∩ Pmin (Z) = Ū w (Z)
 10.6 Fourier expansion of minimal parabolic Eisenstein series 305

fixes the left coset Pmin (Z) · b1 cwb2 where c ∈ D, and without loss of generality,
b1 ∈ Uw , b2 ∈ Ū w .
Since
Pmin (Z) · b1 cwb2 · γ = Pmin (Z) · b1 cwb2 ,
it follows that
b2 γ b2−1 ∈ Uw ∩ Ū w = {1}.
This proves that γ must be the identity matrix and the action is proper.

We now return to the computation of the Fourier coefficient Em (z, s) initiated

at the beginning of this section. In order to simplify presentation of formulae,
we introduce the notation
<
Gw = Pmin (Z) (S L(n, Z) ∩ G w ),
with G w = U DwU as in the Bruhat decomposition in Proposition 10.3.6. The
Bruhat decomposition tells us that
(
Pmin (Z)\S L(n, Z) = Gw .
w∈W

We compute, using Lemma 10.6.3, (10.6.5), and the Bruhat decomposition

above,


Em (z, s) = Is (γ uz) ψm (u) d ∗ u
γ ∈Pmin (Z)\S L(n,Z)
U (Z)\U (R)





= Is (b1 cwb2 uz) ψm (u) d ∗ u
w∈Wn c b1 ∈Uw (Q), b2 ∈Ū w (Q) ∈w U (Z)\U (R)
b1 cwb2 ∈ Gw / w





=
w∈Wn c b1 ∈Uw (Q), b2 ∈Ū w (Q) ∈w =Ū w (Z)
b1 cwb2 ∈ Gw / w
 
× Is (cwb2 uz) ψm (u) d ∗ u
Uw (Z)\Uw (R) Ū w (Z)\Ū w (R)




= ψm (b2 ) · Is (c)
w∈Wn c b1 ∈Uw (Q), b2 ∈Ū w (Q)
b1 cwb2 ∈ Gw / w
 
× Is (wuz) ψm (u) d ∗ u.
Uw (Z)\Uw (R) Ū w (R)

The double integral above can be explicitly computed as follows. By

(10.6.5), every u ∈ U (R) can be written as u = u 1 · u 2 with u 1 ∈ Uw (R) and
306 Langlands Eisenstein series

u 2 ∈ Ū w (R). Clearly ψm (u 1 u 2 ) = ψm (u 1 )ψm (u 2 ). Since wu 1 = u 1 w for some

u 1 ∈ U (R), we may, therefore, write
 
Is (wuz) ψm (u) d ∗ u
Uw (Z)\Uw (R) Ū w (R)
 

= Is u 1 wu 2 z ψm (u 1 u 2 ) d ∗ u 1 d ∗ u 2 .
Uw (Z)\Uw (R) Ū w (R)
 
∗
= ψm (u 1 ) d u 1 · Is (wu 2 z) ψm (u 2 ) d ∗ u 2 . (10.6.6)
Uw (Z)\Uw (R) Ū w (R)

Note that the last integral on the right-hand side of (10.6.6) will be a degenerate
Whittaker function, as in Definition 5.10.1, if w is not the long element of the
Weyl group.
It is instructive to illustrate the double integral (10.6.6) with an example. We
shall consider G L(4, R). In this case
⎛ ⎞
1 ∗ ∗ ∗
⎜ 1 ∗ ∗⎟
U (R) = ⎜ ⎝
⎟.
1 ∗⎠
1
For our example, we will let
⎛ ⎞
1
⎜ 1 ⎟
w=⎜
⎝
⎟.
1⎠
1
In this case,
⎛ ⎞ ⎛ ⎞
1 1 ∗ ∗ ∗
⎜ 1 ∗⎟ ⎜ 1 ∗ ⎟
Uw (R) = ⎜
⎝
⎟, Ū w (R) = ⎜ ⎟,
1 ∗⎠ ⎝ 1 ⎠
1 1
and the double integral (10.6.6) takes the form:
 1  ∞ ∞ ∞ ∞
e−2πim 3 u 3,4 du 3,4 · Is (wu 2 z)ψm (u 2 ) d ∗ u 2 ,
0 −∞ −∞ −∞ −∞

where
⎛ ⎞
1 u 1,2 u 1,3 u 1,4
⎜ 1 u 2,3 ⎟
u2 = ⎜
⎝
⎟,
⎠ d ∗ u 2 = du 1,2 du 1,3 du 1,4 du 2,3 .
1
1
 10.7 Meromorphic equation of Eisenstein series 307

In the special case that w = w0 is the long element

⎛ ⎞
1
⎜ 1 ⎟
⎜ ⎟
w0 = ⎜ ⎟,
⎝ ... ⎠
1
we have Ū w0 (R) = U (R) and Uw0 (R) is trivial. In this case, the double integral
(10.6.6) is precisely the Jacquet Whittaker function given in Section 5.5. If the
(n − 1)-tuple m = (m 1 , m 2 , . . . , m n−1 ) satisfies m i = 0 for all 1 ≤ i ≤ n − 1,
i.e., the character ψm is non-degenerate, then the double integral (10.6.6) will
vanish unless w = w0 is the long element. This is because Uw (Z)\Uw (R) will
be non-trivial and just a direct product of intervals [0, 1], so that

ψm (u 1 ) d ∗ u 1 = 0.
Uw (Z)\Uw (R)
*
In general, the integral ψm (u 1 ) d ∗ u 1 will be either 1 or 0, while
Uw (Z)\Uw (R)
the other integral

Is (wu 2 z) ψm (u 2 ) d ∗ u 2
Ū w (R)

will be a Whittaker function.

10.7 Meromorphic continuation and functional equation

of maximal parabolic Eisenstein series
Let us fix (for this section) the notation
⎛ ⎞
∗ ··· ∗ ∗
⎜ .. .. .. ⎟
⎜ ⎟
P = Pn−1,1 = ⎜ . · · · . . ⎟ ⊂ ,
⎝∗ ··· ∗ ∗⎠
0 ··· 0 1
to be the maximal parabolic subgroup of  whose elements have bottom row
equal to (0, . . . , 0, 1). For s ∈ C, z ∈ hn , consider the maximal parabolic Eisen-
stein series
 ns 
E ∗P (z, s) = π −ns/2  ζ (ns)E P (z, s)
2
where


E P (z, s) = Det(γ z)s , ((s) > 2/n), (10.7.1)
P\
308 Langlands Eisenstein series

and where Det is the determinant function on hn . While it is not yet clear that
(10.7.1) converges absolutely for (s) > 2/n, this will follow directly from
(10.7.4). Note that in (10.7.1), we must put γ z in canonical Iwasawa form (as
in Proposition 1.2.6) before actually taking the determinant. The meromorphic
continuation and functional equation of E P (z, s) can be obtained from the
Poisson summation formula

1

f (m · z) = fˆ(m · (t z)−1 ), (10.7.2)
m∈Z n |Det(z)| m∈Zn

which holds for smooth functions f : Rn → C with sufficient decay at ±∞.

Here
∞ ∞
f˜((x1 , . . . , xn )) = ··· f ((t1 , . . . , tn ))e−2πi(t1 x1 +···+tn xn ) dt1 · · · dtn .
−∞ −∞

In order to be able to apply (10.7.2), it is necessary to rewrite the maximal

parabolic Eisenstein series in the form of an Epstein zeta function. This is done
as follows.
First note that if
   
∗  ∗
γ = , γ = ∈ ,
a1 · · · an a1 · · · an
then there exists p ∈ P with γ  = pγ if and only if (a1 , . . . , an ) = (a1 , . . . , an ).
Consequently, each coset of P\ is uniquely determined by n relatively prime
integers (a1 , . . . , an ).
Furthermore,
⎛ ⎞⎛y ··· y ⎞
1 n−1
  1
⎟⎜ ⎟
xi, j
∗ ⎜ ..
γz = ·⎜ .. ⎟⎜
⎜ . ⎟
⎟
a1 · · · an ⎝ . ⎠⎝ ⎠
y1
1
1
 
∗
= ,
b1 · · · bn
where

b1 = a1 y1 · · · yn−1
b2 = (a1 x1,2 + a2 )y1 · · · yn−2
..
.
bn = (a1 x1,n + a2 x2,n + · · · + an−1 xn−1,n + an ). (10.7.3)
 10.7 Meromorphic equation of Eisenstein series 309

On the other hand, by the Iwasawa decomposition (Proposition 1.2.6),

γ z = τ · k · r In ,

where k ∈ O(n, R), 0 < r ∈ R, In denotes the n × n identity matrix, and τ is

the canonical form for the Iwasawa decomposition. By comparing norms of the
bottom rows, which amounts to the identity (γ z) · t (γ z) = (τ kr In ) · t (τ kr In ) =
τ t τ · r 2 In , one obtains

b12 + · · · + bn2 = r 2 .

Consequently

Det(γ z) = Det(τ ) = |Det(γ )| Det(z)r −n

3 4−n/2
= Det(z) b12 + · · · + bn2 .

It immediately follows that

3 4−ns/2
ζ (ns)E P (z, s) = Det(z)s b12 + · · · + bn2 , (10.7.4)
(a1 ,...,an )∈Zn
(a1 ,...,an )=(0,...,0)

with bi given by (10.7.3) for i = 1, . . . , n. The right-hand side of (10.7.4) is

termed an Epstein zeta function. We multiply by ζ (ns) on the left to convert
the sum on the right to a sum over (a1 , . . . , an ) ∈ Zn , eliminating the relatively
prime condition.
Next, we utilize the Poisson summation formula (10.7.2) to show
that E P (z, s) has a meromorphic continuation and satisfies a functional
equation.

Proposition 10.7.5 The maximal parabolic Eisenstein series E P (z, s) defined

in (10.7.1) has meromorphic continuation to all s ∈ C and satisfies the func-
tional equation
 ns 
E ∗P (z, s) := π −ns/2  ζ (ns)E P (z, s) = E ∗P (t z −1 , 1 − s).
2
Further, E ∗P (z, s) is holomorphic except for simple poles at s = 0, 1.

Proof Fix u > 0. For x = (x1 , . . . , xn ) ∈ Rn , define

f u (x) := e−π (x1 +···+xn )·u .

2 2

Then we have the Fourier transform fˆu (x) = (1/u n/2 ) f u (x/u) . We shall make
use of the Poisson summation formula (10.7.2) with this choice of function f u .
It follows from (10.7.3), (10.7.4), and the integral representation of the Gamma
310 Langlands Eisenstein series

function that
∞ =

>
E ∗P (x, s) = Det(z)
s
f u ((a1 , . . . , an ) · z) − f ((0, . . . , 0))
(a1 ,...,an )∈Zn
0
du
. × u ns/2
u
The proposition follows by breaking the integral into two parts: [0, 1] and
[1, ∞], and then applying the Poisson summation formula (10.7.2) just as we
did in the proof of the functional equation of the Riemann zeta function given
on page 1.

10.8 The L-function associated to a minimal parabolic

Eisenstein series
To obtain the Fourier coefficients of the minimal parabolic Eisenstein series in
a more precise form, we follow the method in Section 3.14. This method can
also be used to obtain the meromorphic continuation and functional equation
of the Eisenstein series. We compute the action of the Hecke operator (9.3.5)
on the I-function. The I-function is an eigenfunction of all the Hecke operators,
and the eigenvalue of the Hecke operator Tm will give us the (m, 1, . . . , 1)th
Fourier coefficient of the minimal parabolic Eisenstein series as in Theorem
9.3.11. Although Theorem 9.3.11 is stated for Maass forms, it can be easily
generalized to Eisenstein series. Note that this method works up to a normalizing
factor. We shall prove the following theorem.
Theorem 10.8.1 Let s = (s1 , . . . , sn−1 ) ∈ Cn−1 with n ≥ 2. Define s − n1 to
be (s1 − n1 , . . . , sn−1 − n1 ). For z ∈ hn , let E(z, s) be the minimal parabolic
Eisenstein series (10.4.1), for  = S L(n, Z), with Fourier expansion



∞

∞

E(z, s) = C(z, s) + ··· A((m 1 , m 2 , . . . , m n−1 ), s)
γ ∈Un−1 (Z)\ m 1 =1 m 2 =1 m n−1 =0
⎛⎛ ⎞ ⎞
|m 1 m 2 · · · m n−1 |
⎜⎜ .. ⎟   ⎟
⎜⎜ . ⎟ γ ⎟
×WJacquet ⎜⎜ ⎟· z, s, ψ1,1,..., mn−1 ⎟
⎝⎝ |m 1 | ⎠ 1 |m n−1 | ⎠

1
as in Section 10.6. Here C(z, s) denotes the degenerate terms in the Fourier
expansion associated to (m 1 , m 2 , . . . , m n−1 ) with m i = 0 for some 1 ≤ i
≤ n − 1. We shall assume E(z, s) is normalized (multiplied by a suitable
 10.8 The L-function associated to parabolic Eisenstein series 311

function of s) so that A((1, 1, . . . , 1), s) = 1. Then for m 1 = 1, 2, 3, . . . ,

⎛⎛ ⎞⎞
c1

⎜⎜ .. ⎟⎟
⎜⎜ . ⎟⎟
A((m 1 , 1, 1, . . . , 1), s) = Is−(1/n) ⎜⎜ ⎟⎟ .
1≤c ,c ,...,c ∈Z
⎝ ⎝ cn−1 ⎠⎠
1 2 n
)
n
c = m 1 cn
=1

Remarks It is easy to see that A((m 1 , 1, 1, . . . , 1), s) = A((m 1 , 1, 1, . . . , 1),

s  ) whenever

Is−(1/n) (y) = Is  −(1/n) (wy) (10.8.2)

for some fixed w in the Weyl group of G L(n, R) and all diagonal matrices
y as in (5.9.1). This is due to the fact that the action of the Weyl group on
the group of diagonal matrices just permutes the diagonal elements so that the
sum over 1 ≤ c1 , c2 , . . . , cn ∈ Z does not change. We shall show that E(z, s)
is actually an eigenfunction of the Hecke operators. The Hecke relations in
Section 9.3 imply that if A(( p k , 1, 1, . . . , 1), s) satisfies the functional equation
s → s  for s, s  given by (10.8.2) and all primes p and all k = 1, 2, . . . , then
A((m 1 , m 2 , . . . , m n−1 ), s) must also satisfy the functional equations (10.8.2)
for all m 1 , m 2 , . . . , m n−1 ∈ Zn−1 . It immediately follows from Theorem 5.9.8
+
j−1
that for s j,k = (nsn−k+i − 1)/2, we have
i=0


n−1   
∗ − 12 −s j,k 1
E (z, s) = π  + s j,k ζ (1 + 2s j,k )(E(z, s) − C(z, s))
j=1 j≤k≤n−1
2
= E ∗ (z, s  ) (10.8.3)

for all s, s  satisfying (10.8.2) It may also be shown that C(z, s) satisfies these
same functional equations.

Proof of Theorem 10.8.1 Recall the definition (see also (5.1.1)),

 n−1
n−1  b sj
Is (z) = yi i, j
i=1 j=1

with

ij if i + j ≤ n,
bi, j =
(n − i)(n − j) if i + j ≥ n.
312 Langlands Eisenstein series

For m 1 ≥ 1, we compute, using (9.3.5),

⎛⎛ ⎞ ⎞
c1 c1,2 · · · c1,n

⎜⎜ c2 · · · c2,n ⎟ ⎟
1 ⎜⎜ ⎟ ⎟
Tm 1 Is (z) = Is ⎜⎜ .. .. ⎟ · z ⎟
m1
(n−1)/2
)
n ⎝⎝ . . ⎠ ⎠
c = m 1
=1 cn
0 ≤ ci, < c (1 ≤ i < ≤ n)
⎛⎛ ⎞⎞
c1

⎜⎜ .. ⎟⎟
1 ⎜⎜ . ⎟⎟
= Is ⎜⎜ ⎟⎟ · Is (z)
m1
(n−1)/2
)
n ⎝⎝ cn−1 ⎠⎠
c = m 1
=1 cn
0 ≤ ci, < c (1 ≤ i < ≤ n)
⎛⎛ ⎞⎞
c1



n ⎜⎜ .. ⎟⎟
(2i−1−n)/2 ⎜⎜ . ⎟⎟
= ci I s ⎜⎜ ⎟⎟ · Is (z)
)
n i=1
⎝⎝ cn−1 ⎠⎠
c =m 1
=1 cn
⎛⎛ ⎞⎞
c1

⎜⎜ .. ⎟⎟
⎜⎜ . ⎟⎟
= Is−(1/n) ⎜⎜ ⎟⎟ · Is (z)
)
n ⎝⎝ cn−1 ⎠⎠
c =m 1
=1 cn
= A((m 1 , 1, . . . , 1), s) · Is (z).

This computation shows that Is (z) is an eigenfunction of the Hecke operators

Tm 1 with eigenvalue A((m 1 , 1, . . . , 1), s). We shall next show that this implies
that E(z, s) is also an eigenfunction of Tm 1 with the same eigenvalue.
Let Sm 1 denote the set of matrices
⎛ ⎞
c1 c1,2 · · · c1,n
⎜ c2 · · · c2,n ⎟
⎜ ⎟
⎜ . .. ⎟
⎝ . . . ⎠
cn
)
n
where c = m 1 and 0 ≤ ci, < c (1 ≤ i < ≤ n).
=1

Lemma 10.8.4 For m 1 = 1, 2, 3, . . . , there exists a one-to-one correspon-

dence between Sm 1 × S L(n, Z) and S L(n, Z) × Sm 1 .

Proof It follows from the Hermite normal form, Theorem 3.11.1, that for any
α  ∈ Sm 1 , γ  ∈ S L(n, Z), there exists a unique α ∈ Sm 1 and γ ∈ S L(n, Z) such
that γ  α  = αγ . The result follows easily from this.

 10.8 The L-function associated to parabolic Eisenstein series 313

It is a consequence of Lemma 10.8.4 that for every α ∈ Sm 1 , γ ∈ Pmin \

there exists a unique α  ∈ Sm 1 , γ  ∈ Pmin \ such that αγ = γ  α  . It now
follows from the definition of the action of the Hecke operator (9.3.5) and
the definition of the minimal parabolic Eisenstein series (10.4.1) that for
m 1 = 1, 2, 3, . . . ,
−(n−1)/2



Tm 1 E Pmin (z, s) = m 1 Is (γ αz)
α∈Sm 1 γ ∈Pmin \
−(n−1)/2



= m1 Is (αγ z)
γ ∈Pmin \ α∈Sm 1


= A((m 1 , 1, 1, . . . , 1), s) · Is (γ z)
γ ∈Pmin \

= A((m 1 , 1, 1, . . . , 1), s) · E Pmin (z, s).

This proves that E Pmin (z, s) is an eigenfunction of the Hecke operators. The
proof of Theorem 10.8.1 is an immediate consequence of Theorem 9.3.11.
Note that although we only proved Theorem 9.3.11 for Maass forms, it can be
easily generalized to Eisenstein series.

Let n ≥ 2. For v = (v1 , . . . , vn−1 ) ∈ Cn−1 , z ∈ hn , let E v (z) = E Pmin (z, v)

denote the minimal parabolic Eisenstein series (10.4.1). Finally, we are now in
a position to compute the L-function associated to E v , denoted L Ev (s), just as
we did in Section 3.14 for the Eisenstein series on S L(2, Z). It follows from
Theorem 10.8.1 that
⎛⎛ c ⎞⎞
1
cn


∞
⎜⎜ ⎟⎟
⎜⎜ .. ⎟⎟
L Ev (s) = m −s
Iv− n1 ⎜⎜ . ⎟⎟
⎜⎜ cn−1 ⎟⎟
m=1 1≤c1 ,c2 ,...,cn ∈Z ⎝⎝ cn ⎠⎠
)
n
c = m 1
=1
⎛⎛ c2 c ···c ⎞⎞
1 2 n−1

∞
⎜⎜ m
⎟⎟
⎜⎜ .. ⎟⎟
= m −s
Iv− n1 ⎜⎜ . ⎟⎟
⎜⎜ c1 ···cn−2 cn−1
2 ⎟⎟
m=1 1≤c1 ,...,cn−1 ∈Z ⎝⎝ ⎠⎠
) 
n−1 m
c m 1
=1
⎛⎛ c1 ⎞⎞
m


∞

∞

∞ ⎜⎜ .. ⎟⎟
⎜⎜ . ⎟⎟
= ··· (mc1 · · · cn−1 )−s Iv− n1 ⎜⎜ ⎟⎟ .
c1 =1 cn−1 =1 m=1
⎝⎝ cn−1 ⎠⎠
m
1

(10.8.5)
The above computation immediately implies the following theorem.
314 Langlands Eisenstein series

Theorem 10.8.6 (L-function associated to minimal parabolic Eisenstein

series) For n ≥ 2, let E v (z) = E Pmin (z, v) denote the minimal parabolic Eisen-
stein series (10.4.1). Then there exist functions λi : Cn−1 → C, satisfying
(λi (v)) = 0 if (vi ) = n1 (i = 1, . . . , n − 1), such that the L-function associ-
ated to E v is just a product of shifted Riemann zeta functions of the form

n
L Ev (s) = ζ (s − λi (v)).
i=1

Furthermore, L Ev (s) satisfies the functional equation

G Ev (s)L Ev (s) = G Ẽ v (1 − s)L Ẽ v (1 − s),

where

n     
s−λi (v) s − λi (v) n
s − λi (v)
π− = π− 2
ns
G Ev (s) = 2   ,
i=1
2 i=1
2

and Ẽ v is the dual Eisenstein series as in Section 9.2.

We tabulate L Ev (s) for the cases n = 2, 3, 4, respectively:

   
1 1
ζ s+v− ζ s−v+ , (if n = 2)
2 2
ζ (s + v1 + 2v2 − 1) ζ (s − 2v1 − v2 + 1) ζ (s + v1 − v2 ), (if n = 3)
   
3 3
ζ s + v1 + 2v2 + 3v3 − ζ s − 3v1 − 2v2 − v3 +
2 2
   
1 1
× ζ s + v1 − 2v2 − v3 + ζ s + v1 + 2v2 − v3 − , (if n = 4).
2 2
For example, in the last case (n = 4) we have
3
λ1 (v) = − v1 − 2v2 − 3v3 ,
2
3
λ2 (v) = 3v1 + 2v2 + v3 − ,
2
1
λ3 (v) = −v1 + 2v2 + v3 − ,
2
1
λ4 (v) = −v1 − 2v2 + v3 + .
2
Remark 10.8.7 The functions λi (i = 1, . . . , n) are uniquely determined
by (10.8.5). It is easy to see that λ1 (v) = ((n − 1)/2) − v1 − 2v2 − · · ·
− (n − 1)vn−1 . Theorem 10.8.6 is very important. It provides a template for
all future functional equations. The reader can immediately check that when
 10.9 Fourier coefficients of Eisenstein series twisted by Maass forms 315

n = 3 the functional equation of L Ev (s) given above is identical to the func-

tional equation of a G L(3)-Maass form of type v obtained in Theorem 6.5.15.
This is not an accident. In fact, the functional equation of L Ev (s) will always
be identical to the functional equation of a symmetric Maass form of type v
because the Whittaker functions in the Fourier expansion of the Maass form
will match up exactly with the Whittaker functions of the Eisenstein series.
The formal proof of the functional equation only uses the analytic properties of
the Whittaker functions. The proof is entirely independent of the values of the
arithmetic Fourier coefficients.
How to use Theorem 10.8.6 as a template for functional equations. Let
us consider the example of twisting a Maass form by a primitive Dirichlet
character χ (mod q). Its L-function will have the same functional equation as
the L-function associated to the Eisenstein series E v twisted by χ which will
simply be

n−1
L Ev (s, χ) = L(s − λi (v), χ).
i=1

Now, the functional equation of the Dirichlet L-function L(s, χ ) is (see

(Davenport, 1974))
 q (s+aχ )/2  s + a  τ (χ )
χ
(s, χ) :=  L(s, χ ) = aχ √ (1 − s, χ̄ ),
π 2 i q
where τ (χ ) denotes the Gauss sum in (7.1.1). It immediately follows that
L Ev (s, χ ) satisfies the functional equation
 q n(s+aχ )/2  n  
s + aχ − λi (ν)
 Ev (s, χ) :=  L Ev (s, χ )
π i=1
2
 
τ (χ ) n
= aχ √  Ẽ v (1 − s, χ̄ ).
i q
This illustrates the method of obtaining functional equations of Maass forms
of various types by carefully examining the template arising from the case of
minimal parabolic Eisenstein series.

10.9 Fourier coefficients of Eisenstein series twisted

by Maass forms
The method presented in Section 10.8 for obtaining the Fourier coefficients of
minimal parabolic Eisenstein series easily extends to more general Eisenstein
316 Langlands Eisenstein series

series which are associated to an arbitrary standard parabolic subgroup P. This

is due to the fact that such an Eisenstein series as given in Definition 10.4.7 can
actually be written in the form


E P (z, s) = Is  (γ z)
γ ∈(P∩)\

for suitable s  as explained in the discussion just prior to Lemma 10.4.6. It

follows as in Theorem 10.8.1 that the normalized (m 1 , 1, 1, . . . , 1)th Fourier
coefficient of E P (z, s) is just

⎛⎛ ⎞⎞
c1

⎜⎜ .. ⎟⎟
⎜⎜ . ⎟⎟
Is  − n1 ⎜⎜ ⎟⎟ .
1≤c1 ,c2 ,...,cn ∈Z
⎝⎝ cn−1 ⎠⎠
)
n
c =m 1 cn
=1

We now consider the more complex situation of Eisenstein series twisted

by Maass forms of lower rank as in Section 10.5. Fix a parabolic subgroup
P = Pn 1 ,...,nr associated to the partition n = n 1 + · · · + n r , with r ≥ 2. Let
φ = (φ1 , . . . , φr ) be a set of r Maass forms where each φi is a Maass form of
type λi = (λi1 , . . . , λini −1 ) ∈ Cni −1 (i = 1, . . . , r ) for S L(n i , Z), as in Defini-
tion 5.1.3.
We shall show, in the next two propositions, that the Langlands Eisen-
stein series E P (z, s, φ), given in Definition 10.5.3, is an eigenfunction of both
the Hecke operators for S L(n, Z), as in (9.3.5), and the invariant differen-
tial operators on G L(n, R), as in Proposition 2.3.3 provided φ is a Hecke
eigenform, i.e., each φi is an eigenfunction of the S L(n i , Z) Hecke operators.
These properties allow one to obtain the meromorphic continuation and func-
tional equation for the non-degenerate part of the Eisenstein series E P (z, s, φ),
just as we did previously in Section 10.8 for the minimal parabolic Eisenstein
series.
Recall that a smooth function F : hn → C is of type ν ∈ Cn−1 if it is an
eigenfunction of all the invariant differential operators (see Section 2.3) with
the same eigenvalues as the function Iν . In this regard, see also Definition 5.1.3.

Proposition 10.9.1 For r ≥ 2, let P = Pn 1 ,...,nr be a standard parabolic sub-

group associated to the partition n = n 1 + · · · + n r . Let φ = (φ1 , . . . , φr ) be
a set of r Maass forms where each φi is a Maass form for S L(n i , Z) of
type λi . Then for s = (s1 , . . . , sr ) ∈ Cr , the Eisenstein series E P (z, s, φ) is an
 10.9 Fourier coefficients of Eisenstein series twisted by Maass forms 317

eigenfunction of type λ + s  where s  ∈ Cn−1 is such that


r
 s
Is  (z) = Det mni (z) i
i=1

)
r
for all z ∈ hn , and λ ∈ Cn−1 is such that φi (mni (z)) is of type λ.
i=1

Proof Since the invariant differential operators commute with the action of
G L(n, R), it is enough to check that

r
  s
φi mni (z) · Det mni (z) i , (10.9.2)
i=1

the generating function of the Eisenstein series, is an eigenfunction. This will

be the case because each Maass form φi is a linear combination of Whittaker
functions Wλi (z i ) of type λi with z i ∈ hni , where


Wλi (z i ) = Iλi wni uz i ψi (u) d ∗ u
Uni (R)

is a Jacquet Whittaker function, for some character ψi of Uni (R), as in (5.5.1).

Here wni is the long element of the Weyl group for G L(n i , R).
⎛ ⎞
wn 1
⎜ wn 2 ⎟
⎜ ⎟
Now, for w = ⎜ .. ⎟,
⎝ . ⎠
wnr


r
  s
Wλi mni (z) Det mni (z) i
i=1
 
r
  s
= Iλi mni (wuz) ψi (u) · Det mni (z) i d ∗ u
i=1
Uni (R)

 
r
  s
= Iλi mni (wuz) ψi (u) · Det mni (wuz) i d ∗ u,
i=1
Uni (R)

since Det(wu) = 1. It immediately follows that the above, and hence, (10.9.2) is
) a
n−1
an eigenfunction because it is obtained from y for suitable a1 , a2 , . . . , an−1
=1
by integrating a set of left translates. To show that E P (z, s, φ) is an eigenfunction
of type λ + s  one uses the fact that Iλ (z) · Is  (z) = Iλ+s  (z).

318 Langlands Eisenstein series

Proposition 10.9.3 With the notation of Proposition 10.9.1, let E P (z, s, φ) be

a Langlands Eisenstein series for S L(n, Z). Let s = (s1 , . . . , sr ) ∈ Cr satisfy
n 1 s1 + · · · + n r sr = 0. Assume that φ is a Hecke eigenform, and let Tm denote
the Hecke operator given in (9.3.5). Then for m = 1, 2, 3, . . . ,
Tm E P (z, s, φ) = λm (s) · E P (z, s, φ)
where

s +η1 s +η2
λm (s) = A1 (c1 )A2 (c2 ) · · · Ar (cr ) · C11 C22 · · · Crsr +ηr ,
1≤C1 ,C2 ,...,Cr ∈ Z
C1 C2 ···Cr =m
where η1 = 0, and ηi = n 1 + n 2 + · · · + n i−1 for i ≥ 1. In the above Ai (Ci )
denotes the eigenvalue of the S L(n i , Z) Hecke operator TCi acting on φi which
may also be viewed as the (Ci , 1, 1, . . . ,1)th Fourier coefficient of φi .
" #$ %
n i −1 terms
Proof For m = 1, 2, . . . , we compute, as in the proof of Theorem 10.8.1:
 
r
 s
Tm φi mni (z) · Det mni (z) i
i=1


r

= m −(n−1)/2 φi mni (cz) · Det(mni (cy))si (10.9.4)
)
n i=1
c =m
=1
0≤ci, <c (1≤i< ≤n)

where
⎛ ⎞
c1 c1,2 · · · c1,n
⎜ c2 · · · c2,n ⎟
⎜ ⎟
c=⎜ .. .. ⎟ .
⎝ . . ⎠
cn
Note that
mni (cz) = mni (c) · mni (z).
Define ηi = n 1 + n 2 + · · · + n i−1 with the convention that η1 = 0. Then for
each i = 1, 2, . . . , r , we have
⎛ ⎞
cηi +1 cηi +1,ηi +2 · · · cηi +1,ηi+1
⎜ cηi +2 · · · cηi +2,ηi+1 ⎟
⎜ ⎟
mni (cy) = ⎜ .. .. ⎟
⎝ . . ⎠
cηi+1
⎛ ⎞
y1 y2 · · · yn−ηi −1
⎜ y1 y2 · · · yn−ηi −2 ⎟
⎜ ⎟
×⎜ .. ⎟.
⎝ . ⎠
y1 y2 · · · yn−ηi+1
 10.10 The constant term 319

To complete the proof, we rewrite the sum on the right-hand side of

(10.9.4) so that for each 1 ≤ i ≤ r , the sum turns into the local Hecke operator
for S L(n i , R) acting on φi . This can be done as follows. Introduce r inte-
)
n
gers 1 ≤ C1 , C2 , . . . Cr satisfying C1 C2 · · · Cr = c = m. Then (10.9.4) is
=1
equal to


r
  s +η  s
m −(n−1)/2 TCi φi mni (z) · Ci i i · Det mni (y) i
C1 C2 ···Cr =m i=1


r
 s +η  s
= m −(n−1)/2 Ai (Ci ) φi mni (z) · Ci i i · Det mni (y) i ,
C1 C2 ···Cr =m i=1

where TCi denotes the Hecke operator on S L(n i , Z) and

ηi = n 1 + n 2 + · · · + n i−1 if i > 1 while η1 = 0.

Propositions 10.9.1 and 10.9.3 allow one to show that the non-degenerate
terms in the Whittaker expansion of general Langlands Eisenstein series for
S L(n, Z) with n ≥ 2 have a meromorphic continuation and satisfy the same
functional equation as the Whittaker functions that occur in the Fourier–
Whittaker expansion.

10.10 The constant term

For n, r ≥ 2, let P = Pn 1 ,...,nr be the standard parabolic subgroup of G L(n, R)
associated to the partition n = n 1 + · · · + n r . Let φ = (φ1 , . . . , φr ) be a set of
r Maass forms where for i = 1, 2, . . . , r , each φi is automorphic for S L(n i , Z).
+r
Let s = (s1 , . . . , sr ) ∈ Cr with n i si = 0. We are interested in determin-
i=1
ing the constant term (in the Fourier expansion) of the Langlands Eisenstein
series


r
  s
E P (s, z, φ) = φi mni (γ z) · Det mni (γ z) i (10.10.1)
γ ∈(P∩)\ i=1

as in Definition 10.5.3. In (Langlands, 1966), the concept of the Fourier expan-

sion along an arbitrary parabolic is introduced for the first time. This is a
more general notion than the usual constant term and is required for the
Langlands spectral decomposition. We shall now define this more general con-
stant term.
320 Langlands Eisenstein series

Definition 10.10.2 (Constant term along an arbitrary parabolic) Con-

sider, E P (s, z, φ), a Langlands Eisenstein series as in (10.10.1).Let P  be any
parabolic subgroup of G L(n, R) with Langlands decomposition P  = N  M 
as in (10.4.4). Then the constant term of E P along the parabolic P  is given
by

E P (ηz, s, φ) d ∗ η,
N  (Z)\N  (R)

where N  (Z) = S L(n, Z) ∩ N  (R).

Theorem 10.10.3 (Langlands) The constant term of E P along a parabolic

P  is zero if P has lower rank than P  or if P and P  have the same rank but
are not associate (as in Definition 10.1.4).

Proof We compute, for P(Z) = P ∩  with  = S L(n, Z),


E P (uz, s, φ) d ∗ u
N  (Z)\N  (R)

 
r
  s
= φi mni (γ uz) · Det mni (γ uz) i d ∗ u
γ ∈P(Z)\ i=1
N  (Z)\N  (R)


 
r

= φi mni (γ n  uz)
γ ∈P(Z)\/N  (Z) n  ∈N  (Z) i=1
N  (Z)\N  (R)
 s
× Det mni (γ n  uz) i d ∗ u

  r

= φi mni (γ uz)
γ ∈P(Z)\/N  (Z) i=1
((γ −1 P(Z)γ ) ∩ N  (Z))\N  (R)
 s
× Det mni (γ uz) i d ∗ u (10.10.4)

because

P(Z)γ N  (Z)u 1 = P(Z)γ N  (Z)u 2

if and only if u 1 u −1 
2 ∈ N (Z) ∩ (γ
−1
P(Z)γ ).
Now, by the Bruhat decomposition (see Propositions 10.3.2, 10.3.6), each

γ ∈ P(Z)\/N  (Z)

can be expressed in the form γ = wγ  where w is in the Weyl group and

γ  ∈ Bn (Z)\N  (Z). Making this replacement, a typical term in the sum on the
 10.11 The constant term of S L(3, Z) Eisenstein series 321

right-hand side of (10.10.4) will be of the form

 
r
  s
φi mni (wγ  uz) · Det mni (wγ  uz) i d ∗ u,
< i=1
((γ  −1 w−1 P(Z)wγ  ) ∩ N  (Z)) N  (R)

which after changing variables u → γ  −1 uγ  becomes

 
r
  s
φi mni (wuγ  z) · Det mni (wuγ  z) i d ∗ u.
< i=1
((w−1 P(Z)w) ∩ N  (Z)) N  (R)

Following Langlands, we define

0
N = (M/A) ∩ (wN  w −1 ).

Then the above integral becomes

  
r

φi mni (u 1 wuz)
< 0 N (Z)\0 N (R)
i=1
((w−1 P(Z)w) ∩ N  (Z)) (w−1 N  (R)w)\N  (R)
 s
× Det mni (u 1 wuz) i d ∗ u 1 d ∗ u.

Since φ is a cusp form, the inner integral vanishes unless 0 N is the identity and
this will only happen if P and P  are associate with w ∈ (P, P  ).

10.11 The constant term of S L(3, Z) Eisenstein series

twisted by S L(2, Z)-Maass forms
There are two maximal parabolic subgroups for G L(3, R). They are P1,2 and
P2,1 , as in Example 10.1.2. The maximal parabolic Eisenstein series, associated
to P2,1 , twisted by an S L(2, Z) Maass form φ, as given in (10.5.5), takes the
explicit form:

 s 

E P2,1 (z, s, φ) = y12 y2 φ(z 2 )  , (10.11.1)
γ
γ ∈P21 \

⎞⎛⎛ ⎞
x3 1
y1 y2 x2
where  = S L(3, Z), z = ⎝ x1 ⎠⎝ 1
y1 ⎠, with z 2 = x2 + i y2 ,
1 1
and the slash operator |γ denotes the action of γ on z.
322 Langlands Eisenstein series

Proposition 10.11.2 The constant terms of the Eisenstein series E P2,1 (z, s, φ)
given in (10.11.1) along the various parabolic subgroups take the form:
⎛⎛ ⎞ ⎞
1 1 1 1 u2 u3
E P2,1 ⎝⎝ 1 u 1 ⎠ z, s, φ ⎠ du 1 du 2 du 3 = 0,
0 0 0 1
⎛⎛ ⎞ ⎞
1 1 1 u3  s
E P2,1 ⎝⎝ 1 u 1 ⎠ z, s, φ ⎠ du 1 du 3 = 2 y12 y2 φ(z 2 ),
0 0 1
⎛⎛ ⎞ ⎞
 
1 1 1 u2 u3
E P2,1 ⎝⎝ 1 ⎠ z, s, φ⎠ du 2 du 3 = 2y11−s y22−2s φ (λ − 1) φ(z 1 ),
φ (λ)
0 0 1
where
   
s +  + ir s +  − ir
φ (s) = π −s   L φ (s),
2 2
(with  = 0 or 1 according as φ is even or odd), is the completed L-function of
the Maass form φ (of type 12 + ir ) as in Proposition 3.13.5, and where

3s − 12 if φ is even
λ =
3s + 12 if φ is odd.

Proof Omitted.

10.12 An application of the theory of Eisenstein series to

the non-vanishing of L-functions on the line (s) = 1
The prime number theorem (Davenport, 1974) states that π (x), the number of
primes less than x, is asymptotic to x/log x as x → ∞. A key ingredient to the
analytic proof of the prime number theorem is the fact that the Riemann zeta
function ζ (s) does not vanish on the line (s) = 1. In (Jacquet and Shalika,
1976/77), a new proof of the non-vanishing of ζ (s) on the line (s) = 1 was
obtained. This proof had two very interesting features.
r It made use of the theory of Eisenstein series.
r The proof could be vastly generalized.

Recently, Sarnak (2004), Gelbart, Lapid and Sarnak (2004) obtained explicit
zero–free regions for general automorphic L-functions by use of the Jacquet–
Shalika method. We shall now present a short exposition of this method by
 10.12 An application of the theory of Eisenstein series 323

considering the classical case of G L(2) and then the not so classical case of
G L(3).

Non-vanishing of ζ (s) on (s) = 1 Recall the Fourier expansion, given

in Theorem 3.1.8, of the S L(2, Z) Eisenstein series
√
2π s y
1
E(z, s) = y + φ(s)y
s 1−s
+ σ1−2s (n)|n|s− 2 K s− 12 (2π |n|y)e2πinx
(s)ζ (2s) n=0

where

√  s − 12 ζ (2s − 1)

φ(s) = π , σs (n) = ds,
(s) ζ (2s) d|n
d>0

and
 ∞
1 du
e− 2 y(u+(1/u)) u s
1
K s (y) = .
2 0 u
If ζ (1 + it0 ) = 0 for some t0 ∈ R, then one easily sees that if

E ∗ (z, s) = π −s (s)ζ (2s)E(z, s) = E ∗ (z, 1 − s),

then E ∗ (z, (1 + t0 )/2) must be a non-constant Maass form of type 14 + t02 . This
is because the constant term of the Eisenstein series E ∗ (z, s) will vanish when
s = (1 + t0 )/2. It is easy to show that E ∗ (z, (1 + t0 )/2) is non-constant, in
particular non-zero, because the sum:

1
σ1−2s (n)|n|s− 2 K s− 12 (2π|n|y)e2πinx
n=0

will be non-zero high in the cusp. The key point is that an Eisenstein series can
never be a cusp form because Eisenstein series are orthogonal to cusp forms,
so the inner product
?    @
1 + t0 1 + t0
E ∗, , E ∗,
2 2
would have to be zero. This contradicts the fact that E (∗, (1 + t0 )/2) is not iden-
tically zero. Since we have obtained a contradiction, our original assumption:
that ζ (s) vanished on the line (s) = 1, must be false!

Non-vanishing of G L(2) L-functions on the line (s) = 1 The argument

described above can be vastly generalized. We shall give another example based
on Proposition 10.11.2. Let φ be a Maass form for S L(2, Z) with associated
L-function L φ (s). Let φ (λ) denote the completed L-function as in Proposition
10.11.2, and define E ∗P2,1 (z, s, φ) = (λ)E ∗P2,1 (z, s, φ). If L φ (s) vanishes on
324 Langlands Eisenstein series

(s) = 1, then there exists a special value of s such that E ∗P2,1 (z, s, φ) will be a
Maass form for S L(3, Z), i.e., its constant term will vanish. One again obtains
a contradiction by taking the inner product of E ∗ with itself.

10.13 Langlands spectral decomposition for S L(3, Z)\h3

We conclude this chapter with the Langlands spectral decomposition for the spe-
cial case of S L(3, Z). In order to state the main result succinctly, we introduce
the following notation. Let u j , ( j = 0, 1, 2, . . . ) denote a basis of normalized
Maass forms for S L(2, Z) with z ∈ h2 . Here u 0 is the constant function, and
each u j is normalized to have Petersson norm one. For each j = 0, 1, 2, . . . ,
and s ∈ C, define

E j (z, s) := E P2,1 (z, s, u j )

as in (10.11.1). We also define for s1 , s2 ∈ C, the minimal parabolic Eisenstein

series

E(z, s1 , s2 ) := E Pmin (z, (s1 , s2 ))

as in (10.4.1). The following theorem generalizes Selberg’s spectral decompo-

sition given in Section 3.16. We follow the proof of M. Thillainatesan and thank
her for allowing us to incorporate it here.

Theorem 10.13.1 (Langlands spectral decomposition) Assume that

φ ∈ L2 (S L(3, Z)\h3 ) is orthogonal to the residues of all the Eisenstein series
and is of sufficiently rapid decay that φ, E converges absolutely for all the
Eisenstein series E. Then the function
 
1
φ(z) − φ, E(∗, s1 , s2 ) E(z, s1 , s2 ) ds1 ds2
(4πi)2 ( 13 ) ( 13 )
∞ 
1

− φ, E j (∗, s) E j (z, s) ds,
2πi j=0 ( 12 )

is a cusp form for S L(3, Z).

Proof Let f (z) be an arbitrary automorphic form on h3 . We shall adopt the

simplifying notation that f 0,0 denotes the constant term of f along the minimal
parabolic, f 2,1 denotes the constant term of f along the maximal parabolic P2,1 ,
and f 1,2 denotes the constant term along P1,2 . In order to prove the Langlands
spectral decomposition theorem for S L(3, Z), it is necesssary to show that
the following identities (10.13.2), (10.13.3), (10.13.4) hold for a Maass form
 10.13 Langlands spectral decomposition for S L(3, Z)\h3 325

φ ∈ L2 (S L(3, Z)\h3 ).

1
φ00 (z) = φ, E(∗, s1 , s2 ) E 00 (z, s1 , s2 ) ds1 ds2 . (10.13.2)
(4πi)2
(si )= 13

Note that E j,00 (z, s) = 0 for j ≥ 1 by Theorem 10.10.3.

 
1
φ21 (z) = φ, E(∗, s1 , s2 ) E 21 (z, s1 , s2 ) ds1 ds2
(4πi)2 ( 13 ) ( 13 )
∞ 
1

− φ, E j (∗, s) E j,21 (z, s) ds, (10.13.3)
2πi j=0 ( 12 )
 
1
φ12 (z) = φ, E(∗, s1 , s2 ) E 12 (z, s1 , s2 ) ds1 ds2
(4πi)2 ( 13 ) ( 13 )
∞ 
1

− φ, E j (∗, s) E j,12 (z, s) ds. (10.13.4)
2πi j=0 ( 12 )

The idea of the proof is to embed h2 in h3 and show that φ21 is invariant under
S L(2, Z) with this embedding. Then, we use a G L(2) spectral decomposition
of φ21 .
Following (Garrett, 2002), for z ∈ h3 = S L(3, R)/S O(3, R), we can write:
⎛ ⎞⎛ ⎞⎛ √ ⎞
1 0 x3 0 1/ l 0√ 0
⎜ ⎟
z = ⎝ 0 1 x1 ⎠ ⎝ z 2 0⎠ ⎝ 0 1/ l 0 ⎠ ,
0 0 1 0 0 1 0 0 l

where x1 , x3 , l ∈ R and l > 0. Note that z 2 ∈ h2 = S L(2, R)/S O(2, R).

It can be shown that the constant term φ21 , as a function of z 2 , is invariant
under γ ∈ S L(2, Z) and has constant term φ00 . So already the proof of the main
theorem is reduced to showing (10.13.3) and (10.13.4).
In the proof of (10.13.3) and (10.13.4) we will need to know φ, E for each
of the Eisenstein series that we have defined. We begin by calculating φ, E
for the minimal parabolic Eisenstein series.

Lemma 10.13.5 We have

φ, E(∗, s1 , s2 ) = φ̃ 00 (2s1 + s2 − 2, s1 + 2s2 − 2),

where φ̃ 00 is the double Mellin transform of the constant term φ00 .

The proof of Lemma 10.13.5 is fairly straightforward and we will omit it. We
will, however, give the calculation of φ, E j (∗, s). It is part of our assumption
326 Langlands Eisenstein series

that the integral converges absolutely.


φ, E j (∗, s) = φ(z) E j (z, s) d ∗ z,
\h3

= φ(z) y12s y2s u j (z 2 ) d ∗ z.
P21 \h3

With the previous notation, l is given by y12 y2 = l 3 . As in Section 1.6, we have

the following change of coordinates.
P21 \h3 ≡ S L(2, Z)\h2 × (R/Z)2 × [0 < l < ∞],
and
3 −3 ∗ dl
d∗z =
l d z 2 d x3 d x1 .
2 l
Continuing the calculation gives
   
3 ∞ 1 1 dl
φ, E j (∗, s) = φ(z) l 3s−3 u j (z 2 ) d ∗ z 2 d x3 d x1 ,
2 0 0 0 l
S L(2,Z)\h2
 ∞ 
3 dl
= φ21 (z 2 , l) l 3s−3 u j (z 2 ) d ∗ z 2 ,
2 0 l
S L(2,Z)\h2
 ∞
3 dl
= φ21 (∗, l), u j  l 3s−3 .
2 0 l
The above calculation proves the following Lemma 10.13.6. When we use the
notation of Garrett (2002), we will write φ21 (z 2 , l) for φ21 (z).
Lemma 10.13.6 For (s)  1,
3
φ, E j (∗, s) = ã j (3s − 3),
2
where ã j (s) is the Mellin transform of
a j (l) = φ21 (∗, l), u j .

As observed earlier, the constant term φ21 considered as a function of z 2 ,

has a Selberg spectral expansion. We have given a proof in Section 3.16 of the
Selberg (S L(2, Z)) spectral decomposition using Mellin transforms. Then

∞ 
1
φ21 (z 2 , l) = φ21 (∗, l), u j  u j (z 2 )+ φ21 (∗, l), E(∗, v) E(z 2 , v) dv,
j=0
4πi ( 12 )

where E(z, s) is the G L(2) Eisenstein series.

 10.13 Langlands spectral decomposition for S L(3, Z)\h3 327

Note that if we apply Mellin inversion to the results of Lemma 10.13.6 and
by our assumptions move the line of integration to ( 12 ), we have the following.

1
a j (l) = φ21 (∗, l), u j  = φ, E j (∗, s) l 3−3s ds. (10.13.7)
πi ( 12 )

Now we can put this together with the spectral decomposition of φ21 (z 2 , l)
to get
∞ 
1

φ21 (z 2 , l) = φ, E j (∗, s) l 3−3s u j (z 2 ) ds
πi j=0 ( 12 )

1
+ φ21 (∗, l), E(∗, v) E(z 2 , v) dv.
4πi ( 12 )

Make the change of variable s → 1 − s in the first integral. Then the equation
above becomes:
∞ 
1

φ21 (z 2 , l) = φ, E j (∗, s) l 3s u j (z 2 ) ds
πi j=0 ( 12 )

1
+ φ21 (∗, l), E(∗, v) E(z 2 , v) dv. (10.13.8)
4πi ( 12 )

We shall need to use Proposition 10.11.2 on the Fourier expansion of the

maximal parabolic Eisenstein series. It follows that we can rewrite (10.13.8) in
the form
∞ 
1

φ21 (z 2 , l) = φ, E j (∗, s) E j,21 (z 2 , l, s) ds
2πi j=0 ( 12 )

1
+ φ21 (∗, l), E(∗, v) E(z 2 , v) dv. (10.13.9)
4πi ( 12 )

Let us now assume the following proposition, whose proof will be deferred
to later.

Proposition 10.13.10 We have


1
φ21 (∗, l), E(∗, v) E(z 2 , v) dv
4πi
( 12 )
 
1
= φ, E(∗, s1 , s2 ) E 21 (z 2 , l, s1 , s2 ) ds1 ds2 .
(4πi)2
( 13 ) ( 13 )
328 Langlands Eisenstein series

Using the above proposition, we can rewrite (10.13.9) in the form

∞ 
1

φ21 (z) = φ, E j (∗, s) E j,21 (z, s) ds
2πi j=0
( 12 )
 
1
+ φ, E(∗, s1 , s2 ) E 21 (z, s1 , s2 ) ds1 ds2 .
(4πi)2
( 13 ) ( 13 )

This proves (10.13.3). This is almost the end because (10.13.4) is proved
using (10.13.3) and a few properties of Eisenstein series. On G L(3), there is an
involution which preserves the Iwasawa form. The involution ι is defined by
⎛ ⎞
1
ι t −1
z = w z w, where w=⎝ 1 ⎠ . (10.13.11)
1
For any automorphic form φ(z), we can define another automorphic form,
also on S L(3, Z), denoted φ̃(z) given by

φ̃(z) = φ(ι z). (10.13.12)

We let Ẽ j (z, s) = E j (ι z, s). It is easy to show that Ẽ(z, s1 , s2 ) = E(z, s2 , s1 ).

It follows that φ21 (ι z) = φ̃ 12 (z) and φ12 (ι z) = φ̃ 21 (z). So, applying (10.13.3)
to φ̃ gives
∞ 
1

φ̃ 21 (z) = φ̃, E j (∗, s) E j,21 (z, s) ds
2πi j=0
( 12 )
 
1
+ φ̃, E(∗, s1 , s2 ) E 21 (z, s1 , s2 ) ds1 ds2 .
(4πi)2
( 13 ) ( 13 )

Since φ̃ 21 (ι z) = φ12 (z), it follows that:

∞ 
1

φ12 (z) := φ̃, E j (∗, s) E j,21 (ι z, s) ds
2πi j=0
( 12 )
 
1
+ φ̃, E(∗, s1 , s2 ) E 21 (ι z, s1 , s2 ) ds1 ds2 .
(4πi)2
( 13 ) ( 13 )

It is easy to see that φ̃, ψ̃ = φ, ψ. It follows that φ̃, E j (∗, s) =
φ, Ẽ j (∗, s). Also, we have the following equalities:

φ̃, E(∗, s1 , s2 ) = φ, Ẽ(∗, s1 , s2 ) = φ, E(∗, s2 , s1 ).

 10.13 Langlands spectral decomposition for S L(3, Z)\h3 329

From these statements, we can continue the calculation of φ12 (z).

∞ 
1

φ12 (z) = φ, Ẽ j (∗, s) Ẽ j,12 (z, s) ds
2πi j=0
( 12 )
 
1
+ φ, E(∗, s2 , s1 ) E 12 (z, s2 , s1 ) ds1 ds2 .
(4πi)2
( 13 ) ( 13 )

Let s → 1 − s in the first integral. Then

∞ 
1

φ12 (z) = φ, Ẽ j (∗, 1 − s) Ẽ j,12 (z, 1 − s) ds
2πi j=0
( 12 )
 
1
+ φ, E(∗, s2 , s1 ) E 12 (z, s2 , s1 ) ds1 ds2 . (10.13.13)
(4πi)2
( 13 ) ( 13 )

Now, the maximal Eisenstein series has the functional equation:

Ẽ j (z, 1 − s) = θ (s) E j (z, s),

where
 j (λ − 1)
θ (s) = .
 j (λ)

The coefficients in L j (v) are real since they come from the Hecke form u j , so
we know that θ (s) = θ(s̄). Applying these observations to (10.13.13) gives
∞ 
1

φ12 (z) = θ(s̄)φ, E j (∗, s) Ẽ j,12 (z, 1 − s) ds
2πi j=0
( 12 )
 
1
+ φ, E(∗, s2 , s1 ) E 12 (z, s2 , s1 ) ds1 ds2 ,
(4πi)2
( 13 ) ( 13 )
∞ 
1

φ12 (z) = φ, E j (∗, s) E j,12 (z, s) ds
2πi j=0
( 12 )
 
1
+ φ, E(∗, s1 , s2 ) E 12 (z, s1 , s2 ) ds1 ds2 .
(4πi)2
( 13 ) ( 13 )

This proves (10.13.4), thus proving the main Theorem 10.13.1.

330 Langlands Eisenstein series

Proof of Proposition 10.13.10 It still remains to prove Proposition 10.13.10.

Let us define for (v1 )  1 and (v2 )  1,

 
v 
E ∗ (z, v1 , v2 ) = 2 y12 y2 1 E(z 2 , v2 ) ,
γ ∈P21 \ γ

a series which is a rearrangement of the terms of the minimal parabolic Eisen-

stein series. We can see this as follows. For (s1 ), (s2 )  1,

2s +s s +2s 
E(z, s1 , s2 ) = y1 1 2 y21 2  ,
γ ∈∞ \ γ


s +(s /2)+3s2 /2 
= y12s1 +s2 y21 2  ,
γ ∈∞ \ γ


 
s1 +(s2 /2) 3s2 /2 
= y12 y2 y2  .
β∈∞ \P21 α∈P21 \ βα

It was shown in Chapter 6 that the function det z = y12 y2 , is invariant under
β ∈ P21 . So, det βz = det z.
Previously we had dropped the dependence of n in the notation, but to avoid
confusion we need to introduce n in the notation. Thus (n) = S L(n, Z) and
∞ (n) is the set of upper-triangular matrices in S L(n, Z) with 1s on the diagonal.
 

 s1 +(s2 /2)
3s2 /2  
E(z, s1 , s2 ) = 2
y1 y2 y2   ,
α∈P21 \(3) β∈∞ (3)\P21 β α

  
 s1 +(s2 /2) 3s2 
= 2 y12 y2 E z2,  ,
2 
α∈P21 \(3) α
 
s2 3s2
= E∗ z, s1 + , .
2 2
In the above we have used ∞ (3)\ P21 ≡ ∞ (2)\ S L(2, Z), a fact which is easy
to prove and also follows as a special case of Lemma 5.3.13.
Now, we will drop the dependence of the notation on n. To prove the proposi-
tion, we want to calculate the following inner product, in the region of absolute
convergence of E ∗ (z, v1 , v2 ). In a manner similar to the calculation of Lemma
10.13.6, we have

φ, E ∗ (∗, v1 , v2 ) = φ(z) E ∗ (z, v1 , v2 ) d ∗ z
\h3

= φ(z) E ∗ (z, v1 , v2 ) d ∗ z,
\h3
 10.13 Langlands spectral decomposition for S L(3, Z)\h3 331

 ∞  1  1 
=3 φ(z)y12v1 y2v1 l −3 E(z 2 , v2 )
0 0 0
S L(2,Z)\h2

dl
× d ∗ z 2 d x3 d x1 ,
l
 ∞ 
dl
=3 φ21 (z 2 , l) l 3v1 −3 E(z 2 , v2 ) d ∗ z 2 ,
0 l
S L(2,Z)\h2
 ∞
dl
=3 φ21 (∗, l), E(∗, v2 ) l 3v1 −3 .
0 l

Let a(l) = φ21 (∗, l), E(∗, v2 ), then we have just shown that

φ, E ∗ (∗, v1 , v2 ) = 3 ã(3v1 − 3).

If we take the inverse Mellin transform of ã(v), then for some σ  0 we

have

1
a(l) = ã(v)l −v dv,
2πi (σ )

3
= ã(3v1 − 3) l 3−3v1 dv1 ,
2πi (σ  )

1
= φ, E ∗ (∗, v1 , v2 ) l 3−3v1 dv1 .
2πi ( 12 )

In the last step we have used the assumption that φ is orthogonal to all the
residues of the Eisenstein series to move the line to (v1 ) = 12 .
We have just shown that

1
φ21 (∗, l), E(∗, v2 ) = φ, E ∗ (∗, v1 , v2 ) l 3−3v1 dv1 . (10.13.14)
2πi
( 12 )

Recall that we want to show


1
φ21 (∗, l), E(∗, v2 ) E(z 2 , v2 ) dv2
4πi
( 12 )
 
1
= φ, E(∗, s1 , s2 ) E 21 (z 2 , l, s1 , s2 ) ds1 ds2 .
(4πi)2
( 13 ) ( 13 )
332 Langlands Eisenstein series

Using (10.13.14), we have that the left-hand side of the above is


1
LHS = φ21 (∗, l), E(∗, v2 ) E(z 2 , v2 ) dv2 ,
4πi ( 12 )
 
1
= φ, E ∗ (∗, v1 , v2 ) l 3(1−v1 ) E(z 2 , v2 ) dv1 dv2 ,
2(2πi)2 ( 12 ) ( 12 )
 
1
= φ, E ∗ (∗, v1 , v2 ) l 3v1 E(z 2 , v2 ) dv1 dv2 ,
2(2πi)2 ( 12 ) ( 12 )

where we have let v1 → 1 − v1 . Note that l 3 = y12 y2 . So,

 
1
LHS = φ, E ∗ (∗, v1 , v2 ) y12v1 y2v1 E(z 2 , v2 ) dv1 dv2 .
2(2πi)2 ( 12 ) ( 12 )

Now let v1 = s1 + (s2 /2) and v2 = 3s2 /2. From the definition of E ∗ (z, v1 , v2 ),
we get the following.
   
3  2 s1 +(s2 /2) 3s2
LHS = φ, E(∗, s1 , s2 ) y1 y2 E z2, ds1 ds2 .
4(2πi)2 ( 13 ) ( 13 ) 2
(10.13.15)

Using the fact that 21 (z 2 , l) is invariant under S L(2, Z) and so has a Fourier
expansion, it is easy to show that for any form (z),


21 (z) = 0,n (z),
n∈Z

where
⎛⎛ ⎞ ⎞
 1  1  1 1 u2 u3
n 1 ,n 2 (z) =  ⎝⎝0 1 u 1 ⎠ z⎠ e(−n 1 u 1 − n 2 u 2 ) du 1 du 2 du 3 .
0 0 0
0 0 1
We will need some results from (Bump, 1984) and will use the notation from
there. In (Bump, 1984), it is shown that the expansion of E 0,n is the sum of
three Whittaker functions indexed by w ∈ V, where V is a certain subset of the
Weyl group W . (In the notation of (Bump, 1984), V = {w1 , w2 , w4 }.) So,
1
(s1 ,s2 )
E 0,n (z, s1 , s2 ) = aw (n, s1 , s2 ) W0,n (z, w). (10.13.16)
2 w∈V

Using the expansion of the G L(2) Eisenstein series, it is easy to show that
the factor (y12 y2 )s1 +(s2 /2) E (z 2 , 3s2 /2) which appears in (10.13.15) is exactly:
 
 2 s1 +(s2 /2) 3s2 1
(s1 ,s2 )
y1 y2 E z2, = aw (n, s1 , s2 ) W0,n (z, w2 ).
2 2 n∈Z 2
 10.13 Langlands spectral decomposition for S L(3, Z)\h3 333

Now in (10.13.15), we will divide the integral into three parts and in each
part let (s1 , s2 ) → (u(s1 ), u(s2 )) for u ∈ U. We choose the subset U of W in
such a way as to get the sum over u to give us E 0,n (z, s1 , s2 ). We can do this
because the Whittaker functions have certain transformation properties. We
(s1 ,s2 )
will be interested in the transformation of W0,n (z, w2 ) under the action of the
Weyl group on (s1 , s2 ); specifically for w ∈ V, there exists a u ∈ U depending
on w, such that
(u(s1 ),u(s2 )) (s1 ,s2 )
W0,n (z, w2 ) = Bwu W0,n (z, w). (10.13.17)
Recall that w ∈ V are those that appear in the expansion of E 21 .
LHS

 
3 (s1 ,s2 )
= 2
φ, E(∗, s1 , s2 ) aw2 (n, s1 , s2 )W0,n (z, w2 ) ds1 ds2 ,
n∈Z
8(2πi) ( 1
) ( 1
)
 
3 3


1
= φ, E(∗, u(s1 ), u(s2 )) aw2 (n, u(s1 ), u(s2 ))
n∈Z u∈U
8(2πi)2
( 13 ) ( 13 )
(u(s1 ),u(s2 ))
× W0,n (z, w2 ) ds1 ds2 ,


 
1
= φ, E(∗, u(s1 ), u(s2 )) aw2 (n, u(s1 ), u(s2 ))
n∈Z w∈V
8(2πi)2
( 13 ) ( 13 )
(s1 ,s2 )
× Bwu W0,n (z, w)ds1 ds2 .
Also note that at (si ) = 13 , we have (u(si )) = 13 .
From Bump again, we can show that for any w ∈ W,
E(z, w(s1 ), w(s2 )) = ϕw (s1 , s2 )E(z, s1 , s2 )
for some ϕw independent of z. This is done as follows. First, let us define, as in
(Bump, 1984), the function G(z, s1 , s2 ) in the following way.
G(z, s1 , s2 ) = B(s1 , s2 ) E(z, s1 , s2 ),
where
   
ζ (3s1 ) ζ (3s2 ) ζ (3s1 + 3s2 − 1) 3s1 3s2
B(s1 , s2 ) =  
4π 3s1 +3s2 −1/2 2 2
 
3s1 + 3s2 − 1
× E(z, s1 , s2 ).
2
Now from (Bump, 1984), we know that G(z, s1 , s2 ) is invariant under the
action of the Weyl group on the pair (s1 , s2 ). In particular, it follows,
B(w(s1 ), w(s2 )) E(z, w(s1 ), w(s2 )) = B(s1 , s2 )E(z, s1 , s2 ). (10.13.18)
334 Langlands Eisenstein series

B(s1 , s2 )
So, ϕw (s1 , s2 ) = . From the definition B(s1 , s2 ) = B(s1 , s2 ).
B(w(s1 ), w(s2 ))
At (si ) = 13 , we have that si = 23 − si and (w(si )) = 13 . It follows that at
(si ) = 13 ,
2
B − s1 , 23 − s2
ϕw (s1 , s2 ) = 2 3
.
B 3
− w(s1 ), 23 − w(s2 )

Consequently,
  2
1

B − s1 , 23 − s2
LHS = φ, E(∗, s1 , s2 ) · 2 3

2(4πi)2 n∈Z w∈V B 3
− u(s1 ), 23 − u(s2 )
si = 13
(s1 ,s2 )
× aw2 (n, u(s1 ), u(s2 ))Bwu W0,n (z, w) ds1 ds2 .

To prove the proposition, it remains to show the following.

Lemma 10.13.19 The constant term E 21 (z, s1 , s2 ), of the Eisenstein series

E(z, s1 , s2 ) is

1

B 23 − s1 , 23 − s2
E 21 (z, s1 , s2 ) = 
2 n∈Z w∈V B 23 − u(s1 ), 23 − u(s2 )
(s1 ,s2 )
× aw2 (n, u(s1 ), u(s2 ))Bwu W0,n (z, w).

Proof We have defined coefficients aw (n, s1 , s2 ) in (10.13.16) such that

1

(s1 ,s2 )
E 21 (z, s1 , s2 ) = aw (n, s1 , s2 ) W0,n (z, w). (10.13.20)
2 n∈Z w∈V

By the linear independence of these Whittaker functions, we are reduced to

showing that

B 23 − s1 , 23 − s2
aw (n, s1 , s2 ) =  2 aw2 (n, u(s1 ), u(s2 ))Bwu ,
B 3 − u(s1 ), 23 − u(s2 )
(10.13.21)

where u depends on w as noted in (10.13.17).

The proof of the above lemma will follow if we put all the definitions and
transformation properties together. We can write equation (10.13.18) using the
 10.13 Langlands spectral decomposition for S L(3, Z)\h3 335

expansion (10.13.16). So for any u ∈ U, we have the equalities below.

B(s1 , s2 )
E 21 (z, s1 , s2 ) = E 21 (z, u(s1 ), u(s2 ))
B(u(s1 ), u(s2 ))

B(s1 , s2 ) (s1 ,s2 )
av (n, s1 , s2 ) W0,n (z, v)
v∈V
B(u(s 1 ), u(s 2 ))

(u(s1 ),u(s2 ))
= aw (n, u(s1 ), u(s2 ))W0,n (z, w).
w∈V

Now if we consider this last equality along with the transformation property
(u(s1 ),u(s2 )) (s1 ,s2 )
W0,n (z, w2 ) = Bwu W0,n (z, w)

and use the fact that the Whittaker functions are linearly independent, we get
the equality:
B(s1 , s2 ) (s1 ,s2 )
aw (n, s1 , s2 )W0,n (z, w)
B(u(s1 ), u(s2 ))
(s1 ,s2 )
= aw2 (n, u(s1 ), u(s2 ))Bwu W0,n (z, w).

So,
B(s1 , s2 )
aw (n, s1 , s2 ) = aw2 (n, u(s1 ), u(s2 ))Bwu . (10.13.22)
B(u(s1 ), u(s2 ))
Now (10.13.21) reduces to showing that

B (u(s1 ), u(s2 )) B 23 − s1 , 23 − s2
= 2 . (10.13.23)
B(s1 , s2 ) B 3 − u(s1 ), 23 − u(s2 )
To prove this and to finish the proof of the lemma, we have to show that the
product
 
2 2
B(s1 , s2 ) B − s1 , − s2
3 3
is invariant under the action of the Weyl group on (s1 , s2 ). Since
 
2 2
B(s1 , s2 ) B − s1 , − s2
3 3
is a product of six gamma functions and six zeta functions and since the quan-
tities

3s1 , 3s2 , 3s1 + 3s2 − 1, 2 − 3s1 , 2 − 3s2 , 3 − 3s1 − 3s2

are permuted by the Weyl group, this completes the proof of Lemma 10.13.19
and Proposition 10.13.10.

336 Langlands Eisenstein series

GL(n)pack functions The following GL(n)pack functions, described in the

appendix, relate to the material in this chapter:
BruhatCVector BruhatForm
BlockMatrix KloostermanBruhatCell
KloostermanCompatibility LanglandsForm
LanglandsIFun LongElement
MPEisensteinGamma MPEisensteinLambdas
MPEisensteinSeries MPExteriorPowerGamma
MPExteriorPowerLFun MPExteriorPowerGamma
MPSymmetricPowerLFun ParabolicQ
 11
Poincaré series and Kloosterman sums

Poincaré series and Kloosterman sums associated to the group S L(3, Z) were
introduced and studied in (Bump, Friedberg and Goldfeld, 1988) following the
point of view of Selberg (1965). A very nice exposition of the G L(2) theory
is given in (Cogdell and Piatetski-Shapiro, 1990). The method was first gen-
eralized to G L(n) in (Friedberg, 1987), (Stevens, 1987). In (Bump, Friedberg
and Goldfeld, 1988) it is shown that the S L(3, Z) Kloosterman sums are hyper
Kloosterman sums associated to suitable algebraic varieties. Non-trivial bounds
were obtained by using Hensel’s lemma and Deligne’s estimates for hyper-
Kloosterman sums (Deligne, 1974) in (Larsen, 1988), and later (Dabrowski
and Fisher, 1997) improved these bounds by also using methods from algebraic
geometry following (Deligne, 1974). Sharp bounds for special types of Kloost-
erman sums were also obtained in (Friedberg, 1987a,c). In (Dabrowski, 1993),
the theory of Kloosterman sums over Chevalley groups is developed. Impor-
tant applications of the theory of G L(n) Kloosterman sums were obtained in
(Jacquet, 2004b) (see also (Ye, 1998)).
Another fundamental direction for research in the theory of Poincaré series
and Kloosterman sums was motivated by the G L(2) Kuznetsov trace formula,
(see (Kuznecov, 1980) and also (Bruggeman, 1978)). Generalizations of the
Kuznetsov trace formula to G L(n), with n ≥ 3 were obtained in (Friedberg,
1987), (Goldfeld, 1987), (Ye, 2000), but they have not yet proved useful for
analytic number theory. The chapter concludes with a new version of the G L(n)
Kuznetsov trace formula derived by Xiaoqing Li.

11.1 Poincaré series for SL(n, Z)

Let n ≥ 2 and let s = (s1 , . . . , sn−1 ) ∈ Cn−1 . Fix a character ψ of the upper
triangular unipotent group Un (R) of n × n matrices with coefficients in R and

337
338 Poincaré series and Kloosterman sums

1s on the diagonal. Note that for fixed integers m 1 , m 2 , . . . , m n−1 , the character
ψ may be defined by
ψ(u) = ψ (u) = e2πi (m 1 u 1 +m 2 u 2 +···+m n−1 u n−1 )
m 1 ,...,m n−1 (11.1.1)
where
⎛ ⎞
1 u n−1
⎜ 1 u n−2 ∗ ⎟
⎜ ⎟
⎜ .. .. ⎟
u=⎜ . . ⎟ ∈ Un (R).
⎜ ⎟
⎝ 1 u1 ⎠
1
The Poincaré series is constructed from the following functions:
r The I (z)–function given in (5.1.1).
s
r A bounded function: eψ : hn → C which is characterized by the property:

eψ (uz) = ψ(u)eψ (z)

for all u ∈ Un (R), and z ∈ hn . Such functions are termed e-functions.
Definition 11.1.2 (Poincaré series) For n ≥ 2, let s ∈ Cn−1 , and fix an e-
function eψ (z) as above. Then the Poincaré series for  = S L(n, Z), denoted
P(z, s, eψ ), is formally given by


P(z, s, eψ ) := Is (γ z)eψ (γ z).
γ ∈Un (Z)\

Remark Since eψ is bounded on hn , it easily follows from Proposition 10.4.3

that the infinite series above defining the Poincaré series converges absolutely
and uniformly on compact subsets of hn provided (si ) is sufficiently large for
every i = 1, 2, . . . , n − 1.

The Poincaré series is characterized by a very useful property. The inner

product of the Poincaré series with a Maass form will give a certain Mellin
transform of an individual Fourier coefficient of the Maass form. This is made
explicit in the following proposition.
Proposition 11.1.3 Let φ be a Maass form for S L(n, Z) with Fourier expan-
sion given by



∞

∞
A(m 1 , . . . , m n−1 )
φ(z) = ···
γ ∈Un−1 (Z)\S L(n−1,Z) m 1 =1 m n−2 =1 m n−1 =0
)
n−1
|m k |k(n−k)/2
k=1
   
γ
× WJacquet M · z, ν, ψ1,...,1, m n−1 ,
1 |m n−1 |
 11.2 Kloosterman sums 339

as in (9.1.2). Then φ, P(∗, s, eψ ) is equal to

∞ ∞ 
A(m 1 , . . . , m n−1 ) n−1
dyi
· · · WJacquet (M y, ν, ψ1,...,1 ) · Is̄ (y) eψ (y) .
)
n−1
yii(n−i)+1
|m k |k(n−k)/2 0 0
i=1
k=1

Proof By the usual unfolding argument, we have

φ, P(∗, s, eψ )

= φ(z) P(z, s, eψ ) d ∗ z
\hn
  
n−1
dyi
= φ(x y)eψ (y) · ψ̄m 1 ,...,m n−1 (x) d x j,k i(n−i)+1
,
1≤ j<k≤n i=1 yi
Un (Z)\hn

where we have used the explicit measure d ∗ z given in Theorem 1.6.1. The inner
d x-integral picks off the (m 1 , . . . , m n−1 )th Fourier coefficient, and the result
follows.

11.2 Kloosterman sums

The Fourier expansion of the Poincaré series can be computed in the same way
the Fourier expansion was computed for minimal parabolic Eisenstein series
in Section 10.6. Note that if the e-function is the constant function, then the
Poincaré series is just an Eisenstein series. The Fourier expansion involves
certain Kloosterman sums which we shall forthwith define.
Let Wn denote the Weyl group of G L(n, R). In Section 10.3 we established
the Bruhat decomposition
(
G L(n, R) = Gw, (where G w = Un w Dn Un ),
w∈Wn

and Dn denotes the subgroup of diagonal matrices in G L(n, R). We let Un (Z)
denote the matrices in Un with integer coefficients. Let c1 , . . . , cn−1 be non-zero
integers, and set
⎛ ⎞
1/cn−1
⎜ cn−1 /cn−2 ⎟
⎜ ⎟
⎜ .. ⎟
c=⎜ . ⎟ (11.2.1)
⎜ ⎟
⎝ c2 /c1 ⎠
c1

We adopt the notation:  = S L(n, Z) and w = (w−1 · t Un (Z) · w) ∩ Un (Z).

340 Poincaré series and Kloosterman sums

Definition 11.2.2 (Kloosterman sums) Fix w ∈ Wn , c ∈ Dn as in (11.2.1),

and ψ, ψ  characters of Un as in (11.1.1). The S L(n, Z) Kloosterman sum is
defined by


Sw (ψ, ψ  , c) := ψ(b1 ) ψ  (b2 ),
γ ∈ Un (Z)\∩G w / w
γ = b1 cwb2

provided this sum is well defined, (i.e., it is independent of the choice of Bruhat
decomposition for γ ). Otherwise, the Kloosterman sum is defined to be zero.

Example 11.2.3 (S L(2, Z) (Kloosterman sum) In this example,

 −1     
c1 0 0 −1 1 ∗
c= , w= , w = = U2 (Z).
0 c1 1 0 1

The Kloosterman
 sum is
 based on  the Bruhat decomposition. It is convenient
1 b1 /c1 1 b2 /c1
to set b1 = , b2 = . In this case, the Bruhat decom-
1 1
 
±1 ∗
position says that every γ ∈ S L(2, Z), γ = can be written in the
0 ±1
form
   −1   
1 b1 /c1 c1 −1 1 b2 /c1
γ =
1 c1 1 1
  
b1 (b1 b2 − 1)/c1
=
c1 b2

where

b1 b2 ≡ 1 (mod c1 ).

The S L(2, Z) Kloosterman sum takes the form

   
1 b1 /c1 1 b1 /c1
Sw (ψ, ψ  , c) = ψ ψ .
b (mod c )
0 1 0 1
1 1
b1 b1 ≡ 1 (mod c1 )
(11.2.4)

If we let M, N ∈ Z such that

   
1 u 1 u
ψ = e2πi Mu , ψ = e2πi N u ,
0 1 0 1
 11.2 Kloosterman sums 341

then the Kloosterman sum in (11.2.4) can be written in the more traditional form

2πi Mβ+N β̄
Sw (ψ, ψ  , c) := S(M, N , c1 ) = e c1
. (11.2.5)
β (mod c1 )
β β̄ ≡ 1 (mod c1 )

Example 11.2.6 (S L(3, Z) long element Kloosterman sum) This example

is taken from (Bump, Friedberg and Goldfeld, 1988). Let
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
1/c2 −1 1 ∗ ∗
c=⎝ c2 /c1 ⎠,w = ⎝ 1 ⎠ , w = ⎝ 1 ∗ ⎠ = U3 (Z).
c1 1 1
Here w is the so-called long element in the Weyl group. Set
⎛ ⎞ ⎛ ⎞
1 α2 α3 1 β2 β3
b1 = ⎝ 1 α1 ⎠ ∈ U3 (Q), b2 = ⎝ 1 β1 ⎠ ∈ U3 (Q).
1 1
We are interested in determining when an element
⎛ ⎞
a1,1 a1,2 a1,3
γ = ⎝ a2,1 a2,2 a2,3 ⎠ ∈ S L(3, Z)
a3,1 a3,2 a3,3
has the Bruhat decomposition
γ = b1 cwb2 . (11.2.7)
It follows from Proposition 10.3.6 that (11.2.7) holds if
c1 = a3,1 = 0, c2 = a2,2 a3,1 − a2,1 a3,2 = 0. (11.2.8)
With this choice of c1 , c2 , the identity (11.2.7) implies that γ must be equal to
⎛ a2,1 a3,2 α2 a2,1 a3,2 α2 β1 ⎞
a3,1 α3 a2,2 α2 − c1 + a3,1 α3 β2 a2,2 α2 β1 − 1
c2 − c1 + c1 α3 β3
⎜ ⎟
⎜ a2,1 a3,2 a2,1 a3,2 β1 ⎟.
⎝ c1 α1 a2,2 − c1 + c1 α1 β2 a2,2 β1 − c1 + c1 α1 β3 ⎠
c1 c1 β2 c1 β3

One may use this to systematically solve for the coefficients of b1 and b2 . In
this manner, one obtains
⎛ ⎞
1 c1 a1,2 − a1,1ca2 3,2 ac1,11
⎜ ⎟
b1 = ⎜
⎝ 1 a2,1 ⎟
c1 ⎠
,
1
⎛ a3,2 a3,3 ⎞
1 c1 c1
⎜ ⎟
b2 = ⎝ 1 c1 a2,3 − a2,1 a3,3 ⎠. (11.2.9)
c2
1
342 Poincaré series and Kloosterman sums

For M = (M1 , M2 ) ∈ Z2 , N = (N1 , N2 ) ∈ Z2 , let us define two characters

ψ M , ψ N by

ψ M (u) = e2πi(M1 u 1 +M2 u 2 ) , ψ N (u) = e2πi(N1 u 1 +N2 u 2 ) ,

⎛ ⎞
1 u2 u3
where u = ⎝ 1 u 1 ⎠ ∈ U3 (R).
1
It follows from (11.2.7), (11.2.8), (11.2.9) that the long element Kloosterman
sum takes the form

Sw (ψ M , ψ N , c)
6M a 7 6 N (c a −a a ) N a 7

M (c a −a a )
2πi 1c 2,1 + 2 1 1,2c 1,1 3,2 2πi 1 1 2,3c 2,1 3,3 + 2c 3,2
= e 1 2 ·e 2 1 ,
a2,1 ,a3,2 (mod c1 )
c1 a2,3 −a2,1 a3,3 (mod c2 )
c1 a1,2 −a1,1 a3,2 (mod c2 )

subject to the constraint that there exist integers a1,2 , a1,3 , a2,2 such that
 
c1 (a1,2 a2,3 −a1,3 a2,2 )+a3,2 a1,3 a2,1 − a1,1 a2,3 + a3,3 a1,1 a2,2 − a1,2 a2,1 =1,

i.e., that Det(γ ) = 1.

It is important to determine the compatibility conditions for which Kloost-
erman sums are well defined and non-zero. The following proposition is taken
from (Friedberg, 1987).

Proposition 11.2.10 (Compatibility condition) Fix n ≥ 2. The Kloosterman

sum Sw (ψ, ψ  , c) given in Definition 11.2.2 is well defined if and only if

ψ(cwuw −1 ) = ψ  (u)

for all u ∈ (w−1 Un (R)w) ∩ Un (R). Hence the Kloosterman sum is non-zero
only in this case.

Proof See Lemma 10.6.3.

Definition 11.2.11 (Kloosterman zeta function) Fix n ≥ 2. Let ψ, ψ  be

two characters of Un (R), c of the form (11.2.1), and Sw (ψ, ψ  , c) the Kloost-
erman sum as in Definition 11.2.2. Let s = (s1 , . . . , sn−1 ) ∈ Cn−1 with (si )
sufficiently large for all i = 1, . . . , n − 1. The Kloosterman zeta function,
Z (ψ, ψ  , s) is defined by the absolutely convergent series


∞
∞
Sw (ψ, ψ  , c)
Z (ψ, ψ  , s) := ··· nsn−1 .
c1 =1
cns1 · · · cn−1
cn−1 =1 1
 11.3 Plücker coordinates and the evaluation of Kloosterman sums 343

Remarks The Kloosterman zeta function was first introduced, for the case of
G L(2), in (Selberg, 1965) who showed the meromorphic continuation in s and
the existence of infinitely many simple poles on the line (s) = 12 occurring at
the eigenvalues of the Laplacian. Selberg used Weil’s bound for G L(2) Kloost-
erman sums to deduce that the Kloosterman zeta function converged absolutely
in the region (s) > 34 . He was able to immediately deduce from this the bound
≥ 163
for the lowest eigenvalue of the Laplacian (see (Selberg, 1965)).
For (s) > 12 + , it was shown in (Goldfeld and Sarnak, 1983) that the
Kloosterman zeta function is

1

|s| 2
O .
(s) − 12
Such results have not been obtained for higher-rank Kloosterman zeta functions.
This would constitute an important research problem.

11.3 Plücker coordinates and the explicit evaluation of

Kloosterman sums
Definition 11.3.1 (Plücker coordinates) Let g ∈ G L(n, R) with n ≥ 2. The
 the n − 1 row vectors ρ1 , . . . , ρn−1 , where, for each
Plücker coordinates of g are
n
1 ≤ k ≤ n − 1, ρk ∈ R k consists of every possible k × k minor formed with
the bottom k rows of the matrix g.

As an example, consider
⎛ ⎞
g1,1 g1,2 g1,3
⎝
g = g2,1 g2,2 g2,3 ⎠ ∈ G L(3, R).
g3,1 g3,2 g3,3
Then the Plücker coordinates are

ρ1 = {g3,1 , g3,2 , g3,3 }

ρ2 = {g2,1 g3,2 − g3,1 g2,2 , g2,1 g3,3 − g3,1 g2,3 , g2,2 g3,3 − g3,2 g2,3 }.

The key theorem needed for the evaluation of Kloosterman sums is the
following (see (Friedberg, 1987)).

Theorem 11.3.2 Let n ≥ 2. If g ∈ S L(n, R), a necessary and sufficient condi-

tion for the coset of g in Un (Z)\S L(n, R) to contain a representative in S L(n, Z)
is that each of the rows ρk (with 1 ≤ k ≤ n − 1) of Plücker coordinates of g
consist of coprime integers.
344 Poincaré series and Kloosterman sums

Proof It is convenient to have a notation for the minors that occur in the
Plücker coordinates. Fix n ≥ 2. For 1 ≤ k ≤ n − 1, let Lk denote the set of all
k-element subsets of {1, 2, . . . , n} with the lexicographical ordering <, where

{ 1 , . . . , k } < 1 , . . . , k

if 1 < 1 . If the first elements are equal, but only in that case, you look at the
second and compare them, whichever is higher, and that is the “greater” set. One
continues naturally in this manner obtaining an ordering which is analogous to
alphabetical order. Then the set Lk can be used to index the minors that occur
in the Plücker coordinate ρk . For example, if { 1 , . . . , k } ∈ Lk , then we may
consider the minor formed with the columns 1 , . . . , k .
For each k with 1 ≤ k < n − 1, we may list the elements of Lk+1 (in some
order) as follows:
- .
Lk+1 = λk,1 , . . . , λk, n 
k+1

with

λk,i = { k,i,1 , . . . , k,i,k+1 }, k,i,1 < k,i,2 < · · · < k,i,k+1 .

This allows us to index the coordinates of R2 −2 by L1 ∪ · · · ∪ Ln−1 . Next,

n

define V ⊆ R2 −2 to be the affine algebraic set of all

n


ν1 , . . . , νn , ν1,2 , ν1,3 , . . . , νn−1,n , . . . , ν2,3,...,n ⊆ R2 −2
n
(11.3.3)

satisfying
⎛ ⎞
νλk,1
⎜ νλk,2 ⎟
⎜ ⎟
⎜ .. ⎟ ∈ Image(T ), (11.3.4)
⎜ . ⎟
⎝ ⎠
νλ  
n
k, k+1
 n
for all 1 ≤ k < n − 1, where T = (ti, j ) (with 1 ≤ i ≤ k+1 , 1 ≤ j ≤ n) is
the linear transformation

(−1)r −1 νλk,i −{ j} if j = k,i,r ∈ λk,i ,
ti, j = (11.3.5)
0 otherwise.
Here λk,i is a set of k integers and λk,i − { j} means to delete the element j
from the set λk,i . Note also that the linear transformation T maps Rn to Rn
by simple left matrix multiplication of T on an n-dimensional column vector
in Rn . Thus, given coordinates νλ (with λ ∈ Lk ) we see that (11.3.4), (11.3.5)
gives a condition which allows us to see which νλ (with λ ∈ Lk+1 ) occur.
 11.3 Plücker coordinates and the evaluation of Kloosterman sums 345

For example, if n = 3, then V is given by the set of all

{ν1 , ν2 , ν3 , ν1,2 , ν1,3 , ν2,3 } subject to the constraint that
⎛ ⎞ ⎛⎛ ⎞⎞
ν1,2 ν2 −ν1 0
⎝ ν1,3 ⎠ ∈ Image ⎝⎝ ν3 0 −ν1 ⎠⎠ ,
ν2,3 0 ν3 −ν2

i.e., such that

ν3 ν1,2 − ν2 ν1,3 + ν1 ν2,3 = 0.

n
−2
We now consider the map M : G L(n, R) → R2 given by

M(g) := {Mλ (g) | λ ∈ L1 ∪ · · · ∪ Ln−1 },

where for λ = { 1 , . . . , k }, the function Mλ (g) denotes the k × k minor formed

from the bottom k rows and columns 1 , . . . , k .
The first step in the proof of Theorem 11.3.2 is to show that the above map
M : G L(n, R) → R2 −2 induces a bijection from Un (R)\G L(n, R) to
n

 
V1 := v ∈ V  vλ = 0 for some λ ∈ Ln−1 .

To see this, note that the map M factors through Un (R)\G L(n, R). Furthermore,
condition (11.3.4) holds for M(g) since Mλ (g) can be computed for λ ∈ Lk in
terms of Mλ (g) with λ ∈ Lk−1 for all 1 < k < n, by expanding the determinant
along the top row. Since Det(g) = 0 it follows that M maps into V1 . Now
given v ∈ V1 , the condition (11.3.4) guarantees that there exists an (n − 1) × n
matrix with appropriate minors. Since vλ = 0 for some λ ∈ Ln−1 , this can be
completed to a matrix g ∈ G L(n, R) with image v. Consequently, the map
M is onto V1 . It remains to show that the map M : Un (R)\G L(n, R) → V1 is
injective.
To see this, note first that by Proposition 10.3.6, if
 (1) 
1 if λ ∈ λ(1)
1 , . . . , λn−1
Mλ (g) =
0 otherwise,

where λi(1) = {n, n − 1, . . . , n − i + 1} for 1 ≤ i ≤ n − 1, then g must lie in

Un (R). Suppose that M(g) = M(h). Then for any λ = { 1 , . . . , k } ∈ Lk we
have by Proposition 10.3.6 that

g −1 (e1 ) ∧ · · · ∧ g −1 (en−k ) ∧ e 1 ∧ · · · ∧ e k
= h −1 (e1 ) ∧ · · · ∧ h −1 (en−k ) ∧ e 1 ∧ · · · ∧ e k .
346 Poincaré series and Kloosterman sums

Hence

Mλ (gh −1 )e1 ∧ · · · ∧ en = g −1 e1 ∧ · · · ∧ g −1 en−k ∧ h −1 e 1 ∧ · · · ∧ h −1 e k

= h −1 e1 ∧ · · · ∧ h −1 en−k ∧ h −1 e 1 ∧ · · · ∧ h −1 e k
= e1 ∧ · · · ∧ en−k ∧ e 1 ∧ · · · ∧ e k .

It follows that

1 if λ = λ(1)
k ,
Mλ (gh −1 ) =
0 if λ = λ(1)
k ,

and, therefore, Un (R)g = Un (R)h.

We now complete the proof of Theorem 11.3.2. Let
 
V2 = ν ∈ V  νλ ∈ Z for all λ, gcd(νλ | λ ∈ Lk ) = 1 for 1 ≤ k ≤ n − 1 .

It is enough to show that the map M : Un (Z)\S L(n, Z) → V2 is surjective. It

follows from the previous arguments that given ν ∈ V2 , there exists a matrix
g ∈ G L(n, R) with the desired minors. We must show that the coset Un (R)g
contains an integral element. Let pλ ∈ Z satisfy


pλ Mλ (g) = 1, (1 ≤ k < n).
λ∈Lk

It follows from this and Proposition 10.3.6 that

 

p λ g e 1 ∧ · · · ∧ g e k = bλ e 1 ∧ · · · ∧ e k , (11.3.6)
λ∈Lk λ∈Lk
λ={ 1 ,..., k } λ={ 1 ,..., k }

where bλ ∈ R and bn−k+1,...,n = 1.

Next, define a matrix h determined by the identities

e1 ∧ · · · ∧ en−1 ∧ h(ei ) = e1 ∧ · · · ∧ en−1 ∧ g(ei ), (11.3.7)

and

e1 ∧ · · · ∧ e j−1 ∧ h(ei ) ∧ e j+1 ∧ · · · ∧ en (11.3.8)

⎛ ⎞

 
= e1 ∧ · · · ∧ e j−1 ∧ h(ei ) ∧ ⎝ pλ g e 1 ∧ · · · ∧ g e n− j ⎠ ,
λ∈Ln− j

for all 1 ≤ i ≤ n, 1 ≤ j ≤ n − 1. Then h is an n × n matrix with integer coef-

ficients since each entry is a sum of minors of g multiplied by the integers pλ .
It remains to show that h ∈ S L(n, Z) and M(g) = M(h), so it suffices to show
that hg −1 ∈ Un (R).
 11.3 Plücker coordinates and the evaluation of Kloosterman sums 347

Now (11.3.7) implies that

n−1
h(ei ) = g(ei ) + a ji e j (11.3.9)
j=1

for a ji ∈ R. If we substitute this in (11.3.8) with j < n, it then follows from

(11.3.6) that


a ji e1 ∧ · · · ∧ en = bλ e1 ∧ · · · ∧ e j−i ∧ g(ei ) ∧ e 1 ∧ · · · ∧ e n− j .
λ∈Ln− j
λ={ j+1,...,n}

Define µ = (µi j ) by hg −1 = I + µ. Since (I + µ)g = h, it follows from

(11.3.9) that
a ji e1 ∧ · · · ∧ en = e1 ∧ · · · ∧ e j−i ∧ µg(ei ) ∧ e j+1 ∧ · · · ∧ en (11.3.10)
for 1 ≤ i ≤ n, 1 ≤ j ≤ n − 1. By comparing (11.3.9), (11.3.10), we obtain


n

gki µ jk = bλ gmi (±1)λ ,
k=1 λ∈Ln− j
λ={ j,...,m̂,...,n},m> j
+
where g(ei ) = g ji e j , and m̂ means that the term m is omitted. In particular,
µ jk = 0 when j ≥ k, which is what was required.

The Plücker relations are defined by a set of quadratic forms in the bottom row
based minor determinants of any n × n matrix in G L(n, R). These forms have
coefficients ±1. They must vanish if the values assigned to symbols representing
the minor determinants come from any square matrix. The number of Plücker
relations grows rapidly with n, because each j × j sub-matrix, with elements
chosen from the bottom j rows and any j columns, also gives rise to a set of
relationships of the given type. In dimension 2 there are no relationships and in
dimension 3 just one, the Cramer’s rule relationship ν1 ν23 − ν2 ν13 + ν3 ν12 = 0.
(Here νλ , where λ is an ordered subset of {1, 2, . . . , n}, is used to represent the
minor determinant based on the bottom |λ| rows and the columns indexed by
the elements of λ.)
Example 11.3.11 (G L(4) Plücker relations) I would like to thank Kevin
Broughan for this example in which ten relations for dimension n = 4 are
derived.
(i) The simplest relation is obtained by expanding the matrix using the bottom
row-based minors of size n − 1 along the bottom row. By Cramer’s rule we
obtain
(−1)1+1 ν1 ν234 + (−1)1+2 ν2 ν134 + (−1)1+3 ν3 ν124 + (−1)1+4 ν4 ν123 = 0.
348 Poincaré series and Kloosterman sums

(ii) Now let the rows of a fixed but arbitrary 4 × 4 matrix be represented by
the vectors a1 , a2 , a3 , a4 . Then because for all vectors v, v ∧ v = 0:
0 = (a3 ∧ a4 ) ∧ (a3 ∧ a4 ) = α e1 ∧ e2 ∧ e3 ∧ e4 ,
where the ei are the standard unit vectors and α is a real constant with value
ν12 ν34 − ν24 ν13 + ν14 ν23 , which is therefore 0.
(iii) The relation ν24 ν123 + ν12 ν234 − ν23 ν124 = 0 will now be derived. First
expand the wedge product of the bottom two rows:
a3 ∧ a4 = ν12 e1 ∧ e2 + ν13 e1 ∧ e3 + ν14 e1 ∧ e4
+ ν23 e2 ∧ e3 + ν24 e2 ∧ e4 + ν34 e3 ∧ e4 .
Then
a3 ∧ a4 = e2 ∧ (−ν12 e1 + ν23 e3 + ν24 e4 )
+ ν13 e1 ∧ e3 + ν14 e1 ∧ e4 + ν34 e3 ∧ e4
= e2 ∧ ω + η,
say, where ω is a 1-form and η a 2-form with e2 not appearing. Then
a2 ∧ a3 ∧ a4 = ν123 e1 ∧ e2 ∧ e3 + ν234 e2 ∧ e3 ∧ e4 + ν124 e1 ∧ e2 ∧ e4
so
a2 ∧ a3 ∧ a4 ∧ ω = (ν24 ν123 + ν12 ν234 − ν23 ν124 )e1 ∧ e2 ∧ e3 ∧ e4 + 0
= λ e1 ∧ e2 ∧ e3 ∧ e4 .
Since each term of η has two of {e1 , e3 , e4 } and each term of ω one of this
set we can write
η ∧ ω = (ν12 ν34 − ν24 ν13 + ν14 ν23 )e1 ∧ e3 ∧ e4 = 0
by the relation derived in (ii) above.
But then
a2 ∧ a3 ∧ a4 ∧ ω = a2 ∧ (a3 ∧ a4 ) ∧ ω
= a2 ∧ (e2 ∧ ω + η) ∧ ω
= a2 ∧ (η ∧ ω) = 0,
so, therefore, λ = 0, and we get ν24 ν123 + ν12 ν234 − ν23 ν124 = 0.
Three similar relationships are derived by factoring out in turn each of the
unit vectors e1 , e3 and e4 .
(iv) The four remaining relations are obtained by applying the dimension 3
relations to each of the four subsets of {1, 2, 3, 4} of column numbers of size 3.
 11.3 Plücker coordinates and the evaluation of Kloosterman sums 349


Example 11.3.12 (The w = In−1 ±1 Kloosterman  sum)
This example was
first worked out in (Friedberg, 1987). For w = In−1 ±1 , we have
⎛ ⎞
1 ∗ 0
⎜ .. .. ⎟
⎜ . .⎟
Uw := (w −1 Un (R)w) ∩ Un (R) = ⎜ ⎟.
⎝ 0⎠
1

It follows from Proposition 11.2.10 (brute force computation) that the Kloost-
erman sum is non-zero only when c1 |c2 , c2 |c3 , . . . , cn−2 |cn−1 .
Next, we show that the map M : G L(n, R) → Rn −2 arising in the proof of
2

Theorem 11.3.2, composed with projection, gives a bijection between the sets
 
γ ∈ Un (Z)\S L(n, Z) ∩ G w / w  diag(γ ) = c

and
- 

S = (νn , νn−1,n , . . . , ν2,3,...,n ) ∈ Zn−1  νn (mod c1 ), νn−1,n (mod c2 ), . . . ,
.
ν2,3,...,n (mod cn−1 ), (νn− j+1,...,n , c j /c j−1 ) = 1 for all 1 ≤ j ≤ n − 1 ,

with c0 = 1.
To see the bijection, note that the image of M composed with projection
is contained in S, since Det(γ ) = 1 implies that the relative primality condi-
tions must hold. Further, the map is one-to-one because the information in S
determines all minors of the form

Mn−k,...,

n−i,...,n
(γ ), i <k (mod ck )

by induction on k for
ck
Mn−k,...,

n−i,...,n
(γ ) = · Mn−k+1,...,

n−i,...,n
(γ ).
ck−1

Since the remaining Mλ (γ ) are determined by Proposition 10.3.6 it follows that

the map is one-to-one. As for surjectivity, it suffices to show that the point in
V arising from a point of S is actually in V2 . This follows by induction since
  ck
gcd νn−k,...,
 0≤i ≤k−1 = .
n−i,...,n
ck−1

The above bijection can be used to prove

350 Poincaré series and Kloosterman sums

Theorem 11.3.13 Let n ≥ 2 and define M = (M1 , . . . , Mn−1 ), and N =

(N1 , . . . , Nn−1 ) to be two (n − 1)–tuples of integers. Let ψ M , ψ N be two
characters of Un (R), as in (11.1.1), determined by the conditions
ψ M (u) = e2πi(M1 u 1 +···+u n−1 Mn−1 ) ,
ψ N (u) = e2πi(N1 u 1 +···+u n−1 Nn−1 ) ,

for all u ∈ Un (R). Let c be given by (11.2.1), and assume w = In−1 ±1 . Sup-
pose that c1 |c2 | · · · |cn−1 and
Mi+1 cn−i cn−i−2
Ni = 2
(11.3.14)
cn−i−1
for i = 1, 2, . . . , n − 2 with c0 = 1. Then
 

M x M x
2πi c 1 /cn−1 + c2 n−1
xn−2
+ ··· +
Mn−1 x2 x1
+
Nn−1 x1
S(ψ M , ψ N , c) = e n−1 n−2 n−2 /cn−3 c1 /c0 c1
,
xi (mod ci )
(xi , ci /ci−1 )=1

where xi x̄i ≡ 1 (mod ci /ci−1 ) for 1 ≤ i ≤ n − 1.

Remark The compatibility condition of Proposition 11.2.10 implies the
divisibility criterion c1 |c2 | · · · |cn−1 together with the identitities (11.3.14).

11.4 Properties of Kloosterman sums

We now proceed to determine some of the more important properties of the
Kloosterman sums.
Proposition 11.4.1 Let n ≥ 2 and define M = (M1 , . . . , Mn−1 ), and
N = (N1 , . . . , Nn−1 ) to be two (n − 1)–tuples of integers. Let ψ M , ψ N be two
characters of Un (R), as in (11.1.1), determined by the conditions
ψ M (u) = e2πi ( M1 u 1 +···+u n−1 Mn−1 ) , ψ N (u) = e2πi ( N1 u 1 +···+u n−1 Nn−1 ) ,
for all u ∈ Un (R). Let w be in the Weyl group of G L(n, Z), let c be given by
(11.2.1), and assume that the Kloosterman sum Sw (ψ M , ψ N , c) is well defined.
Then its value depends only on
Mi (mod ci ), and Ni (mod cn−i ),
for i = 1, 2, . . . , n − 1.
Proof It is clear that the proposition holds for the long element Kloosterman
sum. For non-zero integers c1 , . . . , cn−1 , let U (c1 , . . . , cn−1 ) denote the group
 
U (c1 , . . . , cn−1 ) = u ∈ Un (R)  ci u i ∈ Z ,
where u is as in (11.1.1). Then, by the use of Cramer’s rule, one may show that
any γ ∈ S L(n, Z) ∩ G w has Bruhat decompositions γ = b1 cwb2 = b1 cwb2
 11.4 Properties of Kloosterman sums 351

where

b2 ∈ U (cn−1 , . . . , c1 ), b1 ∈ U (c1 , . . . , cn−1 ).

By choosing the appropriate one of the above decompositions when evaluating

the Kloosterman sum, the result follows.

Proposition 11.4.2 (Multiplicativity condition) Fix n ≥ 2, and let ψ M , ψ N

be characters of Un (R) as in Proposition 11.4.1. Let
⎛ ⎞
1/cn−1
⎜ cn−1 /cn−2 ⎟
⎜ ⎟
⎜ .. ⎟
c=⎜ . ⎟,
⎜ ⎟
⎝ c2 /c1 ⎠
c1
⎛  ⎞
1/cn−1
⎜ 
cn−1 
/cn−2 ⎟
⎜ ⎟
 ⎜ .. ⎟
c =⎜ . ⎟
⎜ ⎟
⎝ c2 /c1 ⎠
c1
with positive integers c1 , c1 , . . . , cn−1 , cn−1

. Let w be in the Weyl group of
G L(n, R). Then there exist characters ψ N  , ψ N  of Un (R) such that

Sw (ψ M , ψ N , cc ) = Sw (ψ M , ψ N  , c) · Sw (ψ M , ψ N  , c ).

Proof This was first proved for n = 3 in (Bump, Friedberg and Goldfeld,
1988). See (Friedberg, 1987), (Stevens, 1987) for the proof.

It was shown in (Friedberg, 1987) that the Kloosterman sums Sw (ψ M , ψ N , c)

are non-zero only if w is of the form
⎛ ⎞
Ii 1
⎜ Ii 2 ⎟
⎜ ⎟
w=⎜ . ⎟
⎝ .. ⎠
Ii
where the Ik are k × k identity matrices and i 1 + · · · + i = n. It was further
shown in (Friedberg, 1987) that if wn is the long element and, in addition,
c1 , . . . , cn−1 are pairwise coprime, then the long element Kloosterman sum
Swn (ψ M , ψ N , c) factors into G L(2) Kloosterman sums.
In the special case that c1 , c2 , . . . , cn−1 are all suitable powers of a fixed
prime p, then the Kloosterman sum Sw (ψ M , ψ N , c), with c given by (11.2.1),
will be associated to an algebraic variety defined over the finite field F p of p
352 Poincaré series and Kloosterman sums

elements. By use of a deep theorem of Deligne (1977), on bounds for hyper

Kloosterman sums of type
 

x +···+x
2πi 1 p n
e ,
x1 ···xn =1
xi ∈F p ,(i=1,...,n)

it is proven in (Friedberg, 1987) (see also Theorem 11.3.13) that for

 
1
w= ,
In−1
then for all 1 ≤ j ≤ n − 1,
⎛ ⎞
⎜ ((n−1)2 /2 j)+  (n− j−1)(n− j−2)/2 ⎟
|Sw (ψ M , ψ N , c)| = O ⎜
⎝c j p ⎟.
⎠
p|cn−1
p | c j

11.5 Fourier expansion of Poincaré series

Let n ≥ 2, z ∈ hn , s ∈ Cn−1 , and, ψ a character of Un (R) as in (11.1.1). Con-
sider an e-function, eψ : hn → C, which is a bounded function characterized
by the property that
eψ (uz) = ψ(u)eψ (z)
for all u ∈ Un (R), and z ∈ h . n

Let


P(z, s, eψ ) := Is (γ z)eψ (γ z)
γ ∈Un (Z)\
be a Poincaré series for S L(n, Z) as in Definition 11.1.2. The main goal of this
section is an explicit computation of the N = (N1 , . . . , Nn−1 ) ∈ Zn−1 Fourier
coefficient 
P(uz, s, eψ )ψ N (u) d ∗ u, (11.5.1)
Un (Z)\Un (R)

with
⎛ ⎞
1 u n−1
⎜ 1 u n−2 u i, j ⎟
⎜ ⎟
⎜ .. .. ⎟
u=⎜ . . ⎟ ∈ Un (R),
⎜ ⎟
⎝ 1 u1 ⎠
1
)
d ∗u = du i, j , and ψ N (u) = e2πi(u 1 N1 +···+u n−1 Nn−1 ) .
1≤i< j≤n
The computation of (11.5.1) requires some additional notation.
 11.5 Fourier expansion of Poincaré series 353

Definition 11.5.2 Fix n ≥ 2. We define the group Vn of diagonal matrices:

⎧ ⎛ ⎞  ⎫
⎪  ⎪
⎨ v1  n ⎬

Vn := v = ⎝ ..
. ⎠ v
 i = ±1∀1 ≤ i ≤ n, Det(v) = vi = 1 .
⎪
⎩  ⎪
⎭
vn  i=1

Definition 11.5.3 Fix n ≥ 2. Let ψ : Un (R) → C be a character of Un (R) as

in (11.1.1). Then for any v ∈ Vn , we define a new character ψ v : Un (R) → C
by the condition

ψ v (u) := ψ(v −1 uv) = ψ(vuv),

for all u ∈ Un (R).

We are now ready to state and prove the main theorem of this section which
was first proved in (Bump, Friedberg and Goldfeld, 1988) for n = 3, and then
more generally in (Friedberg, 1987).

Theorem 11.5.4 (Fourier coefficient of Poincaré series) Fix n ≥ 2, and let

M = (M1 , . . . , Mn−1 ), N = (N1 , . . . , Nn−1 ) ∈ Zn−1 . Then, for ψ = ψ M , and
(si ) sufficiently large (i = 1, . . . , n − 1), the Fourier coefficient (11.5.1) is
given by
 




∞

∞
Sw ψ M , ψ Nv , c Jw z; ψ M , ψ Nv , c
··· ns ,
w∈Wn v∈Vn c1 =1 cn−1 =1
c1ns1 · · · cn−1
n−1

where Sw (ψ M , ψ Nv , c) is the Kloosterman sum as in Definition 11.2.2, and



Jw z; ψ M , ψ Nv , c = Is (wuz) eψ M (wcuz) ψ Nv (u) d ∗ u,
Ūw (R)

with c given by (11.2.1) and Ūw (R) = (w −1 Un (R)w) ∩ Un (R) as in (10.6.4).

Proof It follows from (11.5.1) and the definition of the Poincaré series that
 

P(uz, s, eψ M )ψ N (u)d ∗ u = Is (γ uz)eψ M (γ uz)ψ N (u)d ∗ u.
γ ∈Un (Z)\
Un (Z)\Un (R) Un (Z)\Un (R)

The sum over γ ∈ Un (Z)\ on the right above may be rewritten using the
Bruhat decomposition (see Section 11.2) after taking an additional quotient by
354 Poincaré series and Kloosterman sums

Vn as in Definition 11.5.2. In this manner, one obtains

= .
γ ∈Un (Z)\ w∈Wn v∈Vn γ ∈Un (Z)\(∩G w )/Vn

Choose a set of representatives Rw for Un (Z)\( ∩ G w )/Vn w with


w = w−1 · t Un (Z) · w ∩ Un (Z),

as before. By Proposition 10.3.6, for every γ ∈ Rw , we may choose a Bruhat

decomposition

γ = b1 cwb2 , b1 , b2 ∈ Un (Q),

and c as in (11.2.1). In an anologous manner, we may rewrite

  
= ,
Un (Z)\Un (R) Uw (Z)\Uw (R) Ūw (Z)\Ūw (R)

with Uw , Ū w as in (10.6.4), (10.6.5). The result follows after several computa-

tions such as those preceding and including (10.6.6).

11.6 Kuznetsov’s trace formula for SL(n, Z)

Kuznetsov’s trace formula for S L(2, Z) was established by Kuznetsov and first
published by him in (Kuznecov, 1980). The Kuznetsov trace formula was also
written up by Bruggeman (1978). Later on, it was systematically developed and
heavily used by Iwaniec and his collaborators to:
r extend Bombieri’s mean value theorem on primes to large moduli (see
(Bombieri, Friedlander and Iwaniec, 1987));
r achieve subconvexity bounds of L functions (see (Iwaniec and Kowalski,
2004) and the references there, for example).

Another spectacular application was due to Motohashi (1997) who proved

the asymptotic formula for the fourth moment of the Riemann zeta function
by making use of Kuznetsov’s formula on S L(2, Z). In this section, we will
generalize this valuable tool to the case of S L(n, Z) with n ≥ 2.
The proof of the Kuznetsov trace formula which we give here is a generaliza-
tion of Zagier (unpublished notes) on Kuznetsov’s trace formula on S L(2, Z).
It was worked out by Xiaoqing Li, and I would like to thank her for allowing
me to include it in this section.
 11.6 Kuznetsov’s trace formula for SL(n, Z) 355

Throughout Section 11.6, we identify, for n  2, the generalized Siegel upper

half-plane

hn = S L(n, R)/S O(n, R).

Warning Note that this does not conform to the definition of hn as given in
Definition 1.2.3, although it is a very close approximation. We shall only adopt
this notation temporarily in this section because of the particular applications
in mind, and the fact that the Selberg transform may take a very simple form in
some cases.

For z, z  ∈ hn and every k ∈ Cc∞ (S O(n, R)\S L(n, R)/S O(n, R)), a space
which contains infinitely differentiable compactly supported S O(n, R) bi-
invariant functions on S L(n, R), we define an automorphic kernel


K (z, z  ) = k(z −1 γ z  ). (11.6.1)
γ ∈S L(n,Z)

Clearly

K (z, z  ) ∈ L2 (S L(n, Z)\hn ). (11.6.2)

For M = (M1 , . . . , Mn−1 ), N = (N1 , . . . , Nn−1 ) ∈ Zn−1 , where none of Mi

and Ni , 1  i  n − 1 are zero, we call them non-degenerate, we may com-
pute the (M, N )th Fourier coefficients of K (z, z  ) in two ways. One way is
to use the Bruhat decomposition of G L(n, R), and the other is to use the
spectral decomposition of L2 (S L(n, Z)\hn ). After these two computations are
made, we end up with an identity which is termed the pre-Kuznetsov trace
formula.
To start, define


PM (y, z ) := K (x y, z  )ψ M (x) d ∗ x. (11.6.3)
Un (Z)\Un (R)

Here
⎛ ⎞
1 x1,2 x1,3 ... x1,n
⎜ 1 x2,3 ... x2,n ⎟
⎜ ⎟
⎜ .. .. ⎟
x =⎜ . . ⎟, xi := xn−i,n−i+1 (i = 1, . . . , n − 1),
⎜ ⎟
⎝ 1 xn−1,n ⎠
1
356 Poincaré series and Kloosterman sums

⎛ ⎞
y1 y2 . . . yn−1
⎜ y1 y2 . . . yn−2 ⎟
⎜ ⎟  −1/n
⎜ .. ⎟
y=⎜ . ⎟ · y1n−1 y2n−2 . . . yn−2
2
yn−1 ,
⎜ ⎟
⎝ y1 ⎠
1
(11.6.4)

ψ M (x) = e2πi(M1 x1 +···Mn−1 xn−1 ) , (11.6.5)

and

d∗x = d xi, j . (11.6.6)
1≤i< j≤n

Lemma 11.6.7 The function PM (y, z  ) is a Poincaré series in z  .

Proof We rewrite PM (y, z  ) as


PM (y, z ) = K (x y, z  )ψ M (x) d ∗ x
Un (Z)\Un (R)


= k(z −1 γ z  )ψ M (x) d ∗ x.
γ ∈S L(n,Z)
Un (Z)\Un (R)

Define

eψ M (y, z  ) := k(z −1 z  )ψ M (x) d ∗ x. (11.6.8)
Un (Z)\Un (R)

Then for u ∈ Un (R), we have


eψ M (y, uz  ) = k(z −1 uz  )ψ M (x) d ∗ x
Un (Z)\Un (R)

= k(z −1 z  )ψ M (ux) d ∗ x
Un (Z)\Un (R)

= ψ M (u) k(z −1 z  )ψ M (x) d ∗ x
Un (Z)\Un (R)

= ψ M (u)eψ M (y, z  ),
so that eψ M (y, z  ) is an e-function. It is also clear that PM (y, z  ) ∈
L2 (S L(n, Z)\hn ) as a function of z  . Hence, PM (y, z  ) is a Poincaré series
as in Definition 11.1.2 with s = (0, 0, . . . , 0).

 11.6 Kuznetsov’s trace formula for SL(n, Z) 357

It follows from Theorem 11.5.4 and Lemma 11.6.7 that


PM (y, x  y  )ψ N (x  ) d ∗ x  (11.6.9)
Un (Z)\Un (R)




∞

∞
 
= ··· Sw ψ M , ψ Nv ; c Jw y, y  ; ψ M , ψ Nv , c ,
w∈Wn v∈Vn c1 =1 cn−1 =1

where, after applying (11.6.8),



Jw y, y ; ψ M , ψ N , c = eψ M (y, wcuy  )ψ Nv (u) d ∗ u
 v

Ūw
 
= k(z −1 · wcuy  ) ψ M (x) ψ Nv (u) d ∗ x d ∗ u,
Ūw Un (Z)\Un (R)
(11.6.10)
and Sw (ψ M , ψ Nv ; c) is the Kloosterman sum as in Definition 11.2.2.

The other way of computing the Fourier coefficient (11.6.9) is to make use
of the Langlands spectral decomposition (Langlands, 1966, 1976) (see also
Theorem 10.13.1). Langlands spectral decomposition states that
L2 (S L(n, Z)\hn ) = L2discrete (S L(n, Z)\hn ) ⊕ L2cont (S L(n, Z)\hn )
where
L2discrete (S L(n, Z)\hn ) = L2cusp (S L(n, Z)\hn ) ⊕ L2residue (S L(n, Z)\hn ).
Here L2cusp denotes that the space of Maass forms, L2residue consists of iterated
residues of Eisenstein series twisted by Maass forms, and L2cont is the space
spanned by integrals of Langlands Eisenstein series.
For each k ∈ Cc∞ (S O(n, R)\S L(n, R)/S O(n, R)), we can define an integral
operator L k which acts on f ∈ L2 (S L(n, Z)\hn ), by

(L k f )(z) := f (w)k(z −1 w) d ∗ w
hn


= f (γ w)k(z −1 γ w) d ∗ w
γ ∈
S L(n,Z)\hn

= f (w)K (z, w) d ∗ w.
S L(n,Z)\hn

Then L k commutes with all the invariant differential operators. Furthermore,

we have
358 Poincaré series and Kloosterman sums

Lemma 11.6.11 (Selberg) Fix n ≥ 2. Let φ be a Maass form of type ν ∈ Cn−1

for S L(n, Z) as in Definition 5.1.3. Then

(L k φ)(z) = k̂(ν)φ(z),

where k̂(ν) depends only on k and ν. More precisely,


k̂(ν) = (L k Iν )(I ) = k(w)Iν (w)dw,
hn

where I is the n × n identity matrix and Iν is given in Definition 2.4.1.

Usually, k̂(ν), as in Lemma 11.6.11, is called the Selberg transform of k and


k̄(y) = k(x y) d ∗ x
Un (R)

is called the Harish transform of k. One can think of the Selberg transform
as the Mellin transform of its Harish transform. The Selberg transform has an
inversion formula due to Harish-Chandra, Bhanu-Murthy, Gangolli, etc., see
(Terras, 1988) and the references there.
For simplicity, we introduce the parameters ak for 1 ≤ k ≤ n − 1, which are
defined by

j(n − j)
ak

n− j n−1
+ = b ji νi
2 k=1
2 i=1

for any 1  j  n − 1, where b ji , νi are defined in (5.1.1).

It is easy to check that
3 +j
n− 4

n−1 ( j(n− j)/2)+ ak /2
Iν (z) = Ia (z) := yj k=1
.
j=1

For z ∈ G L(n, R),


h a (z) := pa (k t zk) dk,
S O(n,R)

with

n−1
|z i | 2 (ān−i /2)−(ān−i+1 /2)−1) ,
1
pa (z) =
i=1
 11.6 Kuznetsov’s trace formula for SL(n, Z) 359

where z i is the i × i upper left-hand corner in z for 1  i  n − 1, and where

we normalize the measure dk such that

dk = 1.
S O(n,R)

Proposition 11.6.12 (Inversion formula) For k ∈ L2 (S O(n, R)\hn ), we

have
 
k(w)Ia (w)h a (z t z)
k(z) = wn dwda1 . . . dan−1
|cn (a)|2
(ai )=0 hn
1
i
n−1

k̂(a)h a (z t z)
= wn da1 . . . dan−1 ,
|cn (a)|2
(ai )=0
1
i
n−1

where wn is some constant depending on n, and

  
π π
|cn (a)|−2 = |ai − a j+1 | · tanh |ai − a j+1 | ,
1
i
j
n−1
4 4

with


n
ai = 0.
i=1

Proof See page 88 in (Terras, 1988). Our model hn can be identified with her
model SP n of n × n positive definite matrices of determinant 1 by considering
the map: hn → SP n : z → z t z.

Now, we continue with the derivation of Kuznetsov’s trace formula. Let

φ j , j  1 be an orthonormal basis for the cuspidal spectrum and  j , j  1 be
an orthonormal basis for the residual spectrum. For k ∈ Cc∞ (S O(n, R)\hn ) we
have the following spectral expansion of the kernel function K (z, z  ):



K (z, z  ) = k̂(ν j )φ j (z  )φ j (z) + k̂( j ) j (z  ) j (z)
j1 j1


+ cE k̂(s)E(z  , s)E(z, s) ds,
Eisenstein series E

where c E are constants depending on E.

360 Poincaré series and Kloosterman sums

Next, we compute the Fourier coefficients of the Poincaré series PM (y, z  )

by the above spectral expansion of K . It follows, as in the proof of Lemma 9.1.3
that

PM (y, x  y  )ψ N (x  ) d ∗ x 
Un (Z)\Un (R)

A j (N ) A j (M)
= k̂(ν j ) WJacquet (N ∗ y  , ν j )WJacquet (M ∗ y, ν j )
)
n−1
j1 |Ni |i(n−i)/2 |M j | j(n− j)/2
i, j = 1


ε(N , s) ε(M, s)
+ cE k̂(s) n−1
) )
n−1
Eisenstein series E
|Ni |i(n−i)/2 |Mi |i(n−i)/2
i=1 i=1

× WJacquet (N ∗ y  , s)WJacquet (M ∗ y, s) ds, (11.6.13)

where for each φ j in the cuspidal spectrum, every Eisenstein series E in the
continuous spectrum, and all N = (N1 , . . . , Nn−1 ) ∈ Zn−1 with N1 · · · Nn−1 =
0, we have
⎛ ⎞
N1 · · · |Nn−1 |
⎜ .. ⎟
⎜ . ⎟
∗ ⎜ ⎟
N := ⎜ N1 N2 ⎟,
⎜ ⎟
⎝ N1 ⎠
1

A j (N )
WJacquet (N ∗ y  , ν j ) = φ j (x y  )ψ N (x) d ∗ x,
)
n−1
|Ni |i(n−i)/2 Un (Z)\Un (R)
i=1

and, correspondingly,

ε(N , s) ∗ 
WJacquet (N y , s) = E(x y  , s)ψ N (x) d ∗ x.
)
n−1
|Ni |i(n−i)/2 Un (Z)\Un (R)
i=1

The contribution from the residual spectrum is 0 due to the lack of non-
degenerate Whittaker models for the residual spectrum.

Remark To simplify notation, we have supressed the character ψ1,...,1, Nn−1

|Nn−1 |
in the Jacquet Whittaker functions above.

By comparing (11.6.9) and (11.6.13), we immediately obtain the following

pre-Kuznetsov trace formula.
 11.6 Kuznetsov’s trace formula for SL(n, Z) 361

Proposition 11.6.14 (Pre-Kuznetsov trace formula) Fix n ≥ 2, and let

k ∈ Cc∞ (S O(n, R)\hn ), M = (M1 , . . . , Mn−1 ) ∈ Zn−1 , N = (N1 , . . . , Nn−1 )
∈ Zn−1 be nondegenerate. Then

A j (N ) A j (M)
k̂(ν j ) WJacquet (N ∗ y  , ν j )WJacquet (M ∗ y, ν j )
)
n−1
j1 |Ni |i(n−i)/2 |M j | j(n− j)/2
i, j = 1


ε(N , s) ε(M, s)
+ cE k̂(s) n−1
) )
n−1
Eisenstein series E
|Ni |i(n−i)/2 |Mi |i(n−i)/2
i=1 i=1

× WJacquet (N ∗ y  , s)WJacquet (M ∗ y, s) ds




∞

∞
 
= ··· Sw ψ M , ψ Nv ; c Jw y, y  ; ψ M , ψ Nv , c .
w∈Wn v∈Vn c1 =1 cn−1 =1

It is useful to be able to remove Jacquet’s Whittaker functions in the above

formula. This can be done by employing Stade’s formula (see (Stade, 2002))
for the Mellin transform of products of Whittaker functions on G L(n). In order
to present Stade’s formula succinctly, we introduce parameters αk and βk for
1  k  n − 1, which are defined by

j(n − j)


n− j n−1
+ αk = b ji νi (11.6.15)
2 k=1 i=1

for any 1  j  n − 1, where b ji , νi are defined in (5.1.1). We also set

n−1
αn = − αi . (11.6.16)
i=1

Similar relations hold for βk and νi .

Proposition 11.6.17 (Stade’s formula) For (s)  1 we have

 
n−1
dy j
WJacquet (y, ν) WJacquet (y, ν  )
(n− j)(s− j)
2 π (n− j)s y j
j=1
yj
(R+ )n−1
 ns −1 
n   
n
s + α j + βk
=  ,
2 j=1 k=1
2

where α j and νi are related by (11.6.15), (11.6.16), and similarly for β j

and νi .
362 Poincaré series and Kloosterman sums

Remark Stade proves the above formula for (s) large, but one may obtain
|α j | < 12 by Proposition 12.1.9, and similarly for βk , so the right-hand side of
the formula is analytic for s  1. On the other hand Jacquet (2004a) shows
the integral on the left is absolutely convergent for (s)  1, so by analytic
continuation Stade’s formula holds for s  1.

Definition 11.6.18 Fix n ≥ 2, N = (N1 , . . . , Nn−1 ) ∈ Zn−1 . For j = 1,

2, . . . , let
n   
A j (N ) n
1 + α j + αk
ρ j (N ) =  ,
)
n−1 2
 (n/2) |Ni |i(n−i)/2 j=1 k=1
i=1
n   
εs (N ) n
1 + α j + αk
ηs (N ) = 
)
n−1 2
 (n/2) |Ni |i(n−i)/2 j=1 k=1
i=1

be normalized N th Fourier coefficients of S L(n, Z) Maass cusp forms and

Eisenstein series, respectively.

We are now ready to state and prove the Kuznetsov trace formula.

Theorem 11.6.19 (Kuznetsov trace formula) Fix n ≥ 2, and choose

k ∈ Cc∞ (S O(n, R)\hn ), M = (M1 , . . . , Mn−1 ) ∈ Zn−1 , N = (N1 , . . . , Nn−1 )
∈ Zn−1 satisfying M1 N1 · · · Mn−1 Nn−1 = 0. Then



k̂(ν j )ρ j (N )ρ̄ j (M) + c E k̂(s)ηs (N )η̄s (M) ds
j1 Eisenstein series E




∞

∞
 
= ··· Sw ψ M , ψ Nv ; c Hw ψ Nv , c ,
w∈Wn v∈Vn c1 =1 cn−1 =1

where k̂ is given in Lemma 11.6.11, Wn is the Weyl group of G L(n, R), Vn

is as in Definition 11.5.2, ψ Nv is as in Definition 11.5.3, Sw (ψ M , ψ Nv ; c) is the
Kloosterman sum as in Definition 11.2.2, c E is as in (11.6.13), and ρ j , ηs are
given in Definition 11.6.18.
Furthermore,
  
 v  −1 −1
Hw ψ N , c = k tM x wcut N ψ M (x) ψ Nv (u) d ∗ u d ∗ x
(R+ )n−1 Un (Z)\Un (R) Ūw


n−1
(n− j)(1− j) dt j
× 2 π (n− j) t j
j=1
tj
 11.6 Kuznetsov’s trace formula for SL(n, Z) 363

where t M = t(M ∗ )−1 , t N = t(N ∗ )−1 , and

⎛ ⎞ ⎛ ⎞
M1 · · · |Mn−1 | N1 · · · |Nn−1 |
⎜ . ⎟ ⎜ .. ⎟
⎜ . . ⎟ ⎜ . ⎟
M ∗ := ⎜ ⎟, N ∗ := ⎜ ⎟,
⎝ M1 ⎠ ⎝ N1 ⎠
1 1

⎛ ⎞
t1 · · · tn−1
⎜ .. ⎟
⎜ . ⎟
t := ⎜ ⎟.
⎝ t1 ⎠
1

Proof In the pre-Kuznetsov trace formula, given in Proposition 11.6.14,

choose y, y  so that M ∗ y = N ∗ y  = t. This may be accomplished by
setting

ti = Mi yi = Ni yi , (i = 1, 2, . . . , n − 1).

Next, multiply both sides of the pre-Kuznetsov trace formula by


n−1
(n− j)(1− j) − j(n− j) dt j
δn π n− j t j 2
j=1
tj

and then integrate over (R+ )n−1 . Theorem 11.6.19 immediately follows from
Proposition 11.6.17.

Concluding Remarks
(i) The test function k can be extended to a larger space, say, Harish-
Chandra’s Schwartz space, as long as the convergence of both sides is not a
problem.
(ii) The nice feature of the above Kuznetsov type formula is that the residual
spectrum does not appear, while its appearance is inevitable in the Selberg–
Arthur trace formula. In this sense, the Kuznetsov formula is more handy
to treat the cuspidal spectrum.
(iii) In the case of G L(2), Kuznetsov also derived an inversion formula so that
one can put any good test function on the Kloosterman sum side. Such an
inversion formula on G L(n) does not exist yet, although it is certainly a
very interesting research problem.
364 Poincaré series and Kloosterman sums

GL(n)pack functions The following GL(n)pack functions, described in the

appendix, relate to the material in this chapter:
KloostermanBruhatCell KloostermanCompatibility KloostermanSum
PluckerCoordinates PluckerInverse PluckerRelations
Whittaker.
 12
Rankin–Selberg convolutions

The Rankin–Selberg convolution for L-functions associated to automorphic

forms on G L(2) was independently discovered by Rankin (1939) and Selberg
(1940). The method was discussed in Section 7.2 in connection with the
Gelbart–Jacquet lift and a generalization to G L(3) was given in Section 7.4. A
much more general interpretation of the original Rankin–Selberg convolution
for G L(2) × G L(2) in the framework of adeles and automorphic representa-
tions was first obtained in (Jacquet, 1972). The theory was subsequently further
generalized in (Jacquet and Shalika, 1981).
The Rankin–Selberg convolution for the case G L(n) × G L(n  ),
(1 ≤ n < n  ), requires new ideas. Note that this includes G L(1) × G L(n  )
which is essentially the Godement–Jacquet L-function whose holomorphic
continuation and functional equation was first obtained by Godement and
Jacquet (1972). A sketch of the general Rankin–Selberg convolution for
G L(n) × G L(n  ) in classical language was given in (Jacquet, 1981). The
theory was further extended in the context of automorphic representations
in (Jacquet, Piatetski-Shapiro and Shalika, 1983). In this chapter, we shall
present an elementary and self contained account of both the meromorphic
continuation and functional equation of Rankin–Selberg L-functions associated
to G L(n) × G L(n  ). In particular, this will give the meromorphic continuation
and functional equation of the Godement–Jacquet L-function.
The fact that Rankin–Selberg L-functions have Euler products is a con-
sequence of the uniqueness of Whittaker functions for local fields. The non-
Archimedean theory was worked out in (Jacquet, Piatetski-Shapiro and Shalika,
1983) while the Archimedean local factors were computed in (Jacquet and
Shalika, 1981, 1990).
Explicit computations of the local factors in the G L(n) × G L(n  ) Rankin–
Selberg convolution were obtained by Bump (1987). Bump started with a

365
366 Rankin–Selberg convolutions

theorem of Shintani (1976) which relates the Fourier coefficients of Maass forms
with Schur polynomials as was discussed in Section 7.4, and ties everything
together with an old identity of Cauchy. Such formulae were originally con-
jectured by Langlands (1970) and generalizations were found by Kato (1978)
and Casselman and Shalika (1980). We shall follow this approach as in (Bump,
1984, 1987, 2004). Alternatively, a basic reference for an adelic treatment of the
Rankin–Selberg theory is Cogdell’s, analytic theory of L-functions for G L n , in
(Bernstein and Gelbart, 2003).
The Rankin–Selberg convolution is one of the most important constructions
in the theory of L-functions. Naturally it has had inumerable generalizations.
The excellent survey paper of Bump (to appear) gives a panoramic overview
of the entire subject. We give applications of the Rankin–Selberg convolution
method towards the generalized Ramanujan and Selberg conjectures. In par-
ticular, the chapter concludes with the theorem of Luo, Rudnick and Sarnak
(1995, 1999) which has been used to obtain the current best bounds for the
Ramanujan and Selberg conjectures.

12.1 The GL(n) × GL(n) convolution

Fix an integer n ≥ 2. Let f, g be Maass forms for S L(n, Z) of type ν f ,
νg ∈ Cn−1 , respectively, with Fourier expansions:

∞

∞
A(m 1 , . . . , m n−1 )
f (z) = ···
γ ∈Un−1 (Z)\S L(n−1,Z) m 1 =1 m n−2 =1 m n−1 =0
)
n−1
|m k |k(n−k)/2
k=1

   
γ
× WJacquet M · z, ν f , ψ1,...,1,m n−1 /|m n−1 | ,
1

∞

∞
B(m 1 , . . . , m n−1 )
g(z) = ···
γ ∈Un−1 (Z)\S L(n−1,Z) m 1 =1 m n−2 =1 m n−1 =0
)
n−1
|m k |k(n−k)/2
k=1

   
γ
× WJacquet M · z, νg , ψ1,...,1, m n−1 /|m n−1 | ,
1
(12.1.1)

as in (9.1.2).
 12.1 The GL(n) × GL(n) convolution 367

Definition 12.1.2 (Rankin–Selberg L-function) For n ≥ 2, let f, g be two

Maass forms for S L(n, Z) as in (12.1.1). Let s ∈ C. Then the Rankin–Selberg
L-function, denoted L f ×g (s), is defined by


∞
∞
A(m 1 , . . . , m n−1 ) · B(m 1 , . . . , m n−1 )
L f ×g (s) = ζ (ns) ···  n−1 n−2 s ,
m 1 =1 m n−1 =1 m 1 m 2 · · · m n−1
which converges absolutely provided (s) is sufficiently large.
This definition conforms with the notation of Theorem 7.4.9.
Theorem 12.1.3 (Euler product) Fix n ≥ 2. Let f, g be two Maass forms for
S L(n, Z) as in (12.1.1) with Euler products

∞
A(m, 1, . . . , 1)  n
L f (s) = s
= (1 − α p,i p −s )−1 ,
m=1
m p i=1

∞
B(m, 1, . . . , 1)  n
L g (s) = s
= (1 − β p,i p −s )−1 ,
m=1
m p i=1

then L f ×g will have an Euler product of the form:


n 
n
L f ×g (s) = (1 − α p,i β p, j p −s )−1 .
p i=1 j=1

Proof The proof follows as in Section 7.4. The key point is that the identity
(7.4.14) generalizes. In this case, we have

A p k1 , p k2 , . . . , p kn−1 = Sk1 ,...,kn−1 (α p,1 , α p,2 , . . . , α p,n ),
and Theorem 12.1.3 is a consequence of Cauchy’s identity, Proposition 7.4.20.

Our main result is the following generalization of Theorems 7.2.4 and 7.4.9.
Theorem 12.1.4 (Functional equation) For n ≥ 2, let f, g be two Maass
forms of types ν f , νg for S L(n, Z), as in (12.1.1), whose associated L-functions
L f , L g , satisfy the functional equations:
  
n −s+λi (ν f ) s − λi (ν f )
 f (s) := π 2  L f (s) =  f˜ (1 − s)
i=1
2
  
n −s+λ j (νg ) s − λ j (νg )
g (s) := π 2  L g (s) = g̃ (1 − s),
j=1
2

as in Theorem 10.8.6 and Remark 10.8.7 where f˜, g̃, are the dual Maass forms
(see Section 9.2). Then the Rankin–Selberg L-function, L f ×g (s) (see Defini-
tion 12.1.2) has a meromorphic continuation to all s ∈ C with at most a simple
368 Rankin–Selberg convolutions

pole at s = 1 with residue proportional to  f, g, the Petersson inner product

of f with g.
Furthermore, L f ×g (s) satisfies the functional equation


n  n −s+λi (ν f )+λ j (νg ) s − λi (ν f ) − λ j (νg )
 f ×g (s) := π 2  L f ×g (s)
i=1 j=1
2
=  f˜×g̃ (1 − s).

Remark It follows from (10.8.5) and Remark 10.8.7 that the powers of π
occurring in the above functional equations take the much simpler form:

n
−s+λi (ν) 
n 
n −s+λi (ν f )+λ j (νg )
= π −ns/2 , = π −n
2
π 2 π 2 s/2
.
i=1 i=1 j=1

Proof This Rankin–Selberg convolution requires the maximal parabolic

Eisenstein series E P (z, s), with P = Pn−1,1 , as defined in (10.7.1). The proof
of Theorem 12.1.4 is a generalization of the proof of Theorem 7.4.9. The
idea is to compute the inner product of f · ḡ with E P (z, s̄). It follows that
for  = S L(n, Z),

 f · ḡ, E P (∗, s̄) = f (z)g(z) · E P (z, s̄) d ∗ z,
\hn

with d ∗ z, the invariant measure as given in Proposition 1.5.3. If we now apply

the unfolding trick, we obtain

 f · ḡ, E P (∗, s̄) = f (z)g(z) · Det(z)s d ∗ z,
P\hn

Now
⎛ ⎞
∗ ··· ∗ ∗
⎜ .. .. .. ⎟
⎜ ··· .⎟
P = ⎜. . ⎟ ⊂ ,
⎝∗ ··· ∗ ∗⎠
0 ··· 0 1
is generated by matrices of type
⎛ ⎞ ⎛ ⎞
0 0 ··· 0 r1
⎜ .. ⎟ ⎜ .. .. .. ⎟
⎜ .⎟ ⎜.
⎜ mn−1 ⎟, ⎜ ··· . . ⎟
⎟,
⎜ ⎟ ⎝0
⎝ 0⎠ ··· 0 rn−1 ⎠
0 ··· 0 1 0 ··· 0 1
 12.1 The GL(n) × GL(n) convolution 369

where mn−1 ⊂ S L(n − 1, Z). Consequently

(  
γ
· P\hn ∼
= Un (Z)\hn , (12.1.5)
γ ∈U (Z)\S L(n−1,Z)
1
n−1

a result obtained earlier in Lemma 5.3.13. Since the Fourier expansion of f

over S L(n − 1, Z), and Det(z) is invariant
given in (12.1.1) is written asa sum 
γ
under left multiplication by with γ ∈ S L(n − 1, Z), we may unfold
1
further using (12.1.5) to obtain

 f · ḡ, E P (∗, s̄)


∞

∞
A(m 1 , . . . , m n−1 )
= ···
m 1 =1 m n−2 =1 m n−1 =0
)
n−1
Un (Z)\hn |m k |k(n−k)/2
k=1
 2πi (m 1 x1 +···+m n−1 xn−1 )
× WJacquet M y, ν f , ψ1,...,1 e · ḡ(z) Det(z)s d ∗ z
∞ ∞
∞
∞
A(m 1 , . . . , m n−1 )B(m 1 , . . . , m n−1 )
= ··· ···
m 1 =1 m n−2 =1 m n−1 =0
)
n−1
0 0 |m k |k(n−k)
k=1

× WJacquet (M y, ν f , ψ1,...,1 ) · W̄Jacquet (M y, νg , ψ1,...,1 ) Det(y)s d ∗ y.

Recall from Proposition 1.5.3 that


n−1
dyk
d∗ y = yk−k(n−k) .
k=1
yk

It follows that

ζ (ns) f · ḡ, E P (∗, s̄)

∞ ∞
= L f ×g (s) · · · WJacquet (y, ν f , ψ1,...,1 )W̄Jacquet (y, νg , ψ1,...,1 )Det(y)s d ∗ y
0 0

= L f ×g (s)G ν f ,νg (s),

say.
The meromorphic continuation and functional equation of L f ×g (s) now
follows from the meromorphic continuation and functional equation of
E P (z, s), (given in Proposition 10.7.5) provided we know the meromorphic
continuation and functional equation of G ν f ,νg (s), the Mellin transform of the
product of Whittaker functions. This latter functional equation was obtained in
a very explicit form by Stade (2001) which allows one to prove the functional
equation stated in Theorem 12.1.4.
370 Rankin–Selberg convolutions

A much simpler way to obtain the explicit functional equation of L f ×g (s)

is by using Remark 10.8.7. In this approach, the functional equation of the
minimal parabolic Eisenstein series is used as a template. By formally taking
the Rankin-Selberg convolution of the minimal parabolic Eisensteins series
E ν f (s) and E νg (s), one readily sees that

n 
n

L Eν f ×Eνg (s) = ζ (ns) ζ s − λi (ν f ) − λ j (νg ) .
i=1 j=1

The correct form of the functional equation can immediately be seen. Of course,
one needs to deal with the difficult convergence problems when taking Rankin–
Selberg convolutions which are not of rapid decay, but the techniques for dealing
with this are now well known (Zagier, 1982), (Liemann, 1993).

One of the original applications of the Rankin-Selberg method (see (Rankin,

1939), (Selberg, 1940)) was to obtain strong bounds for the Fourier coefficients
of automorphic forms. If A(m 1 , . . . , m n−1 ) denotes the Fourier coefficient of
a Maass form on S L(n, Z) as in (12.1.1), then we have already obtained the
bound

n−1
|A(m 1 , . . . , m n−1 )|  |m k |k(n−k)/2
k=1

in Lemma 9.1.3. For example, this bound gives

A(m, 1, . . . , 1)  m (n−1)/2 ,

which unfortunately grows exponentially with n. By using the Rankin–Selberg

convolution for S L(n, Z) we can eliminate the dependence on n in the exponent.
The following bound was first obtained in a much more general setting in
(Jacquet and Shalika, 1981).

Proposition 12.1.6 For n ≥ 2, let A(m 1 , . . . , m n−1 ) denote the Fourier

coefficient of a Maass form on S L(n, Z) as in (12.1.1). Then there exists a
constant C > 0 such
1
|A(m, 1, . . . , 1)| ≤ C · |m| 2 .

Proof In view of the addendum to Theorem 9.3.11, it is enough to prove that

1
|A(1, . . . , 1, m)| ≤ C · |m| 2 . For σ > 0, consider the identity
σ+i∞ 
1 xs 1
(log x) −1 if x > 1,
ds = ( −1)!
2πi s 0 if 0 < x ≤ 1,
σ −i∞
 12.1 The GL(n) × GL(n) convolution 371

which is easily proved by shifting the line of integration to the left and apply-
ing Cauchy’s residue theorem. In view of the fact that the Rankin–Selberg
L-function, L f × f (s), specified in Definition 12.1.2 has positive coefficients
and |A(1, . . . , 1, m)|2 occurs as a coefficient of m −s , one obtains for fixed σ,
sufficiently large and all |m| sufficiently large that
σ+i∞
1 (2m)s
|A(1, . . . , 1, m)| 
2
L f × f (s) ds. (12.1.7)
2πi s
σ −i∞

Here one chooses σ sufficiently large so that L f × f (σ + it) converges absolutely

for all t ∈ R. Also, one chooses sufficiently large so that |L f × f (s)|  |s| −2
for (s) > 0. This can be done, because by standard methods, the functional
equation of theorem 12.1.4 implies that L f × f (s) has polynomial growth prop-
erties in the strip (s) > 0. Proposition 12.1.6 can be proved by shifting
the line of integration in (12.1.7) to the left and noting that L f × f (s) has a
simple pole at s = 1 with residue R, say. By Cauchy’s theorem, it follows that
|A(1, . . . , m)|2  R · 2m + O (1) .

Remark 12.1.8 Note that the method for proving Proposition 12.1.6 also
shows that A(m 1 , m 2 , . . . , m n−1 ) behaves like a constant on average. If we
write

∞
b(m)
L f × f (s) =
m=1
ms

as a Dirichlet series, then one may show that as x → ∞,

b(m) ∼ c · x
m≤x

for some constant c > 0.

Another application of the Rankin–Selberg G L(n) × G L(n) convolution is

to obtain a bound on the eigenvalues of a Maass form. This result was again
first obtained in a much more general setting in (Jacquet and Shalika, 1981).

Proposition 12.1.9 For n ≥ 2, let f be a Maass form for S L(n, Z) of type

(ν1 , . . . , νn−1 ) ∈ Cn−1 as in (12.1.1). Then
1
(λi (ν)) ≤
2
for i = 1, 2, . . . , n − 1, with λi defined as in Theorem 10.8.6 and
Remark 10.8.7.
372 Rankin–Selberg convolutions

Proof By Landau’s lemma (Iwaniec and Kowalski, 2004), a Dirichlet series

must be absolutely convergent up to its first pole. In the case of L f × f (s) this
implies that the Dirichlet series for L f × f (s) converges absolutely for (s) > 1.
We also know that  f × f (s), given in Theorem 12.1.4, has a meromorphic con-
tinuation to all s ∈ C with at most a simple pole at s = 1. The above remarks
immediately imply that the Gamma factors in  f × f (s) cannot have poles
for (s) > 1. One may then check, using Remark 10.8.7, that this implies
Proposition 12.1.9.

12.2 The GL(n) × GL(n + 1) convolution

Fix n ≥ 2, and let f, g be Maass forms of type ν f ∈ Cn−1 , νg ∈ Cn , for
S L(n, Z), S L(n + 1, Z), respectively, with Fourier expansions (see (9.1.2)):



∞

∞
A(m 1 , . . . , m n−1 )
f (z) = ···
γ ∈Un−1 (Z)\S L(n−1,Z) m 1 =1 m n−2 =1 m n−1 =0
)
n−1
|m k |k(n−k)/2
k=1
⎛⎛ ⎞ ⎞
m 1 · · · |m n−1 |
⎜⎜ .. ⎟   ⎟
⎜⎜ . ⎟ γ ⎟
× WJacquet ⎜⎜ ⎟· z, ν f , ψ1,...,1, m n−1 ⎟,
⎝⎝ m1 ⎠ 1 |m n−1 | ⎠
1
(12.2.1)



∞

∞
B(m 1 , . . . , m n )
g(z) = ···
)
n
γ ∈Un (Z)\S L(n,Z) m 1 =1 m n−1 =1 m n =0 |m k |k(n+1−k)/2
k=1
⎛⎛ ⎞ ⎞
m 1 · · · |m n |
⎜⎜ .. ⎟  ⎟
⎜⎜ . ⎟ γ ⎟
× WJacquet ⎜⎜ ⎟· z, νg , ψ1,...,1, mn ⎟.
⎝⎝ m1 ⎠ 1 |m n |
⎠
1

Definition 12.2.2 (Rankin–Selberg L-function) For n ≥ 2, let f, g be two

Maass forms for S L(n, Z), S L(n + 1, Z), respectively, as in (12.2.1). Let
s ∈ C. Then the Rankin–Selberg L-function, denoted L f ×g (s), is defined by


∞
∞
A(m 2 , . . . , m n ) · B(m 1 , . . . , m n )
L f ×g (s) = ···  n n−1 s ,
m 1 =1 m n =1 m1 m2 · · · mn

which converges absolutely provided (s) is sufficiently large.

 12.2 The GL(n) × GL(n + 1) convolution 373

There are a number of significant differences between the

G L(n) × G L(n + 1) Rankin–Selberg convolution and the G L(n) × G L(n)
convolution given in Section 12.1. For example, the Riemann zeta function
does not appear in Definition 12.2.2 as it did in Definition 12.1.2. Even
more surprising is the fact that the G L(n) × G L(n + 1) convolution does
not involve an Eisenstein series and the functional equation comes from the
invariance of f, g under reflections of the Weyl group instead of the functional
equation of an Eisenstein series. This explains the lack of the appearance of
the Riemann zeta function in Definition 12.2.2. Another explanation comes
from the theory of Schur polynomials and the applications to Hecke operators
and Euler products. The convolution L-function L f ×g (s) has a simple Euler
product whose explicit construction will be deferred to Section 12.3.
We now explain the G L(n) × G L(n + 1) convolution. The key idea is to
embed G L(n) in G L(n + 1) into the upper left-hand corner and then integrate
against a power of the determinant. Accordingly, let z = x y ∈ hn with
⎛ ⎞ ⎛ ⎞
1 xn−1 x1,3 ··· x1,n y1 y2 · · · yn−1
⎜ 1 xn−2 ··· x2,n ⎟ ⎜ y1 y2 · · · yn−2 ⎟
⎜ ⎟ ⎜ ⎟
⎜ .. .. .. ⎟, ⎜ .. ⎟
x= ⎜ . . . ⎟ y =⎜ . ⎟,
⎜ ⎟ ⎜ ⎟
⎝ 1 x1 ⎠ ⎝ y1 ⎠
1 1

as in Section 9.1, and consider

  
 1 z
|Det(z)|s− 2 d ∗ z.
1
f · ḡ, |Det(∗)|s̄− 2 = f (z) · g
1
S L(n,Z)\hn

1
The power Det(z)s− 2 , above, was chosen to make the final formulae as simple
as possible. In view of the Fourier expansion of g given in (12.2.1), we may
apply the unfolding trick to g and obtain
 1
f · ḡ, |Det(∗)|s̄− 2

∞
∞
B(m 1 , . . . , m n ) 
= ··· f (z)
)
n
m 1 =1 m n−1 =1 m n =0 |m k |k(n+1−k)/2 Un (Z)\hn
k=1
⎛⎛ ⎞ ⎞
m 1 · · · |m n |
⎜⎜ .. ⎟  ⎟ s− 1
⎜⎜ ⎟ z ⎟
× W̄Jacquet⎜⎜ . ⎟ , νg , ψ1,...,1, |mmnn | ⎟ Det(z) 2 d ∗ z.
⎝⎝ m1 ⎠ 1 ⎠
1
374 Rankin–Selberg convolutions

Next, make the transformation

z → m −1
1 In z,

where In is the n × n identity matrix. Since f (z) and d ∗ z are invariant under
scalar multiplication, it follows that
 1
f · ḡ, |Det(∗)|s̄− 2

∞
∞
B(m 1 , . . . , m n ) 
= ··· f (z)
)
n
m 1 =1 m n−1 =1 m n =0 |m k |k(n+1−k)/2 Un (Z)\hn
k=1
⎛⎛ ⎞ ⎞
m 2 · · · m n y1 · · · yn−1
⎜⎜ m · · · m y · · · y ⎟ ⎟
⎜⎜ 2 n−1 1 n−2 ⎟ ⎟
⎜⎜ ⎟ ⎟
⎜⎜ .. ⎟ ⎟
× WJacquet ⎜ ⎜⎜
⎜ . ⎟, νg , ψ1,...,1 ⎟
⎟ ⎟
⎜⎜ m 2 y1 ⎟ ⎟
⎜⎜ ⎟ ⎟
⎝⎝ 1 ⎠ ⎠
1
−n (s− 12 )  
× e−2πi (m 2 x1 +m 3 x2 +···+m n xn−1 ) m 1 Det(y)s− 12 d ∗ z

∞
∞
A(m 2 , . . . , m n ) B(m 1 , . . . , m n )
= ···  n n−1 s
m 1 =1 m n−1 =1 m n =0 m 1 m 2 · · · |m n |
∞ ∞   
 y
, νg , ψ1,...,1 Det(y)s− 2 d ∗ y,
1
× · · · WJacquet y, ν f , ψ1,...,1 W̄Jacquet
1
0 0
(12.2.3)
because the above integral in d ∗ x picks off the (m 2 , m 3 , . . . , m n )th Fourier
) −k(n−k)
n−1
coefficient (Theorem 5.3.2) and d ∗ y = yk dyk /yk as in (1.5.4).
k=1
Define
 1
 f ×g (s) := f · ḡ, |Det(∗)|s̄− 2 . (12.2.4)
Theorem 12.2.5 (Functional equation) Let f, g be Maass forms asso-
ciated to S L(n, Z), S L(n + 1, Z), respectively, with Fourier expansions
given by (12.2.1). Then the Rankin–Selberg L-function, L f ×g (s), defined in
Definition 12.2.2 has holomorphic continuation to all s ∈ C and satisfies the
functional equation
 f ×g (s) =  f˜×g̃ (1 − s),
where  f ×g (s) is defined in (12.2.4), (12.2.3), and f˜, g̃ denote the dual Maass
forms as in Section 9.2.
 12.2 The GL(n) × GL(n + 1) convolution 375

Proof The functional equation is a consequence of the substitution

z −→ w · t (z −1 ) · w −1 := ι z,

where w is the long element as in Proposition 9.2.1. It follows from (9.2.2) that
⎛ ⎞⎛ ⎞
1 δx1 y1 y2 · · · yn−1
⎜ 1 −x2 ∗ ⎟⎜ y2 y3 · · · yn−2 ⎟
⎜ ⎟⎜ ⎟
ι ⎜ . . ⎟ ⎜ . ⎟
z=⎜ .. .. ⎟⎜ .. ⎟,
⎜ ⎟⎜ ⎟
⎝ 1 −xn−1 ⎠ ⎝ yn−1 ⎠
1 1
!n/2"+1
where δ = (−1) . It follows from Proposition 9.2.1, and the fact that f, g
are automorphic that
 ι   
˜f (ι z) = f (z), z z
g̃ =g .
1 1
Consequently,
 1
 f ×g (s) = f · ḡ, |Det(∗)|s̄− 2
  
z
|Det(z)|s− 2 d ∗ z
1
= f (z) · g
1
S L(n,Z)\hn
  ι 
z
f˜(ι z) · g̃ |Det(z)|s− 2 d ∗ z.
1
= (12.2.6)
1
S L(n,Z)\hn

But d ∗ z is fixed under the transformation z → t (z −1 ) while

|Det(t (z −1 ))|s− 2 = |Det(z)| 2 −s .

1 1
(12.2.7)

It follows from the above remarks, after applying the substitution (12.2.7) to
(12.2.6), that formally
  
wz
|Det(z)| 2 −s d ∗ z
1
 f ×g (s) = f (wz) · g̃
˜
1
S L(n,Z)\hn
  
z
|Det(z)| 2 −s d ∗ z
1
= f˜(z) · g̃
1
S L(n,Z)\hn

=  f˜×g̃ (1 − s).

The above formal proof can be made rigorous by breaking the integrals from 0 to
∞ into two pieces [0, 1], [1, ∞] and applying Riemann’s method for obtaining
376 Rankin–Selberg convolutions

the functional equation of the Riemann zeta function. We will not pursue this
here.

12.3 The G L(n) × G L(n  ) convolution with n < n 


Fix 2 ≤ n < n  − 1. Let f, g be Maass forms of type ν f ∈ Cn−1 , νg ∈ Cn −1 ,
for S L(n, Z), S L(n  , Z), respectively, with Fourier expansions (see (9.1.2)):

∞

∞
A(m 1 , . . . , m n−1 )
f (z) = ···
γ ∈Un−1 (Z)\S L(n−1,Z) m 1 =1 m n−2 =1 m n−1 =0
)
n−1
|m k |k(n−k)/2
k=1
   
γ
× WJacquet M · z, ν f , ψ1,...,1, mn−1 ,
1 |m n−1 |

(12.3.1)



∞

∞
B(m 1 , . . . , m n  −1 )
g(z) = ··· n) −1

γ ∈Un  −1 (Z)\S L(n  −1,Z) m 1 =1 m n  −2 =1 m n  −1 =0 |m k |k(n −k)/2
k=1

  
γ
× WJacquet M · z, νg , ψ1,...,1, m n  −1 .
1 |mn −1 |

The G L(n) × G L(n + 1) Rankin–Selberg method discussed in Section 12.2

does not naturally generalize to arbitrary n  > n. The best way to proceed in
the more general situation is to apply a projection operator before taking the
inner product. We follow (Bump, 1987), (Bump, to appear) and the exposition in
Cogdell’s chapter “On the analytic theory of L-functions for G L n ” in (Bernstein
and Gelbart, 2003).

Definition 12.3.2 (Projection operator) Fix integers 2 ≤ n < n  − 1. We


introduce the projection operator Pnn which acts on Maass forms for
S L(n  , Z) and maps them to cuspidal automorphic functions for the parabolic
subgroup

Pn,1 (Z) = Pn,1 (R) ∩ S L(n + 1, Z)

where
 
G L(n, R) ∗
Pn,1 (R) = ⊂ G L(n + 1, R).
0 1
 12.3 The G L(n) × G L(n  ) convolution with n < n  377

Let g be a Maass form for S L(n  , Z) as in (12.3.1). For z ∈ Pn,1 (R) we define
⎛⎛ ⎞⎞
u 1,n+2 ··· u 1,n 
⎜⎜ .. .. ⎟⎟
⎜⎜ z . ··· . ⎟ ⎟
⎜⎜ ⎟⎟
  ⎜⎜⎜⎜ .. ⎟⎟⎟
1 1
⎟
n −(n  −n−1)/2 ⎜⎜ u · · · . ⎟⎟
Pn (g)(z) := |Det(z)| · · · g ⎜⎜ n+1,n+2
⎟⎟
⎜⎜ .. .. ⎟⎟
⎜⎜ 1 . . ⎟ ⎟
0 0 ⎜⎜ ⎟⎟
⎜⎜ . ⎟⎟
⎝⎝ .. u 
n −1,n 
⎠⎠
1


× e−2πi (u n+1,n+2 +u n+2,n+3 +···+u n −1,n ) du i, j .
n+2≤ j≤n 
1≤i< j
 
It is clear that if p ∈ Pn,1 (Z), then Pnn (g)( pz) = Pnn (g)(z).

Lemma 12.3.3 Fix 2 ≤ n < n  − 1, and let g be given by (12.3.1). Then for
z ∈ Pn,1 (R), we have


Pnn (g)(z)




∞

∞
B(1, . . . , 1, m n  −n , . . . , m n  −1 )
= |Det(z)|−(n −n−1)/2 ··· n)
 −1
γ ∈Un (Z)\S L(n,Z) m n  −n =1 m n  −1 =1
|m k |k(n  −k)/2
k=n  −n
⎛⎛ ⎞ ⎞
m n  −n · · · m n  −1
⎜⎜ m n  −n · · · m n  −2 ⎟  ⎟
⎜⎜ ⎟ ⎟
⎜⎜ .. ⎟ γz ⎟
× WJacquet ⎜⎜ . ⎟· , νg , ψ1,...,1 ⎟,
⎜⎜ ⎟ In  −n ⎟
⎝⎝ m n  −n ⎠ ⎠
In  −n
where Ir denotes the r × r identity matrix.

Proof Let z ∈ Pn,1 (R). Since Pnn (g(z)) is invariant under left multiplication
by Pn,1 (Z) it follows from Section 5.3 that it has a Fourier expansion as in
Theorem 5.3.2. After some computation, the proof follows as in the proof of
Theorem 9.4.7.

With these preliminaries out of the way, we may proceed to describe the
Rankin–Selberg convolution. Let f, g be Maass forms for S L(n, Z), S L(n  , Z),
respectively as in (12.3.1). The requisite convolution is given by the
1
inner product,  f · Pnn  (g), |Det(∗)|s̄− 2 , taken over the fundamental domain
378 Rankin–Selberg convolutions


S L(n, Z)\G L(n, R). Lemma 12.3.3 allows us to unravel Pnn (g) in this inner
product. It follows that
 1
f · Pnn  (g), |Det(∗)|s̄− 2

∞
∞
B(1, . . . , 1, m n  −n , . . . , m n  −1 )
= f (z) ··· n)
 −1
m n  −n =1 m n  −1 =1
Un (Z)\G L(n,R) |m k |k(n  −k)/2
k=n  −n
⎛⎛ ⎞ ⎞
m n  −n · · · m n  −1
⎜⎜ m n  −n · · · m n  −2 ⎟   ⎟
⎜⎜ ⎟ ⎟
⎜⎜ .. ⎟ z ⎟
× W Jacquet ⎜⎜ . ⎟· , νg , ψ1,...,1 ⎟
⎜⎜ ⎟ In  −n ⎟
⎝⎝ m n  −n ⎠ ⎠
In  −n
n  −n
× |Det(z)|s− 2 d ∗ z.

In the above integral, we may make the change of variables z → m −1 n  −n In z,

noting that f and d ∗ z are invariant under this transformation. After this, the d ∗ x
integral will pick off the (m n  −n+1 , . . . , m n  −1 ) coefficient of f . It follows that
 1
f · Pnn  (g), |Det(∗)|s̄− 2

∞
∞
A(m n  −n+1 , . . . , m n  −1 ) B(1, . . . , 1, m n  −n , . . . , m n  −1 )
= ···  n s
m n  −n =1 m n  −1 =1  −n+1 · · · m n  −1
m n  −n m nn−1
∞ ∞   
y
× ··· WJacquet (y, ν f , ψ1,...,1 )W̄Jacquet , νg , ψ1,...,1
In  −n
0 0
n  −n
× |Det(z)|s− 2 d ∗ y.

By analogy with the Rankin–Selberg constructions given in Sections 12.1,

12.2, it is natural, after the above computations, to make the following definition.

Definition 12.3.4 (Rankin–Selberg L–function) Fix 2 ≤ n < n  . Let f, g

be two Maass forms for S L(n, Z), S L(n  , Z), respectively, as in (12.3.1). Let
s ∈ C. Then the Rankin–Selberg L-function, denoted L f ×g (s), is defined by


∞
∞
A(m 2 , . . . , m n ) · B(m 1 , . . . , m n , 1, . . . , 1)
L f ×g (s) = ···  n n−2 s ,
m 1 =1 m n =1 m1 m2 · · · mn
which converges absolutely provided (s) is sufficiently large. Here, the
coefficient B(m 1 , . . . , m n , 1, . . . , 1), has precisely n  − n − 1 ones on the
" #$ %
n  −n−1
right-hand side.
 12.3 The G L(n) × G L(n  ) convolution with n < n  379

Remark We may also take n = 1 in Definition 12.3.4 by letting f be the con-

stant function. In this case the Rankin–Selberg L-function is just the Godement–
Jacquet L-function as in Section 9.4. In fact, all constructions and proofs in this
section will work when n = 1.

Theorem 12.3.5 (Euler product) Fix 2 ≤ n < n  . Let f, g be two Maass

forms for S L(n, Z), S L(n  , Z), respectively, as in (12.3.1) with Euler products

∞
A(m, 1, . . . , 1)  n
L f (s) = = (1 − α p,i p −s )−1 ,
m=1
ms p i=1

∞
B(m, 1, . . . , 1)  n
L g (s) = s
= (1 − β p,i p −s )−1 ,
m=1
m p i=1

then L f ×g will have an Euler product of the form:


n 
n
L f ×g (s) = (1 − α p,i β p, j p −s )−1 .
p i=1 j=1

Proof The proof follows the ideas in Section 7.4, but requires a modification
of Cauchy’s identity (Proposition 7.4.20). For details, see (Bump, 1984, 1987,
to appear).

Finally, we consider the meromorphic continuation and functional equation.

Theorem 12.3.6 (Functional equation) Fix 2 ≤ n < n  . Let f, g be two

Maass forms for S L(n, Z), S L(n  , Z), respectively, as in (12.3.1), whose asso-
ciated L-functions L f , L g satisfy the functional equations:
  
n −s+λi (ν f ) s − λi (ν f )
 f (s) := π 2  L f (s) =  f˜ (1 − s)
i=1
2
n  
−s+λ j (νg ) s − λ j (νg )
g (s) := π 2  L g (s) = g̃ (1 − s),
j=1
2

as in Remark 10.8.7 where f˜, g̃ are the dual Maass forms (see Section 9.2). Then
the Rankin–Selberg L-function, L f ×g (s) (see Definition 12.3.4) has a holomor-
phic continuation to all s ∈ C. Furthermore, L f ×g (s) satisfies the functional
equation


n  n −s+λi (ν f )+λ j (νg ) s − λi (ν f ) − λ j (νg )
 f ×g (s) := π 2  L f ×g (s)
i=1 j=1
2
=  f˜×g̃ (1 − s).
380 Rankin–Selberg convolutions

Remark Note that as in the remark after Theorem 12.1.4, the power of π in

the above theorem simplifies to π −nn s/2 .

Proof This theorem has essentially been proved in the case n  = n + 1 in

Theorem 12.2.5, so we need only consider n < n  − 1.
Let z ∈ hn . The functional equation is a consequence of the substitution

z −→ w · t (z −1 ) · w−1 := ι z,

where w is the long element as in Proposition 9.2.1. It follows from (9.2.2)

that
⎛ ⎞⎛ ⎞
1 δx1 y1 y2 · · · yn−1
⎜ 1 −x2 ∗ ⎟⎜ y2 y3 · · · yn−2 ⎟
⎜ ⎟⎜ ⎟
ι ⎜ . . ⎟ ⎜ . ⎟
z=⎜ .. .. ⎟⎜ .. ⎟,
⎜ ⎟⎜ ⎟
⎝ 1 −xn−1 ⎠ ⎝ yn−1 ⎠
1 1
!n/2"+1
where δ = (−1) . It follows from Proposition 9.2.1 and the fact that f, g
are automorphic that

f˜ (ι z) = f (z),

and
 ι   
z z
g̃ =g .
In  −n In  −n

Consequently,
 1 
 f ×g (s) = f · Pnn  (g), |Det(∗)|s̄− 2
  
z
|Det(z)|s− 2 d ∗ z
1
= f (z) · Pnn  (g)
I n  −n
S L(n,Z)\G L(n,R)
  ι 
z
f˜(ι z) · Pnn  (g̃) |Det(z)|s− 2 d ∗ z.
1
=
In  −n
S L(n,Z)\G L(n,R)
(12.3.7)

But d ∗ z is fixed under the transformation z → t (z −1 ) while

  1  1
Det(t (z −1 ))s− 2 = Det(z) 2 −s . (12.3.8)
 12.4 Generalized Ramanujan conjecture 381

It follows from the above remarks, after applying the substitution (12.3.8) to
(12.3.7), that formally
  
 z
|Det(z)| 2 −s d ∗ z
1
 f ×g (s) = f˜(wz) · ι ◦ Pnn  ◦ ι (g̃)
1
S L(n,Z)\G L(n,R)
  
 z
|Det(z)| 2 −s d ∗ z
1
= f˜(z) · ι ◦ Pnn  ◦ ι (g̃)
1
S L(n,Z)\G L(n,R)

=
˜ f˜×g̃ (1 − s),

where
  1
˜ f˜×g̃ (s) = f˜ · ι ◦ Pnn  ◦ ι (g̃),
 |Det(∗)|s̄− 2 .

The above formal proof can be made rigorous by breaking the integrals from 0 to
∞ into two pieces [0, 1], [1, ∞] and applying Riemann’s method for obtaining
the functional equation of the Riemann zeta function.
Once the form of the functional equation is obtained, the precise Gamma
factors in the functional equation can be deduced by using the functional equa-
tion of the minimal parabolic Eisenstein series as a template as we did at the
end of Section 12.1. We leave these computations to the reader.

12.4 Generalized Ramanujan conjecture

Let f be a Maass form of type ν = (ν1 , . . . , νn−1 ) ∈ Cn−1 for S L(n, Z) with
Fourier expansion as given in (9.1.2):



∞

∞
A(m 1 , . . . , m n−1 )
f (z) = ···
γ ∈Un−1 (Z)\S L(n−1,Z) m 1 =1 m n−2 =1 m n−1 =0
)
n−1
|m k |k(n−k)/2
k=1
   
γ
× WJacquet M · z, ν, ψ1,...,1, mn−1 , (12.4.1)
1 |m n−1 |

where
⎛ ⎞
m 1 · · · m n−2 · |m n−1 |
⎜ .. ⎟
⎜ . ⎟
⎜ ⎟
M =⎜ m1m2 ⎟, A(m 1 , . . . , m n−1 ) ∈ C,
⎜ ⎟
⎝ m1 ⎠
1
382 Rankin–Selberg convolutions

and
⎛⎛ ⎞⎞
1 u n−1
⎜⎜ 1 u n−2 ∗⎟ ⎟
⎜⎜ ⎟⎟
⎜⎜ ⎟⎟
ψ1,...,1, ⎜⎜ ..
.
..
. ⎟⎟ = e2πi (u 1 +···+u n−2 +u n−1 ) .
⎜⎜ ⎟⎟
⎝⎝ 1 u 1 ⎠⎠
1
We further assume that f is normalized so that A(1, . . . , 1) = 1 and that f is an
eigenfunction of the Hecke operators given in (9.3.5). It was shown in (9.4.2)
that the Godement–Jacquet L-function

∞
A(n, 1, . . . , 1)
L f (s) =
n=1
ns
has an Euler product given by

L f (s) = 1 − A( p, . . . , 1) p −s + A(1, p, . . . , 1) p −2s −
p
−1
· · · + (−1)n−1 A(1, . . . , p) p (1−n)s + (−1)n p −ns

n
= (1 − α p,i p −s )−1 , (12.4.2)
p i=1

where α p,i ∈ C for i = 1, 2, . . . , n.

Conjecture 12.4.3 (Ramanujan conjecture at finite primes) For n ≥ 2, let
f be a Maass form for S L(n, Z) as in (12.4.1). The generalized Ramanujan
conjecture asserts that α p,i given in (12.4.2) satisfy
|α p,i | = 1,
for all primes p and i = 1, . . . , n. Equivalently, for every rational prime p we
have the bound
|A( p, 1, . . . , 1)| ≤ n.

This conjecture has been proved for holomorphic modular forms on G L(2)
in (Deligne, 1974) (see Section 3.6) but is still a major unsolved problem for
Maass forms. Deligne shows that every holomorphic modular form (Hecke
eigenform) over Q, say, is associated to an algebraic variety and that the pth
Fourier coefficient can be interpreted in terms of the number of points on this
variety over F p , the finite field of p elements. From this point of view, the
Ramanujan conjecture is equivalent to the Riemann hypothesis for varieties
over finite fields first conjectured in (Weil, 1949) and, in a stunning tour de
 12.4 Generalized Ramanujan conjecture 383

force, finally proved in (Deligne, 1974). The problem with trying to generalize
this approach to non-holomorphic automorphic forms is that there seem to be no
visible connections between the theory of Maass forms and algebraic geometry.
There is yet another fundamental conjecture which was originally formu-
lated by Selberg (1965) (see also Section 3.7). Classically this is known as the
Selberg eigenvalue conjecture. For the cognoscenti, it is clear from the adelic
point of view that the Selberg eigenvalue conjecture is really the generalized
Ramanujan conjecture at the infinite prime. In the next section, we will give a
more elementary explanation of why these two conjectures can be placed on an
equal footing. Here is the generalized Selberg eigenvalue conjecture.
Conjecture 12.4.4 (Selberg eigenvalue conjecture) For n ≥ 2, let f (z) be a
Maass form of type (ν1 , ν2 , . . . , νn−1 ) ∈ Cn−1 for S L(n, Z) as in (12.4.1). Then
1
(νi ) = , (λi (ν)) = 0,
n
for i = 1, 2, . . . , n − 1, with λi as defined in Theorem 10.8.6 and
Remark 10.8.7.
Remark The first to observe that the classical Ramanujan conjecture con-
cerning the Fourier coefficients of (z) can be reformulated to a very general
conjecture on G L(n) appears to have been Satake (1966). The generalized
Ramanujan conjecture has been established for automorphic cusp forms of
G L(n, F), of algebraic type satisfying a Galois conjugacy condition, where F
is a complex multiplication field (see (Harris and Taylor, 2001)). The proof is
a remarkable tour-de-force combining the Arthur–Selberg trace formula and
the theory of Shimura varieties. The generalized Ramanujan conjecture is still
unproven for Maass forms for S L(n, Z) of the type considered in this book.
While the Selberg eigenvalue Conjecture 12.4.4 is not hard to prove for
S L(2, Z) (see Theorem 3.7.2) it is still an unsolved problem for congruence
subgroups (see Conjecture 3.7.1); and, of course, Conjecture 12.4.4 can be easily
generalized to higher level congruence subgroups of S L(n, Z) with n ≥ 2.
In (Kim and Sarnak, 2003), one may find the current world record for both the
Ramanujan conjecture and Selberg eigenvalue conjecture for G L(2) (this
includes the case of congruence subgroups of higher level). The precise bounds
obtained are
7
|α p,1 |, |α p,2 | ≤ p 64 ,

and correspondingly,
7
(ν) ≤ .
64
384 Rankin–Selberg convolutions

In (Selberg, 1965) the bound ≤ 1/4 instead of ≤ 7/64 as in (Kim and Sarnak,
2003) was attained. Selberg’s result was slightly improved to < 1/4 in (Gelbart
and Jacquet, 1978). Further improvements were given in (Serre, 1981) (see also
1 1
(Serre, 1977)), who obtained |α p,i | ≤ p 5 , Shahidi (1988), |α p,i | < p 5 , Duke
1
and Iwaniec (1989), found a different proof of |α p,i | ≤ 2 p 5 , Bump, Duke,
5
Hoffstein and Iwaniec (1992), |α p,i | ≤ p 28 . In the case of G L(n) with n ≥ 2,
Jacquet and Shalika (1981) obtained the bound < 1/2. This bound was proved
for S L(n, Z) in Propositions 12.1.6, 12.1.9. For the finite primes p, Serre (1981)
suggested that one could do better by a clever use of the Rankin–Selberg L-
function. This led to the bound
1 1
≤ − 2 (12.4.5)
2 dn + 1
for Maass forms on G L(n) over a number field of degree d. In (Luo, Rudnick and
Sarnak, 1995, 1999) the dependence on d was removed. We give an exposition
of the Luo, Rudnick and Sarnak method (over Q) in the next section.

12.5 The Luo–Rudnick–Sarnak bound for the

generalized Ramanujan conjecture
As an application of the Rankin–Selberg method, we shall give a proof of the
bound (12.4.5) obtained in (Luo, Rudnick and Sarnak, 1995, 1999). In order
to simplify the exposition, we work over Q, instead of a number field, so that
d = 1. Let us restate the theorem.
Theorem 12.5.1 (Luo–Rudnick–Sarnak) Fix n ≥ 2. Let f be a Maass form
of type (ν1 , . . . , νn−1 ) for S L(n, Z) which is an eigenfunction of the Hecke
operators as in (12.4.1). Then for α p,i as in (12.4.2), we have
1
− 1 1 1
|α p,i | ≤ p 2 n 2 +1 , (λi (ν)) ≤ −
2 n2 + 1
for all primes p, 1 ≤ i ≤ n, and 1 ≤ j ≤ n − 1. Here λi is defined in
Theorem 10.8.6 and Remark 10.8.7.

Remark In (Luo, Rudnick and Sarnak, 1999) the above bound was obtained
for Maass forms on G L(n) over an arbitrary number field.
Proof Recall the definition of L f (s) given in (12.4.2).

n 
L f (s) = (1 − α p,i p −s )−1 .
i=1 p
 12.5 The Luo–Rudnick–Sarnak bound for Ramanujan conjecture 385

Now, let χ be a primitive Dirichlet character (mod q). Consider f ⊗ χ, which

has associated L-function
 n
L f ⊗χ (s) = (1 − α p,i χ ( p) p −s )−1
p i=1

We may now take the Rankin–Selberg convolution of f with f ⊗ χ . This leads

to the Rankin–Selberg L-function

n 
n
L ( f ⊗χ)× f (s) = (1 − α p,i α p, j χ ( p) p −s )−1 .
p i=1 j=1

Further, as in the proof of Theorem 12.1.4, Remark 10.8.7, and the comments
after it, the function L ( f ⊗χ )× f (s) satisfies the functional equation

( f ⊗χ)× f (s)


n  n   s+aχ −λi (ν)−λ j (ν)
q 2 s − λi (ν) − λ j (ν) + aχ
=  L ( f ⊗χ )× f (s)
i=1 j=1
π 2
= χ · ( f˜⊗χ̄ )× f˜ (1 − s), (12.5.2)

where
 √ 2 
i aχ q n 0 if χ (−1) = 1,
χ = , aχ =
τ (χ ) 1 if χ (−1) = −1.

It follows from the methods used to prove Theorem 12.1.4 that ( f ⊗χ )× f (s)
has a holomorphic continuation to all s ∈ C except for simple poles at s = 1
and s = 0, the latter simple poles can only occur if χ ≡ 1 is the trivial character.
Note that at finite level there can be finitely many χ for which this has a pole
(of course, this impacts nothing). This result was also proved in much greater
generality in the combination of papers: (Shahidi, 1981, 1985), (Jacquet and
Shalika, 1981, 1990), (Jacquet, Piatetski-Shapiro and Shalika, 1983), (Moeglin
and Waldspurger, 1989).
We also require, for any fixed prime p0 , the modified Rankin–Selberg L-
function

n 
n

1 − α p0 ,i α p0 , j χ ( p0 ) p0−s , (12.5.3)
p
L ( f0 ⊗χ)× f (s) = L ( f ⊗χ )× f (s) ·
i=1 j=1

which is the same as L ( f ⊗χ )× f (s) except that the Euler factor at p0 has been
removed.
The key idea in the proof of Theorem 12.5.1 is the following lemma.
386 Rankin–Selberg convolutions

Lemma 12.5.4 Fix n ≥ 2 and fix p0 to be either 1 or a rational prime.

Let f be a Maass form for S L(n, Z) as above. Then for any real number
β > 1 − (2/(n 2 + 1)), there exist infinitely many primitive Dirichlet charac-
ters χ such that χ ( p0 ) = 1, χ (−1) = 1, and
p
L ( f0 ⊗χ )× f (β) = 0.

We defer the proof of this lemma until later and continue with the proof of
Theorem 12.5.1.
It follows from (12.5.3) that

n 
n
 −1
1 − α p0 ,i α p0 , j χ ( p0 ) p0−s
p
L ( f ⊗χ)× f (s) = L ( f0 ⊗χ )× f (s) · .
i=1 j=1

Assume χ(−1) = 1, so aχ = 0, and the Gamma factors in the functional equa-

tion take the form (12.5.5). Since ( f ⊗χ )× f (s) is holomorphic for χ non-trivial,
we see that any pole of

n 
n
 −1
1 − α p0 ,i α p0 , j χ ( p0 ) p0−s
i=1 j=1

or



n 
n
s − λi (ν) − λ j (ν)
 , (12.5.5)
i=1 j=1
2
p
must be a zero of L ( f0 ⊗χ )× f (s).
Assume that χ ( p0 ) = 1 and for some 1 ≤ i ≤ n we have |α p0 ,i |2 = p β with
β > 1 − (2/(n 2 + 1)). Then (1 − |α p0 ,i |2 p0−s )−1 has a pole at s = β. Sim-
ilarly, assume that (λi (ν)) = β for some i = 1, . . . , n − 1, and for some
β > 1 − n 22+1 . Then the Gamma factor
n  
s − 2(λi (ν))

i=1
2

would have a pole at s = 2β. It immediately follows, in both these cases,

p
that L ( f0 ⊗χ )× f (s) would have to have a zero in a region which contradicts
Lemma 12.5.4. This proves Theorem 12.5.1. It also explains why the Ramanu-
jan Conjecture 12.4.3 and the eigenvalue Conjecture 12.4.4 can be placed on
an equal footing.

Remark The above proof yields a slightly better bound for the case of
G L(2) Maass forms when combined with the Gelbart–Jacquet lift. For example,
 12.5 The Luo–Rudnick–Sarnak bound for Ramanujan conjecture 387

if we start with a Maass form of type ν for S L(2, Z) with L-function

)) 2
(1 − α p,i p −s )−1 , then the Gelbart–Jacquet lift will yield a degree 3 Euler
p i=1
product as in Lemma 7.3.5 associated to a Maass form of type (2ν/3, 2ν/3)
for S L(3, Z). When combined with the proof of Theorem 12.5.1, one gets the
bounds
1 1
|α p,i | ≤ p 5 (i = 1, 2), |(ν)| ≤ ,
5
a result first obtained by Shahidi (1988) by quite different methods.
Proof of Lemma 12.5.4 We first prove Lemma 12.5.4 in the simpler case
when p0 = 1, as in (Luo, Rudnick and Sarnak, 1995). Afterwards, we sketch
the proof in the more general case when p0 is a prime, as in (Luo, Rudnick and
Sarnak, 1999).

With the notation of (12.5.3), for a primitive character χ (mod q), the func-
tional equation (12.5.2) may be rewritten in the form.
 q (n 2 /2)−n 2 s
p p
L ( f0 ⊗χ )× f (s) = χ G p0 (s) L ( f0¯⊗χ̄)× f¯ (1 − s), (12.5.6)
π
where
 
) 1−s−λi (ν)−λ j (ν)  −1
n )n
 2
1 − α p0 ,i α p0 , j p0−(1−s)
i=1 j=1
G p0 (s) =   −1 .
)n ) n
α p0 ,i α p0 , j p0−s
s−λi (ν)−λ j (ν)
 2
1−
i=1 j=1

It follows easily from the Euler product (Theorem 12.1.3) and the coefficient
bound in Proposition 12.1.6 that for β ≥ 1, we have L ( f ⊗χ )× f (β) = 0. This was
proved in much greater generality in (Shahidi, 1981). Consequently, we may
assume that 1 − (2/(n 2 + 1)) < β < 1. The key idea of the proof is to show
that for  > 0, and Q sufficiently large that


∗ p
L ( f0 ⊗χ )× f (β)  Q 2− , (12.5.7)
q∼Q χ (mod q)
+∗
where means that the sum ranges over primitive characters χ (mod q)
+
satisfying χ ( p0 ) = χ (−1) = +1, and means we sum over primes
q∼Q
Q ≤ q ≤ 2Q. The lower bound (12.5.7) immediately implies Lemma 12.5.4.
To prove (12.5.7), we require an auxilliary compactly supported smooth
function h : [A, B] → R, where 0 < A < B, and h(y) ≥ 0, and in addition
 ∞
h(y)
dy = 1.
0 y
388 Rankin–Selberg convolutions

Define
∞
dy
h̃(s) := h(y)y s ,
y
0

to be the Mellin transform of h, and


1 ds
h 1 (y) := h̃(s)y −s .
2πi s
(s)=2

Then, by Mellin inversion,

 ∞
h(x)
h 1 (y) = d x.
y x
It follows that

0 ≤ h 1 (y) ≤ 1; h 1 (y) = 1; if 0 < y ≤ A,

while

h 1 (y) = 0, if y ≥ B.

Next, for y > 0, define


1 ds
h 2 (y) := h̃(−s)G p0 (−s + β)y −s .
2πi s
Re(s)=2

We shall forthwith show that h 2 satisfies the following bounds.

h 2 (y)  M y −M , for y ≥ 1 and any positive integer M, (12.5.8)

1−β0 −β−
h 2 (y)  1 + y , for 0 < y ≤ 1 and any  > 0, (12.5.9)

where

β0 = 2 max (λi (ν)),

1≤i≤n

and, we may recall that the λi (ν) occur in the Gamma factors of G p0 (s) as in
(12.5.6). To prove (12.5.8) shift the line of integration to the right to the line
(s) = M. The result follows since G p0 (s) has at most polynomial growth in
s in fixed vertical strips while h̃(−s) has rapid decay. To prove (12.5.9), shift
the line of integration to the left. We may assume β0 + β − 1 < 0, otherwise,
(12.5.9) is obvious. After shifting to the left, we will pick up the first simple
pole at (s) = β0 + β − 1. The bound (12.5.9) follows immediately.
The next step in the proof is the derivation of an approximate functional
equation for L ( f ⊗χ )× f (s) of the type previously derived in Section 8.3. The
 12.5 The Luo–Rudnick–Sarnak bound for Ramanujan conjecture 389

asymmetric form of the functional equation (12.5.6) then leads to an asymmetric

approximate functional equation.
Let

∞
b(m)
L ( f ⊗χ)× f (s) = χ (m).
m=1
ms

It follows that for any Y > 1,


1 ds
I(β, Y ) := h̃(s) L ( f ⊗χ)× f (s + β) Y s
2πi s
(s)=2


∞   s
b(m)χ (m) 1 Y ds
= h̃(s)
m=1
mβ 2πi m s
(s)=2

∞
b(m) m 
= β
χ (m)h 1 . (12.5.10)
m=1
m Y

On the other hand, we may shift the line of integration to the left, apply the
functional equation (12.5.6), and then let s → −s, to obtain

1 ds
I(β, Y ) := L ( f ⊗χ )× f (β) + h̃(s) L ( f ⊗χ )× f (s + β) Y s ,
2πi s
(s)=−1
(12.5.11)

where the integral on the right-hand side in (12.5.11) is equal to

 q n22 (1−2β) 
−s
χ
2
π πn Y ds
h̃(−s) G p0 (−s + β)L ( f¯⊗χ̄ )× f¯ (1 − β + s)
2πi q n2 s
(s)=1

which is equal to

 n 2 
 q  n22 (1−2β)
∞
b̄(m)χ̄ (m) π
χ h 2 mY · .
π m=1
m 1−β q

Combining (12.5.10), (12.5.11), and the above calculation, we obtain, for any
Y > 1, the approximate functional equation

∞
b(m) m 
L ( f ⊗χ )× f (β) = χ (m) · h 1
m=1
mβ Y


 q  n22 (1−2β)
∞
b̄(m) mY π n
2

− χ χ̄ (m) · h 2 .
π m=1
m 1−β q n2
390 Rankin–Selberg convolutions

This may be rewritten in the equivalent form

∞
b(m) m 
L ( f ⊗χ )× f (β) = β
χ (m) · h 1
m=1
m Y


τ (χ )n q −n
2 2
β
∞
b̄(m) mY π n
2

− n2
χ̄ (m) · h 2 . (12.5.12)
π 2 (1−β)
m=1
m 1−β q n2

The final step in the proof of Lemma 12.5.4 uses the approximate func-
tional equation (12.5.12) to deduce (12.5.7). We shall restrict ourselves to
prime q so that all non-trivial characters are automatically primitive. We
require:
⎧
⎪
⎪ m ≡ 0 (mod q)

⎨0,
χ (m) = q−1
− 1, m ≡ ±1 (mod q) (12.5.13)
⎪
⎪
2
χ (mod q) ⎩−1, otherwise.
χ=χ0 , χ(−1) = +1

We then use (12.5.12) and (12.5.13) to estimate

L ( f ⊗χ )× f (β). (12.5.14)
q∼Q χ (mod q)
χ =χ0 , χ (−1) = +1

The contribution of the first sum on the right-hand side of (12.5.12) to the
average (12.5.14) is



∞
b(m) m 
χ (m) · h 1
q∼Q χ (mod q) m=1
mβ Y
χ=χ0 , χ (−1) = +1

q −1
b(m)  m 

b(m)  m 
= h1 − h1 .
q∼Q
2 m≡±1 (mod q) m β Y q∼Q (m,q)=1
mβ Y

(12.5.15)

We will show that the main contribution to (12.5.15) comes from the term
m = 1. All other terms will give a much smaller contribution. In fact, the
contribution of the term m = 1 is

q −1 1
q −1
h1 = .
q∼Q
2 Y q∼Q
2

The sum over m ≡ 1 (mod q), m = 1 will contribute

 

q −1
∞
b(1 + dq) 1 + dq
∞
b(m)m    m 
h 1  Q h 1 ,
q∼Q
2 d=1 (1 + dq)β Y m=1
m (β) Y
 12.5 The Luo–Rudnick–Sarnak bound for Ramanujan conjecture 391

where we have used the fact that for m = 1, the number of different
representations m = 1 + dq = 1 + d  q  is O(m  ). By the properties of the
Rankin–Selberg convolution L f × f (s), (see Section 12.1, and in particular,
Remark 12.1.8) we see that b(m) ≥ 0 of all m = 1, 2, . . . , and


b(m) ∼ c f x, (x → ∞)
m≤x

+ +
∞
b(m)m 
for some c f > 0. It follows that m (β)
|h 1 (m/Y )|  QY 1−β+ and,
q∼Q m=1
therefore,

q −1
b(m)  m 
h1  QY 1−β+ .
q∼Q
2 m≡±1 (mod q)
mβ Y
m=1

By the same type of computation, one also obtains a bound for the last sum on
the right-hand side of (12.5.15):


b(m)  m 
h1  QY 1−β+ .
q∼Q (m,q)=1
mβ Y

Finally, we consider the contribution of the second term on the right-hand

side of (12.5.12). For this we make use of a deep bound of Deligne (1974) for
hyper-Kloosterman sums. Define
 

2πi
x1 +···+xn
K n (r, q) := e q
,
x1 x2 ···xn ≡r (mod q)

to be the hyper-Kloosterman sum. Then Deligne proved that

K n (r, q)  q (n−1)/2 . (12.5.16)

We shall use (12.5.16) to prove that for m ∈ Z and q  |m that

χ̄ (m)τ (χ )n  q (n +1)/2 .
2 2
(12.5.17)
χ (mod q)
χ=χ0 ,χ(−1)=+1

Indeed, we have the identity

2 q − 1 2
χ̄ (m)τ (χ )n = K n 2 (m, q) + K n 2 (−m, q) − (−1)n ,
χ (mod q)
2
χ =χ0 ,χ(−1)=+1

from which (12.5.17) immediately follows.

392 Rankin–Selberg convolutions

We now return to the computation of the contribution of the second term on

the right-hand side of (12.5.12). This contribution is




τ (χ )n q −n β

2 2 ∞
b̄(m) mY π n
2

χ̄(m) · h 2
q∼Q χ (mod q)
π (n 2 /2)(1−β) m=1 m 1−β q n2
χ=χ0 ,χ (−1)=+1



q n 2+1 −n 2 β
|b(m)|  mY π n 
2
∞ 2

 · h 2 
n2
q∼Q π 2
(1−β)
m=1
m 1−β  q n2 



q n 2+1 −n 2 β  
∞
r Y π n  β dr
2 2

  f2 r
n2
q∼Q π 2
(1−β)  q n2  r
1

q
n 2 +1
2
 n2
q∼Q π 2 (1+β)
· Yβ
2
1+ n 2+1
 Q · Y −β .
Collecting together all the previous computations with the approximate func-
tional equation (12.5.12) yields the asymptotic formula



q −1  n 2 +1
L ( f ⊗χ)× f (β) = + O QY 1−β+ + Q 1+ 2 · Y −β .
q∼Q χ (mod q) q∼Q
2
χ=χ0 ,χ(−1)=+1

Choosing Y ∼ Q (n +1)/2 , we obtain Lemma 12.5.4 when p0 = 1.

2

The case when p0 is a prime is more difficult to deal with. In this situtation
we replace (12.5.13) with



∗ = Nq if m = 1,
χ (m) = (12.5.18)
β|q χ (mod β) ≥0 otherwise,
χ ( p0 )=χ(−1)=1

where


∗
Nq = 1,
β|q χ (mod β)
χ ( p0 )=χ(−1)=1

and  ∗ means that we sum over primitive characters only.

In this variation of the method, we do not assume that q is prime. If q0 is the
largest integer dividing q with the property that (q0 , m) = 1, then


∗
χ (m) ≥ 0,
β|q0 χ (mod β)
χ ( p0 )=χ(−1)=1

because the sum is going over all elements in a group.

 12.6 Strong multiplicity one theorem 393

The key point is that one needs to know that Nq is large for many values of
q. More precisely, one requires that for every  > 0 and all Q sufficiently large


Nq  Q 2− . (12.5.19)
Q<q<2Q

This may be proved using a construction of Rohrlich (1989) together with a

variation of the Bombieri–Vinogradov theorem due to Murty and Murty (1987).
With (12.5.18) and (12.5.19) in place, the proof of Lemma 12.5.4 proceeds as
before. See (Luo, Rudnick and Sarnak, 1999) for precise details.
The case of a finite place and a number field is quite a bit more complicated.
The reason is that the condition that χ be 1 at a finite place p, imposes a
strong condition on the set of such χ s. To get around this one resorts to the
construction of special χ s which have conductors which are highly divisible
(Rohrlich, 1989), and in fact form a very sparse sequence. In this case in order
to execute the averaging one needs to use the positivity of the coefficients of
the Rankin–Selberg L-function L(s, f × f˜).

12.6 Strong multiplicity one theorem

As a final application of the Rankin–Selberg method, we give a proof of the
strong multiplicity one theorem, originally due to Jacquet and Shalika (1981).

Theorem 12.6.1 (Strong multiplicty one) Let f, g be two Maass forms for
S L(n, Z) as in (12.1.1) with Fourier coefficients

A(m 1 , . . . , m n−1 ), B(m 1 , . . . , m n−1 ),

respectively, with m 1 ≥ 1, . . . , m n−2 ≥ 1, m n−1 = 0. If

A( p, 1, . . . , 1) = B( p, 1, . . . , 1)

for all but finitely many primes p then f = g.

Proof If f = g, then the inner product  f, g = 0. By Theorem 12.1.4, the

Rankin–Selberg L-function L f ×g (s) (see Definition 12.1.2) has a meromorphic
continuation to all s ∈ C with at most a simple pole at s = 1 with residue
proportional to  f, g. By our assumption that f = g, it follows that L f ×g (s) is
entire and has no pole at s = 1. This is a contradiction because we are assuming
that A( p, 1, . . . , 1) = B( p, 1, . . . , 1) for all but finitely many primes p, which
implies by Theorem 12.1.3 (up to a finite number of Euler factors) that L f ×g (s)
is a Dirichlet series with positive coefficients, so it must have a pole. If it did
394 Rankin–Selberg convolutions

not have a pole then by standard techniques in analytic number theory the sum
+
A( p, 1, . . . , 1) · B( p, 1, . . . , 1) would be small as x → ∞.

p≤x

Theorem 12.6.1 can actually be proved in a much stronger form, i.e., it

still holds provided A( p, 1, . . . , 1) = B( p, 1, . . . , 1) for only finitely many
primes p. By using the logarithmic derivative L f ×g (s)/L f ×g (s), Moreno
(1985) obtained a general explicit bound on the number of primes required.
 13
Langlands conjectures

About 25 years ago I was discussing analytic number theory with Jean-Pierre
Serre. I distinctly recall how he went to the blackboard and wrote down the
Euler product

n
(1 − α p,i p −s )−1
p i=1

corresponding to an automorphic form on G L(n), and then pointed out that one
of the most important problems in the theory of L-functions was to obtain the
analytic properties of the higher kth symmetric power L-functions given by
    −1
1 − α p,i1 α p,i2 · · · α p,ik p −s .
p 1≤i 1 ≤i 2 ≤···≤i k ≤n

He then explained that if one knew that these L-functions (for k = 1, 2, 3, . . . )

were all holomorphic for (s) > 1 then it would easily follow from Landau’s
lemma (a Dirichlet series converges absolutely up to its first pole, see (Iwaniec-
Kowalski, 2004)) that
|α p,i | = 1
for all primes p and all i = 1, 2, . . . , n. This, of course, is the famous general-
ized Ramanujan conjecture discussed in Section 12.4.
In January 1967, while at Princeton University, Langlands hand wrote a 17
page letter to André Weil. The letter outlines what are now commonly known
as the “Langlands conjectures.” Weil suggested that the letter be typed and it
then circulated widely. It is now available at the Sunsite webpage:

http://www.sunsite.ubc.ca/DigitalMathArchive/Langlands/functoriality.html

The conjecture about the kth symmetric power L-functions shown to me by

Serre is a special case of Langlands more general conjectures. One may also

395
396 Langlands conjectures

consider the kth exterior power L-function given by

   −1
1 − α p,i1 α p,i2 · · · α p,ik p −s .
p i 1 <i 2 <···<i k

Langlands predicts that the kth

symmetric power  L-function is associated to
+
an automorphic form on G L 1 while the kth exterior power
1≤i 1 ≤i 2 ≤···≤i k ≤n


+
L-function is associated to an automorphic form on G L 1 ,
1<i 1 <i 2 <···<i k ≤n
i.e., it is on G L(M) where M is simply given by the number of Euler factors
in the Euler product. Although these conjectures are easy to state in terms of
Euler products, it is very hard to get a grip on them from this point of view.
The great insight of Langlands, in his letter to Weil, is to show that each L-
function associated to an automorphic form on G L(n), say, is also associated
to a certain representation of an infinite dimensional Lie group, and by taking
tensor powers or exterior powers of the representation one may validate the
predictions.
Langlands came to his conjectures by carefully studying Eisenstein series
and Artin L-functions. For example, we have shown in Theorem 10.8.6 that the
L-functions associated to a minimal parabolic Eisenstein series for S L(n, Z) are
simply a product of shifted Riemann zeta functions. It is not hard to construct
the higher symmetric and exterior powers which will again be products of
other shifted Riemann zeta functions. One can then validate the Langlands
conjectures by showing (with the method of templates, after Remark 10.8.7),
that the higher symmetric and exterior products satisfy the expected functional
equations.
Another compelling explanation for Langlands general conjectures, as
explained by Langlands himself, is the striking analogy with Artin L-functions
(Langlands, 1970). We will explore this analogy in the next two sections.
Langlands conjectures (see (Arthur, 2003), (Sarnak, to appear)) show that
all automorphic forms should be encoded in the G L(n) automorphic spectrum.
It also follows from (Arthur, 1989, 2002) that the decomposition of a general
group may be reduced to the study of G L(n). So for the theory of L-functions,
the group G L(n) plays an especially important role.
The conjectures we have described up to now (symmetric and exterior powers
of automorphic representations) are examples of the so-called “global Lang-
lands functoriality conjectures over number fields.”
 13.1 Artin L-functions 397

Progress on the Langlands global functoriality conjectures over number

fields has been slow, but with spectacular recent developments. Here is an
account of the major developments up to the present.
r (Gelbart-Jacquet, 1978) The symmetric square lift from G L(2) to G L(3) .
(see Section 7.3)
r (Ramakrishnan, 2002) The tensor product lift: G L(2) × G L(2) → G L(4).
r (Kim and Shahidi, 2000, 2002) Tensor product lift: G L(2) × G L(3)
→ G L(6), the symmetric cube lift: G L(2) → G L(4).
r (Kim, 2003), (Kim and Shahidi, 2000) The exterior square lift: G L(4)
→ G L(6), symmetric fourth power lift: G L(2) → G L(5).
r (Cogdell, Kim, Piatetski-Shapiro and Shahidi, 2001, 2004) Lifting from split
classical groups to G L(n).

The local Langlands conjectures (Carayol, 1992, 2000) have seen much
greater advances. In (Laumon, Rapoport and Stuhler, 1993), the local Lang-
lands conjecture was proved for local fields of prime characteristic. This was
followed by Harris and Taylor (2001) who gave a proof for characteristic zero
and then Henniart (2000) gave a simplified proof. In (Drinfeld, 1989) a proof
of Langlands functoriality conjecture was obtained for G L(2) over a function
field. Finally Lafforgue (2002) established Langlands conjectures for G L(n)
(with n ≥ 2) in the function field case.
Other references for Langlands conjectures include (Borel, 1979), (Bump,
1997), (Gelbart, 1984), (Arthur, 2003), (Bernstein and Gelbart, 2003), (Moreno,
2005).

13.1 Artin L-functions

Let K be an algebraic number field of finite degree (Galois extension) over
another number field k with Galois group G = Gal(K /k). Let

ρ : G → G L(V )

be a representation of G into a finite dimensional complex vector space V of

dimension n. Then ρ is a homomorphism from G into the group G L(V ) of
isomorphisms of V into itself. The group G L(V ) may be identified with
G L(n, C).
1
Example 13.1.1 (cubic field) Let K = Q(2 3 , e2πi/3 ), and k = Q. Then the
Galois group G = Gal(K /Q) is S3 , the symmetric group of all permutations of
398 Langlands conjectures

three objects. A representation ρ : G → G L(3, R) can be explicitly given as

follows. For α ∈ K and g ∈ G, we denote the action of g on α by α g . Similarly,
for α, β, γ ∈ K we define the binary operation ◦ by
⎛ ⎞ ⎛ g⎞
α α
⎝ β ⎠ ◦ g := ⎝ β g ⎠ .
γ γg

Now let
1 1 1
α = 23 , β = e2πi/3 · 2 3 , γ = e4πi/3 · 2 3 .

For each g ∈ G, we may define a matrix ρ(g) ∈ G L(3, R) by the identity

⎛ ⎞ ⎛ ⎞
α α
⎝ β ⎠ ◦ g = ρ(g) · ⎝ β ⎠
γ γ

where · denotes multiplication of a matrix by a vector. The matrices ρ(g), g ∈

G are just the six permutation matrices
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
1 0 0 0 1 0 0 0 1
⎝0 1 0⎠, ⎝0 0 1⎠, ⎝1 0 0⎠
0 0 1 1 0 0 0 1 0
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
0 1 0 0 0 1 1 0 0
⎝1 0 0⎠, ⎝0 1 0⎠, ⎝0 0 1⎠.
0 0 1 1 0 0 0 1 0

Artin L-function – preliminary definition

Let K /k be a Galois extension with Galois group G. Let ρ : G → G L(V )
be a representation of G as above. An Artin L-function is a meromorphic
function of a complex variable s denoted L(s, ρ, K /k) attached to this data.
The precise definition gives a realization of L(s, ρ, K /k) as an Euler product
and requires the interplay between the representation ρ and prime numbers,
which is determined by the Frobenius automorphism which we now discuss.

Definition 13.1.2 (Frobenius automorphism of a finite field extension) Let

Fq , Fq  be finite fields of prime power orders q, q  , respectively, where q|q  . The
map x → x q fixes Fq and permutes the elements of Fq  in such a way as to give
an automorphism of Fq  . This map is defined to be the Frobenius automorphism
of Fq  /Fq . It is a particular element of Gal(Fq  /Fq ).
 13.1 Artin L-functions 399

Let K /k be a Galois extension of number fields of degree n. Set O K , Ok to

be the ring of integers of K , k, respectively. Let p be a prime ideal of Ok . Then
the ideal pO K factors into powers of prime ideals Pi (i = 1, . . . , r ) of O K as
follows:

pO K = P1e1 · · · Prer . (13.1.3)

If we apply g ∈ G to this, we get

 g e  e
pO K = P1 1 · · · Prg r . (13.1.4)

By unique factorization, the two factorizations in (13.1.3), (13.1.4) must be

the same. By varying g ∈ G this implies that all the e j must be equal to some
integer e ≥ 1. Therefore,

pO K = P1e · · · Pre

and n = e f r. We say p is ramified if e > 1, otherwise, it is unramified.

Let P be one of the primes Pi , (i = 1, . . . , r ) which occur in the above
factorization. We define the decomposition group DP , as the set
 
DP := g ∈ G  P g = P .

The decomposition group tells us how pO K splits. If we write G in terms of

cosets of DP :
(

G= DP · gi ,
i=1

where = [G : DP ] then the distinct conjugate divisors to P are just the divisors
P gi , (1 ≤ i ≤ ). It follows that = r and the order of DP is e f. For unramified
primes this is just f .
Consider the residue fields FP := O K /P and Fp := Ok /p, respectively.
Then FP , Fp are finite fields.It is not hard to see that the elements of the
decomposition group DP are automorphisms of FP /Fp. Indeed, if g ∈ DP and

α≡β (mod P),

then α g ≡ β g (mod P g ). But P g = P so that

α g ≡ β g (mod P).

Consequently, elements of DP take congruence classes (mod P) to congruence

classes (mod P), which gives a homomorphism from DP into Gal(FP /Fp). Let
 
IP := g ∈ G  x g ≡ x (mod P), ∀x ∈ O K ,
400 Langlands conjectures

denote the inertia group of P. It turns out that the map

DP /IP → Gal(FP /Fp)
is onto. It follows that there is an element of DP which maps onto the Frobenius
automorphism in Gal(FP /Fp). Naturally, we shall call this element (denoted
FrP ) the Frobenius automorphism also. It is characterized by
x FrP ≡ x N (p) (mod P) (13.1.5)
for all x ∈ O K . Here N (p) denotes the absolute norm of p which is the cardinal-
ity of the finite field Fp. Note that the Frobenius automorphism is well defined
only modulo the inertia.
The prime p ∈ Ok is unramified in K /k if and only if IP is trivial (contains
only the identity element) for any prime P occurring in the factorization of
pO K . Similarly, the prime pO K factors into n prime ideals (splits completely)
in K /k if and only if DP = IP = {1}.
Lemma 13.1.6 Let K /k be a Galois extension of number fields with Galois
group G. Let g ∈ G and let P in O K be a prime above the prime p in Ok . If
FrP is the Frobenius automorphism determined by (13.1.5) then
Fr g = g −1 · FrP · g.
P

If DP , DP g denote the decomposition groups of P, P g , respectively, then

DP g = g −1 · DP · g.

Proof If we apply g ∈ G to (13.1.5), we otain

x FrP ·g ≡ (x g ) N (p) (mod P g ),
−1
for all integers x ∈ O K . If we replace x by x g , it follows that
−1
·FrP ·g
xg ≡ x N (p) (mod P g ).

Since the Galois group G is transitive on primes P lying over p, it follows

that all the Frobenius elements FrP are conjugate. Thus, attached to the prime
p of Ok is a conjugacy class of Frobenius elements in Gal(K /k). We are now
ready to define the Artin L-function.
Definition 13.1.7 (Artin L-function) Let K /k be a Galois extension of num-
ber fields with Galois group G. For a finite-dimensional complex vector space
V , let ρ : G → G L(V ) be a representation. Let s ∈ C with (s) sufficiently
 13.1 Artin L-functions 401

large. The Artin L-function (denoted L(s, ρ, K /k)) attached to this data is given
by the Euler product
  −1    −1
Det I − ρ (FrP ) N (p)−s Det I − ρ (FrP ) V IP N (p)−s ,
p unramified p ramified

where I is the identity matrix and P is any prime above p. Furthermore, for p
ramified, the quotient DP /IP acts on the subspace V IP of V on which IP acts
trivially. The notation ρ (FrP ) |V IP means that the action of ρ (FrP ) is restricted
to V IP .

Note The indicated determinant in Definition 13.1.7 is well defined (inde-

pendent of the choice of P above p) by Lemma 13.1.6, since the determinant
only depends on the conjugacy class of FrP . To see this note that for any g ∈ G,
since the determinant is a multiplicative function and Det(g −1 · g) = 1,
 
Det I − ρ (FrP ) N (p)−s = Det g −1 · (I − ρ(FrP )N (p)−s ) · g

= Det I − (g −1 · ρ(FrP ) · g)N (p)−s

= Det I − ρ(FrP g )N (p)−s . (13.1.8)

Here, the last equality follows from Lemma 13.1.6.

These L-functions were first introduced in (Artin, 1923, 1930) (see also
(Roquette, 2000)). Artin made the following famous conjecture.

Conjecture 13.1.9 (Artin’s conjecture) Let K /k be a Galois extension of

number fields with Galois group G. For a finite-dimensional complex vector
space V , let ρ : G → G L(V ) be a representation. If the representation ρ is
irreducible and not the trivial representation, then L(s, ρ, K /k) is an entire
function of s ∈ C.

Artin himself proved Conjecture 13.1.9 when ρ is one-dimensional in (Artin,

1927). To quote from (Langlands, 1970):

Artin’s method is to show that in spite of the differences in the definitions the function
L(s, ρ, K /F) attached to a one-dimensional ρ is equal to a Hecke L-function L(s, χ)
where χ = χ (ρ) is a character of F ∗ \I F . He employed all the available resources
of class field theory, and went beyond them, for the equality of L(s, ρ).

It is not hard to show (see Heilbronn’s article in (Cassels and Fröhlich, 1986))
that Artin’s L-function L(s, ρ, K /k) can be expressed as a product of rational
powers of abelian L-functions of Hecke’s type, where the abelian L-functions
are associated to intermediate fields k ⊆  ⊆ K with K /  abelian. A major
advance on Artin’s conjecture in the case when Gal(K /k) is not an abelian
402 Langlands conjectures

group was made by Brauer (1947) who proved that all irreducible represen-
tations ρ of the Galois group G can be expressed as Z-linear combinations
of induced representations of one-dimensional representations on subgroups
of G. As a consequence, he showed that L(s, ρ, K /k) can be expressed as a
product of integer powers of abelian L-functions of Hecke’s type. This proved
that Artin’s L-functions extended to meromorphic functions in the entire com-
plex s plane and satisfied a functional equation. The problem was that Brauer
could not exclude negative integral powers so that Artin’s conjecture was still
unproven.
Further advances on Artin’s Conjecture 13.1.19 did not come until Langlands
changed the entire landscape of research around this problem by making the
striking conjecture that Artin’s L-functions should be L-functions associated to
Maass forms on G L(n).
When n = 2 and the image of ρ in P G L(2, C) is a solvable group, Artin’s
conjecture was solved in (Langlands, 1980). The ideas in this paper played a
crucial role in Taylor and Wiles (1995) proof of Fermat’s Last Theorem. In
(Langlands, 1980) the conjecture is proved for tetrahedral and some octahedral
representations and in (Tunnell, 1981) the results are extended to all octahedral
representations. When n = 2 and the projective image is not solvable, the only
possibility is that the projective image is isomorphic to the alternating group A5 .
These representations are called icosahedral because A5 is the symmetric group
of the icosahedron. Joe Buhler’s Harvard Ph.D. thesis (see (Buhler, 1978)) gave
the first example where Artin’s conjecture was proved for an icosahedral repre-
sentation. The book (Frey, 1994) proves Artin’s conjecture for seven icosahedral
representations (none of which are twists of each other). In (Buzzard and Stein,
2002), the conjecture is proved for eight more examples. A further advance
was made in (Buzzard, Dickinson, Shepherd-Barron and Taylor, 2001) who
proved Artin’s conjecture for an infinite class of icosahedral Galois represen-
tations which were disjoint from the previous examples.Very little is known
for n > 2.

13.2 Langlands functoriality

The converse Theorem 3.15.3 of Hecke–Maass and the Gelbart–Jacquet sym-
metric square lift (Theorem 7.3.2) from G L(2) to G L(3) have been the principle
motivation for writing this book. These results constitute one of the first impor-
tant proofs of a special case of Langlands conjectures.
The Gelbart–Jacquet lift, or more accurately: the Gelbart–Jacquet functorial
image, is an instance of Langlands functoriality. While it is beyond the scope
of this book to give a definition of functoriality in the most general scenario,
 13.2 Langlands functoriality 403

we shall attempt to motivate the definition and give a feeling for the important
program that Langlands has created.
The key new idea introduced by Langlands in his 1967 letter to Weil, and also
introduced independently in (Gelfand, Graev and Pyatetskii-Shapiro, 1990) is
the notion of an automorphic representation. We have intensively studied
Maass forms for S L(n, Z). These are examples of automorphic forms. The
leap to automorphic representation is a major advance in the subject with
profound implications.It was first intensively researched, for the case of G L(2),
in (Jacquet and Langlands, 1970).
In the interests of notational simplicity and an attempt by the writer to
explain in as simple a manner as possible the ideas behind the functorial-
ity conjectures of Langlands, we shall restrict our discussion, for n ≥ 2, to
the group S L(n, Z) which acts on G L(n, R) by left matrix multiplication. An
automorphic form is then a Maass form for S L(n, Z) as in Definition 5.1.3, a
Langlands Eisenstein series for S L(n, Z) as in Chapter 10, or the residue of such
an Eisenstein series. By Langlands spectral theorem (see (Langlands, 1966) and
also Theorem 10.13.1 for the case of G L(3)) these automorphic forms generate
the C-vector space
 < 
Vn := L2 S L(n, Z) G L(n, R) O(n, R) · R× .

We introduce the right regular representation which maps g ∈ G L(n, R) to

the endomorphism, F → ρ(g)F of Vn , and is defined by

ρ : G L(n, R) → End(Vn ),

where

(ρ(g)F)(z) := F(z · g)

for all F ∈ Vn , z ∈ hn = G L(n, R)/(O(n, R) · R× ), and all g ∈ G L(n, R).

Remark This is a representation into the endomorphisms of an infinite

dimensional vector space! The space of Maass forms (cusp forms) is invari-
ant under this representation. It decomposes into an infinite direct sum of irre-
ducible invariant subspaces. If π is the representation on one of these invariant
subspaces then π is termed an automorphic cuspidal representation and corre-
sponds to a Maass form. The L-function associated to π is then the L-function
associated to the Maass form as in Definition 9.4.3, i.e., it is a Godement–Jacquet
L-function.

At this point Langlands made a remarkable hypothesis which may be viewed

as a naive form of his functoriality conjecture. He assumed:
404 Langlands conjectures

r that the properties of an automorphic representation mimic the properties

of a Galois representation;
r that the properties of an L-function associated to an automorphic form
(automorphic representation) mimic the properties of an Artin L-function.

We shall illustrate these hypotheses with some simple examples. Let K /k

be a Galois extension of number fields with Galois group G. For a finite-
dimensional complex vector space V , let ρ : G → G L(V ) be a representation,
and consider the Artin L-function L(s, ρ, K /k) given in Definition 13.1.7. If
we let g ∈ G L(n, R) and then use the first two identities in (13.1.8), it follows
that, for p unramified, we may diagonalize the matrix ρ (FrP ), i.e.,
⎛ ⎞
λp,1
⎜ .. ⎟
g −1 · ρ (FrP ) · g = ⎝ . ⎠,
λp,n
where λp,1 , λp,2 , . . . , λp,n are the eigenvalues. A similar result holds when p is
ramified. The Euler product for the Artin L-function L(s, ρ, K /k) then takes
the form
 n
L(s, ρ, K /k) = (1 − λp,i N (p)−s )−1 .
p i=1

Now, it is possible to combine two Galois representations (of the type ρ

above) and create a new Galois representation just as when we multiply two
numbers to create a new number. In fact, there are many ways to do this. Such
laws of composition have interesting images on the L-function side. One may
think of the L-functions associated to Maass forms as the basic atoms which
can be combined in various ways to form molecules, i.e., more complicated
L-functions.
Let ρ : G → G L(V ), ρ  : G → G L(V  ) be two Galois representations of
the Galois group G = Gal(K /k) where V, V  are vector spaces (over C) of
dimensions n, n  , respectively. Let

n
L(s, ρ, K /k) = (1 − λp,i N (p)−s )−1 ,
p i=1


n
 −1

L(s, ρ , K /k) = 1 − λp,i N (p)−s ,
p j=1

be the corresponding Artin L-functions. An interesting way to combine ρ, ρ  is

to form the tensor product ρ ⊗ ρ  which is a representation of G into G L(V ⊗
V  ). Every element of V ⊗ V  is a linear combination of terms of the form
 13.2 Langlands functoriality 405

v ⊗ v  with v ∈ V, v  ∈ V  . Then ρ ⊗ ρ  is defined by letting

(ρ ⊗ ρ  )(g)(v ⊗ v  ) := ρ(g)(v) ⊗ ρ(g)(v  )
for all g ∈ G, v ∈ V, v  ∈ V  . Without loss of generality, we may assume
that for some g ∈ G, the representations ρ, ρ  have been diagonalized so
that ρ(g), ρ  (g) correspond to diagonal matrices in G L(n, C), G L(n  , C),
respectively, where
⎛ ⎞ ⎛  ⎞
λ1 λ1
⎜ .. ⎟ ⎜ .. ⎟
ρ(g) = ⎝ . ⎠, ρ  (g) = ⎝ . ⎠. (13.2.1)
λn λn 
So, if (e1 , . . . , en ) and (e1 , . . . , en  ) denote the standard bases for V, V  , respec-
tively, then it is clear that ρ(g) maps ei → λi ei while ρ  (g) maps ej → λj ej .
It then follows that the tensor product of the two representations (ρ ⊗ ρ  )(g)
maps
ei ⊗ ej → λi λj ei ⊗ ej (for 1 ≤ i ≤ n, 1 ≤ j ≤ n  ).
Consequently, we have shown that

n 
n

L(s, ρ ⊗ ρ , K /k) = (1 − λp,i λp, j N (p)−s )−1 .
p i=1 j=1

But L(s, ρ ⊗ ρ  , K /k) is another Artin L-function!It is also the Rankin–Selberg

convolution as defined by the Euler product representation in Theorem 12.3.6.
So if Langlands hypothesis (that automorphic representations mimic Galois
representations) is correct it would have to follow that the Rankin–Selberg
convolution of two automorphic representations is again automorphic. This is
an important example of Langlands functoriality.
Another example comes from taking the symmetric product V ∨ V  of two
vector spaces V, V  defined over C. The vector space V ∨ V  then consists of
all linear combinations of terms of the form v ∨ v  with v ∈ V, v  ∈ V  , and
where ∨ satisfies the rules
v ∨ v  = v  ∨ v,
(a1 v1 + a2 v2 ) ∨ v  = a1 v1 ∨ v  + a2 v2 ∨ v  ,
for all v ∈ V, v  ∈ V  , and a1 , a2 ∈ C. If ρ : G → G L(V ), ρ  : G → G L(V  )
are two Galois representations of G = Gal(K /k), then we may consider the
symmetric product representation ρ ∨ ρ  which is a representation of G into
G L(V ∨ V  ). Then ρ ∨ ρ  is defined by letting
(ρ ∨ ρ  )(g)(v ∨ v  ) := ρ(g)(v) ∨ ρ(g)(v  )
406 Langlands conjectures

for all g ∈ G, v ∈ V, v  ∈ V  . Let us consider the special case that V = V  and

ρ = ρ  where V has the basis e1 , . . . , en . We may view V ∨ V as a subspace
of V ⊗ V with basis elements ei ⊗ e j where 1 ≤ i ≤ j ≤ n. If we assume as
before that ρ(g), ρ  (g) correspond to diagonal matrices as in (13.2.1) then, in
this case, we may consider the symmetric square representation ρ ∨ ρ. It is
easy to see that
 
L(s, ρ ∨ ρ, K /k) = (1 − λp,i λp, j N (p)−s )−1 .
p 1≤i≤ j≤n

Other examples of this type can be given by considering, for example, the
exterior product of two vector spaces (see Section 5.6), and then forming the
exterior product of two Galois representations or by taking higher symmetric
or exterior powers.
Yet another type of interesting operation that can be done with representa-
tions is to consider induced representations. This corresponds to induction from
a Galois subgroup (see (Bump, 1997). Langlands derived from this process his
theory of base change (Langlands, 1980).
An even deeper theorem in Galois representations is Artin’s reciprocity law
(Artin, 1927) (see also the introductory article (Lenstra-Stevenhagen, 2000))
which generalized Gauss’ law of quadratic reciprocity, and included all known
reciprocity laws up to that time. Langlands formulated an even more general
version of Artin reciprocity in the framework of automorphic representations.
Of course, Langlands did not stop here. This was the starting point. For a gen-
eral connected reductive algebraic group G he introduced, in (Langlands, 1970),
the dual group or what is now known as the L-group, L G.He then formulated
his now famous principle of functoriality which states that given any two con-
nected reductive algebraic groups H, G and an L-homomorphism L H → L G
then this should determine a transfer or lifting of automorphic representations
of H to automorphic representations of G.
In fact, the modern theory of automorphic forms allows us to parameterize
each automorphic representation of a reductive group such as H by a set of
semisimple conjugacy classes {cv } in LH , a complex Lie group, where v runs
over almost all the places of the defining global field. The functoriality prin-
ciple then roughly states that for each “L–homomorphism” f LH → LG, the
collection { f (cv )} defines, in fact, an automorphic representation of G, or more
precisely a “packet” of automorphic representations of G.
 List of symbols

,  d∗z gl(n, R)
 , ⊗ (Rn )  G L(2, R)
 ,  (Rn ) 1 G L(3, R)
!, " 2 G L(n, R)
⊗ (Rn ) (z) G L(n, Z)
∧ Det G L(V )
∨ eψ Gw
⊗ E(z, s) Gw
An 1 ,...,nr E ∗ (z, s) γa
A(m 1 , . . . , m n−1 ) E(z, s) ∞
α p,i E(z, s, χ) (N )
αp E ∗ (z, s, χ) 0 (N )
Bn E P (z, s) w
bi, j E ∗P (z, s) h(d)
B(s1 , s2 ) E Pmin (z, s) h
C G () E P (z, s, φ) hn
C+∞ Em (z, s) ♥
χ (n) Ei j In
D d IP
Dn fˆ Ir
Dn f˜ Is
Dα Fq Iν (z)
Di j F(z, h, q) Iν (z, Pn 1 ,...,nr )
DP FrP Ks
d(n) G ν (s) K ν (y)
d(z, z ) G̃ ν (s) k(u)
d ∗u gcd k(z)

407
408 List of symbols

K (z, w) O(n, R) S O(n, R)

K (z, z  ) O K , Ok Sλ
L b(s, ψ n ) ωi S(T )
L χ (s) (P) σi
L(s, ρ, K /k) (P, P  ) σs (n)
L E(∗,w) (s) PM (y, z  ) σa
L Eν (s) Pr a,b
L f (s) P̂n,r T (L)
L f˜ (s) Pn 1 ,...,nr Tg
L f ×g (s) φ̂ m (z) Tn
Lk ψm τ (χ )
Lk+1 ψ(u) θ (z)
L(s, ∨k ) P(z, s, eψ ) θχ (z)
L2 (S L(2, Z)\h2 ) Q U
√
L2cusp (S L(2, Z)\h2 ) Q( −D) U (L)
L2cont (S L(2, Z)\h2 ) qw Un (R)
L2 (S L(n, Z)\hn ) R Un (Z)
λD Rr,δ Uw
nb (s) R, Ũ w
 E(∗,w) (s) ρ(g) u(z, z  )
 f (s) S µ(n)
 f ×g (s) Sn v j,k
 (Rn ) Sn Vn
M S(n, c) W (z)
M(n, R) Sk1 ,k2 (x1 , x2 , x3 ) W (z, ν, ψ)
Mn 1 ,...,nr Sk (x1 , . . . , xn ) WJacquet (z, ν, ψ)
∗
Mn 1 ,...,nr Sw (ψ, ψ  , c) WJacquet (z, ν, ψ)
M(g) Sw (ψ M , ψ N , c) w0
Mλ (g) S L(2, R) wn
mP S L(3, R) Wn
ni S L(n, R) Z
Ni S L(2, Z) Zn
Nn 1 ,...,nr S L(3, Z) Z (ψ, ψ  , s)
O(2, R) S L(n, Z) ζ (s)
 Appendix The GL(n)pack Manual
Kevin A. Broughan

A.1 Introduction
This appendix is the manual for a set of functions written to assist the reader to
understand and apply the theorems on GL(n, R) set out in the main part of the
book. The software for the package is provided over the world wide web at

http://www.math.waikato.ac.nz/∼kab

and is in the form of a standard Mathematica add-on package. To use the

functions in the package you will need to have a version of Mathematica at
level 4.0 or higher.

A.1.1 Installation
First connect to the website given in the paragraph above and click on the link
for GL(n)pack listed under “Research” to get to the GL(n)pack home page.
Instructions on downloading the files for the package will be given on the home
page. If you have an earlier version of GL(n)pack first delete the file gln.m,
the documentation gln.pdf and the validation program glnval.nb. The name of
the file containing the package is gln.m. To install, if you have access to the file
system for programs on your computer, place a copy of the file in the standard
repository for Mathematica packages – this directory is called “Applications”
on some systems. You can then type gln.m and then press the Shift and Enter
keys to load the package. You may need administrator or super-user status to
complete this installation. Alternatively place the package file gln.m anywhere
in your own file system where it is safe and accessible.

Instructions for Windows systems The package can be loaded by typing

409
410 Appendix The GL(n)pack Manual

SetDirectory[”c:\your\directory\path”];
<<gln.m;ResetDirectory[]<Shift/Enter>

or

Get[”gln.m”,Path->{”c:\your\directory\path”}]
<Shift/Enter>

where the path-name in quotes should be replaced by the actual path-name of

directories and subdirectories which specify where the package has been placed
on a given computer.

Instructions for Unix/Linux systems These are the same as for Windows,
but the path-name syntax style should be like /usr/home/your/subdirectory.

Instructions for Macintosh systems These are the same as for Windows,
but the path-name syntax style should be like HD:Users:ham:Documents:.

All systems The package should load printing a message. The functions of
GL(n)pack are then available to any Mathematica notebook you subsequently
open.

A.1.2 About this manual

This appendix contains a list of all of the functions available in the pack-
age GL(n)pack followed by a manual entry for each function in alphabet-
ical order. Many functions contain the transcript of an example and refer-
ence to the part of the text to which the function relates, as well as lists of
related functions. Each function has both a Mathematica style long name and a
3 letter/digit abbreviated name. Either can be used, but the error messages and
usage information are all in terms of the long names. To obtain information
about bug fixes and updates to GL(n)pack consult the website for the package:
given in A.I. above.

A.1.3 Assistance for users new to computers or Mathematica

On the GL(n)pack website (see above) there are links giving tutorial and other
information for those people who want access to the package but are new
or relatively new to computers. The manual entries assume familiarity with
Mathematica, so some may require extra help. Alternatively sit down with
someone familiar with Mathematica to see it at work.
 A.1 Introduction 411

There are many issues to do with computer algebra and mathematical soft-
ware that will arise in any serious evaluation or use of Mathematica and
GL(n)pack. A comment on one aspect: GL(n)pack function arguments are first
evaluated and then checked for correct data type. If the user calls a function
with an incorrect number of arguments or an argument of incorrect type, rather
than issue a warning and proceeding to compute (default Mathematica style),
GL(n)pack prints an error message, aborts the evaluation and returns the user
to the top-level, no matter how deeply nested the function which makes the
erroneous call happens to be placed. This is a tool for assisting users to debug
programs which include calls to this package.

A.1.4 Mathematica functions

Each GL(n)pack function is a standard Mathematica function and so will work
harmoniously with built-in Mathematica functions and user functions. Useful
standard functions include those for defining functions (Module and Block for
example) list and matrix manipulation and operations (“.” represents matrix
multiplication), special functions (such as the BesselK), plotting functions
and the linear algebra add-on package. Note however that a formatted matrix,
returned by MatrixForm, is not recognized by Mathematica as a matrix. A
matrix in Mathematica is just a list of lists of equal length.

A.1.5 The data type CRE (Canonical Rational Expression)

Many GL(n)pack functions take symbolic arguments which are either explicit
integers or real or complex numbers (exact or floating point) or mathematical
expressions which could evaluate to numbers. These expressions are expected
to be in the class sometimes called “Canonical Rational Expressions” or CREs.
This class of expression is defined as follows: members are rational functions
with numerical coefficients and with symbolic variables, any number of which
may be replaced by function calls, or functions which are not evaluated (“noun
forms”) or functions of any finite number of arguments each of which can
be, recursively, a CRE. Some package functions will accept lists of CREs or
matrices with CRE elements. This should cover most user needs, but notice
it excludes simple types like matrices with elements which are matrices. (The
single exception is the GL(n)pack function MakeBlockMatrix, which takes as
argument a matrix with matrix elements.) If a user is unsure regarding the data
type of a mathematical expression, the GL(n)pack function CreQ can be used.
See the manual entry.
412 Appendix The GL(n)pack Manual

A.1.6 The algorithms in this package

The reader who uses this package may notice that many functions
appear to run almost instantaneously, for example MakeBlockMatrix or
HeckeMultiplicativeSplit. Others however take considerable time to complete,
minutes or hours rather than seconds. This is often because the underlying algo-
rithm employed is exponential, or in at least one instance, more than exponential.
Improvements in this completion time are of course possible: The GL(n)pack
code is interpreted, so there may be speed-ups attainable using the Mathematica
function Compile, even though it has a restricted domain of application. Exist-
ing algorithms could be replaced by faster algorithms. The existing algorithms
could be re-implemented in a compile-load-and-go language such as C++ or
Fortran, or an interactive language allowing for compilation such as Common
Lisp. This latter would be the best choice, because execution speed for com-
piled code is quite comparable to that of the two former choices, but its range
of data types is vast, certainly sufficient for all of the package. Functions which
will slow significantly as the dimension increases include GetCasimirOperator,
ApplyCasimirOperator, KloostermanSum, MPSymmetricPowerLFun, and
SpecialWeylGroup.

A.1.7 Acknowledgements
Kevin Broughan had assistance from Columbia University and the University
of Waikato and a number of very helpful individuals while writing the package.
These included Ross Barnett, Mike Eastwood, Sol Friedberg, David Jabon (for
the Smith Normal Form code), Jeff Mozzochi and Eric Stade, in addition to
Dorian Goldfeld, who’s detailed text and explicit approach made it possible.
Their contribution is gratefully acknowledged.
 A.2 Functions for GL(n)pack 413

A.2 Functions for GL(n)pack

ApplyCasimirOperator[m,expr,iwa] (aco): The operator acts on a CRE.
BruhatCVector[a] (bcv): The minors (c1 , . . . , cn−1 ).
BruhatForm[a] (bru): The four Bruhat factors of a symbolic matrix.
BlockMatrix[a,rows,cols] (blm): Extract a general sub-block of a matrix.
CartanForm[a] (car): The two Cartan factors of a numeric matrix.
ConstantMatrix[c,m,n] (com): Construct a constant matrix of given size.
CreQ[e] (crq): Check a Canonical Rational Expression.
DiagonalToMatrix[d] (d2m): Convert a list to a diagonal matrix.
EisensteinFourierCoefficient[z,s,n] (efc): The GL(2) Fourier series nth term.
EisensteinSeriesTerm[z,s,ab] (est): The nth term of the series for GL(2).
ElementaryMatrix[n,i,j,c] (elm): Construct a specified elementary matrix.
FunctionalEquation[vs, i] (feq): Generate the affine parameter maps.
GetCasimirOperator[m,n,“x”,“y”,“f”] (gco): The Casimir operators.
GetMatrixElement[a, i, j] (gme): Return a specified element.
GlnVersion[] (glv): Print the date of the current version.
HeckeCoefficientSum[m, ms, “x”](hcs): The right-hand side of the sum.
HeckeEigenvalues[m,n,“a”] (hev): The values of (λm ) for G L(n).
HeckeMultiplicativeSplit[m](hms): Prepare a Hecke Fourier coefficient.
HeckeOperator[n, z,“f”] (hop): of nth order for forms on hn .
HeckePowerSum[e, es,“B”](hps): Exponents for the Hecke sum at any prime.
HermiteFormLower[a] (hfl): The lower left Hermite form.
HermiteFormUpper[a] (hfu): The upper left Hermite form.
IFun[ν,z] (ifn): The power function Iν (z).
InsertMatrixElement[e,i,j,a] (ime): Insert an expression in a given matrix.
IwasawaForm([a] (iwf): The product of the Iwasawa factors of a matrix.
IwasawaXMatrix[w] (ixm): Get the x matrix from the Iwasawa form.
IwasawaXVariables[w] (ixv): Get the x variables from the Iwasawa form.
IwasawaYMatrix[z] (iym): Get the y matrix from the Iwasawa form.
IwasawaYVariables[z] (iyv): Get the y variables from the Iwasawa form.
IwasawaQ[z] (iwq): Test to see if a matrix is in Iwasawa form.
KloostermanBruhatCell[a,x,c,w,y] (kbc): Solve a = x.c.w.y for x and y.
KloostermanCompatibility[t1,t2,c,w] (klc): Relations for a valid sum.
KloostermanSum[t1,t2,c,w] (kls): Compute an explicit Kloosterman sum.
LanglandsForm[p,d] (llf): The three matrices of the decomposition.
LanglandsIFun(g,d,s) [lif]: Summand for the Eisenstein series.
LeadingMatrixBlock[a,i,j] (lmb): Extract a leading sub-block of a matrix.
LongElement[n] (lel): Construct the matrix called the long element.
LowerTriangleToMatrix[l] (ltm): Construct a lower triangular matrix.
414 Appendix The GL(n)pack Manual

MakeBlockMatrix[mlist] (mbm): Construct a matrix from submatrices.

MakeMatrix[“x”,m,n] (mkm): Make a matrix with indexed elements.
MakeXMatrix[n,“x”] (mxm): Construct a symbolic unimodular matrix.
MakeXVariables[n,“x”] (mxv): Construct a list of Iwasawa x variables.
MakeYMatrix[n,“y”] (mym): Construct a symbolic diagonal matrix.
MakeYVariables[n,“y”] (myv): Construct Iwasawa y variables.
MakeZMatrix[n,“x”,“y”] (mzm): Construct a symbolic Iwasawa z matrix.
MakeZVariables[n,“x”,“y”] (mzv): A list of the x and y variables.
MatrixColumn[m,j] (mcl): Extract a column of a given matrix.
MatrixDiagonal[a] (mdl): Extract the diagonal of a matrix.
MatrixJoinHorizontal[a,b] (mjh): Join two matrices horizontally.
MatrixJoinVertical[a,b] (mjv): Join two matrices vertically.
MatrixLowerTriangle[a] (mlt): Extract the lower triangular elements.
MatrixRow[m,i] (mro): Extract a row of a matrix.
MatrixUpperTriangle[a] (mut): Extract the upper triangular elements.
ModularGenerators[n] (mog): Construct the generators for S L(n, Z).
MPEisensteinGamma[s,v] (eig): Gamma factors for a parabolic series.
MPEisensteinLambdas[v] (eil): The λi (v) shifts.
MPEisensteinSeries[s,v] (eis): Minimal parabolic Eisenstein series.
MPExteriorPowerGamma[s,v,k] (epg): Exterior power gamma factors.
MPExteriorPowerLFun[s,v,k] (epl): Minimal parabolic exterior power.
MPSymmetricPowerGamma[s,v,k] (spg): Symmetric power gamma.
MPSymmetricPowerLFun[s,v,k] (spf): Minimal parabolic symmetric power.
NColumns[a] (ncl): The column dimension of a matrix.
NRows[a] (nro): The row dimension of a matrix.
ParabolicQ[p,d] (paq): Test a matrix for membership in a given subgroup.
PluckerCoordinates[a] (plc): Compute the bottom row-based minors.
PluckerInverse[Ms] (pli): Compute a matrix with given minors.
PluckerRelations[n,v] (plr): Compute quadratic relations between minors.
RamanujanSum[n,c] (rsm): Evaluate the Ramanujan sum s(n, c).
RemoveMatrixColumn[a,j] (rmc): Remove a matrix column.
RemoveMatrixRow[a,i] (rmr): Remove a row of a matrix non-destructively.
SchurPolynomial[k,x] (spl): The Schur multinomial Sk (x1 , . . . , xn ).
SmithElementaryDivisors[a] (sed): Smith form elementary divisors.
SmithForm[a] (smf): Compute the Smith form of an integer matrix.
SmithInvariantFactors[a] (sif): Smith form invariant factors.
SpecialWeylGroup[n] (swg): Weyl integer rotation group with det 1 elements.
SubscriptedForm[e] (suf): Print arrays with integer arguments as subscripts.
SwapMatrixColumns[a,i,j] (smc): Return a new matrix.
 A.2 Functions for GL(n)pack 415

SwapMatrixRows[a,i,j] (smr): Return a new matrix with rows swapped.

TailingMatrixBlock[a,i,j] (tmb): Extract a tailing matrix block.
UpperTriangleToMatrix[u] (utm): Form an upper triangular matrix.
VolumeBall[r,n] (vbl): The volume of a ball in n-dimensions.
VolumeFormDiagonal[“a”,n] (vfd): The volume form for diagonal matrices.
VolumeFormGln[“g”,n] (vfg): The volume form for G L(n).
VolumeFormHn[“x”,“y”,n] (vfh): Volume form for the upper half-plane.
VolumeFormUnimodular[“x”,n) (vfu): The form for the unimodular group.
VolumeHn[n] (vhn): The volume of the generalized upper half-plane.
VolumeSphere[r,n] (vsp): The volume of a sphere in n-dimensions.
Wedge[f1 , · · · , fn ] (weg): The wedge product and the d operator.
WeylGenerator[n,i,j] (wge): Each matrix generator for the Weyl group.
WeylGroup[n] (wgr): Compute the Weyl group of permutation matrices.
Whittaker[z,v,psi] (wit): Compute the function WJacquet symbolically.
WhittakerGamma[v] (wig): Gamma factors for the Whittaker function.
WMatrix[n] (wmx): The long element matrix with determinant 1.
ZeroMatrix[m,n] (zmx): The zero matrix of given dimensions.
416 Appendix The GL(n)pack Manual

A.3 Function descriptions and examples

 ApplyCasimirOperator (aco)
This function computes the Casimir operator acting on an arbitrary expression,
or undefined function, of a matrix argument in Iwasawa form, and optionally
other parameters, with respect to the Iwasawa variables, which must be specified
by giving the matrix in Iwasawa form.
To simply compute the operator it is easier to use the function GetCasimir-
Operator. Because this function uses symbolic differentiation, only functions
or expressions which can be differentiated correctly and without error by Math-
ematica may be used as valid arguments. Note also that no check is made that
the user has entered valid Mathematica variables. Since all arguments are eval-
uated it is good practice to use the function Clear to ensure arguments which
should evaluate to themselves do so.
See Proposition 2.3.3, Example 2.3.4 and Proposition 2.3.5.

ApplyCasimirOperator[m, expr, iwa] −→ value

m is a positive integer with value 2 or more being the order of the operator,
expr is a Mathematica expression, normally in the Iwasawa matrix or variables
and other parameters, which can be symbolically differentiated,
iwa is a numeric or symbolic matrix in Iwasawa form,
value is an expression or number being the result of applying the Casimir
operator in the Iwasawa variables to the expression.

Example

See also: GetCasimirOperator, MakeZMatrix, IwasawaForm, IwasawaQ.

 A.3 Function descriptions and examples 417

 BlockMatrix (blm)
This function returns a specified sub-block of a matrix. The entries of the sub-
block must be contiguous.
Block matrices are used in a number of places, but most especially in
Chapter 10 on Langlands Eisenstein series.

BlockMatrix[a, rows, columns] −→ b

a is a matrix of CREs,
rows is a list of two valid row indices for a, being the first and last sub-block
rows,
columns is a list of two valid column indices for a, being the first and last
sub-block columns,
b is the sub-block of a with the specified first and last row and column sub-block
indices.

Example

See also: LeadingMatrixBlock, TailingMatrixBlock, MakeBlockMatrix, Make-

Matrix.

 BruhatCVector (bcv)
In the explicit Bruhat decomposition of a non-singular matrix a, the diagonal
matrix c has a special form, each element being the ratio of absolute values of
minor determinants (ci ) of the original matrix a with the element in the (i, i)th
418 Appendix The GL(n)pack Manual

position being cn−i+1 /cn−i for 2 ≤ i ≤ n − 1 with the (n, n)th element being
c1 and the 1, 1th, det(w)det(a)/cn−1 . This function returns those ci .
See Section 10.3 and Proposition 10.3.6.

BruhatCVector[a] −→ c

a is a non-singular n × n square matrix of CREs,

c is a list of n − 1 CREs {c1 , . . . , cn−1 }.

Example

See also: BruhatForm, LanglandsForm.

 BruhatForm (bru)
This function finds the factors of a non-singular matrix, which may have entries
which are polynomial, rational or algebraic expressions, so that the matrix can
be expressed as the product of an upper triangular matrix with 1s on the leading
diagonal (unipotent), a diagonal matrix, a permutation matrix (with a single 1
in each row and column), and a second unipotent matrix. When an additional
constraint, namely, that the transpose of the permutation matrix times the second
upper triangular matrix is lower triangular, then the factors are unique. This is
the so-called Bruhat decomposition.
See Section 10.3 Proposition 10.3.6. The decomposition is used in
Section 10.6 to derive the Fourier expansion of a minimal parabolic Eisenstein
series.
 A.3 Function descriptions and examples 419

BruhatForm[a] −→ {u1 ,c,w,u2 }

a is a non-singular square CRE matrix,

u1 is an upper triangular unipotent matrix,
c is a diagonal matrix,
w is a permutation matrix,
u2 is an upper triangular unipotent matrix.

Example

See also: CartanForm, HermiteForm, IwasawaForm, SmithForm, Langlands-

Form.

 CartanForm (car)
This function gives a form of the Cartan decomposition of a numeric real
non-singular square matrix, namely the factorization a = k · ex p(x) where k
is orthogonal and x symmetric. It follows from this that the transpose of an
invertible matrix satisfies an equation

t
a = k.a.k

for some orthogonal matrix k, where t a is the transpose of a. The function is

restricted to numeric matrices because eigenvectors and eigenvalues are used.

CartanForm[a] −→ {k, s}

a is a non-singular real numeric square matrix,

k is an orthogonal matrix,
s is the matrix exponential of a symmetric matrix.
420 Appendix The GL(n)pack Manual

Example

See also: BruhatForm, IwasawaForm, LanglandsForm, HermiteFormLower,

HermiteFormUpper, SmithForm.

 ConstantMatrix (com)
This function constructs a constant matrix with specified element value.
This function can be used together with other functions to construct matrices.

ConstantMatrix[c, m, n] −→ a

c is a CRE, m is an integer with m ≥ 1,

n is an integer with n ≥ 1,
a is an m by n matrix having each element equal to c.
See also: ZeroMatrix, ElementaryMatrix, MatrixJoinHorizontal, MatrixJoin-
Vertical.

 CreQ (crq)
This function checks to see if its argument evaluates to a so-called Canonical
Rational Expression (CRE), i.e. a number (real or complex, exact or floating
 A.3 Function descriptions and examples 421

point) or rational function in one or many variables with numerical coefficients,

where any number of the variables can be replaced by function calls, including
calls to undefined (so-called noun) functions of one or many arguments with
arguments being canonical rational expressions. This is the data type expected
by GL(n)pack functions.
See the introduction to the appendix.

CreQ[e] −→ P
e is a Mathematica expression,
P is True if e is a CRE and False otherwise.

Example

See also: ParabolicQ, KloostermanSumQ.

 DiagonalToMatrix (d2m)
This function takes a list and constructs a matrix with the list elements as the
diagonal entries.
Diagonal matrices appear in many places, including in the Iwasawa and
Bruhat decompositions.

DiagonalToMatrix[di] −→ a

di is a non-empty list of CREs,

a is a square matrix of size the length of di, with zeros in off-diagonal positions,
and with the diagonal entries being the elements of di and in the same order.

See also: MatrixDiagonal.

422 Appendix The GL(n)pack Manual

 EisensteinFourierCoefficient (efc)
This function returns the nth term of the Fourier expansion of an Eisenstein
series for GL(2), with an explicit integer specified for n.
See Section 3.1, especially Theorem 3.1.8.

EisensteinFourierCoefficient[z, s, n] −→ v

z is a CRE, s is a CRE,
n is an integer being the index of the nth coefficient,
v is a complex number or symbolic expression representing the nth Fourier
term of the Eisenstein Fourier expansion for GL(2) with parameters
z and s.

Example

See also: EisensteinSeriesTerm, IFun, LanglandsIFun.

 EisensteinSeriesTerm (est)
This function returns the term of the Eisenstein series E(z, s) for GL(2), namely
the summand of:
1
ys
E(z, s) =
2 a,b∈Z,(a,b)=1 |az + b|2s

with explicit values for the integers a, b.

See Definition 3.1.2.

EisensteinSeriesTerm[z, s, ab] −→ v

z is a CRE, s is a CRE,
ab is a list of two integers {a, b}, at least one of which must be non-zero,
v is a complex number or symbolic expression representing the term of the
Eisenstein series for GL(2) with parameters z, s, a, b.
 A.3 Function descriptions and examples 423

Example

See also: EisensteinFourierCoefficient, LanglandsIFun.

 ElementaryMatrix (elm)
This function returns a square matrix having 1s along the leading diagonal and
with a given element in a specified off-diagonal position.

ElementaryMatrix[n, i, j, c] −→ e

n is a strictly positive integer being the size of the matrix,

i is a strictly positive integer being the row index of the off-diagonal entries,
j is a strictly positive integer with i = j, representing the column index of the
off-diagonal entries,
c is a CRE to be placed at the (i, j)th position,
[e] is an n by n matrix with 1s on the leading diagonal and all other elements
zero, except in the (i, j)th position where it is c.

Example

See also: SwapMatrixRows, SwapMatrixColumns.

424 Appendix The GL(n)pack Manual

 FunctionalEquation (feq)
This function, for each index i, returns a list of affine combinations of its
variables representing the ith functional equation for the Jacquet–Whittaker
function of order n ≥ 2.
See Section 5.9, especially equations (5.9.5), (5.9.6) and Example 5.9.7.

FunctionalEquation[v, i] −→ vp

v is a list of CREs of length n − 1,

i is a strictly positive integer with 1 ≤ i ≤ n − 1 being the index of the functional
equation,
vp is a list of CREs representing the transformations required of the variables
vp for the ith functional equation.

Example

See also: Whittaker, WhittakerGamma, WhittakerStar.

 GetCasimirOperator (gco)
This function computes the Casimir operator acting on an arbitrary noun func-
tion and with respect to the Iwasawa variables. Note that this function makes
an explicit brute-force evaluation of the operator, so is not fast, especially for
n ≥ 3.
See Proposition 2.3.3, Example 2.3.4 and Section 6.1.

GetCasimirOperator[m,n,“x”,“y”,“f”] −→ Operator

m is positive integer with value 2 or more being the order of the operator,
n is a positive integer with value 2 or more being the dimension of the Iwasawa
form,
“x” is a string, being the name of the symbol such that the variables in the upper
triangle of the matrix given by the Iwasawa decomposition are x[i, j],
“y” is a string, being the name of the symbol such that the terms in the first
n − 1 positions of the leading diagonal of the Iwasawa decomposition
are
y[1] · · · y[n − 1], y[1] · · · y[n − 2], . . . , y[1],
 A.3 Function descriptions and examples 425

“f” is a string being the name of a function of noun form (i.e. it should not be
defined as an explicit Mathematica function or correspond to the name of
an existing function) which will appear as partially differentiated by the
computed Casimir operator, Operator is an expression in the variables
(x[i, j], 1 ≤ i < j ≤ n), (y[i], 1 ≤ i ≤ n − 1)
and the partial derivatives of the function with name “f” with respect to
argument slots of f arranged in the order (x1,1 , x1,2 , . . . , y1 , . . . , yn−1 ).

Example

See also: IwasawaForm, ApplyCasimirOperator.

 GetMatrixElement (gme)
The specified element of a matrix is returned.
GetMatrixElement[a, i, j] −→ e
a is a matrix of CREs,
i is the row index of the element,
j is the column index of the element,
e is the (i, j)th element of a.
See also: MatrixColumn, MatrixRow, MatrixBlock.
426 Appendix The GL(n)pack Manual

 GlnVersion (glv)
This function prints out the date of the version of GL(n)pack which is being
used, followed by the version of Mathematica. It has no argument, but the
brackets must be given.
GlnVersion[] −→ True

 HeckeCoefficientSum (hcs)
This function takes a natural number m, a list of natural numbers
{m 1 , . . . , m n−1 } and a string for a function name and finds the terms in the
sum right-hand side


λm A(m 1 , . . . , m n−1 ) = A(c0 m 1 /c1 , c1 m 2 /c2 , . . . , cn−2 m n−1 /cn−1 )
)n−1
where the summation is over all (ci ) such that i=0 ci = m and ci |m i , 1 ≤ i ≤
n − 1.
See Theorem 9.3.11 and equation (9.3.17).

HeckeCoefficientSum[m, ms, “A”] −→ s

m is a natural number (i.e. a strictly positive integer) representing the index of
the eigenvalue λm ,
ms is a list of natural numbers representing the multi-index of the Fourier
coefficient A,
“A” is a string giving the name of a noun function for the Fourier coefficient,
s is a term or sum of terms being the right-hand side of the expansion
λm A(m1 , . . .).
Example

See also: HeckeOperator, SchurPolynomial, HeckePowerSum, HeckeEigen-

value.

 HeckeEigenvalue (hev)
This function returns the value of the mth eigenvalue of the ring of HeckeOpera-
tors acting on square integrable automorphic forms f (z) for hn . Note that when
the Euler product of a Maass form is known, the Fourier coefficients which
appear in the expressions for the eigenvalues (the A in λm = A(m, 1, . . . , 1))
 A.3 Function descriptions and examples 427

can be expressed in terms of Schur polynomials in the parameters which appear

in the Euler product.
See Section 9.3, especially Theorem 9.3.11.

HeckeEigenvalue[m, n, a] −→ λm

m is a natural number representing the index of the eigenvalue λm ,

n is a positive integer of size two or more being the dimension of GL(n),
a is a string representing the name of a function of n − 1 integers being the
Fourier coefficients of a given Maass form which is an eigenfunction of all
of the Hecke operators,
λm is an expression representing the mth Hecke eigenvalue.

Example

See also: HeckeOperator, HeckeMultiplicativeSplit, SchurPolynomial, Hecke-

PowerSum.

 HeckeMultiplicativeSplit (hms)
This function takes a list of natural numbers {m 1 , . . . , m n−1 }, finds the primes
and their powers that divide any of the m i , and returns a list of lists of those
primes and their powers. The purpose of this function is the evaluation of
the Hecke Fourier coeffients of a Maass form in terms of Schur polynomials
when the Euler product coeffients of the form are known. If p1 , . . . , pr are
the primes and ki, j is the maximum power of pi dividing m j , then the Fourier
coefficient

r
 k k
A(m 1 , . . . , m n−1 ) = A pi i,1 , . . . , pi i,n−1 .
i=1

See Theorem 9.3.11 and equation (7.4.14).

HeckeMultiplicativeSplit[m] −→ list

m is a list of natural numbers representing the multi-index of the Fourier coef-

ficient,
list consists of sublists, each being a prime pi and a list of n − 1 powers of that
prime ki,j .
428 Appendix The GL(n)pack Manual

Example

See also: HeckeOperator, SchurPolynomial, HeckeEigenvalue, HeckeCoeffi-

cientSum.

 HeckeOperator (hop)
This function computes the nth order Hecke operator which acts on square
integrable forms on hn .
See Section 9.3, especially formula (9.3.5).
HeckeOperator[n, z, f] −→ Tn (f(z))
n is a natural number being the order of the operator,
z is a square matrix of CREs of size n,
f is a string being the name of a function of a square matrix of size n,
Tn (f(z)) is an expression representing the action of the nth Hecke operator on
the matrix function f(z).
Example

See also: HeckeEigenvalue.

 A.3 Function descriptions and examples 429

 HeckePowerSum (hps)
This function takes a natural number m, a list of natural numbers
{m 1 , . . . , m n−1 } and a string for a function name and finds the powers of any
fixed prime in the sum right-hand side


λm A(m 1 , . . . , m n−1 ) = A(c0 m 1 /c1 , c1 m 2 /c2 , . . . , cn−2 m n−1 /cn−1 )
)n−1
where the summation is over all (ci ) such that i=0 ci = m and ci |m i , 1 ≤ i ≤
n − 1 in case m and each of the m i is a power of a fixed prime. The powers
that appear in the expansion are the same for any prime. The purpose of this
function is to simplify the study of the multiplicative properties of the Fourier
coefficients.
See Theorem 9.3.11 and equation (9.3.17).

HeckePowerSum[a, as, “B”] −→ list

a is a non-negative integer, being the power a of any prime p such that pa is the
index of the eigenvalue λpa ,
as is a list of non-negative integers representing the powers of a fixed prime
which appear in the multi-index of a Fourier coefficient,
list consists of a sum of terms B[b1,i , . . . , bn−1,i ] such that the corresponding
term in the Hecke sum would have a value A(pb1,i , . . .).

Example

See also: HeckeOperator, SchurPolynomial, HeckeEigenvalue, HeckeCoeffi-

cientSplit,

 HermiteFormLower (hfl)
This function computes the lower left Hermite form h of a non-singular integer
matrix a, and a unimodular matrix l such that a = lh. This Hermite form is a
lower triangular integer matrix with strictly positive elements on the diagonal
of increasing size, and such that each element in the column below a diagonal
entry is non-negative and less than the diagonal entry.
See Theorem 3.11.1.
430 Appendix The GL(n)pack Manual

HermiteFormLower[a] −→ {l, h}

a is a non-singular integer matrix,

l is a unimodular matrix,
h is a lower triangular integer matrix, being the Hermite form of a.

Example

See also: HermiteFormUpper, SmithForm.

 HermiteFormUpper (hfu)
This function computes the upper Hermite form h of a non-singular integer
matrix a, and a unimodular matrix l such that a = lh. This Hermite form is an
upper triangular integer matrix with strictly positive elements on the diagonal
of increasing size, and such that each element in the column above a diagonal
entry is non-negative and less than the diagonal entry.
See Theorem 3.11.1.

HermiteFormUpper[a] −→ {l, h}

a is a non-singular integer matrix,

l is a unimodular matrix,
h is an upper triangular integer matrix, being the Hermite form of a.
 A.3 Function descriptions and examples 431

Example

See also: HermiteFormLower, SmithForm.

 IFun (ifn)
This function returns the value
 n−1
n−1  b νj
Iν (z) = yi i, j
i=1 j=1

where bi, j = i j if i + j ≤ n and (n − i)(n − j) if i + j > n. The n × n matrix z

is real and non-singular, or has CRE elements which could evaluate to a real non-
singular matrix. The variables yi are those in the Iwasawa decomposition of z.
See Definition 2.4.1 and equation (5.1.1).
IFun[ν,z] −→ v
ν is a list of n − 1 CREs,
z is an n × n non-singular matrix of CREs,
v is the product Iν (z).
432 Appendix The GL(n)pack Manual

Example

See also: IwasawaForm, IwasawaYVariables.

 InsertMatrixElement (ime)
An element is inserted into a matrix returning a new matrix and leaving the
original unchanged.
InsertMatrixElement[e, i, j, a] −→ b
e is a CRE being the element to be inserted,
i is the row index of the position where the element is to be inserted,
j is the column index of the position where the element is to be inserted,
a is the original matrix of CREs,
b is a new matrix, being equal to a but with e in the (i, j)th position.
See also: DiagonalToMatrix, MatrixJoinHorizontal, MatrixJoinVertical.

 IwasawaForm (iwf)
This function computes the Iwasawa form of a non-singular real matrix a.
This consists of the product of an upper triangular unipotent matrix x and a
diagonal matrix y with strictly positive diagonal entries such that, for some
non-singular integer matrix u, real orthogonal matrix o and constant diagonal
matrix δ, a = u.x.y.o.δ. This function returns a single matrix z = x.y.
See Section 1.2.
IwasawaForm[a] −→ z
a is a non-singular square matrix of CREs,
z is an upper-triangular matrix with positive diagonal entries, being the Iwasawa
form of a.
 A.3 Function descriptions and examples 433

Example
The Iwasawa decomposition of the matrix
 
a b
c d

is found.

See also: IwasawaQ, MakeZMatrix, IwasawaXMatrix, IwasawaYMatrix,

IwasawaXVariables, IwasawaYVariables.

 IwasawaXMatrix (ixm)
This function returns the unipotent matrix x corresponding to the decomposition
z = x.y of a matrix z in Iwasawa form.
See Proposition 1.2.6 and Example 1.2.4

IwasawaXMatrix[w] −→ x

w is a square non-singular matrix of CREs which must be in Iwasawa form,

x an upper-triangular matrix with 1s on the diagonal and values xi,j in each (i, j)
position above the diagonal.

Example
In this example the x-matrix, x-variables, y-matrix and y-variables are extracted
from a generic matrix in Iwasawa form.
434 Appendix The GL(n)pack Manual

See also: IwasawaForm, IwasawaXMatrix, IwasawaYVariables, IwasawaY-

Matrix.

 IwasawaXVariables (ixv)
This function returns the x-variables from a matrix z = x.y in Iwasawa form.
These are the elements in the strict upper triangle of the matrix x in row order.
See Definition 1.2.3 and Proposition 1.2.6.
IwasawaXVariables[w] −→ l
w is a square non-singular matrix of CREs which must be in Iwasawa form,
l is a list of the form {x1,2 , . . . , x1,n , x2,3 , . . . , xn−1,n }.
See also: IwasawaForm, IwasawaXMatrix, IwasawaYVariables, IwasawaY-
Matrix.

 IwasawaYMatrix (iym)
This function returns the y-matrix from the decomposition z = x.y of a matrix
z in Iwasawa form.
 A.3 Function descriptions and examples 435

See Definition 1.2.3 and Proposition 1.2.6.

IwasawaYMatrix[z] −→ y
z is a square non-singular matrix of CREs which must be in Iwasawa form,
y a diagonal matrix where the ith diagonal slot has the value y1 · · · yn−i for
1 ≤ i ≤ n − 1 where the (n, n)th position has the value 1.
See also: IwasawaForm, IwasawaXMatrix, IwasawaYVariables, IwasawaX-
Variables.

 IwasawaYVariables (iyv)
This function returns a list of the y-variables from the Iwasawa decomposition
of a matrix z = x.y.
See Definition 1.2.3 and Proposition 1.2.6.
IwasawaYVariables[z] −→ L
z is a square non-singular matrix of CREs which must be in Iwasawa form,
L a list {y1 , . . . , yn−1 } of the y-variables of the Iwasawa form.
See also: IwasawaForm, IwasawaYMatrix, IwasawaXVariables, IwasawaX-
Matrix.

 IwasawaQ (iwq)
This function tests a Mathematica form or expression to see whether it is a
non-singular square matrix in Iwasawa form.
See Section 1.2.
IwasawaQ[z] −→ value
z is a Mathematica form,
value is True if z is a matrix of CREs in Iwasawa form, False otherwise.
See also: IwasawaForm, MakeZMatrix.

 KloostermanBruhatCell (kbc)
This function takes an explicit permutation matrix w with all other arguments
being symbolic. It returns rules which solve for x and y in the square matrix
Bruhat decomposition equation a = x.c.w.y assuming c is in “Friedberg form”,
x and y are unipotent and y satisfies t w.t y.w is upper triangular. These rules
are not unique.
See Chapter 11, especially Section 11.2. Also Lemma 10.6.3.
436 Appendix The GL(n)pack Manual

KloostermanBruhatCell[a,x,c,w,y] −→ rules
a is a symbol which will be used as the name of an n × n matrix,
x is a symbol which will be used as the name of a unipotent matrix,
c is a symbol which will be used as the name of an array c[i] representng a list
of n − 1 non-zero integers specifying the diagonal of a matrix. (Note that
the 1st element of the diagonal represents the term det(w)/c[n − 1], the
second c[n − 1]/c[n − 2] and so on down to the last c[1] as in the notation
of (11.2.1).),
w is an n × n matrix which is zero except for a single 1 in each row and column,
being an explicit element of the Weyl Group Wn ,
y is a symbol which will be used as the name of a unipotent matrix which satisfies
t t
w. y.w is upper triangular making the decomposition unique, given a,
rules is a list of rules of the form x[i, j] → eij or y[i, j] → eij where the eij are
expressions in the a[i, j] and c[i].
Example

See also: BruhatForm, BruhatCVector, KloostermanCompatibility, Klooster-

manSum.

 KloostermanCompatibility (klc)
This function takes an explicit permutation matrix w, with remaining arguments
symbolic, and returns a list of values, each element being a different type
of constraint applicable to any valid Kloosterman sum based on w. The first
element is a list of forms restricting the characters. The second is a set of
divisibility relations restricting the values of the diagonal matrix c. And the third
is the set of minor relations. A typical approach to forming Kloosterman sums
would be to first run this function, determine a valid set or sets of parameters
from the symbolic output, and then run KloostermanSum using explicit integer
values of valid parameters.
 A.3 Function descriptions and examples 437

See Chapter 11, Proposition 11.2.10, Lemma 10.6.3.

KloostermanCompatibility[t1,t2, c, w, v] −→ {characters, divisibilities, minors}
+n−1
t1 is a symbol representing the character e2π i i=1 t1[i,i+1] of Un (R),
t2 is a symbol representing another character of Un (R),
c is a symbol representing the diagonal of a matrix in Friedberg notation. (Note
that the 1st element of the diagonal is the term det(w)/cn−1 , the second
cn−1 /cn−2 and so on down to the last c1 as in the notation of (11.2.1).),
w is an n × n matrix which is zero except for a single 1 in each row and column,
representing an explicit element of the Weyl Group Wn ,
v is a symbol representing the generic name of any bottom row-based minor,
characters is a list of expressions relating the elements of t1, t2, and the ci .
Each expression must vanish if an explicit Kloosterman sum is to be valid,
divisibility is a list of lists, each sublist being of the form {ci , 1} or {ci , cj }. The
former means valid sums must have the ci = ±1. The latter means they
must have ci |cj ,
minors is a list of rules of the form v[{j1 , j2 , . . . , ji }] → ci or 0, giving the
constraints on minors.
Example

See also: KloostermanBruhatCell, BruhatForm, BruhatCVector, Kloosterman-

Sum, PluckerRelations, PluckerCoordinates, PluckerInverse.

 KloostermanSum (kls)
This function computes the generalized Kloosterman sum for SL(n, Z) for
n ≥ 2 as given by Definition 11.2.2. When n = 2 this coincides with the clas-
sical Kloosterman sum. More generally the sum is


S(θ1 , θ2 , c, w) := θ1 (b1 )θ2 (b2 )
γ =b1 cwb2
438 Appendix The GL(n)pack Manual

where

γ ∈ Un (Z)\S L(n, Z) ∩ G w / w

and w = t w.t U n (Z).w ∩ Un (Z) and G w is the Bruhat cell associated to the
permutation matrix w. Since these sums are only well defined for some par-
ticular compatible values of the arguments the user is advised to first run
KloostermanCompatibility with an explicit w to determine those values.
) 2
Note that the complexity of the algorithm is O( 1≤i≤n−1 |ci |n ) = O(cn ) where
c = max|ci |.
See Chapter 11.

KloostermanSum[t1, t2, c, w] −→ value

t1 is a list of n − 1 integers representing a character of Un (R),

t2 is a list of n − 1 integers representing another character of Un (R),
c is a list of n − 1 non-zero integers specifying the diagonal of a matrix. (Note
that the 1st element of the matrix is det(w)/cn−1 , the second cn−1 /cn−2 and
so on down to the last c1 as in the notation of (11.2.1).),
w is an n × n matrix which is zero except for a single 1 in each row and column,
representing an explicit element of the Weyl subroup Wn of GL(n, R),
value is a sum of complex exponentials being a Kloosterman sum when it is
well defined.

Example
This n = 4 example shows how KloostermanCompatibility should be run after
selecting a permutation matrix. Then KloostermanSum is called with compat-
ible arguments.
 A.3 Function descriptions and examples 439

Example
This first illustrates commutativity of the LongElement sums (c/f (Friedberg
(1987), its Proposition 2.5)), then Proposition 11.4.1 and finally is given an
example of a classical sum showing it is real.

See also: BruhatForm, BruhatCVector, KloostermanCompatibility, Klooster-

manBruhatCell, PluckerCoordinates, PluckerInverse, PluckerRelations.
440 Appendix The GL(n)pack Manual

 LanglandsForm (llf)
This function returns a list of the three matrices of the Langlands decomposition
of a square matrix in a parabolic subgroup specified by a partition of the matrix
dimension.
See Section 10.2.

LanglandsForm[p, d] −→ {u, c, m}

p is a square matrix of CREs,

d is a list of r positive integers of length at most n with sum n,
u is a unipotent block upper triangular matrix,
c is a diagonal matrix with r diagonal blocks each being a positive constant
times the identity,
m is a block diagonal matrix with each diagonal block having determi-
nant ±1.

Example

See also: ParabolicQ, LanglandsIFun.

 A.3 Function descriptions and examples 441

 LanglandsIFun (lif)
This function computes the summand for Langlands’ Eisenstein series with
respect to a specified parabolic subgroup.
See Chapter 10, Definition 10.4.5.

LanglandsIFun[g, p, s] −→ Is (g.z)

g is a non-singular matrix of CREs in the parabolic subgroup specified by the

second argument,
p is a list of r positive integers representing a partition of the matrix dimension,
+
s is list of r CREs such that ri=1 di si = 0,
Is (g.z) is the summand for the Langlands Eisenstein series.

Example

See also: LanglandsForm.

 LeadingMatrixBlock (lmb)
This function extracts a leading matrix block of specified dimensions.

LeadingMatrixBlock[a, i, j] −→ b

a is a matrix of CREs, i is a valid row index for a,

j is a valid column index for a,
b is the leading block of a with i rows and j columns.

See also: BlockMatrix, TailingMatrixBlock, GetMatrixElement.

442 Appendix The GL(n)pack Manual

 LongElement (lel)
This function constructs the so-called long element of the group GL(n,Z), a
matrix with 1s along the reverse leading diagonal and 0s elsewhere.
See Chapter 5.

LongElement[n] −→ w

n is a strictly positive integer,

w is an n by n matrix with 1s down the reversed leading diagonal and 0s
elsewhere.

Example

See also: WMatrix, ModularGenerators.

 LowerTriangleToMatrix (ltm)
This function takes a list of lists of increasing length and forms a matrix with
zeros in the upper triangle and the given lists constituting the rows of the lower
triangle.

LowerTriangleToMatrix[l] −→ a

l is a list of lists of CREs of strictly increasing length representing the elements

of a lower triangular submatrix including the diagonal. The first has length
1 and each successive sublist has length 1 more than that preceding sublist,
a is a full matrix with 0 in each upper triangular position.

Example
 A.3 Function descriptions and examples 443

See also: UpperTriangleToMatrix.

 MakeBlockMatrix (mbm)
This function takes a list of lists of matrices and creates a single matrix wherein
the jth matrix element of the ith sublist constitutes the (i, j)th sub-block of this
matrix. In order that this construction succeed, the original matrices must have
compatible numbers of rows and columns, i.e. the matrices in each sublist must
have the same number of rows for that sublist and for each j the jth matrix in
each sublist must have the same number of columns. In spite of this restriction,
the function is a tool for building matrices rapidly when they have a natural
block structure.

MakeBlockMatrix[A] −→ B

A is a list of lists of equal length of matrix elements, each matrix having CRE
elements,
B is a single matrix with sub-blocks being the individual matrices in A.

Examples

See also: ConstantMatrix, ZeroMatrix, LongElement, WeylGroup, Special-

WeylGroup.

 MakeMatrix (mkm)
This function returns a symbolic matrix of given dimensions.
444 Appendix The GL(n)pack Manual

MakeMatrix[“a”, m, n] −→ A

“a” is a string being the name of the generic symbolic matrix element variable
a[i, j],
m is a strictly positive integer representing the number of rows of A,
n is a strictly positive integer representing the number of columns of A,
A is a symbolic matrix with (i, j)th entry a[i, j].

Example

See also: MakeYMatrix, MakeZMatrix, MakeBlockMatrix, ZeroMatrix,

ConstantMatrix, InsertMatrixElement, WeylGroup, ModularGenerators,
LongElement, WMatrix, SpecialWeylGroup.

 MakeXMatrix (mxm)
This function returns a symbolic upper triangular square matrix of given dimen-
sion with 1s on the leading diagonal, i.e. a unipotent matrix.
See Definition 1.2.3.

MakeXMatrix[n, “x”] −→ u

n is a strictly positive integer representing the size of the matrix,

“x” is a string being the name of the generic symbolic matrix element variable
x[i, j],
u is an upper-triangular symbolic matrix with 1s on the leading diagonal.
 A.3 Function descriptions and examples 445

Example

See also: MakeXVariables, MakeYMatrix, MakeYVariables, MakeZMatrix,

MakeZVariables.

 MakeXVariables (mxv)
This function returns a list of the x-variables which appear in the symbolic
generic Iwasawa form of a square matrix of given dimension.
See Definition 1.2.3.
MakeXVariables[n, “x”] −→ l
n is a strictly positive integer representing the size of the matrix,
“x” is a string being the name of the generic list element variable x[i, j],
l is a list of the x-variables in order of increasing row index.
See also: MakeXMatrix, MakeYMatrix, MakeYVariables, MakeZMatrix.
446 Appendix The GL(n)pack Manual

 MakeYMatrix (mym)
This function returns a symbolic diagonal matrix of given dimension with val-
ues on the leading diagonal being the product of the y-variables of a matrix
expressed in Iwasawa form.
See Definition 1.2.3 and the manual entry for MakeXMatrix.
MakeYMatrix[n, “y”] −→ d
n is a strictly positive integer representing the size of the matrix,
“y” is a string being the name of the generic symbolic matrix element variable
y[i] such that the jth diagonal element is the product y[1]y[2] · · · y[n − j],
d is a diagonal matrix.
See also: MakeXMatrix, MakeXVariables, MakeYVariables, MakeZMatrix.

MakeYVariables (myv)
This function returns a symbolic list of the n − 1 y-variables which would occur
in the Iwasawa form of a matrix of size n × n.
See Definition 1.2.3 and the manual entry for MakeXMatrix.
MakeYVariables[n, “y”] −→ l
n is a strictly positive integer representing the size of the matrix,
“y” is a string being the name of the generic variable y[i],
l is a list of the form {y[1], . . . , y[n − 1]}.
See also: MakeXMatrix, MakeXVariables, MakeYMatrix, MakeZMatrix,
MakeZVariables.

 MakeZMatrix (mzm)
This function returns a symbolic upper triangular square matrix of given dimen-
sion being in generic Iwasawa form.
See Example 1.2.4 and the manual entry for MakeXMatrix.
MakeZMatrix[n, “x”,“y”] −→ u
n is a strictly positive integer representing the size of the matrix,
“x” is a string being the name of the generic symbolic Iwasawa x-variable
x[i, j],
“y” is a string being the name of the generic symbolic Iwasawa y-variable y[i],
u is an upper-triangular symbolic matrix with (i, j)th element having the form
x[i, j]y[1] · · · y[n − j].
 A.3 Function descriptions and examples 447

See also: MakeXMatrix, MakeXVariables, MakeYMatrix, MakeYVariables,

MakeZVariables.

 MakeZVariables (mzv)
This function returns a list of the variables which occur in the Iwasawa form
for a matrix with generic symbolic entries and of given size.
See the manual entry for MakeXMatrix.
MakeZVariables[n, “x”, “y”] −→ l
n is a strictly positive integer representing the size of the matrix,
“x” is a string being the name of the generic symbolic matrix element x[i, j]
with i > j,
“y” is a string being the name of the generic symbolic matrix element y[i],
l is a list of the Iwasawa variables with the x-variables first in order of increasing
row index followed by the y-variables:
{x[1, 2], . . . , x[1, n], x[2, 3], . . . , x[n − 1, n], y[1], . . . , y[n − 1]}.
See also: MakeXMatrix, MakeXVariables, MakeYMatrix, MakeYVariables,
MakeZMatrix.

 MatrixColumn (mcl)
This function returns a given column of a matrix.
MatrixColumn[m, j] −→ c
m is a matrix of CREs,
j is a valid column index for m,
c is the jth column of m returned as a list.
See also: MatrixRow.

 MatrixDiagonal (mdl)
This function extracts the diagonal of a matrix.
MatrixDiagonal[a] −→ d
a is a square matrix of CREs,
d is a list, being the diagonal entries of a in the same order.
See also: DiagonalToMatrix.
448 Appendix The GL(n)pack Manual

 MatrixJoinHorizontal (mjh)
This function assembles a new matrix by placing one matrix to the right of
another compatible matrix.
MatrixJoinHorizontal[a, b] −→ c
a is a matrix of CREs,
b is a matrix with the same number of rows as a,
c is a matrix with block decomposition c = [a|b].
Example

See also: MatrixJoinVertical.

 MatrixJoinVertical (mjv)
This function assembles a new matrix by placing one matrix above another
compatible matrix.
MatrixJoinVertical[a, b] −→ c
a is a matrix of CREs,
b is a matrix with the same number of columns as a,
c is a matrix with block decomposition having a above b.
See also: MatrixJoinHorizontal.

 MatrixLowerTriangle (mlt)
This function extracts the elements in the lower triangle of a square matrix,
including the diagonal, and returns them as a list of lists.
MatrixLowerTriangle[a] −→ t
a is a square matrix of CREs,
t is a list of lists where the ith element of the jth list represents the (j, i)th
element of a.
See also: MatrixUpperTriangle, LowerTriangleToMatrix.
 A.3 Function descriptions and examples 449

 MatrixRow (mro)
This function returns a given row of a matrix.

MatrixRow[m, i] −→ r

m is a matrix of CREs,
i is a row index of m,
r is a list representing the ith row of m.

See also: MatrixColumn.

 MatrixUpperTriangle (mut)
This function extracts the elements in the upper triangle, including the diagonal,
of a square matrix and returns a list of lists of the elements from each row.

MatrixUpperTriangle[a] −→ t

a is a square matrix of CREs,

t is a list of lists of elements with the ith element of the jth list being the
(j, i + j − 1)th element of a.

Example

See also: MatrixLowerTriangle, UpperTriangleToMatrix.

 ModularGenerators (mog)
This function returns a list of two matrix generators for the subgroup of the
group of integer matrixes with determinant 1, i.e. generators of SL(n, Z).
See Chapter 5, especially Section 5.9.
450 Appendix The GL(n)pack Manual

ModularGenerators[n] −→ g

n is a positive integer with n ≥ 2,

g is a list of two n by n matrices which will generate SL(n, Z).

Example

See also: WeylGenerator, WeylGroup, SpecialWeylGroup, WMatrix, Long-

Element.

 MPEisensteinGamma (eig)
This function computes the gamma factors for the minimal parabolic Eisenstein
series
n  
s − λi (v)
G Ev (s) = π −ns/2  .
i=1
2

See Chapter 10, Theorem 10.8.6.

MPEisensteinGamma[s,v] −→ G

s is a CRE representing a complex number,

v is a list of n − 1 CREs with n ≥ 2 representing complex parameters,
G is the gamma factor for the minimal parabolic Eisenstein series functional
equation.

Example

See also: MPEisensteinLambdas, MPEisensteinSeries, MPExteriorPower-

Gamma, MPExteriorPowerLFun, MPSymmetricPowerLFun, MPSymmetric-
PowerGamma.
 A.3 Function descriptions and examples 451

 MPEisensteinLambdas (eil)
This function computes the functions λi (v) : Cn−1 → C such that the L-
function associated with the minimal parabolic Eisenstein series L Ev (s) is a
product of shifted zeta values

n
L Ev (s) = ζ (s − λi (v)).
i=1

See Chapter 10, (10.4.1) and Theorem 10.8.6.

MPEisensteinLambdas[v] −→ L

v is a list of n − 1 CREs with n ≥ 2 representing complex parameters,

L is a list of affine expressions in the elements of v representing the functions
λi (v).

Example

See also: MPEisensteinSeries, MPEisensteinGamma, MPExteriorPower-

Gamma, MPExteriorPowerLFun, MPSymmetricPowerLFun, MPSymmetric-
PowerGamma.

 MPEisensteinSeries (eis)
This function computes the L-function associated with the minimal parabolic
Eisenstein series E v (z) as a product of shifted zeta values

n
L Ev (s) = ζ (s − λi (v)).
i=1

See Chapter 10, (10.4.1) and Theorem 10.8.6.

MPEisensteinSeries[s,v] −→ Z

s is a CRE representing a complex number,

v is a list of n − 1 CREs with n ≥ 2 representing complex parameters,
Z is a product of n values of the Riemann zeta function at shifted arguments.
452 Appendix The GL(n)pack Manual

Example

See also: MPEisensteinLambdas, MPEisensteinGamma, MPExteriorPower-

Gamma, MPExteriorPowerLFun, MPSymmetricPowerLFun, MPSymmetric-
PowerGamma.

 MPExteriorPowerGamma (epg)
This function returns the gamma factors for the kth symmetric L-function asso-
ciated with a minimal parabolic Eisenstein series.
See the introduction to Chapter 13.

MPExteriorPowerGamma[s,v,k] −→ G

s is a CRE representing a complex number,

v is a list of n − 1 CREs with n ≥ 2 representing complex parameters,
k is a natural number k ≥ 1 representing the order of the exterior power,
G is the gamma factor for the functional equation of the exterior power.

Example

See also: MPEisensteinLambdas, MPEisensteinSeries, MPEisensteinGamma,

MPExteriorPowerLFun, MPSymmetricPowerLFun, MPSymmetricPower-
Gamma.

 MPExteriorPowerLFun (epl)
This function returns the kth exterior power of the L-function of a minimal
parabolic Eisenstein series as a product of zeta values.
See the introduction to Chapter 13.
This function can be used to show that exterior power L-functions satisfy a
functional equation.
 A.3 Function descriptions and examples 453

MPExteriorPowerLFun[s,v,k] −→ Z

s is a CRE representing a complex number,

v is a list of n − 1 CREs with n ≥ 2 representing complex parameters,
k is a natural number k ≥ 1 representing the order of the exterior power,
Z is a product of Riemann zeta function values.

Example

See also: MPEisensteinLambdas, MPEisensteinSeries, MPEisensteinGamma,

MPExteriorPowerGamma, MPSymmetricPowerLFun, MPSymmetricPower-
Gamma.

 MPSymmetricPowerLFun (spf)
This function returns the kth symmetric power of the L-function of a minimal
parabolic Eisenstein series as a product of zeta values.
See the introduction to Chapter 13.
This can be used to show that symmetric power L-functions satisfy a func-
tional equation.

MPSymmetricPowerLFun[s,v,k] −→ Z

s is a CRE representing a complex number,

v is a list of n − 1 CREs with n ≥ 2 representing complex parameters,
k is a natural number k ≥ 1 representing the order of the exterior power,
Z is a product of Riemann zeta function values.

Example
454 Appendix The GL(n)pack Manual

See also: MPEisensteinLambdas, MPEisensteinSeries, MPEisensteinGamma,

MPExteriorPowerGamma, MPExteriorPowerLFun, MPSymmetricPower-
Gamma.

 MPSymmetricPowerGamma (spg)
This function returns the gamma factors for the kth symmetric L-function asso-
ciated with a mimimal parabolic Eisenstein series.
See the introduction to Chapter 13.

MPSymmetricPowerGamma[s,v,k] −→ G

s is a CRE representing a complex number,

v is a list of n − 1 CREs with n ≥ 2 representing complex parameters,
k is a natural number k ≥ 1 representing the order of the exterior power,
G is the gamma factors for the kth symmetric power L-function.

Example

See also: MPEisensteinLambdas, MPEisensteinSeries, MPEisensteinGamma,

MPExteriorPowerGamma, MPExteriorPowerLFun, MPSymmetricPowerL-
Fun.

 NColumns (ncl)
This function gives the number of columns of a matrix.

NColumns[a] −→ n

a is a matrix of CREs,
n is the number of columns of a.

See also: NRows.

 A.3 Function descriptions and examples 455

 NRows (nro)
This function gives the number of rows of a matrix.

NRows[a] −→ m

a is a matrix of CREs,
m is the number of rows of a.

See also: NColumns.

 ParabolicQ (paq)
This function tests a square matrix to see whether it is in a given parabolic
subgroup as specified by a non-trivial partition of the matrix dimension.
See Chapter 10, especially Section 10.1.

ParabolicQ[a, d] −→ ans

a is an n × n square matrix with entries which are CREs,

d is list of at most n positive integers with sum n,
ans is True if a is in the specified subgroup and False otherwise.

Example

See also: LanglandsForm, LanglandsIFun.

456 Appendix The GL(n)pack Manual

 PluckerCoordinates (plc)
This function takes an n × n square matrix and returns a list of lists of the so-
called Plücker coordinates, namely the values of all of the bottom j × j minors
with 1 ≤ j ≤ n − 1.
See Chapter 11, Section 11.3, Theorem 11.3.1.
PluckerCoordinates[a] −→ value
a is an n × n matrix of CREs,
value is a list of lists being the values of all of the j × j minor determinants with
1 ≤ j ≤ n − 1 based on the bottom row and taking elements from the bot-
tom j rows. The jth sublist has the j × j minors in lexical order of the column
indices.
Example

See also: PluckerRelations, PluckerInverse, KloostermanSum.

 PluckerInverse (pli)
This function takes a list of lists of integers, which could be the Plücker
coordinates arising from a square matrix, and returns such a matrix having
determinant 1. The matrix is not unique but PluckerInverse followed by
 A.3 Function descriptions and examples 457

PluckerCoordinates gives the identity, provided the list of lists of integers

is compatible, i.e. arises from some matrix.
See Section 11.3.

PluckerInverse[Ms] −→ a

Ms is a list of n − 1 sublists of integers Ms = {{M1 , . . . , Mn }, {M12 , . . .}, . . .},

representing the Plücker coordinates of a matrix in lexical order,
a is an integer matrix having those Plücker coordinates or False in case they
are incompatible.

Example

See also: PluckerCoordinates, PluckerRelations, KloostermanSum, Klooster-

manCompatibility.

 PluckerRelations (plr)
This function computes all the known quadratic relationships between the
minors of a generic square n × n matrix known as the Plücker coordinates.
See Chapter 11.
458 Appendix The GL(n)pack Manual

In the case that n = 2 there are none and for n = 3 one. For n > 3 the
number grows dramatically. No claim is made that this function returns, for
any given n, a complete set of independent relationships. By “complete”
is meant sufficient to guarantee the coordinates arise from a member of
S L(n, Z).

PluckerRelations[n, v] −→ relations

n is a positive integer with n ≥ 2,

v is a symbol representing the generic name used for the Plücker coordi-
nates so v[{i1 , . . . , ij }] is the matrix minor based on the last j rows and
the columns indexed by i1 , . . . , ij with these indices in strictly increasing
order,
relations is a list of quadratic expressions with coefficients ±1 in the v[{· · ·}]s,
which vanish whenever the values of the v[{· · ·}]s come from the minors
of an n × n matrix.

Example

See also: PluckerCoordinates, PluckerInverse, KloostermanSum.

 A.3 Function descriptions and examples 459

 RamanujanSum (rsm)
This function computes the Ramanujan sum s(n, c) for explicit natural number
values of n, c, namely


c
s(n, c) = e2πi(r/c) .
r =1,(r,c)=1

See Definition 3.1.4 and Proposition 3.1.7.

RamanujanSum[n,c] −→ s
n is a strictly positive integer, c is a strictly positive integer,
s is an integer being the Ramanujan sum.
Example

 RemoveMatrixColumn (rmc)
A given row is removed from a matrix, creating a new matrix and leaving the
original unchanged.
RemoveMatrixColumn[a, j] −→ b
a is a matrix of CREs,
j is a valid column index of a,
b is a matrix with all columns identical to a except the jth which is missing.
See also: RemoveMatrixRow.

 RemoveMatrixRow (rmr)
A given row is removed from a matrix, leaving the original unchanged.
RemoveMatrixRow[a, i] −→ b
a is a matrix of CREs,
i is a valid row index of a,
b is a matrix with all rows identical to a except the ith which is missing.
See also: RemoveMatrixColumn.
460 Appendix The GL(n)pack Manual

 SchurPolynomial (spl)
This function computes the Schur polynomial in n variables x1 , . . . , xn with
n − 1 exponents k1 , . . . , kn−1 , that is to say the ratio of the determinant of a
k +···+ki−1 +n−i
matrix with (i, j)th element 1 for i = n and x j 1 for 1 ≤ i ≤ n − 1,
to the determinant of the matrix which is 1 for i = n and x j for 1 ≤ i ≤ n − 1.
n−i

See Section 7.4.

SchurPolynomial[x, k] −→ Sk (x1 , . . . , xn )
x is list of n CREs,
k is a list of n − 1 CREs,
Sk (x1 , . . . , xn ) is the Schur polynomial.
Example

See also: HeckeMultiplicativeSplit.

 SmithElementaryDivisors (sed)
This function computes the elementary divisors of a non-singular n × n integer
matrix a, i.e. for each j with 1 ≤ j ≤ n, the gcd d j (a) of all of the j × j
minor determinants. If s j is the jth diagonal entry of the Smith form then
s j = d j (a)/d j−1 (a).
SmithElementaryDivisors[a] −→ l
a is a non-singular n × n integer matrix,
l is a list of the n Smith form elementary divisors of a in the order
{d1 (a), . . . , dn (a)}.
See also: SmithForm, SmithInvariantFactors, HermiteFormUpper, Hermite-
FormLower.
 A.3 Function descriptions and examples 461

 SmithForm (smf)
This function returns the Smith form diagonal matrix d of a square non-singular
matrix a with integer entries. This matrix d satisfies 0 < di,i and di,i | di+1,i+1
for all i ≤ n − 1. It also returns unimodular matrixes l, r such that a = l.d.r .
See Theorem 3.11.2.
SmithForm[a] −→ {l, d, r}
a is non-singular integer matrix,
l is a unimodular matrix,
d is a diagonal matrix, being the Smith Form of a, r is a unimodular matrix.
Example
The Smith form of a 4 by 4 matrix is computed and the result checked.

See also: SmithElementaryDivisors, SmithInvariantFactors, HermiteForm-

Lower, HermiteFormUpper.

 SmithInvariantFactors (sif)
This function computes the invariant factors of the Smith form of a non-singular
integer matrix a. These are all of the prime powers which appear in the diagonal
entries of the Smith form of a.
SmithInvariantFactors[a] −→ l
a is an n × n non-singular integer matrix,
l is a list of prime powers.
See also: SmithForm, SmithElementaryDivisors, HermiteFormUpper,
HermiteFormLower.
462 Appendix The GL(n)pack Manual

 SpecialWeylGroup (swg)
This function, for each natural number n, returns the group of n × n matrices
with each entry being 0 or ±1, and having determinant 1. There are 2n−1 n! such
matrices.
See Sections 6.3 and 6.5.

SpecialWeylGroup[n] −→ g

n is a natural number representing the matrix dimension,

g is a list of n × n matrices representing the Weyl group.

Example

See also: WeylGroup, WMatrix, WeylGenerator, ModularGenerators, Long-

Element.
 A.3 Function descriptions and examples 463

 SubscriptedForm (suf)
This function takes a Mathematica expression and prints it out in such a way that
subexpressions of the form x[n1 , n2 , . . . , nj ], where the ni are explicit integers,
are printed in the style
xn1 ,n2 ,...,nj .
The value of this function is for improving the look of expressions for inspec-
tion and should not be used otherwise. Compare the Mathematica function
MatrixForm. Not all expressions can be subscripted using this function.
SubscriptedForm[e] −→ f
e is a Mathematica expression,
f is a subscripted rendition of the same expression.

Example

 SwapMatrixColumns (smc)
Two columns of a matrix are exchanged creating a new matrix and leaving the
original unchanged.
464 Appendix The GL(n)pack Manual

SwapMatrixColumns[a, i, j] −→ b
a is a matrix of CREs,
i is a valid column index for a,
j is a valid column index for a,
b is a matrix equal to a except the ith and jth columns have been exchanged.
See also: SwapMatrixRows, ElementaryMatrix.

 SwapMatrixRows (smr)
Two rows of a matrix are exchanged creating a new matrix and leaving the
original unchanged.
SwapMatrixRows[a, i, j] −→ b
a is a matrix of CREs,
i is a valid row index for a,
j is a valid row index for a,
b is a matrix equal to a except the ith and jth rows have been exchanged.
See also: SwapMatrixColumns, ElementaryMatrix.

 TailingMatrixBlock (tmb)
This function returns a tailing matrix block of specified dimensions leaving the
original matrix unchanged.
TailingMatrixBlock[a, i, j] −→ b
a is a matrix of CREs,
i is a positive integer less than the number of rows of a,
j is a positive integer less than the number of columns of a,
b is the tailing block of a with i rows and j columns.
See also: LeadingMatrixBlock, BlockMatrix.

 UpperTriangleToMatrix (utm)
This function takes a list of lists of strictly decreasing length and forms a matrix
with zeros in the lower triangle and with the given lists constituting the rows of
the upper triangle. The length of the matrix is the length of the first sublist. The
last sublist has length 1 and each successive sublist has length one less than the
preceding sublist.
UpperTriangleToMatrix[u] −→ a
u is a list of lists of CREs of decreasing length representing the elements of an
upper triangular submatrix including the diagonal,
 A.3 Function descriptions and examples 465

a is a full matrix with 0 in each lower-triangular position.

See also: MatrixUpperTriangle, LowerTriangleToMatrix.

 VolumeBall (vbl)
This function computes the volume of an n-dimensional ball with given radius.

VolumeBall[r, n] −→ Vol

r is a CRE representing the radius of the ball,

n is a positive integer, being the dimension of the ball,
Vol is the n-dimensional volume of the ball.

Example

See also: VolumeSphere, VolumeHn.

 VolumeFormDiagonal (vfd)
This function computes the differential volume form for the set of diagonal
matrices
A
n
dai ,
i=1

where the product is the wedge product.

See Sections 1.4 and 1.5.

VolumeFormDiagonal[“a”, n] −→ Form

“a” is a string which will be the name of a one-dimensional array symbol,

n is a positive integer representing the dimension of the form,
Form is the diagonal volume form based on the variables a[i].

See also: VolumeFormGln, VolumeFormHn, VolumeFormUnimodular.

466 Appendix The GL(n)pack Manual

 VolumeFormGln (vfg)
This function computes the differential volume form for the matrix group
G L(n, R) using the wedge product.
See Sections 1.4 and 1.5 and Proposition 1.4.3.
VolumeFormGln[“g”, n] −→ Form
“g” is a string which will be the name of a two-dimensional array symbol,
n is a positive integer representing the dimension of the matrices,
Form is the diagonal volume form based on the variables g[i, j].
Example

See also: VolumeFormHn, VolumeFormDiagonal, VolumeFormUnimodular.

 VolumeFormHn (vfh)
This function computes the differential volume form for the generalized upper
half-plane.
See Definition 1.2.3 and Proposition 1.5.3.
VolumeFormHn[“x”, “y”, n] −→ Form
“x” is a string which will be the name of a two-dimensional array symbol,
“y” is a string which will be used as the name of a one-dimensional array
symbol,
n is a positive integer representing the dimension of the matrices which appear
in the Iwasawa decomposition,
Form is the volume form based on the variables x[i, j], y[j].
Example

See also: VolumeFormGln, VolumeFormDiagonal, VolumeFormUnimodular.

 VolumeFormUnimodular (vfu)
This function computes the differential volume form for the group of unimod-
ular matrices, i.e. real upper-triangular with 1s along the leading diagonal.
See Sections 1.4 and 1.5.
 A.3 Function descriptions and examples 467

VolumeFormHn[“x”, n] −→ Form
“x” is a string being the name of an array symbol,
n is a positive integer representing the dimension of the matrices,
Form is the volume form based on the variables x[i, j].
Example

See also: VolumeFormHn, VolumeFormDiagonal, VolumeFormGln.

 VolumeHn (vhn)
This function computes the volume of the generalized upper half-plane using
the volume element VolumeFormHn.
See Example 1.5.2 and Proposition 1.5.3.
VolumeHn[n] −→ Vol
n is an integer with n ≥ 2 representing the order of the upper half-plane, being
the size of the matrices appearing in the Iwasawa form,
Vol is a real number.
See also: VolumeBall, VolumeSphere.

 VolumeSphere (vsp)
This function computes the n-dimensional volume of the sphere S n in Rn+1 .
VolumeSphere[r, n] −→ Vol
r is a CRE being the radius of the sphere,
n is the dimension of the sphere,
Vol is the volume of the sphere computed using n-dimensional Lebesgue
measure.
See also: VolumeBall, VolumeHn.

 Wedge,d
This function computes the Wedge product of any finite number of functions or
differential forms in an arbitrary number of dimensions. It works with the dif-
ferential form operator d. Note that these functions have a different construction
from others in GL(n)pack , and have limited error control. An alternative to the
function Wedge is the infix operator which may be entered into Mathematica
468 Appendix The GL(n)pack Manual

by typing a backslash, and open square bracket, the word “Wedge” and then a
closing square bracket. It prints like circumflex, but is not the same. Note that
wedge products of vectors are not currently supported.
See Sections 1.4, 1.5, 5.6, 5.7 and 5.8.
Wedge[f1 , f2 , . . . , fn ] −→ value
fi is an expression or a form,
value is the wedge product of the functions or forms fi .
Example
In this example the function Wedge is used in conjunction with the differential
form generator function d. Note that symbols, such as a, can be declared to be
constant explicitly by setting, d[a] = 0.

See also: VolumeFormGln, VolumeFormHn.

 WeylGenerator (wge)
This function returns a set of matrix generators for the Weyl subgroup of the
group of integer matrices with determiant ±1, which consists of all matrices
with exactly one ±1 in each row and column. A single call returns a single
generator.
 A.3 Function descriptions and examples 469

See Chapter 6.
Also see the manual entry for SpecialWeylGroup.

WeylGenerator[n,i,j] −→ g

n is a positive integer with n ≥ 2,

i is a positive integer with i ≤ n,
j is a positive integer with i = j ≤ n,
g a matrix with 1s along the leading diagonal and zeros elsewhere, except in the
(i, j)th position where the value is -1 and (j, i)th position where the value
is 1 and where the corresponding (i, i)th and (j, j)th diagonal elements are
0.

Example

See also: ModularGenerators, LongElement.

 WeylGroup (wgr)
This function returns, for each whole number n, a list of all of the Weyl group
of n by n permutation matrices.
See the proof of Proposition 1.5.3.

WeylGroup[n] −→ {m1 , m2 , . . . , mk }

n is a positive integer with n ≥ 1,

mj is a matrix with a single 1 in each row and column.

Example
470 Appendix The GL(n)pack Manual

See also: WeylGenerators, SpecialWeylGroup, LongElement.

 Whittaker (wit)
This function computes a symbolic interated integral representatin of the gen-
eralized Jacquet Whittaker function WJacquet (also written W J ) of order n,
for n ≥ 2, as defined by Equation (5.5.1). See Proposition 3.4.6, Section 3.4,
Equation (5.5.1) and Equation (5.5.5). The algorithm uses the recursive rep-
resentation of the Whittaker function defined by Stade (1990, Theorem 2.1)
related to that used in the book as follows. Let W S and W S∗ be Stade’s Whittaker
and Whittaker starred functions respectively and let  represent the gamma
factors for either form. Then

Q · W S∗ =  · W J = W J∗ = W S

where

n−1
−µ j
Q = Iν (y) yj ,
j=1

n− j
µj = r j,k ,
k=1

k+ j−1 

nνi j
r j,k = − .
i=k
2 2

Whittaker[z, v, m, u] −→ {coef, char, gam, value}

z is an n × n non-singular CRE matrix,

v is a list of n − 1 CREs,
m is a list +of n − 1 CREs (mi ) representing a character
ψm (x) = e2π i( 1≤i≤n−1 mi xi,i+1 )) ,
u is a symbol which will be used to form the dummy variables in the iterated
integral,
coef is the coefficient cν,m defined in Proposition 5.5.2,
char is the value of the character ψm (x) for z = x.y the Iwasawa form,
gam is a product of terms as returned by WhittakerGamma,
value is a symbolic expression being the value of the Whittaker function at My
where z = x.y with parameters v and character ψ1,1,1,..,1 . In this expres-
sion the K-Bessel function at a complex argument and parameter Kν (z), is
represented by the noun function K[ν, z].
 A.3 Function descriptions and examples 471

Example

See also: WhittakerGamma.

 WhittakerGamma (wig)
This function returns the gamma factors for the generalized Jacquet Whittaker
function. See Definition 5.9.2. Note that although this definition differs from
that in (Stade, 1990), the gamma factor that it represents is the same.
WhittakerGamma[v] −→ value
v is a list of n − 1 CREs which, if any are numerical, satisfy vi > 1/n,
value is an expression, being the product of the gamma factors for the Whittaker
function of order n.
See also: Whittaker.
472 Appendix The GL(n)pack Manual

 WMatrix (wmx)
This function returns the so-called w-matrix, with (−1)!n/2" in the (1, n)th
position and 1 in every other reversed diagonal position, a member of SL(n, Z).
See Section 5.5.
WMatrix[n] −→ w
n is a strictly positive integer with n ≥ 2,
w is an n by n matrix with each element 0, except the (1, n)th which is (−1)!n/2"
and every (i, n − i + 1)th which is 1 for 2 ≤ i ≤ n.
Example

See also: LongElement.

 ZeroMatrix (zmx)
This function returns a zero matrix of given dimensions.
ZeroMatrix[m,n] −→ Z
m is a strictly positive integer representing the number of matrix rows,
n is a strictly positive integer representing the number of matrix columns,
Z is a zero matrix with m rows and n columns.
See also: ConstantMatrix.
 References

Ahlfors L. (1966), Complex Analysis, Int. Series in Pure and Applied Math. New York,
McGraw–Hill Book Company, pp. 133–7.
Arthur J. (1979), Eisenstein series and the trace formula, in Automorphic forms, rep-
resentations and L-functions (Proc. Sympos. Pure Math., Oregon State University,
Corvallis, OR, 1977), Proc. Sympos. Pure Math., Vol. XXXIII, Providence, RI,
Amer. Math. Soc., Part 1, pp. 253–74.
(1989), Unipotent automorphic representations: conjectures, Orbites unipotentes et
représentations, II, Astérisque No. 171–2, 13–71.
(2002), A note on the automorphic Langlands group, Dedicated to Robert V. Moody,
Canad. Math. Bull. 45 (4), 466–82.
(2003), The principle of functoriality. Mathematical challenges of the 21st century,
(Los Angeles, CA, 2000). Bull. Amer. Math. Soc. (N.S.) 40 (1), 39–53.
Artin E. (1923), Über eine neue Art von L-Reihen, Hamb. Math. Abh. 3, 89–108.
(Collected papers, Edited by Serge Lang and John T. Tate. Reprint of the 1965
original. New York–Berlin, Springer-Verlag, (1982), pp. 105–24).
(1927), Beweis des allgemeinen Reziprozitätsgesetzes, Hamb. Math. Abh., 353–63.
(Collected papers, Edited by Serge Lang and John T. Tate. Reprint of the 1965
original. New York–Berlin, Springer-Verlag, (1982), pp. 131–41).
(1930), Zur Theorie der L-Reihen mit allgemeinen Gruppencharakteren, Hamb. Math.
Abh. 8, 292–306. (Collected papers, Edited by Serge Lang and John T. Tate. Reprint
of the 1965 original. New York–Berlin, Springer-Verlag, (1982) pp. 165–79).
Arveson W. (2002), A short course on spectral theory, Graduate Texts in Mathematics
209, New York, Springer-Verlag.
Atkin A. O. and Li W. C. (1978), Twists of newforms and pseudo-eigenvalues of W -
operators, Invent. Math. 48 (3), 221–43.
Aupetit B. (1991), A primer on spectral theory, Universitext. New York, Springer-Verlag.
Baker Alan. (1971), Imaginary quadratic fields with class number “2”, Ann. Math. (2)
94, 139–52.
Baker Andrew. (2002) Matrix groups, An introduction to Lie group theory, Springer
Undergraduate Mathematics Series, London, Springer-Verlag.
Banks W. (1997), Twisted symmetric-square L-functions and the nonexistence of Siegel
zeros on GL(3), Duke Math. J. 87 (2), 343–53.

473
474 References

Bernstein J. (2002), Meromorphic Continuation of Eisenstein Series, IAS/Park City

Lecture Notes, Park City, UT.
Bernstein J. and Gelbart S. (Editors) (2003), An Introduction to the Langlands Program,
Boston, Birkhäuser.
Bombieri E., Friedlander J. B. and Iwaniec H. (1987), Primes in arithmetic progressions
to large moduli. II, Ann. Math. 277 (3), 361–93.
Boothby W. M. (1986), An introduction to differentiable manifolds and Riemannian
geometry, Second edition, Pure and Applied Mathematics 120, Orlando, FL, Aca-
demic Press.
Borel A. (2001), Essays in the History of Lie Groups and Algebraic Groups, History of
Math. 21, Amer. Math. Soc., London Math. Soc., pp. 5–6.
(1966), Automorphic forms, in Algebraic Groups and Discontinuous Groups, Proc.
Sympos. Pure Math. Vol. IX, Amer. Math. Soc., pp. 199–210.
(1979), Automorphic L-functions. Automorphic forms, representations and L-
functions, (Proc. Sympos. Pure Math., Oregon State University, Corvallis, OR,
1977), Part 2, Proc. Sympos. Pure Math., Vol. XXXIII, Providence, RI, Amer.
Math. Soc., pp. 27–61.
Borel A. and Harish-Chandra (1962), Arithmetic subgroups of algebraic groups, Ann.
Math. (2) 75, 485–535.
Bourbaki N. (1998a), Algebra I, Chapters 1–3, Translated from the French. Reprint
of the 1989 English translation, Elements of Mathematics, Berlin, Springer-
Verlag.
(1998b), Lie groups and Lie algebras, Chapters 1–3, Translated from the French,
Reprint of the 1989 English translation, Elements of Mathematics, Berlin, Springer-
Verlag.
(2003), Algebra II, Chapters 4–7, Translated from the 1981 French edition by P. M.
Cohn and J. Howie, Reprint of the 1990 English edition. Elements of Mathematics,
Berlin, Springer-Verlag.
Brauer R. (1947), On Artin’s L-series with general group characters, Ann. Math. (2) 48,
502–14.
Brown W. (1988), A second course in linear algebra, A Wiley-Interscience Publication,
New York, John Wiley.
Bruggeman R. W. (1978), Fourier coefficients of cusp forms, Invent. Math. 45 (1), 1–18.
Buhler J. (1978), Icosahedral Galois representations, Lecture Notes in Mathematics
654, Berlin–New York, Springer-Verlag.
Bump D. (1984), Automorphic Forms on G L(3,R), Lecture Notes in Mathematics 1083,
Springer-Verlag.
(1987), The Rankin–Selberg method: A survey, in Number Theory, Trace Formulas
and Discrete Groups, Symposium in honor of Atle Selberg, Oslo, Norway Edited
by K. E. Aubert, E. Bombieri, D. Goldfeld, Academic Press, pp. 49–109.
(1997), Automorphic forms and representations, Cambridge Studies in Advanced
Mathematics 55, Cambridge, Cambridge University Press.
(2004), Lie Groups, Graduate Texts in Mathematics 225, Springer-Verlag.
(to appear), The Rankin–Selberg method: An introduction and survey.
Bump D., Duke W., Hoffstein J. and Iwaniec H. (1992), An estimate for the Hecke
eigenvalues of Maass forms, Int. Math. Res. Notices 4, 75–81.
 References 475

Bump D. and Friedberg S. (1990), The exterior square automorphic L-functions on

G L(n), Festschrift in honor of I. I. Piatetski-Shapiro on the occasion of his
sixtieth birthday, Part II (Ramat Aviv, 1989), Israel Math. Conf. Proc., 3, Weiz-
mann, Jerusalem, pp. 47–65.
Bump D., Friedberg S. and Goldfeld D. (1988), Poincaré series and Kloosterman sums
for S L(3,Z), Acta Arith. 50, 31–89.
Bump D. and Ginzburg D. (1992), The symmetric square L-functions on G L(r ), Ann.
Math. 136, 137–205.
Bump D., Ginzburg D. and Hoffstein J. (1996), The symmetric cube, Invent. Math. 125
(3), 413–49.
Buzzard K., Dickinson M., Shepherd-Barron N. and Taylor R. (2001), On icosahedral
Artin representations, Duke Math. J. 109 (2), 283–318.
Buzzard K. and Stein W. (2002), A mod five approach to modularity of icosahedral
Galois representations, Pacific J. Math. 203 (2), 265–82.
Capelli A. (1890), Sur les opérations dans la théorie des formes algébriques, Math.
Annalen 37, 1–27.
Carayol H. (1992), Variétés de Drinfeld compactes, d’aprés Laumon, Rapoport et
Stuhler, (French. French summary) [Compact Drinfeld varieties, after Laumon,
Rapoport and Stuhler] Séminaire Bourbaki, Vol. 1991/92. Astérisque No. 206,
Exp. No. 756, 5, 369–409.
(2000), Preuve de la conjecture de Langlands locale pour G L n : travaux de Harris-
Taylor et Henniart, (French, French summary) [Proof of the local Langlands con-
jecture for G L n : works of Harris and Taylor and of Henniart] Séminaire Bourbaki,
Vol. 1998/99. Astérisque No. 266, Exp. No. 857, 4, 191–243.
Casselman W. and Shalika J. (1980), The unramified principal series of p-adic groups
II: The Whittaker function, Compositio Math. 41, 207–31.
Cassels J. W. S. and Fröhlich A. (Eds.) (1986), Algebraic number theory, Proceedings of
the instructional conference held at the University of Sussex, Brighton, September
1–17, 1965, Reprint of the 1967 original, London, Academic Press [Harcourt Brace
Jovanovich, Publishers].
Chandrasekharan K. and Narasimhan R. (1962), Functional equations with multiple
Gamma factors and the average order of arithmetical functions, Ann. Math. 76,
93–136.
Cogdell J. W. and Piatetski-Shapiro I. I. (1990), The arithmetic and spectral analysis of
Poincaré series, Perspectives in Mathematics, 13, Boston, MA, Academic Press.
(1994), Converse theorems for G L n , Publ. Math. IHES 79, 157–214.
(1999), Converse theorems for G L n , II, J. Reine Angew. Math. 507, 165–88.
(2001), Converse theorems for G L n and their applications to liftings, Cohomology of
Arithmetic Groups, Automorphic Forms, and L-functions, Mumbai 1998, Tata Inst.
of Fund. Res., Narosa, pp. 1–34.
Cogdell J. W., Kim H. H., Piatetski-Shapiro I. I. and Shahidi F. (2001), On lifting from
classical groups to G L N , Inst. Hautes Études Sci. Publ. Math. 93, 5–30.
(2004), Functoriality for the classical groups, Inst. Hautes Études Sci. Publ. Math.
99, 163–233.
Cohen H. (1993). A Course in Computational Algebraic Number Theory, Graduate Texts
in Mathematics, Springer-Verlag.
476 References

Conrey J. B. and Ghosh A. (1993), On the Selberg class of Dirichlet series: small degrees,
Duke Math. J. 72, 673–93.
Conway J. B. (1973), Functions of One Complex Variable, Graduate Texts in Mathe-
matics, Springer-Verlag.
Curtis M. L. (1984), Matrix groups, Second edition, Universitext, New York, Springer-
Verlag.
Dabrowski R. (1993), Kloosterman sums for Chevalley groups, Trans. Amer. Math. Soc.
337 (2), 757–69.
Dabrowski R. and Fisher B. (1997), A stationary phase formula for exponential sums
over Z/ p m Z and applications to G L(3)-Kloosterman sums, Acta Arith. 80 (1),
1–48.
Davenport H. (1974), Multiplicative Number Theory, Markham Publishing (1967), Sec-
ond Edition (Revised by H. Montgomery) Graduate Texts in Mathematics, Springer-
Verlag.
Deligne P. (1974), La conjecture de Weil. I, (French) Inst. Hautes Études Sci. Publ. Math.
43, 273–307.
(1977), Application de la formule des traces aux sommes trigonométriques, in SGA4 12 ,
Springer Lecture Notes 569, 168–232.
Deuring M. (1933), Imaginäre quadratische Zahlkörper mit der Klassenzahl (1), Math.
Zeit. 37, 405–15.
Drinfeld V. G. (1989), Cohomology of compactified moduli varieties of F-sheaves of
rank 2, (Russian) Zap. Nauchn. Sem. Leningrad. Otdel. Mat. Inst. Steklov. (LOMI)
162 (1987), Avtomorfn. Funkts. i Teor. Chisel. III, 107–158, 189; translation in J.
Soviet Math. 46 (2), 1789–821.
Duke W. and Iwaniec H. (1989), Estimates for coefficients of L-functions, I, Automorphic
Forms and Analytic Number Theory (Montreal, PQ, 1989), CRM, pp. 43–47.
Edelen D. G. B. (1985), Applied exterior calculus, A Wiley-Interscience Publication,
New York, John Wiley.
Frey G. (editor) (1994), On Artin’s Conjecture for Odd 2-dimensional Representations,
Berlin, Springer-Verlag.
Friedberg S. (1987a), Poincaré series, Kloosterman sums, trace formulas, and automor-
phic forms for G L(n), Séminaire de Théorie des Nombres, Paris 1985–86, Progr.
Math. 71, Boston, MA, Birkhäuser, 53–66.
(1987b), A global approach to the Rankin–Selberg convolution for G L(3,Z), Trans.
Amer. Math. Soc. 300, 159–78.
(1987c), Poincaré series for G L(n): Fourier expansion, Kloosterman sums, and
algebreo-geometric estimates, Math. Zeit. 196 (2), 165–88.
Friedberg S. and Goldfeld D. (1993), Mellin transforms of Whittaker functions, Bull.
Soc. Math. France, 121, 91–107.
Garrett P. (2002), Volume of S L(n,Z)\S L(n,R) and Spn (Z)\Spn (R), preprint.
Gauss C. F. (1801), Disquisitiones Arithmeticae, Göttingen; English translation by
A. Clarke, revised by W. Waterhouse, New Haven, Yale University Press, 1966;
reprinted by Springer-Verlag, 1986.
Gelbart S. (1984), An elementary introduction to the Langlands program, Bull. Amer.
Math. Soc. (N.S.) 10 (2), 177–219.
Gelbart S. and Jacquet H. (1978), A relation between automorphic representations of
G L(2) and G L(3), Ann. Sci. École Norm. Sup. 4e série 11, 471–552.
 References 477

Gelbart S., Lapid E. and Sarnak P. (2004), A new method for lower bounds of L-functions,
(English. English, French summary) C. R. Math. Acad. Sci. Paris 339 (2), 91–4.
Gelbart S. and Shahidi F. (1988), Analytic properties of automorphic L-functions, Per-
spectives in Mathematics 6, Boston, MA, Academic Press.
Gelfand I. M., Graev M. I. and Pyatetskii-Shapiro I. I. (1990), Representation theory
and automorphic functions, Translated from the Russian by K. A. Hirsch. Reprint
of the 1969 edition. Generalized Functions 6, Boston, MA, Academic Press.
Godement R. (1966), The spectral decomposition of cusp-forms, Algebraic Groups
and Discontinuous Subgroups (Proc. Sympos. Pure Math., Boulder, CO, 1965),
pp. 225–34 Amer. Math. Soc., Providence, RI.
Godement R. and Jacquet H. (1972), Zeta functions of simple algebras, Lecture Notes
in Mathematics, Vol. 260. Berlin–New York, Springer-Verlag.
Goldfeld D. (1974), A simple proof of Siegel’s theorem, Proc. Nat. Acad. Sci. USA. 71,
1055.
(1976), The class number of quadratic fields and the conjecture of Birch and
Swinnerton-Dyer, Ann. Scuola Norm. Sup. Pisa (4) 3, 624–63.
(1985), Gauss’ class number problem for imaginary quadratic fields, Bull. Amer.
Math. Soc. 13, 23–37.
(1987), Kloosterman zeta functions for G L(n,Z ), Proceedings of the International
Congress of Mathematicians, Vol. 1, 2 (Berkeley, CA, 1986), Providence, RI, Amer.
Math. Soc., pp. 417–24.
Goldfeld D., Hoffstein J., Lieman D. (1994), An effective zero free region, in the
Appendix of Coefficients of Maass forms and the Siegel zero, Ann. of Math. 140,
177–81.
(2004), The Gauss class number problem for imaginary quadratic fields, in Heegner
Points and Rankin L-series, MSRI Publications 49, 25–36.
Goldfeld D. and Satnak P. (1983), Sums of Kloosterman sums, Invent. Math. 71 (2),
243–50.
Gross B. and Zagier D. B. (1986), Heegner points and derivatives of L-series, Invent.
Math. 84, 225–320.
Halmos P. (1974), Measure Theory, Graduate Texts in Mathematics 18, Springer-Verlag,
250–62.
Hamburger H. (1921), Über die Riemannsche Funktionalgleichung der ζ –Funktion,
Math. Zeit. 10, 240–54.
Harcos G. (2002), Uniform approximate functional equations for principal L-functions,
IMRN 18, 923–32.
Harish-Chandra (1959), Automorphic forms on a semi-simple Lie group, Proc. Nat.
Acad. Sci. U.S.A. 45, 570–3.
(1966), Discrete series for semisimple Lie groups, II. Explicit determination of the
characters, Acta Math. 116, 1–111.
(1968), Automorphic forms on semisimple Lie groups, Notes by J. G. M. Mars. Lecture
Notes in Mathematics, No. 62, Berlin–New York, Springer-Verlag.
Harris M. and Taylor R. (2001), The geometry and cohomology of some simple Shimura
varieties, With an appendix by Vladimir G. Berkovich, Ann. Math. 151, Princeton,
NJ, Princeton University Press.
Hecke E. (1920), Eine neue Art von Zetafunktionen und ihre Beziehungen zur Verteilung
der Primzahlen, Zweite Mitteilung, Math. Zeit. 6, 11–51.
478 References

(1936), Über die Bestimmung Dirichletscher Reihen durch ihre Funktionalgleichung,

Ann. Math. 113, 664–99.
(1937a), Über Modulfunktionen und die Dirichletschen Reihen mit Eulerscher Pro-
duktententwicklung, I, Ann. Math. 114, 1–28.
(1937b), Über Modulfunktionen und die Dirichletschen Reihen mit Eulerscher Pro-
duktententwicklung, II, Ann. Math. 114, 316–51.
Heegner K. (1952) Diophantine Analysis und Modulfunktionen, Math. Zeit. 56, 227–53.
Heilbronn H. (1934), On the class number in imaginary quadratic fields, Quarterly J.
Math. 5, 150–60.
Hejhal D. (1976), The Selberg Trace Formula for P S L(2,R), Volume 1, Lecture Notes
in Mathematics 548, Springer-Verlag.
(1983), The Selberg Trace Formula for P S L(2,R), Volume 2, Lecture Notes in Math-
ematics 1001, Springer-Verlag.
Henniart G. (2000), Une preuve simple des conjectures de Langlands pour G L(n) sur un
corps p-adique, (French) [A simple proof of the Langlands conjectures for G L(n)
over a p-adic field] Invent. Math. 139 (2), 439–55.
(2002), Progrès récents en fonctorialité de Langlands, (French) [Recent results on
Langlands functoriality] Seminaire Bourbaki, Vol. 2000/2001. Astérisque No. 282
Exp. No. 890, ix, 301–22.
Hewitt E. and Ross K. A. (1979), Abstract harmonic analysis, Vol. I. Structure of topolog-
ical groups, integration theory, group representations, Second edition, Grundlehren
der Mathematischen Wissenschaften (Fundamental Principles of Mathematical
Sciences), 115, Berlin–New York, Springer-Verlag.
Hoffstein J. and Lockhart P. (1994), Coefficients of Maass forms and the Siegel zero,
Ann. Math. 140, 161–81.
Hoffstein J. and Murty M. R. (1989), L-series of automorphic forms on GL(3,R),
Théorie des Nombres (Quebec, PQ, 1987), Berlin, de Gruyter, pp. 398–408.
Hoffstein J. and Ramakrishnan D. (1995), Siegel zeros and cusp forms, IMRN 6, 279–
308.
Ivić A. (1995), An approximate functional equation for a class of Dirichlet series, J. Anal.
3, 241–52.
Iwaniec H. (1990), Small eigenvalues of Laplacian for 0 (N ), Acta Arith. 56, 65–82.
(1995), Introduction to the Spectral Theory of Automorphic Forms, Revista
Matemática Iberoamericana.
Iwaniec H. and Kowalski E. (2004), Analytic Number Theory, Amer. Math. Soc. Colloq.
Publications, Vol. 53.
Iwasawa K. (1949), On some types of topological groups, Ann. Math. 50, 507–58.
Jacquet H. (1967), Fonctions de Whittaker associes aux groupes de Chevalley, (French)
Bull. Soc. Math. France 95, 243–309.
(1972), Automorphic Forms on G L(2), Part II, Lecture Notes in Mathematics No.
278, Springer-Verlag.
(1981), Dirichlet series for the group G L(N ), in Automorphic Forms, Representation
Theory and Arithmetic, Tata Inst. of Fund. Res., Springer-Verlag, pp. 155–64.
(1997), Note on the analytic continuation of Eisenstein series, An appendix to
“Theoretical aspects of the trace formula for G L(2)” [in Representation the-
ory and automorphic forms (Edinburgh, 1996), 355–405, Proc. Sympos. Pure
Math., 61, Amer. Math. Soc., Providence, RI, 1997; MR1476505 (98k:11062)] by
 References 479

A. W. Knapp. Representation theory and automorphic forms (Edinburgh, 1996),

pp. 407–412, Proc. Sympos. Pure Math., 61, Amer. Math. Soc., Providence, RI.
(2004a), Integral Representation of Whittaker Functions, Contributions to automor-
phic forms, geometry, and number theory, Baltimore, MD, Johns Hopkins Univer-
sity Press, 373–419.
(2004b), Kloosterman identities over a quadratic extension, Ann. Math., no. 2, 755–
79.
Jacquet H. and Langlands R. P. (1970), Automorphic forms on G L(2), Lecture Notes in
Mathematics, Vol. 114. Berlin–New York, Springer-Verlag.
Jacquet H., Piatetski-Shapiro I. I. and Shalika J. (1979), Automorphic forms on G L(3),
I, II, Ann. Math. 109, 169–258.
(1983), Rankin–Selberg convolutions, Amer. J. Math. 105, 367–464.
Jacquet H. and Shalika J. (1976/77), A non-vanishing theorem for zeta functions of GLn ,
Invent. Math. 38 (1), 1–16.
(1981), On Euler products and the classification of automorphic representations,
Amer. J. Math. I: 103, 499–588.
(1990), Rankin–Selberg convolutions: Archimedean theory, in Festschrift in Honor
of I. I. Piatetski-Shapiro, Part I, Jerusalem, Weizmann Science Press, pp. 125–207.
Kaczorowski J. and Perelli A. (1999), On the structure of the Selberg class, I: 0 ≤ d ≤ 1,
Acta Math. 182, 207–41.
(2002), On the structure of the Selberg class, V: 1 < d < 5/3, Invent. Math. 150,
485–516.
Kato S. (1978), On an explicit formula for class-1 Whittaker functions on split reductive
groups over p-adic fields, preprint, University of Tokyo.
Kim H. (2003), Functoriality for the exterior square of G L 4 and the symmetric fourth
of G L 2 , J. Amer. Math. Soc. 16, 139–83.
Kim H. and Sarnak P. (2003), Appendix 2 in Functoriality for the exterior square of
G L 4 and the symmetric fourth of G L 2 , J. Amer. Math. Soc. 16 (1), 139–83.
Kim H. and Shahidi F. (2000), Functorial products for G L 2 × G L 3 and functorial
symmetric cube for G L 2 , C. R. Acad. Sci. Paris SÉr. I Math. 331 (8), 599–604.
(2002), Functorial products for G L 2 × G L 3 and the symmetric cube for G L 2 , Ann.
Math. (2), 155 (3), 837–93.
Korkine A. and Zolotareff G. (1873), Sur les formes quadratiques, Ann. Math., 6, 366–
89.
Kowalski E. (2003), Elementary theory of L-functions, I, II, in An Introduction to the
Langlands Program, Edited by J. Bernstein and S. Gelbart, Birkhäuser, pp. 1–71.
Kuznecov N. V. (1980), The Petersson conjecture for cusp forms of weight zero and the
Linnik conjecture. Sums of Kloosterman sums, (Russian) Mat. Sb. (N.S.) 111 (153)
(3), 334–83, 479.
Lafforgue L. (2002), Chtoucas de Drinfeld et correspondance de Langlands, (French)
[Drinfeld shtukas and Langlands correspondence] Invent. Math. 147 (1), 1–241.
Landau E. (1918), Über die Klassenzahl imaginär-quadratischer Zahlkörper, Göttinger
Nachtr., pp. 285–95.
(1935), Bemurkungen zum Heilbronnschen Satz, Acta Arith. I, 1–18.
Lang S. (1969), Analysis II, Mass. Addison-Wesley.
(2002), Algebra, Revised third edition, Graduate Texts in Mathematics 211, New
York, Springer-Verlag.
480 References

Langlands R. P. (1966), Eisenstein series, and Volume of the fundamental domain for
some arithmetical subgroups of Chevalley groups, in Proc. Symp. Pure Math.,
Vol. IX, Algebraic Groups and Discontinuous Subgroups, Amer. Math. Soc.,
pp. 235–57.
(1970), Problems in the theory of automorphic forms, in Lecture Notes in Mathematics
170, Springer-Verlag.
(1971), Euler Products, Yale University Press.
(1976), On the functional equations satisfied by Eisenstein series, Lecture Notes in
Mathematics 544, Springer-Verlag.
(1980), Base Change for G L(2), Princeton, NJ, Princeton University Press.
Larsen M. (1988), Estimation of S L(3,Z) Kloosterman sums, in the Appendix
to Poincaré series and Kloosterman sums for S L(3,Z), Acta Arith. 50,
86–9.
Laumon G., Rapoport M. and Stuhler U. (1993), D-elliptic sheaves and the Langlands
correspondence, Invent. Math. 113 (2), 217–338.
Lavrik A. F. (1966), Funktional equations of Dirichlet L-functions, Soviet Math. Dokl.
7, 1471–3.
Lax P. D. and Phillips R. S. (1976), Scattering Theory for Automorphic Functions, Ann.
Math., No. 87. Princeton, NJ, Princeton University Press, pp. 7–11.
Lenstra H. W. and Stevenhagen P. (2000), Artin reciprocity and Mersenne primes, Nieuw
Arch. Wisk. (5) 1, 44–54.
Li W. C. (1975), Newforms and functional equations, Ann. Math. 212, 285–315.
(1979), L-series of Rankin type and their functional equations, Ann. Math. 244 (2),
135–66.
Lieman D. B. (1993), The GL(3) Rankin–Selberg convolution for functions not of rapid
decay, Duke Math. J. 69 (1), 219–42.
Lindenstrauss E. and Venkatesh A. (to appear), Existence and Weyl’s law for spherical
cusp forms.
Luo W., Rudnick Z. and Sarnak P. (1995), On Selberg’s eigenvalue conjecture, Geom.
Funct. Anal. 5 (2), 387–401.
(1999), On the generalized Ramanujan conjecture for G L(n), Automorphic forms,
automorphic representations, and arithmetic (Fort Worth, TX, 1996), Proc. Sympos.
Pure Math. 66, Part 2, Amer. Math. Soc., Providence, RI, 301–10.
Maass H. (1949), Über eine neue Art von nichtanalytischen automorphen Funktionen
und die Bestimmung Dirichletscher Reihen durch Funktionalgleichungen, Ann.
Math. 122, 141–83.
(1964), Lectures on Modular Functions of One Complex Variable, Tata Inst. Fund.
Research, Bombay.
Macdonald I. G. (1979), Symmetric Functions and Hall Polynomials, Oxford, Clarendon
Press, Ch. 1, Sect. 4, pp. 32–4.
Manin J. I. and Pančiškin A. A. (1977), Convolutions of Hecke series, and their values
at lattice points, (Russian) Mat. Sb. (N.S.) 104(146) (4), 617–51.
Miller S. D. (2001), On the existence and temperedness of cusp forms for S L 3 (Z),
J. Reine Angew. Math. 533, 127–69.
Miller S. D. and Schmid W. (2004), Summation formulas, from Poisson and Voronoi
to the present, in Noncommutative Harmonic Analysis, Progr. Math. 220, Boston,
MA, Birkhäuser, 419–40.
 References 481

Moeglin C. and Waldspurger J. L. (1995), Spectral Decomposition and Eisenstein Series,

Cambridge University Press.
(1989), Le spectre résiduel de G L(n), Ann. Sci. École Norm. Sup. (4) 22, 605–74.
Mordell L. J. (1934), On the Riemann hypothesis and imaginary quadratic fields with a
given class number, J. London Math. Soc. 9, 289–98.
Molteni G. (2002), Upper and lower bounds at s = 1 for certain Dirichlet series with
Euler product, Duke Math. J. 111 (1), 133–58.
Moreno C. (1985), Analytic proof of the strong multiplicity one theorem, Amer. J. Math.
107 (1), 163–206.
(2005), Advanced analytic number theory: L-functions, Mathematical Surveys and
Monographs 115, Providence, RI, Amer. Math. Soc.
Motohashi Y. (1997), Spectral theory of the Riemann zeta-function, Cambridge Tracts
in Mathematics 127, Cambridge, Cambridge University Press.
Müller W. (2004), Weyl’s law for the cuspidal spectrum of S L n , C. R. Math. Acad. Sci.
Paris 338 (5), 347–52.
Munkres J. R. (1991), Analysis on manifolds, Advanced Book Program, Addison-Wesley
Publishing Company, Redwood City, CA.
Murty M. R. (1994), Selberg’s conjectures and Artin L-functions, Bull. Amer. Math.
Soc. (N.S.) 31 (1), 1–14.
Murty M. R. and Murty V. K. (1987), A variant of the Bombieri–Vinogradov theorem,
Number theory (Montreal, Que., 1985), CMS Conf. Proc., Vol. 7, Amer. Math. Soc.
Providence, RI.
Neunhöffer H. (1973), Über die analytische Fortsetzung von Poincaréreihen, S.-B.
Heidelberger Akad. Wiss. Math.-Natur. Kl., pp. 33–90.
Osborne M. S. and Warner G. (1981), The theory of Eisenstein systems, Pure and Applied
Math. 99, Academic Press.
Philips R. S. and Sarnak P. (1985), On cusp forms for co-finite subgroups of PSL (2, R),
Invent. Math. 80 (2), 339–64.
Phragmén E. and Lindelöf E. (1908), Sur une extension d’un principe classique de
l’analyse, Acta Math. 31, 381–406.
Piatetski-Shapiro I. I. (1975), Euler subgroups. Lie groups and their representations,
(Proc. Summer School, Bolyai Jnos Math. Soc., Budapest, 1971), New York,
Halsted, pp. 597–620.
Ramakrishnan D. (2002), Modularity of solvable Artin representations of GO(4)-type,
Int. Math. Res. Not., Number 1, 1–54.
Ramakrishnan D. and Wang S. (2003), On the exceptional zeros of Rankin–Selberg
L-functions, Compositio Math. 135 (2), 211–44.
Rankin R. (1939), Contributions to the theory of Ramanujan’s function τ (n) and similar
arithmetic functions, I and II, Proc. Cambridge Phil. Soc. 35, 351–6, 357–73.
Richert H. E. (1957), Über Dirichletreihen mit Funktionalgleichung, Publ. Inst. Math.
Acad. Serbe Sci. 11, 73–124.
Riemann B. (1859), Ueber die Anzahl der Primzahlen unter einer gegebener Grösse,
Monatsberichte der Berliner Akad, Werke (2nd ed.), 145–53.
Roelcke W. (1956), Analytische Fortsetzung der Eisensteinreihen zu den parabolischen
Spitzen von Grenzkreisgruppen erster Art, Ann. Math. 132, 121–9.
Rohrlich D. (1989), Nonvanishing of L-functions for G L(2), Invent. Math. 97, 383–
401.
482 References

Roquette P. (2000), On the history of Artin’s L-functions and conductors. Seven letters
from Artin to Hasse in the year 1930, Mitteilungen der Mathematischen Gesellschaft
in Hamburg, Band 19, pp. 5–50.
Sarnak P. (1990), Some Applications of Modular Forms, Cambridge University Press
99.
(2004), Nonvanishing of L-functions on (s) = 1, Contributions to Automorphic
forms, Geometry, and Number Theory, Baltimore, MD, Johns Hopkins University
Press, pp. 719–32.
(to appear), The generalized Ramanujan conjectures.
Satake I., Spherical functions and Ramanujan conjecture, (1966) Algebraic Groups and
Discontinuous Subgroups (Proc. Sympos. Pure Math., Boulder, CO, 1965), Amer.
Math. Soc., Providence, RI, 258–64.
Selberg A. (1940), Bemerkungen über eine Dirichlesche reihe, die mit der theorie der
modulformer nahe verbunden ist, Arch. Math. Naturvid. 43, 47–50.
(1956), Harmonic analysis and discontinuous groups in weakly symmetric Rieman-
nian spaces with applications to Dirichlet series, J. Indian Math. Soc. 20, 47–87.
(1960), On discontinuous groups in higher dimensional symmetric spaces, in Contri-
butions to Function Theory, Tata Institute, Bombay, 147–64.
(1963), Discontinuous groups and harmonic analysis, Proc. Int. Congr. Mathemati-
cians (Stockholm, 1962), pp. 177–89.
(1965), On the estimation of Fourier coefficients of modular forms, Proc. Sympos.
Pure Math., Vol. VIII, Amer. Math. Soc., Providence, RI, 1–15.
(1991), Old and new conjectures and results about a class of Dirichlet series, Collected
Papers (Vol. II), Springer-Verlag, 47–63.
Serre J. P. (1977), Modular forms of weight one and Galois representations, Algebraic
number fields: L-functions and Galois properties (Proc. Symp. Durham University,
1975) Academic Press, 193–268.
(1981), Letter to J. M. Deshouillers.
Shahidi F. (1981), On certain L-functions, Amer. J. Math. 103, 297–355.
(1985), Local coefficients as Artin factors for real groups, Duke Math. J. 52 (4),
973–1007.
(1988), On the Ramanujan conjecture and finiteness of poles for certain L-functions,
Ann. Math. 127, 547–84.
(1989), Third symmetric power L-functions for G L(2), Comp. Math. 70, 245–73.
(1990a), On multiplicativity of local factors, in Festschrift in honor of I. I. Piatetski-
Shapiro, Part II, Israel Math. Conf. Proc. 3, Jerusalem, Weizmann, 279–89.
(1990b), A proof of Langlands conjecture on Plancherel measures; complementary
series for p-adic groups, Ann. Math. 132, 273–330.
(1992), Twisted endoscopy and reducibility of induced representations for p-adic
groups, Duke Math. J. (1) 66, 1–41.
Shalika J. A. (1973), On the multiplicity of the spectrum of the space of cusp forms of
G L n , Bull. Amer. Math. Soc. 79, 454–61.
(1974), The multiplicity one theorem for GL(n), Ann. Math. 100, 171–93.
Shimura G. (1971), Arithmetic Theory of Automorphic Functions, Princeton University
Press.
(1973), On modular forms of half-integral weight, Ann. Math. 97, 440–81.
 References 483

(1975), On the holomorphy of certain Dirichlet series, Proc. London Math. Soc. 31
(3), 79–98.
Shintani T. (1976), On an explicit formula for class-1 “Whittaker Functions” on G L(n)
over p-adic fields, Proc. Japan Acad. 52.
Siegel C. L. (1935), Über die Classenzahl quadratischer Zahlkörper, Acta Arith. I, 83–6.
(1936), The volume of the fundamental domain for some infinite groups, Trans. Amer.
Math. Soc. 30, 209–18.
(1939), Einheiten quadratischer Formen, Hamb. Abh. Math. 13, 209–39.
(1980), Advanced Analytic Number Theory, Tata Institute Fund. Research, Bombay.
Soundararajan K. (2004), Strong multiplicity one for the Selberg class, Canad. Math.
Bull. 47 (3), 468–74.
Stade E. (1990), On explicit integral formulas for G L(n,R)-Whittaker functions, Duke
Math. J. 60 (2), 313–362.
(2001), Mellin transforms of GL(n,R) Whittaker functions, Amer. J. Math. 123 (1),
121–61.
(2002), Archimedean L-factors on G L(n) × G L(n) and generalized Barnes integrals,
Israel J. Math. 127, 201–19.
Stark H. (1967), A complete determination of the complex quadratic fields of class
number one, Mich. Math. J. 14, 1–27.
(1972), A transcendence theorem for class number problems I, II, Ann. Math. (2) 94
(1971), 153–73 and 96, 174–209.
Stevens G. (1987), Poincaré series on G L(r ) and Kloostermann sums, Ann. Math. 277
(1), 25–51.
Tate J. (1968), Fourier analysis in number fields and Hecke’s zeta function, Thesis,
Princeton (1950), also appears in Algebraic Number Theory, edited by J. W. Cassels
and A, Frohlich, New York, Academic Press, pp. 305–47.
Tatuzawa T. (1951), On a theorem of Siegel, Japan J. Math. 21, 163–78.
Taylor R. and Wiles A. (1995), Ring-theoretic properties of certain Hecke algebras,
Ann. Math. (2) 141 (3), 553–72.
Terras A. (1985), Harmonic Analysis on Symmetric Spaces and Applications I, Berlin,
Springer–Verlag.
(1988), Harmonic Analysis on Symmetric Spaces and Applications II, Berlin,
Springer–Verlag.
Titchmarsh E. C. (1986), The Theory of the Riemann Zeta Function, Second Edition
revised by D. R. Heath-Brown, Oxford Science Publications, Oxford University
Press.
Tunnell J. (1981), Artin’s conjecture for representations of octahedral type, Bull. Amer.
Math. Soc. (N.S.) 5 (2), 173–75.
Venkov A. B. (1983), Spectral theory of automorphic functions, A translation of Trudy
Mat. Inst. Steklov. 153 (1981), Proc. Steklov Inst. Math. 1982, no. 4(153).
Vinogradov I. and Takhtadzhyan L. (1982), Theory of Eisenstein series for the group
S L(3,R) and its application to a binary problem, J. Soviet Math. 18, 293–324.
Wallach N. (1988), Real Reductive Groups, I, Academic Press.
Watkins M. (2004), Class numbers of imaginary quadratic fields, Math. Comp. 73 (246),
907–38.
Weil A. (1946), Sur quelques résultats de Siegel, Brasiliensis Math. 1, 21–39.
484 References

(1949), Numbers of solutions of equations in finite fields, Bull. Amer. Math. Soc. 55,
497–508.
(1967), Über die Bestimmung Dirichletscher Reihen durch Funktionalgleichungen,
Ann. Math. 168, 149–56.
Weyl H. (1939), Classical Groups, their Invariants and their Representations, Princeton
University Press, Ch. 7, sect. 6, 202–3.
Whittaker E. T. and Watson G. N. (1935), A Course in Modern Analysis, Cambridge
University Press.
Ye Y. (2000), A Kuznetsov formula for Kloosterman sums on G L n , Ramanujan J. 4 (4),
385–95.
(1998), Exponential sums for G L(n) and their applications to base change, J. Number
Theory 68 (1), 112–30.
Zagier D. (1982), The Rankin–Selberg method for automorphic functions which are not
of rapid decay, J. Fac. Sci. University of Tokyo Sect. IA Math. 28, 415–37.
(unpublished), Unpublished notes on Kuznetsov’s formula on S L(2,Z).
 Index

abelian L-function, 401, 402 Bessel function, 65, 67, 72, 134, 411, 470
Abel transform, 101 bi-invariant function, 355
acts continuously, 3, 6, 74, 163, 266 BlockMatrix, 417
additive character, 63, 64, 66, 117 Bohr–Mollerup theorem, 155
additive twist, 206 Bombieri’s theorem, 354, 393
adjoint, 166, 168, 269, 270, 277 bracket, 40, 45, 47
algebra, 39, 40, 42, 43, 44, 47, 50, 114, 139 BruhatCVector, 418
algebraic integer, 60 Bruhat decomposition, 292–294, 303, 305,
algebraic variety, 351, 382 320, 339–341, 350, 353–355, 417, 421,
antiautomorphism, 76, 77, 80, 164, 167, 266 435
ApplyCasimirOperator, 416 BruhatForm, 418
approximate functional equation, 241–245, Bump’s double Dirichlet series, 186–193
388–392
arithmetic group, 235, 285 CartanForm, 419
Arthur–Selberg trace formula, 383 Casimir operator, 47–50, 416
Artin’s conjecture, 236, 401, 402 Cauchy’s determinant, 231, 232, 234
Artin L-function, 396–401, 404, 405 Cauchy’s identity, 229, 231, 232, 234, 367, 379
Artin reciprocity law 406 Cauchy–Schwartz type inequality, 137
associate parabolic center, 9, 11, 25, 44, 46, 289
associative algebra, 39, 40, 43, 45 center of the universal enveloping algebra,
automorphic, 70, 88, 96, 119, 120, 161, 181, 46–50, 114
188, 195, 196, 201, 216, 217, 218, 222, central character, 162
319, 375, 380, 405 character, 115, 116, 128–130, 144, 145, 152,
automorphic condition, 161, 216, 262 155, 195, 207, 216, 217, 218, 221, 250,
automorphic form, 54, 76, 81, 89, 114, 161, 307, 317, 337–342, 350–353, 360, 401,
166, 216, 218, 222, 223, 253, 257, 263, 437, 438, 470
268, 328, 365, 370, 383, 395, 396, character sum, 207
403–406, 426 characteristic zero, 397
automorphic kernel, 355 Chevalley group, 27, 114, 337
automorphic relation, 222 class field theory, 401
automorphic cuspidal representation, 403 class number, 245, 248, 249, 255
automorphic representation, 365, 396, class number one, 246
403–406 commensurator, 74, 164, 266, 267
commutativity of the Hecke ring, 76
base change, 406 compatibility condition, 342, 350
basis, 16, 38–41, 135–137, 195, 324, 359, 406 composition of differential operators, 52

485
486 Index

congruence class, 399 differential one-form, 21–22

congruence subgroup, 54, 69–71, 91, 99, 216, differential operator, 38, 42, 43, 46–50, 131,
253, 256, 383 153, 168
conjugacy class, 400, 401, 406 Dirac δ-sequence, 104
ConstantMatrix, 420 Dirichlet character 194, 195, 208, 214, 221,
constant term, 59, 94–97, 100, 107, 114, 286, 315
319–326, 334 primitive, 206, 208, 214, 215, 216, 218, 222,
continuous spectrum, 95, 100, 106, 360 237, 244, 251, 256, 385, 386
converse theorem, 54, 91, 194, 195, 208, 210, quadratic, 248, 249, 250, 252
214, 215, 221, 235, 402 Dirichlet series 54, 85, 91, 92, 172, 192, 194,
for S L(2, Z), 91, 194 235–238, 241, 243, 246, 250, 253, 257,
for S L(3, Z), 194–197 278, 371–372, 393, 395
convexity bound, 238–241, 248, 250 discrete, 4–6, 15
convolution, 223, 377 discrete group action, 38, 39
convolution L-function, 210, 373 discrete spectrum, 110
convolution operator, 106, 112, 113, 214 discrete subgroup, 5, 6, 19, 74–76, 164, 266
coset decomposition, 75, 82, 167, 207 discriminant, 245–251
coset representative, 5, 16, 76, 90, 118, 126, disjoint partition, 80, 165, 267
159, 260 Disquisitiones, 245
coset space, 23, 24, 303, 304 divisor function, 57
Cramer’s rule, 347, 350 double coset, 303
CRE, 411 double Dirichlet series, 186, 187, 193
CreQ, 420 double Mellin transform, 155, 156, 173,
critical strip, 239, 241 182–186, 325
cubic field, 397 doubly periodic function, 155
cusp, 70, 95, 106, 108, 216 dual, 161, 178, 182, 224
cusp form, 68, 69, 94, 108, 110, 112, 216, 256, dual form, 197, 243, 314
257, 302, 321, 323, 324, 403 dual Maass form, 161, 174, 214, 227, 257,
cuspidal automorphic function, 376 261, 262, 271, 279, 367, 374, 379, 406
cuspidal function, 99, 107, 109, 114
cuspidality condition, 114, 161, 263 EBV, 91–94, 194, 196, 215, 221, 223
cyclic group, 195, 216 e-function 338, 339, 352, 356
cylinder ∞ \h2 , 108 eigenfunction, 38, 39, 50–55, 70, 83–85,
90–93, 95, 104, 110, 114, 115, 117, 128,
d, 467 150, 156, 168, 169, 172, 187, 209,
decomposition group, 399, 400 214–216, 229, 253, 271, 272, 277–279,
degenerate Whittaker function, 150–152, 306 285, 310–313, 316, 317, 382, 384, 427
degree, 231–232, 237, 238, 384, 387, 397, 399 eigenspace, 71
degree two Euler product, 215 eigenvalue, 38, 55, 62, 70–73, 85, 92–95, 104,
–function, 107 110, 112, 117, 128, 162, 169, 264, 272,
Det, 20, 22, 37, 47, 78, 79, 141, 224, 308, 345, 312, 316, 318, 343, 371, 404, 420, 426,
349, 369, 401 427, 429
determinant, 4, 8, 15, 16, 20, 22, 47, 224, eigenvalues of Hecke operators, 310
229–234, 265, 266, 291, 299, 303, 308, eigenvalues of the Laplacian, 343
345–347, 359, 373, 401, 417, 440, 449, eigenvalue conjecture, 386
456, 460, 462 EisensteinFourierCoefficient, 422
diagonal matrix, 10–13, 18, 22, 26, 66, 79, EisensteinSeriesTerm, 422
132, 143, 146, 162, 167, 168, 263, 270, Eisenstein series, 54–60, 90, 94, 100, 105,
294, 417–419, 432, 435, 440, 446, 461 106, 112, 182, 193, 210–212, 218, 219,
DiagonalToMatrix, 421 222–229, 257, 285, 297, 300, 302, 310,
differential equation, 66–67, 70 313, 314, 315, 316, 317, 322, 323, 326,
differential form, 35, 467–468 328, 332, 360, 362, 373, 422
 Index 487

constant term, 286, 321, 322, 323, 334 flag, 291, 292
convergence, 224, 300 Fourier transform, 1, 2, 30, 59, 102, 104,
for S L(2, Z), 112, 225, 313, 323 106–107, 113, 309
for S L(3, Z), 112, 321 Fourier expansion, 54, 56, 58–70, 95, 97, 114,
for S L(n, Z), 302, 318, 319, 396, 403 118, 119, 158, 173, 200, 216, 219, 225,
functional equation, 97, 229, 307, 310, 329, 228, 278, 339, 352, 369, 373
370 of an Eisenstein series, 54, 58–62, 95, 225,
Fourier expansion, 56, 58, 59, 60, 62, 225, 228, 286, 303, 310, 319, 323, 327, 332,
286, 303, 339, 422 418, 422
general parabolic, 295 of a Maass form, 63, 64, 69, 70, 114, 128,
Langlands, 285, 286, 301, 316, 318–320, 129, 159, 214, 260, 261, 282, 302, 315,
357, 403, 417, 441 366, 372, 374, 376, 381
L-function associated to, 89, 90, 310, 314 Fourier transform, 1, 2, 30, 59, 102, 104, 106,
meromorphic continuation, 224, 285, 307, 107, 113, 309
310 Fourier Whittaker expansion, 66–73, 89,
maximal parabolic, 223, 295, 297, 300, 307, 163, 174, 179, 180, 224, 225, 279, 286,
308, 309, 321, 327, 368 319
minimal parabolic, 295, 296, 297, 303, 310, Frobenius automorphism, 398–400
313, 314, 315, 316, 324, 325, 327, 330, function field, 397
339, 370, 381, 418, 450, 451–454 functional analysis, 83, 168, 271
residue, 324, 331, 357, 403 FunctionalEquation, 424
template, 370 functional equation, 1–3, 38, 54
twisted by Maass forms, 286, 301, 302, 315, of Eisenstein series, 97, 212, 229, 307, 310,
316, 321, 357 329, 370
ElementaryMatrix, 423 of K-Bessel function, 134
elementary row and column operations, 79 of L-functions, 54, 62, 86, 91, 178, 182, 187,
endomorphism ♥, 105, 106, 108 188, 194, 195, 201, 202, 206, 208–211,
endomorphism J , 100, 105 213–217, 221, 227, 240–245, 251, 252,
Epstein zeta function, 308, 309 278, 314, 315, 365, 367–371, 374, 379,
equivalence class, 4 385, 387–392, 396, 402, 452, 453
equivalent, 4, 6, 7, 216 of the Riemann zeta function, 2, 3, 54, 90,
Euclidean topology, 20 214, 310, 376, 381
Euler product, 1, 54, 76, 85, 90, 95, 212, 215, for the Selberg class, 245
229, 235–237, 250, 253, 255–257, 365, template, 314
367, 373, 379, 382, 387, 395, 396, 398, of Whittaker functions, 114, 134, 145, 319,
401, 404, 405, 426, 427 424
even Maass form, 73, 74, 86, 87, 90, 91, 99, functoriality, 396, 397, 402–406
100, 106 fundamental discriminant, 249, 250
even primitive character, 208 fundamental domain, 6–8, 15, 27, 31, 34, 36,
explicit Bruhat decomposition, 303 55, 70, 71, 117, 377
exponential sum, 56 fundamental unit, 60
exterior algebra, 139
exterior power, 134–137, 396, 406, Galois conjugacy, 383
452–454 Galois conjugate, 339, 400
exterior square, 397 Galois extension, 397–401, 404
exterior square lift, 397 Galois group 397–402, 404
exterior power L-function, 396, 452 Galois representation, 402, 404–406, 474, 475,
482
finite field, 69, 351, 382, 398–400 Galois subgroup, 404
finite order, 236 Gamma factor, 211, 213, 231, 245, 372, 383,
finite prime, 382, 384 386, 388, 414, 415, 450, 452, 454, 470,
finite propagation speed, 111, 112 471, 475
488 Index

Gamma function, 155, 157, 196, 208, 243, HeckePowerSum, 411

272, 335 Hecke ring, 75, 76, 80, 81, 164, 168, 266, 267,
Gauss, 245, 246, 249, 406 271
Gauss class number problem, 246, 248 Hecke–Maass converse theorem, 91, 92
Gauss sum, 195, 217, 315 HermiteFormLower, 429
Gelbart–Jacquet lift, 213–215, 221, 235, 253, HermiteFormUpper, 430
365, 386, 387, 402 Hermite normal form, 78–81, 165, 268, 312
Generalized Ramanujan conjecture, 381–384, high in the cusp, 113, 323
395 higher rank, 100, 111, 129, 285
generalized upper half-plane, 9, 10, 15, 24, 51, Hilbert space, 62, 84, 110
153, 259, 299, 415, 466, 467 holomorphic modular form, 69, 215, 253, 256,
general linear group, 6, 19 382
GetCasimirOperator, 416, 424 hyper-Kloosterman sum, 337, 352, 391
GetMatrixElement, 425, 441 hyperbolic distance, 101, 103
global Langlands conjecture, 396, 397 hyperbolic Fourier expansion 54, 59, 62
GL(n)pack, 409 hyperbolic Laplacian, 55, 100, 104, 111
GlnVersion, 413 hyperbolic matrix, 60
Godement–Jacquet L-function, 174, 178, 187, hyperbolic plane, 111, 112
188, 194, 206, 214, 215, 277, 279, 282,
365, 379, 382, 403 Is function, 51, 52, 55
Green’s theorem, 62 Iν function, 138, 139, 144, 146, 150, 151
Grössencharakter, 54, 61, 62, 91 icosahedral representation, 402
group action, 38, 39 identity endomorphism, 104, 107
growth, 66–69, 114, 239, 242, 247, 250, identity matrix, 9–14, 31, 34, 43, 47, 52, 79,
388 116, 125, 145, 263, 289, 292, 305, 309,
growth condition, 69, 156, 247, 248 358, 374, 377, 401
growth properties, 128, 129, 159, 266, 239, ideal, 41, 60–62, 245, 399, 400
241, 260, 371 ideal class group, 245
IFun, 431
Haar integral, 21 imprimitive, 195, 208
Haar measure 19–21, 35 imaginary quadratic field, 245–249, 255
Hadamard factorization, 251 induced representation, 402, 406
half-plane, 172, 278 inertia group, 400
Harish-Chandra character, 115 infinite prime, 383
Harish transform, 358 inner product, 62, 81, 95, 96, 135, 136, 156,
harmonic function, 63 166, 210, 211, 226, 227, 255, 269, 277,
HeckeCoefficientSum, 426–429 323, 324, 330, 368, 376, 377, 378,
HeckeEigenvalue, 426–429 393
Hecke eigenfunction, 54, 83, 84, 92, 95, 168, InsertMatrixElement, 432
169, 172, 187, 214, 215, 229, 253, 271, integral operator, 101, 103, 357
272, 277, 279, 310–313, 316, 332, 384, invariant differential operator, 38, 39, 44, 48,
427 50, 55, 64, 114, 128, 150, 156, 209, 271,
Hecke eigenform, 316, 318, 382 316, 317, 357
Hecke grössencharakter, 54, 61, 62, 91 invariant measure, 23–28, 32, 55, 74, 81, 103,
Hecke L-function, 54, 401 104, 117, 166, 227, 269, 297, 368
HeckeMultiplicativeSplit, 427 invariant subspace, 403
Hecke–Maass converse theorem, 92 inverse Abel transform, 101
HeckeOperator, 428 inverse Mellin transform, 96, 202, 331
Hecke operator, 54, 73–77, 80–84, 90, 99, 105, inversion formula, 94, 358, 359, 363
112, 163–172, 266–272, 277, 310, 313, involution, 161, 263, 328
318, 319, 428 irreducible, 401
 Index 489

irreducible representation, 402 Laplacian, 39, 56, 82, 93, 100, 110–112, 216,
Iwasawa decomposition, 8–16, 24, 28, 31, 299, 285, 343
309, 424, 431, 433, 435, 466 Laurent expansion, 158
IwasawaForm, 432 laws of composition, 420
IwasawaQ, 435 LeadingMatrixBlock, 441
IwasawaXMatrix, 431 left equivalent, 78, 79
IwasawaXVariables, 434 left Haar measure, 20, 21
IwasawaYMatrix, 434 left invariant measure, 24, 27, 74, 81, 117,
IwasawaYVariables, 435 166, 269
Iwasawa form, 146, 150, 175, 185, 281, 299, level, 70, 216, 217, 383, 385
308, 328, 413, 416, 424, 432–435, Levi component, 289–292
445–447, 467, 470 L-function, 54, 76, 114, 172–178, 187, 188,
210, 239–243, 250, 255–257, 277, 278,
Jacobi identity, 40 281, 282, 322, 323, 365, 395–396
Jacobi theta function, 2, 221–223 associated to an Eisenstein series, 89, 90,
Jacobian, 124 286, 313–315, 451–454
Jacquet Whittaker function, 139, 142, 154, associated to a Maass form, 91, 174, 177,
159, 160, 175, 260, 280, 307, 360, 470, 194, 213–215, 235, 257, 279, 322, 323,
471 367, 372, 403
Jacquet’s Whittaker function, 129, 130, 136, Artin, 396–401, 404, 405
141, 144, 150, 152, 361 Dirichlet, 236, 237, 244, 249, 251, 252, 253,
256, 315
K-Bessel function, 65, 134 Hecke, 62, 401, 402
KloostermanBruhatCell, 435, 436 L-group, 406
KloostermanCompatibility, 436, 437, 438 Lie algebra, 19, 39–42, 44, 45, 50
KloostermanSum, 436–438 Lie algebra of G L(n, R), 19, 42, 50
Kloosterman sum, 337–343, 349–353, 357, Lie bracket, 41–43
362, 363, 391, 436–438 Lie group, 39, 42, 270, 396, 404
Kloosterman zeta function, 342, 343 lifting, 397, 404
Kronecker symbol, 222, 248 local Langlands conjecture, 397
Kuznetsov trace formula, 337, 354, 355, locally compact, 19–23
360–363 locally compact Hausdorff topological group,
19, 20, 21, 23
Landau’s lemma, 372, 395 LongElement, 439, 442
Langlands, 27, 195, 236, 285, 286, 366, 395, long element, 150, 224, 306, 307, 317, 341,
397, 401, 402, 403, 406 350, 351, 375, 380, 442
LanglandsIFun, 441 long element Kloosterman sum, 351, 342
LanglandsForm, 440 LowerTriangleToMatrix, 442, 448
Langlands conjecture, 195, 395–397, 402 Luo–Rudnick–Sarnak theorem, 366, 384, 387,
Langlands decomposition, 288–290, 298–301, 393
320, 440
Langlands Eisenstein series, 285, 286, 301, Maass form, 54, 62, 83, 86, 112, 216, 235,
316, 318, 319, 320, 357, 403, 417, 256, 257, 286, 310, 313, 382, 383, 403,
441 426, 427
Langlands functoriality, 396, 397, 402, 405 even and odd, 73, 74, 86, 87, 88, 90, 100,
Langlands global functoriality conjectures, 105, 106, 162, 265
396, 397 for S L(2, Z), 62, 63, 67, 69–74, 86, 91, 92,
Langlands–Shahidi method, 286 95, 210, 221, 253, 255, 256, 321–323, 387
Langlands spectral decomposition, 286, 319, for S L(3, Z), 99, 159–163, 172, 174, 176,
324, 357 182, 187, 188, 194, 196, 197, 201, 213,
Laplace operator, 38, 39, 70, 73, 95 214, 215, 227, 229, 255, 302, 324, 387
490 Index

Maass form (cont.) Mellin transform, 2, 86, 94, 96, 155–157,

for S L(n, Z), 99, 114–118, 129, 259–265, 182–186, 173–178, 202, 278–282, 325,
271, 277, 279, 282, 283, 316, 319, 338, 326, 331, 338, 358, 361, 369, 388
358, 366, 367, 370–384, 393, 403 minimal parabolic Eisenstein series, 295, 296,
Fourier expansion, 69, 118, 128, 129, 282, 297, 303, 310, 313–316, 324, 325, 327,
366 330, 339, 370, 381, 418, 450–454
L-functions of, 86, 91, 177, 194, 215, 256, minimal standard parabolic subgroup, 295
279, 379, 402, 403, 404 minor, 294, 343–349, 417, 436, 437,
of type ν, 62, 63, 83 456–460
MakeBlockMatrix, 427 modular form, 69, 215, 253, 256, 382
MakeMatrix, 443 ModularGenerators, 449, 450
MakeXMatrix, 444 modular group, 80
MakeXVariables, 445 Moebius function, 57
MakeYMatrix, 446 MPEisensteinGamma, 450
MakeYVariables, 446 MPEisensteinLambdas, 451
MakeZMatrix, 446 MPEisensteinSeries, 451
MakeZVariables, 447 MPExteriorPowerGamma, 452
Mathematica, 409–412 MPExteriorPowerLFun, 453
matrix, 3 MPSymmetricPowerGamma, 454
diagonal, 10–13, 18, 22, 26, 60, 79, 132, MPSymmetricPowerLFun, 453
143, 146, 162, 167, 168, 263, 270, 294, multiplicativity condition, 351
417, 418, 419, 432, 435, 436, 440, 446, multiplicativity of Fourier coefficients, 168,
461 271
identity, 9, 11–14, 31, 34, 43, 47, 52, 79, multiplicativity relations, 83, 168, 173, 195,
116, 125, 145, 149, 263, 289, 292, 305, 278
361, 358, 374, 377, 401 Multiplicity one theorem, 66, 68, 128, 129,
lower triangular, 11, 12, 418, 429, 442 145, 159, 260, 393
minor, 294, 343–349, 417, 436, 437,
456–460 NColumns, 454
orthogonal, 8, 150, 298, 299, 419 newform, 216
triangular, 8 non-holomorphic automorphic form, 54, 383
unipotent, 151, 259, 289, 292, 294, 337, non-self dual, 257
418, 419, 432–436, 440, 444 non-vanishing, 286, 322, 323
upper triangular, 8, 10–12, 15, 17, 25, 78, norm, 16, 69, 95, 134, 136, 138, 309, 400
79, 115, 116, 118, 128, 148, 151, 159, normal operator, 166, 269
232, 259, 273, 286, 292, 294, 303, 337, non-degenerate Whittaker model, 360
418, 419, 430, 432, 435, 436, 440, 442, NRows, 455
444, 446, 464 number field, 256, 384, 393, 396–401, 404
MatrixColumn, 447
MatrixDiagonal, 447 octahedral representation, 402
MatrixJoinHorizontal, 448 odd Maass form, 73, 74, 86–88, 100, 105, 162,
MatrixJoinVertical, 448 263, 265
MatrixLowerTriangle, 448 odd primitive character, 208
MatrixRow, 449 old form, 217
MatrixUpperTriangle, 449 operator, 38–55, 73, 82, 103, 106–112, 162,
maximal compact subgroup, 28, 33, 99 263, 321
maximal parabolic subgroup, 300, 307, adjoint, 166, 269, 277
321 Laplace, 38, 39, 70, 73, 95
maximal parabolic Eisenstein series, 223, 297, self adjoint, 156
300, 307–309, 321, 327, 368 symmetric, 62
maximum principle, 238, 239 orbit, 4, 121, 122
Mellin inversion, 94, 96, 97, 327, 388 orthogonal group O(n, R), 9, 46, 47, 298
 Index 491

orthonormal basis of Maass forms, 95 Ramanujan–Petersson conjecture, 68, 69

orthonormality, 237 Ramanujan hypothesis, 236
RamanujanSum, 459
parabolic, 223, 286, 288, 319, 320 Ramanujan sum, 56, 57, 459
ParabolicQ, 455 ramified, 399–401, 404
parabolic subgroup, 286–291, 295–302, 316, rank, 50, 65, 99, 100, 111–114, 129, 285–287,
319–322, 376, 440, 441, 455 301, 316, 320, 343
partition, 80, 165, 267, 286–288, 298, 316, rank of a parabolic subgroup, 287
319, 440, 441, 455 Rankin–Selberg convolution, 210, 211,
periodic function, 1, 38, 61, 118, 155, 195, 275 221–223, 227, 229, 256, 257, 258, 366,
Petersson inner product, 81, 95, 166, 227, 255, 368, 370, 377, 385, 391, 405
269, 277, 368 for S L(2, Z), 226
Petersson norm, 69, 72, 324 for S L(3, Z), 226
Petersson’s conjecture, 68, 69 for G L(n) × G L(n), 365, 366, 371, 373
Phragmén–Lindelöf theorem, 238–240 for G L(n) × G L(n + 1), 372, 373
PluckerCoordinates, 456 for G L(n) × G L(n  ), with n < n  , 376
Plücker coordinates, 343, 344, 456–458 Rankin–Selberg L-function, 256, 286, 365,
PluckerInverse, 456 367, 371, 372, 374, 378, 379, 385,
PluckerRelations, 457 393
Plücker relations, 347 rapid decay, 1, 66, 95, 324, 370, 388
Poincaré series, 337–339, 352, 353, 356, 360 rapid growth, 66
point pair invariant, 102, 103, 112 rapidly decreasing, 114
Poisson summation formula, 1–3, 27, 30, 37, real quadratic field, 54, 62, 91
286, 308–310 reciprocity law, 406
polynomial algebra, 50 reductive group, 129, 285, 404
polynomial growth, 66–68, 247, 371, 388 regularity of distribution, 236
pre-Kuznetsov trace formula, 355, 360, 361, RemoveMatrixColumn, 459
363 RemoveMatrixRow, 459
primitive, 195 representation, 1, 2, 24, 140, 154, 176, 212,
primitive character, 3, 208, 221, 387, 390, 215, 223, 224, 236, 252, 281, 302, 309,
392 365, 391, 396–406, 470
primitive Dirichlet character, 195, 206, 208, residual spectrum, 359, 360, 363
214, 215, 218, 220, 237, 244, 251, 253, residue field, 399
256, 315, 385, 386 Riemann hypothesis, 69, 235, 237, 246, 249,
principal congruence subgroup, 70 382
principal determination, 3 Riemann hypothesis for varieties, 398
principal ideal, 60, 61, 245 Riemann zeta function, 2, 3, 27, 38, 54, 85, 89,
principal value, 239 90, 214, 237, 249, 310, 314, 322, 354,
principle of functoriality, 406 373, 376, 381, 396, 451, 453
projection operator, 376 right cosets, 74–77, 82, 164, 167, 266, 270
right equivalent, 77
quadratic form, 245, 347 right Haar measure, 20
quadratic number field, 60, 62, 91, 245–249, right regular representation, 403
251, 255, 256 rotationally invariant, 32
quadratic reciprocity, 406
quotient space, 16, 19, 38, 74, 303 scaling matrix, 216
SchurPolynomial, 460
Ramanujan bound, 241 Schur polynomial, 229, 230, 232, 366, 373,
Ramanujan conjecture, 68, 381–383, 386, 427, 460
395 Selberg class, 235–241, 245, 256
Ramanujan conjecture on average, 236, 241 Selberg conjecture, 366
Ramanujan -function, 68, 69 Selberg eigenvalue conjecture, 70, 383
492 Index

Selberg spectral decomposition, 54, 94, 95, symmetric group, 25, 135, 292, 397, 402
285 symmetric kth, power L-function, 235, 395,
Selberg transform, 102, 355, 358 453, 454
self-adjoint, 81, 103, 106, 108, 110–112, 156, symmetric Maass form, 161, 261, 265, 315
166, 269 symmetric operator, 62
self dual, 213 symmetric power L-function, 256, 286, 395,
semigroup, 76, 80, 164, 266, 267 396, 453
Siegel set, 15, 17, 18, 108, 113, 117, 129, 131 symmetric product, 405
Siegel zero, 235, 245–249, 251–258 symmetric square, 241
Siegel zero lemma, 251, 255–258 symmetric square lift, 253–257, 397, 398
Siegel’s theorem, 249, 251, 253 symmetric square representation, 406
simple reflection, 145, 148
simultaneous eigenfunctions, 83, 277 TailingMatrixBlock, 464
skew symmetry, 40, 234 Taylor series, 43
SmithElementaryDivisors, 460 template, 314, 315, 370, 381, 396
SmithForm, 461 tensor product, 41, 134, 137, 397, 404, 405
SmithInvariantFactors, 461 tensor product lift, 397
Smith normal form, 54, 77–79, 81, 165, 268 tetrahedral representation, 402
smooth function, 32, 38, 39, 44, 46, 50, 55, 62, theta function, 2, 221–223
94, 102, 106, 110, 115, 117, 128, 242, time reversal symmetry, 111–112
308, 316, 387 topological group, 19–21, 23
solvable group, 402 topological space, 3–6, 14, 19, 20, 74–76, 163,
SpecialWeylGroup, 462 266
special orthogonal group S O(n, R), 16, 17, 27, trace formula, 99–101, 337, 354, 355, 359,
28, 33, 35, 139, 325, 355, 357–362 360–363, 383
spectral decomposition, 54, 94, 95, 285, 286, transfer, 406
319, 324–327, 355, 357 transpose-inverse, 30, 37
sphere (n, dimensional), 27, 32, 467 transpose of a matrix, 9, 80, 164, 267, 418, 419
spherical coordinates, 32 transposition, 26, 80, 81, 83, 135, 167, 269
split classical group, 397 trivial character, 217, 223, 385
split semisimple group, 99 trivial representation, 401
splits completely, 400 twist, 91, 194, 195, 206, 208, 214, 237, 402
square integrable, 55, 70, 81, 94, 106, 268, twisted L-function, 214, 215, 217, 221
426, 428 type, 9, 62, 74, 83, 84, 86, 90–92, 99, 100,
stability group, 216 114–117, 128, 130, 134, 137, 145,
stabilizer, 108, 121, 291 160–163, 176, 177, 179, 180, 182, 188,
Stade’s formula, 361, 362 193, 197, 210, 211, 213, 214, 217, 225,
standard L-function, 173 227, 228, 236, 241, 253, 255, 260,
standard parabolic subgroup, 286–288, 295, 262–265, 281, 282, 283, 315–317, 322,
297, 301, 316, 319 323, 352, 358, 366–368, 371, 372, 376,
Stirling, 157, 240, 243 381, 383, 384, 387, 401, 402, 404
Stirling’s formula, 240
strong multiplicity one, 393 unfolding, 222, 227, 339, 368, 373
subconvexity bounds, 354 unimodular, 20, 21, 429, 430, 461, 466
SubscriptedForm, 463 unipotent matrix, 151, 259, 418, 419, 432,
super diagonal, 129, 159, 175, 259, 409 433, 436, 444
SwapMatrixColumns, 463, 464 unipotent radical, 289–291
SwapMatrixRows, 464 unique factorization, 246, 399
symmetric, 6, 62, 162, 264, 266, 419 unit, 60
symmetric cube, 235, 286, 297 universal enveloping algebra, 9, 40–47, 50,
symmetric fourth power, 235, 297 114
 Index 493

unramified, 399–401, 404 weight, 68, 69, 222, 223

UpperTriangleToMatrix, 464 Weil’s converse theorem, 91
upper triangular, 232, 263, 442, 464 WeylGenerator, 468, 469
upper triangular matrix, 8, 10–12, 15, 17, 25, WeylGroup, 469
78, 79, 115, 116, 118, 128, 148, 151, 159, Weyl group, 25, 26, 144, 145, 150, 151, 178,
232, 259, 273, 286, 292, 294, 303, 337, 224, 303, 306, 320, 332–335, 341, 436,
418, 419, 430, 432, 435, 436, 440, 442, 437, 462, 469
444, 446, 464 Weyl group of G L(n, R), 311, 317, 339,
upper half-plane, 3, 4, 9, 10, 15, 24, 51, 55, 59, 350, 351, 362
101, 153, 299, 356, 415, 466, 467 Weyl group of associate parabolic, 288
Weyl’s law, 99, 110, 112
Vandermonde matrix, 231 WhittakerGamma, 470, 471
variational principle, 110 Whittaker, 470
vector space, 39–42, 50, 62, 74, 81, 134–136, Whittaker model, 173, 360
166, 269, 397, 400, 401, 403–406 Whittaker function, 92, 112–114, 117,
volume, 27–32, 296, 465, 466, 467 128–130, 133, 139–145, 149–152, 154,
VolumeBall, 465 159, 172, 175, 192, 260, 276, 280, 306,
VolumeFormDiagonal, 465 307, 315, 317, 319, 332–335, 360, 361,
VolumeFormGln, 466 365, 369, 424, 470, 471
VolumeFormHn, 466 Whittaker function of type ν, 63–68, 130,
VolumeFormUnimodular, 466 156
VolumeHn, 467 WMatrix, 472
VolumeSphere, 467
ZeroMatrix, 472
wave equation, 106, 111, 112 zeta function, 1, 248–250, 335, 342, 343
weak Weyl law, 110 Epstein, 308, 309
Wedge, 467, 468 Riemann, 1–3, 27, 38, 54, 85, 89, 90, 214,
wedge product, 22, 35, 135, 137, 138, 294, 237, 249, 310, 314, 322, 354, 373, 376,
348, 465–468 381, 396, 451, 453