Riemann'S Zeta Function and Beyond: Abstract

Download as pdf or txt
Download as pdf or txt
You are on page 1of 54

BULLETIN (New Series) OF THE

AMERICAN MATHEMATICAL SOCIETY


Volume 00, Number 0, Pages 000–000
S 0273-0979(XX)0000-0

RIEMANN’S ZETA FUNCTION AND BEYOND

STEPHEN S. GELBART AND STEPHEN D. MILLER

Abstract. In recent years L-functions and their analytic properties have as-
sumed a central role in number theory and automorphic forms. In this ex-
pository article, we describe the two major methods for proving the analytic
continuation and functional equations of L-functions: the method of integral
representations, and the method of Fourier expansions of Eisenstein series.
Special attention is paid to technical properties, such as boundedness in verti-
cal strips; these are essential in applying the converse theorem, a powerful tool
that uses analytic properties of L-functions to establish cases of Langlands
functoriality conjectures. We conclude by describing striking recent results
which rest upon the analytic properties of L-functions.

Dedicated to Ilya Piatetski-Shapiro, with admiration

Contents
1. Introduction 2
The Two Methods 3
2. Riemann’s Integral Representation (1859) 4
2.1. Mellin Transforms of Theta Functions 5
2.2. Hecke’s Treatment of Number Fields (1916) 7
2.3. Hamburger’s Converse Theorem (1921) 8
2.4. The Phragmen-Lindelöf Principle and Convexity Bounds 10
3. Modular Forms and the Converse Theorem 12
3.1. Hecke (1936) 12
3.2. Weil’s Converse Theorem (1967) 16
3.3. Maass Forms (1949) 17
3.4. Hecke Operators 17
4. L-functions from Eisenstein Series (1962-) 18
4.1. Selberg’s Analytic Continuation 19
5. Generalizations to Adele Groups 21
6. Tate’s Thesis (1950) 22
7. Automorphic forms on GL(n) 24
7.1. Jacquet-Langlands (1970) 27
7.2. Godement-Jacquet (1972) 30
7.3. Jacquet-Piatetski-Shapiro-Shalika (1979) 31
8. Langlands-Shahidi (1967-) 32
8.1. An Outline of the Method 32
8.2. GL(2) Example 35

The first author was partially supported by the Minerva Foundation, and the second author
was supported by NSF grant DMS-0122799.

c 0000 (copyright holder)


1
2 STEPHEN S. GELBART AND STEPHEN D. MILLER

8.3. Boundedness in Vertical Strips and Non-vanishing 36


9. The Langlands Program (1970-) 37
9.1. The Converse Theorem of Cogdell-Piatetski-Shapiro (1999) 37
9.2. Examples of Langlands L-functions: Symmetric Powers 39
9.3. Recent Examples of Langlands Functoriality (2000-) 40
9.4. Applications to Number Theory (2001-) 42
References 46

1. Introduction
In 1859 Riemann published his only paper1 in number theory, a short ten-page
note which dramatically introduced the use of complex analysis into the subject.
Riemann’s main goal was to outline the eventual proof of the Prime Number The-
orem
x
π(x) = # {primes p ≤ x} ∼ , x → ∞,
log x
log x
i.e. lim π(x) = 1,
x→∞ x
by counting the primes using complex integration (the proof was completed half
a century later by Hadamard and de la Vallee Poussin). Along this path he first
shows that his ζ-function, initially defined in the half-plane Re (s) > 1 by

X 1 Y 1
ζ(s) = = ,
n=1
ns 1 − p1s
p (prime)

has a meromorphic continuation to C. Secondly, he proposes what has remained


as perhaps the most-famous unsolved problem of our day:

The Riemann Hypothesis: ζ(s) 6= 0 for Re s > 1/2.

For more on the history of ζ and Riemann’s work, the reader may consult [15, 30,
36, 179]. Our role here is not so much to focus on the zeroes of ζ(s), but in some
sense rather on its poles. In particular, our emphasis will be on explaining how we
know that ζ(s) extends meromorphically to the entire complex plane, and satisfies
the functional equation
s
s
ξ(s) := π − 2 Γ ζ(s) = ξ(1 − s) .
2
It is one purpose of this paper to give two separate treatments of this assertion. We
want also to characterize the ζ-function as satisfying the following three classical
properties (which are simpler to state in terms of ξ(s), the completed ζ-function).
• Entirety (E): ξ(s) has a meromorphic continuation to the entire complex
plane, with simple poles at s = 0 and 1.
• Functional Equation (FE): ξ(s) = ξ(1 − s).

1See [145], and [36], for translations.


RIEMANN’S ZETA FUNCTION AND BEYOND 3

• Boundedness in Vertical strips (BV): ξ(s) + 1s + 1−s


1
is bounded in any
strip of the form −∞ < a < Re (s) < b < ∞ (i.e. ξ(s) is bounded in
vertical strips away from its two poles).
A second purpose is to overview how these treatments and properties extend to
L-functions assigned to more general groups such as GL(n), the group of invertible
n × n matrices (the function ζ(s) is attached, we shall see, to GL(1)). A major
motivating factor for studying these analytic conditions (especially the technical
BV) is that they have become crucial in applications to the Langlands Functoriality
Conjectures, where they are precisely needed in the “Converse Theorem,” which
relates L-functions to automorphic forms (see Theorems 3.1, 3.2, and Section 7.3).
More to the point, the study and usefulness of L-functions has pervaded many
branches of number theory, wherein complex analysis has become an unexpectedly-
powerful tool. In Section 9 we discuss the connections with some of the most
dramatic recent developments, including the modularity of elliptic curves, progress
towards the Ramanujan conjectures, and the results of Kim and Shahidi. The
two treatments we describe are, in fact, the major methods used for deriving the
analytic properties of L-functions.

The Two Methods. A first method (Section 2) of analytic continuation is Rie-


mann’s, initiated in 1859. In fact, it was one of several different, though similar,
proofs known to Riemann; Hamburger, and later Hecke, moved the theory along
remarkably following this line of attack. Almost a century later, Tate (Section 6)
recast this method in the modern language of adeles in his celebrated 1950 Ph.D.
thesis [173], another famous and important treatment of ζ-functions. The second
– and lesser known – method is via Selberg’s “constant term” in the theory of
Eisenstein series (Section 4.1). This theory, too, has an important expansion: the
Langlands-Shahidi method (Section 8).
As we shall see, both methods take advantage of various (and sometimes hidden)
group structures related to the ζ-function. They also suggest a wide generalization
of the methods: first to handle Dirichlet L-functions, ζ-functions of number fields,
and then quite general L-functions on a wide variety of groups. In Section 3 we begin
by explaining this through the connection between modular forms and L-functions.
In fact, this nexus has been fundamentally important in resolving many classical
problems in number theory. After surveying the classical theory of Hecke, we turn
to the modern innovations of Langlands. We have in mind ideas of Weil, Langlands,
Jacquet, Godement, Piatetski-Shapiro, Shalika, Shahidi, and others. For broader
and deeper recent reports on the nature of L-functions and the application of their
analytic properties, see, for example, [70, 154].
The reader will notice that we have left out many important properties of the
Riemann ζ-function, some related to the most famous question of all, the Riemann
Hypothesis, which can be naturally restated in terms of ξ as
All ρ such that ξ(ρ) = 0 have Re ρ = 1/2.
This is because we are primarily interested in results related to the three properties
E, BV, and FE. As indicated above, we are also only following the development of
a few approaches (see [176] for many more, though which mainly follow Riemann).
Also, to lessen the burden on the reader unfamiliar with adeles, we will more or less
describe the historical development in chronological order, first treating the classical
4 STEPHEN S. GELBART AND STEPHEN D. MILLER

results of Riemann, Hecke, Selberg, and Weil, before their respective generalizations
to adele groups.
To wit, the paper is organized in three parts. The first, Sections 2-4, gives
the background on the classical theory: Section 2 discusses Riemann’s theory of
the ζ-function and its analytic properties; Section 3 focuses on Hecke’s theory of
modular forms and L-functions; and Section 4 centers on Selberg’s theory of non-
holomorphic Eisenstein series. The second part of the paper redescribes these topics
in more modern, adelic terms. Section 5 leads off with a short introduction to the
adeles. Sections 6, 7, and 8 then give a parallel discussion of the respective topics
of Sections 2, 3, and 4, but in a much more general context. Finally, the last part
of the paper is Section 9, where we recount some recent results and applications of
the analytic properties of L-functions. Sections 8 and 9 are quite linked, in that
many of the recent developments and analytic properties used in section 9 come
from the Langlands-Shahidi method, the topic of Section 8. However, the latter
is quite technical, and we have made an attempt to make Section 9 nonetheless
accessible without it.
A word is in order about what we don’t cover. Because our theme is the analytic
properties of L-functions, we have left out a couple of important and timely topics
that lie somewhat outside our focus. Chiefly among these are some developments
towards the Langlands conjectures, for example the work of Lafforgue [92] over
function fields. This is mainly because the analytic properties of L-functions in
the function field setting were long ago established by Grothendieck (see [79]), and
are of a significantly different nature. Some resources to learn more about these
additional topics include [1, 2, 4, 37, 43, 88, 102, 103, 105, 106, 146].
Before starting, we wish to thank J. Bernstein, J. Cogdell, A. Cohen, K. Conrad,
W. Duke, H. Dym, E. Lapid, A. Lubotzky, B. Mazur, S.J. Miller, A. Reznikov, B.
Samuels, P. Sarnak, G. Schectman, F. Shahidi, and the referee for many very helpful
comments.

2. Riemann’s Integral Representation (1859)


As we mentioned in the introduction, Riemann wrote only a single, ten-page
paper in number theory [145]. In it he not only initiated the study of ζ(s) as
a function of a complex variable, but also introduced the Riemann Hypothesis
and outlined the eventual proof of the Prime Number Theorem! At the core of
Riemann’s paper is the Poisson summation formula
X X
(2.1) f (n) = fb(n) ,
n∈Z n∈Z

which relates the sum over the integers of a function f and its Fourier transform
Z
(2.2) fb(r) = f (x) e−2 π i r x dx .
R

The Poisson summation formula is valid for functions f with suitable regularity
properties, such as Schwartz functions: smooth functions which, along with all
1
their derivatives, decay faster than any power of |x| as |x| → ∞. However, by
temporarily neglecting such details, one can in fact quickly see why the Poisson
summation formula implies the functional equation for ζ(s), at least on a formal
RIEMANN’S ZETA FUNCTION AND BEYOND 5

level. Indeed, let f (x) = |x|−s , so that


Z
fb(r) := |x|−s e−2 π i r x dx
(2.3) R
= |r|s−1 G(s) ,
where
Z
(2.4) G(s) = |x|−s e−2 π i x dx .
R
Using the convention that |0|s = 0, we can already see from the Poisson Summation
Formula that
(2.5) 2 ζ(s) = 2 G(s) ζ(1 − s) ,
a functional equation relating ζ(s) to ζ(1 − s). In fact the integral (2.4) is a variant
of the classical Γ-integral
Z ∞
(2.6) Γ(s) = e−x xs−1 dx , Re s > 0
0
and can be shown to equal
π (s−1)/2 Γ( 1−s
2 )
(2.7) G(s) = −s/2 s ,
π Γ( 2 )
at least in the range 0 < Re s < 1 (see [51] or [30, p. 73]). Thus the functional
equation (2.5) is formally identical to Riemann’s functional equation
(2.8) ξ(s) = π −s/2 Γ( 2s ) ζ(s) = ξ(1 − s) .
Of course, neither sum defining ζ(s) in (2.5) converges when the other does, much
less in the range 0 < Re s < 1 where we computed G(s). Indeed, the functional
equation cannot be provenPin the absence of some form of analytic continuation
∞ −s
beyond the region where n=1 n converges. The argument sketched here for
the functional equation seems to have been first considered by Eisenstein, who
succeeded in proving the functional equation not for ζ(s) itself, but for a closely
related Dirichlet L-function (for these, see (2.12) and [30]). André Weil has written
historical accounts [179, 180] which suggest that Riemann was himself motivated
by Eisenstein’s papers to analyze ζ(s) by Poisson summation. The rigorous details
omitted from the above formal summation argument can be found in [116, §5].
2.1. Mellin Transforms of Theta Functions. Riemann’s own, rigorous argu-
ment proceeds by applying the Poisson summation formula (2.1) to the Gaussian
2
f (x) = e−π x t , t > 0, whose Fourier transform is
1 2
fb(r) = √ e− π r /t .
t
The Gaussian is a Schwartz function, and can be legitimately inserted in the Poisson
summation formula. Its specific choice is not absolutely essential, but rather a
matter of convenience, as we will see in Section 6. However, it was an inspired
selection by Riemann, in that it is connected to the theory of modular forms (see
Section 3.1). Thus Riemann’s contribution to the functional equation went far
beyond simply making a formal argument rigorous – it launched the link between
modular forms and L-functions that remains at the forefront of much mathematical
activity a century and a half later.
6 STEPHEN S. GELBART AND STEPHEN D. MILLER

2
By applying Poisson summation to f (x) = e−π x t one thus obtains Jacobi’s
transformation identity
1
(2.9) θ(i t) = √ θ( ti ) ,
t
where

1 X π i n2 τ 1 X 2
θ(τ ) = e = + eπin τ
2 2
n∈Z n=1
(more later in Section 3.1 on θ as a function of a complex variable for Im τ > 0).
Riemann then obtained an integral representation for ξ(s) = π −s/2 Γ( 2s )ζ(s) as
follows:
Z ∞
Γ(s) = ts−1 e−t dt , Re s > 0
0
∞ Z
X ∞
π −s Γ(s) ζ(2s) = (π n2 )−s ts−1 e−t dt , Re s > 1/2
n=1 0
Z ∞
= ts−1 (θ(it) − 21 ) dt
0

∞ 1 1
ts
Z Z
= t s−1 (θ(it) − 12 ) dt + ts−1 θ(it) dt −
1 0 2s 0
Z ∞ Z ∞
= ts−1 (θ(it) − 12 ) dt + t−s−1 θ( ti ) dt − 1
2s
1 1
Z ∞
1 1
= (ts−1 + t1/2−s−1 ) (θ(it) − 12 ) dt − − .
1 2s 1 − 2s
Indeed, replacing s by s/2, the above expression reads
Z ∞
− 2s 1 1 1
s
(2.10) π Γ( 2 ) ζ(s) = (ts/2−1 + t(1−s)/2−1 )(θ(it) − ) dt − − .
1 2 s 1 − s
The integral representation (2.10) allows us to conclude the main analytic properties
mentioned in the introduction:
Theorem 2.1. The function
s
ξ(s) = π − 2 Γ( 2s ) ζ(s)
satisfies properties E, BV, and FE of Section 1.
Proof: We first note that
∞ ∞
1 X 2 X e−π t
(2.11) θ(it) − = e−π n t ≤ e−π n t = = O(e−π t )
2 n=1 n=1
1 − e−π t
for t ≥ 1.2 Since
Z ∞ Z ∞
s −π t
|t e | dt ≤ t b e−π t dt < ∞
1 1
for Re s ≤ b, the integral in (2.10) converges – for any value of s – to an entire
function which is bounded for s in vertical strips. Thus ξ(s) is meromorphic with
2The notation A = O(B) indicates that there exists some absolute constant C > 0 such that
|A| ≤ C · B.
RIEMANN’S ZETA FUNCTION AND BEYOND 7

only simple poles at s = 0 and 1, demonstrating properties E and BV. Having


established that (2.10) gives an analytic continuation, we may conclude that ξ(s) =
ξ(1 − s) (property FE) because of the symmetry present in (2.10). 

2.2. Hecke’s Treatment of Number Fields (1916). In this section we shall


briefly describe Hecke’s generalization [60] of Riemann’s work to certain zeta func-
tions associated to number fields (that is, finite extensions of Q). These subsume
Riemann’s ζ-function, as well as the related Dirichlet L-functions. The latter are
simply Dirichlet series

X χ(n)
(2.12) L(s, χ) = .
n=1
ns
Here χ is a “Dirichlet character”, meaning a non-trivial function χ : Z → C which:
(i) is periodic modulo some integer N ; (ii) obeys χ(nm) = χ(n)χ(m) (“complete
multiplicativity”); and (iii) vanishes on integers sharing a common factor with N .
A Dirichlet character can equally be thought of as a homomorphism from (Z/N Z)∗
to C∗ , extended to Z as a periodic function that vanishes on {n | (n, N ) > 1}. The
Dirichlet L-functions L(s, χ) satisfy the properties E, BV, and FE analogous to
those of ζ(s) (which corresponds to the trivial character); for a complete discussion
and precise analog of Theorem 2.1, see [30].
Our goal here is to describe the generalizations of ζ(s) and L(s, χ) that are the
objects of Hecke’s work, in some sense following the earlier exposition in [42]. This
will necessitate some algebraic background; accordingly this section requires some
familiarity with the concepts involved. However, it is not essential to the rest of the
paper, and readers may wish to skip directly to Section 2.3, or instead to consult
[93, 142] for definitions and examples.
To describe Hecke’s accomplishment, we need to recall some of the local and
global terminology involved. Let F be a number field, and OF its ring of integers.
We will refer to a non-archimedean place v of F as a prime ideal P ⊂ OF . A
fractional ideal of OF is an OF -submodule U such that xU ⊂ OF for some x ∈ F ∗ .
All fractional ideals are invertible (i.e. there exists a fractional ideal P−1 such that
PP−1 = OF ), and all fractional ideals factor uniquely into products of positive
and negative powers of prime ideals. We let ordP (x) denote the exponent of P
occurring in the unique factorization of the principal ideal xOF , and set
|x|v = |x|P = (N P) −ordP (x) ,
where N P is the number of elements in the finite field OF /P. Any real embedding
σ : F → R of F gives rise to a “real” infinite place via the norm |x|v = |σ(x)|;
complex places are defined analogously, and the real and complex places together
comprise the archimedean places of F . For each place of F , the norm | · |v gives a
different completion Fv of F . For example, when F = Q, F∞ = R and Fp = Qp ,
the p-adic numbers (see Section 5 for much more on this theme).
We now come to the generalization of a Dirichlet character to the number field
setting: a Hecke character (also known as a Grössencharacter). We shall think of
one as the product of family of homomorphisms χv : Fv∗ → C∗ , one for each place
of F : Y
χ(x) = χv (x) .
v
8 STEPHEN S. GELBART AND STEPHEN D. MILLER

Two constraints must be made on the family: firstly that χ be trivial on F ∗ , i.e.
for any x ∈ F ⊂ Fv∗
Y
χ(x) = χv (x) = 1 ,
v
and secondly that all but a finite number of the χv be unramified, i.e. trivial
on {x ∈ Fv∗ | |x|v = 1}. If v is such an unramified place, corresponding to a
prime ideal P, χ(P) is defined as χv ($v ), where $v is an element of Fv such that
|$v |v = N P−1 (a “uniformizing parameter” for Fv ). This definition can of course
be extended to ordinary ideals U of OF , provided they are products of prime ideals
corresponding to places where χv is unramified. Hecke’s (abelian) L-series for the
character χ is then defined as the Dirichlet series
X χ(U) Y −1
(2.13) L(s, χ) = s
= 1 − χ(P) (N P)−s .
(N U)
P

Here U is summed over these ordinary ideals of OF just mentioned, and the product
is only over the prime ideals corresponding to these unramified places.
When χ is the trivial character,
P i.e., −sχv = 1 for all v, then L(s, χ) specializes to
be the Dedekind zeta-function (N U) of F . For F = Q this reduces to ζ(s),
and if χ is instead of finite order, LF (s, χ) becomes the Dirichlet L-function (2.12).
Using very clever and intricate arguments, Hecke was able to express his L-series in
terms of generalized “θ-functions”, and to then derive their analytic continuation,
functional equation, and boundedness in vertical strips, a la Riemann.
2.3. Hamburger’s Converse Theorem (1921). Now let us return to the Rie-
mann ζ-function. The next point of the theory is that the Functional Equation
for ζ(s) nearly characterizes it. Indeed, Hamburger [55] showed in 1921 that any
Dirichlet series satisfying ζ’s functional equation and suitable regularity conditions
is necessarily a constant multiple of ζ(s). We state Hamburger’s Theorem 2.4 at
the end of this section, but first begin by describing these conditions, which are
closely related to Boundedness in Vertical strips. In fact, our main motivation in
describing Hamburger’s theorem here is to explain the role of BV and the related
“finite order” conditions; in the modern picture, these are crucial for applications
involving the Converse Theorem (see Sections 3.1, 3.2, and 7.3).
Recall that property BV was stated earlier in terms of the function ξ(s) =
π −s/2 Γ( 2s )ζ(s). One may ask how the individual factors themselves behave as
|Im s| → ∞. Clearly
(2.14) |π −s/2 | = π −Re(s)/2 ,
while Stirling’s formula states that

|Γ(σ + it)| ∼ 2π |t|σ−1/2 e−π |t|/2 ,
(2.15)
uniformly for a ≤ σ ≤ b, |t| → ∞.
Yet the size of |ζ(s)| in the critical strip is quite difficult to pin down. In fact, one
of the central unsolved problems in analytic number theory is the following and its
generalizations.

The Lindelöf Hypothesis: For any fixed ε > 0 and σ ≥ 1/2,


(2.16) ζ(σ + i t) = O(|t|ε ) as |t| → ∞.
RIEMANN’S ZETA FUNCTION AND BEYOND 9

The implied constant in the O-notation here depends implicitly on the value of
ε. In particular, |ζ(1/2 + it)| = O(|t|ε ) for |t| large (this case turns out to be
equivalent to (2.16) via the Phragmen-Lindelöf Principle, Proposition 2.5). The
Lindelöf Hypothesis is implied by the Riemann Hypothesis, and conversely implies
that very few zeros disobey it (see [176, §13]).
Note that by (2.14), (2.15), and the Functional Equation, the behavior for
Re (s) ≤ 1/2 is given by
1/2−σ
t
(2.17) |ζ(σ + it)| ∼ |ζ(1 − σ − it)| , σ fixed, |t| large.

The Lindelöf conjecture is far out of reach, but we can easily prove (weaker) poly-
nomial bounds.
1
Proposition 2.2. ζ(s) − s−1 = O(|s|) for Re s ≥ 1/2.
Proof:
For Re s > 1,
∞ Z ∞
1 X
(2.18) ζ(s) − = n−s − x−s dx
s−1 n=1 1

X∞ Z n+1
(2.19) = (n−s − x−s ) dx .
n=1 n

The integrand in (2.19) is bounded by


Z x
−s −s
|n − x | = s t−s−1 dt ≤ |s| n− Re s − 1 .
n
We conclude that
1
ζ(s) − ≤ |s| ζ(Re s + 1) ,
s−1
1
and so (2.19) gives an analytic continuation of ζ(s) − s−1 to the region Re s > 0.
1
In particular, |ζ(s) − s−1 | ≤ |s| ζ(3/2) for Re s ≥ 1/2. 

Definition: An entire function f (s) is of order ρ if


ρ+
(2.20) f (s) = O(e|s| ) for any  > 0.
It will turn out that the ζ-function and (conjecturally) all L-functions connected
to automorphic forms have order 1. However, many other generalizations of zeta
functions (such as Selberg’s Zeta functions) in fact have order greater than 1.
Proposition 2.3. The function
s
s (s − 1) π − 2 Γ( 2s ) ζ(s)
is (entire and) of order 1.
Proof: By the Functional Equation, it suffices to consider Re s ≥ 1/2. We
have already seen this function is Entire in Theorem 2.1. Another form of Stirling’s
Formula gives that

√ 1 √ 1
(2.21) Γ(s) ∼ 2π e−s s s− 2 = 2π e−s+(s− 2 ) log s , Re s ≥ 1/2, |s| → ∞,
10 STEPHEN S. GELBART AND STEPHEN D. MILLER

and hence Γ(s) = O(e M |s| log |s| ) for some M > 0. Thus by (2.14), (2.21), and
Proposition 2.2
M
s (s − 1) π −s/2 Γ( 2s ) ζ(s) = O(|s|3 e 2 |s| log |s| ).
log |s|
Since for any ε > 0, |s|ε → 0 as |s| → ∞, we conclude
1+ε
s (s − 1) π −s/2 Γ( 2s ) ζ(s) = O(e|s| ).


We note that s (s − 1) ξ(s) is not of any order ρ < 1, as can be seen from (2.14)
and (2.21) as s → ∞ along the real numbers – where ζ(s) is always greater than 1.
3
Theorem 2.4.P(Hamburger’s Converse Theorem )
∞ −s
P∞
Let h(s) = n=1 an n and g(s) = n=1 bn n−s be absolutely convergent for
Re s > 1, and suppose that both (s − 1)h(s) and (s − 1)g(s) are entire functions of
finite order. Assume the functional equation
s 1−s
(2.22) π − 2 Γ( 2s ) h(s) = π − 2 Γ( 1−s
2 ) g(1 − s) .
Then in fact h(s) = g(s) = a1 ζ(s).
This is the theorem which says that ζ(s) is uniquely determined by its func-
tional equation (subject to certain regularity conditions). Hamburger’s theorem
was greatly generalized and enlightened by Hecke approximately 15 years later.
We will in fact later show how to derive Theorem 2.4 from Hecke’s method (see the
discussion after Theorem 3.1.) See also [133, 138].
The original proof of Hamburger’s Theorem relies on the Mellin transform and
inversion formulas; that is, if
Z ∞
−s 1
(2.23) π Γ(s) ζ(2s) = ts−1 (θ(it) − ) dt ,
0 2
then
Z
1 1
t−s π −s Γ(s) ζ(2s) ds

(2.24) θ(it) − =
2 2πi Re(s) = c
for sufficiently large c > 0. Using the Phragmen-Lindelöf principle (Proposi-
tion 2.5), plus the regularity conditions of g(s) and h(s), one can fairly directly
show that every ak is equal to a1 ; that is, h(s) = a1 ζ(s). By the way, it is
of course known that (s − 1)ζ(s) is entire and of order 1. This is because both
s(s − 1)π −s/2 Γ(s/2)ζ(s) and sΓ(s/2)
1
are entire and of order 1 (Theorem 2.1).
2.4. The Phragmen-Lindelöf Principle and Convexity Bounds. A standard
fact from complex analysis, the Phragmen-Lindelöf Principle, can be used to obtain
estimates on ζ(s) in vertical strips from ones on their edges:
Proposition 2.5. (Phragmen-Lindelöf ). Let f (s) be meromorphic in the strip
U = {s | a ≤ Re (s) ≤ b}, a, b ∈ R, with at most finitely-many poles. Suppose that
f (s) satisfies the finite order inequality
 A
f (s) = O e|s| , for some A > 0 ,

3Actually Hamburger proved a more general statement, allowing for an arbitrary, finite number
of poles (see [55],[176, p.31]).
RIEMANN’S ZETA FUNCTION AND BEYOND 11

on U for |Im s| large, and obeys the estimate


f (σ + it) = O(|t|M ) for Re s = a, b, |t| → ∞.
Then
f (σ + it) = O(|t|M ) for a ≤ Re s ≤ b, |t| → ∞
as well.
See [94] for a detailed exposition and proof of Proposition 2.5. An immediate
application of the Phragmen-Lindelöf Principle is to provide a standard bound for
ζ(s) and other L-functions in the critical strip. As an example, let us note the
following bound towards the Lindelöf conjecture:
Lemma 2.1. For any ε > 0,
(2.25) ζ(1/2 + i t) = Oε (t1/4+ε ) , |t| → ∞
where the implied constant depends on ε.
Note that this a sizeable improvement on the trivial bound in Proposition 2.2
towards (2.16).
Proof of Lemma 2.1: First we observe that

X
| ζ(1 + ε + it) | ≤ | n−1 − ε −i t | = ζ(1 + ε) ,
n=1
which is a positive constant. By (2.17), which comes from the functional equation,
| ζ(−ε − it) | = Oε (|t|1/2+ε ) , |t| → ∞ .
Now, set f (s) = ζ(s)ζ(1 − s), a = −ε, b = 1 + ε, and M = 1/2 + ε. Because of the
discussion at the very end of Section 2.3, the conditions of Proposition 2.5 are met;
we conclude |ζ(1/2 + it)ζ(1/2 − it)| = Oε (|t|1/2+ε ) as |t| → ∞. To finish the proof
we replace ε by 2ε,
Pand observe that ζ(s̄) = ζ(s) because of the Schwartz reflection
principle (ζ(s) = n−s is real for s > 1). 

The estimate (2.25) for ζ(s) has been improved many times over; however, for
general L-functions, the bounds given by the above argument are usually the best
known. Because (2.25) interpolates between the bounds at Re s = −1 − ε and ε,
results given by this argument are known as the convexity bounds for L-functions.
A very important problem is to improve these by breaking convexity; even slight
improvements to the convexity bounds for more general L-functions – still falling
far short of Lindelöf’s conjecture – have had many profound applications. Let’s
consider, for example, the possible ways of writing a positive integer as the sum of
three squares. Gauss’ famous condition asserts that the equation
(2.26) x2 + y 2 + z 2 = n
is solvable by some (x, y, z) ∈ Z3 if and only if n is not of the form 4a (8b + 7)
for some integers a, b ≥ 0 (see, for example, [160]). Linnik conjectured that the
solutions to (2.26) are randomly distributed in the sense that the sets
 
(x, y, z)
(2.27) Dn = √ x2 + y 2 + z 2 = n , x, y, z ∈ Z
n
become equidistributed in the sphere S 2 ⊂ R3 as n 6= 4a (8b + 7) increases. This
was in fact proven by W. Duke (see [32, 33, 67]), and can be shown to follow quite
12 STEPHEN S. GELBART AND STEPHEN D. MILLER

directly from subconvexity estimates on automorphic L-functions ([34]), although


this was not Duke’s original argument. For a survey of recent results on sub-
convexity bounds, see [70].
The proof of Lemma 2.1 shows the strength of the finite-order condition. For it
allows us to conclude that ξ(s) decays rapidly as |Im s| → ∞ (and uniformly so in
vertical strips), given only the functional equation and the absolute convergence of
ζ(s) for Re (s) large. This will be useful in the proofs of Theorems 2.4 and 3.1. To
wrap up this section, let’s formally state this for future use.
Lemma 2.2. Assume the conditions of Theorem 2.4 (notably Entirety, Functional
Equation, and the finite order hypothesis). Then both sides of (2.22) are Bounded
in Vertical strips.
We remark that the conclusion of the Lemma does not depend particularly on
the exact form of the Functional Equation (2.22); similar conclusions follow when
the functional equation involves different configurations of Γ-functions and powers
of π.
Proof: The assumption of absolute convergence implies that
X∞
|h(σ + it)| ≤ |an | n−σ < ∞ , σ > 1.
n=1

Then for any ε > 0, |h(s)| is uniformly bounded in the range Re s ≥ 1 + ε, as is


|g(s)| by symmetry. Using the Functional Equation, we see that both
|h(σ + it)| , |g(σ + it)| = O(|t|1/2−σ ) , |t| → ∞
for σ < −ε, and uniformly so in vertical strips (see (2.17)).
We are assuming that (s − 1)g(s) and (s − 1)h(s) are of finite order, so the
Phragmen-Lindelöf Principle (Proposition 2.5) applies. This shows that
|g(σ + it)| , |h(σ + it)| = O(|t|1/2+ε ) , for − ε < σ < 1 + ε.
Thus we have shown that in any vertical strip a ≤ Re s ≤ b, both g(s) and h(s)
are bounded by |Im s|M for some M > 0, as |Im s| → ∞. Stirling’s estimate (2.15)
shows that (2.22) decays rapidly as |Im s| → ∞ in the strip a ≤ Re s ≤ b, and
hence is bounded there. 

3. Modular Forms and the Converse Theorem


3.1. Hecke (1936). As already suggested, Hamburger’s Converse Theorem did
not become completely understood until greatly generalized by Hecke in 1936 ([60,
paper #33], [61]); to describe it, we thus encounter the notion of the space of
modular forms to which functions like θ belong. Note that

1 X 2
θ(τ ) = eπin τ
2 n = −∞
is holomorphic in the upper half plane Im (τ ) > 0; moreover, because it satisfies
(2.9) (when Re τ = 0), clearly
 
−1  τ 1/2
(3.1) θ = θ(τ ) , θ(τ + 2) = θ(τ ) .
τ i
RIEMANN’S ZETA FUNCTION AND BEYOND 13

These two equations say that θ(τ ) is a modular form of weight 21 for the group
generated by τ 7→ τ + 2 and τ 7→ − τ1 . More generally, a modular form of weight
k > 0 and multiplier condition C for the group of substitutions generated by τ 7→
τ + λ and τ 7→ − τ1 is a holomorphic function f (τ ) on the upper half plane satisfying
• (i) f (τ + λ) = f (τ ),
• (ii) f (− τ1 ) = C ( τi )k f (τ ), and
2πiτ P∞ 2πinτ
• (iii) f (τ ) has a Taylor expansion in e λ (cf. (i)): f (τ ) = n=0 an e λ ,
i.e., f is “holomorphic at ∞”.
We denote the space of such f by M (λ, k, C); f is a cusp form if a0 = 0. For
example, the space M (2, 12 , 1) is one dimensional, and consists of multiples of the
θ-function.
Now, given a sequence of complex numbers a0 , a1 , a2 , . . . with an = O(nd ) for
some d > 0, and given λ > 0, k > 0, C = ±1, set

X an
(3.2) φ(s) = ,
n=1
ns
 −s

(3.3) Φ(s) = Γ(s) φ(s) ,
λ
and

2πinτ
X
(3.4) f (τ ) = an e λ .
n=0
(The O-condition on the an ensures that φ(s) converges for Re s > d + 1, and that

f (s) is holomorphic in the upper half plane. In fact, f (τ ) − a0 = O(e− λ Im τ ) –
see (2.11).)
Theorem 3.1. (Hecke’s Converse Theorem) The following two conditions are
equivalent:
(A) Φ(s) + as0 + Ca
k−s is an entire function which is bounded in vertical strips
0

(EBV), and satisfies Φ(s) = C Φ(k − s) (FE);


(B) f belongs to M (λ, k, C).
We will come to the proof Theorem 3.1 shortly, but first wish to explain the
connection to the results of Riemann and Hamburger. Riemann’s Theorem 2.1 is
an example of the direction (B)⇒(A). In the other direction, set
X
φ(s) = ζ(2s) = (n2 )−s ,
n≥1
C = 1 , k = 1/2 , a0 = 1/2 , and λ = 2 .
In this special case, the direction (A)⇒(B) of Theorem 3.1 asserts that θ(τ ) obeys
the modular relations (3.1). Similarly, Theorem 2.4 can be derived from this direc-
tion of Theorem 3.1 as well. For simplicity, suppose that the coefficients an and bn
in the statement are equal (these are not the same an involved here). Then assump-
tions of Theorem 2.4 actually match the properties of ζ and ξ needed in (A). They
guarantee that Φ( 2s ) = π −s/2 Γ( 2s )h(s) is holomorphic in Re s > 0, except perhaps
for a simple pole at s = 1. By the functional equation (2.22), Φ(s) has an analytic
continuation to C except for potential simple poles at s = 0 and 1/2. Because of
(2.22) the residues of Φ(s) at those points are negatives of each other, and thus
14 STEPHEN S. GELBART AND STEPHEN D. MILLER

Φ(s) + as0 + k−s


a0
is Entire, where a0 is the residue of Ψ(s) at s = k. Lemma 2.2
shows the finite order assumption implies that Φ(s) satisfies the BV condition of
(A) as well. Theorem 3.1 therefore produces a modular form f in M (2, 1/2, 1),
which is a one-dimensional space spanned by θ(τ ). So f must in fact be a multiple
of the θ-function, and we conclude that the original Dirichlet series in Theorem 2.4
are multiples of ζ.
Proof of Theorem 3.1: As in Hamburger’s proof of Theorem 2.4, the proof
begins by using Mellin inversion (see (2.24)):
Z
1
(3.5) f (ix) − a0 = x−s Φ(s) ds ,
2πi σ = c
for x > 0, where σ = Re (s), and c is chosen large enough to be in the domain of
absolute convergence of φ(s) (since we are assuming that an = O(nd ), we may take
any c > d + 1).
Assume now (A). We first want to first argue that we can push the line of
integration to the left, past σ = 0, picking up residues of C a0 x−k at s = k ≤ c and
−a0 at s = 0:
Z
−k 1
(3.6) f (ix) − C a0 x = x−s Φ(s) ds .
2π i σ=k−c<0
To see this, we need to show that the integral of Φ(s) over the horizontal paths
[k − c ± ir, c ± ir] tend to zero as r → ∞. We shall use the Boundedness in Vertical
strips assumption to prove the integrand decays rapidly there; in fact, this contour
shift here is the primary importance of the BV property. The assumption that
an = O(nd ) implies that φ(s) converges absolutely for Re s ≥ c > d + 1, where

X
(3.7) |φ(s)| ≤ |an | n−c = O(1) .
n=1
1+ε
Stirling’s asymptotics (2.21) show that Φ(s) satisfies the order-one estimate O(e|s| )
in the region Re s ≥ c. By the functional equation, Φ(s) does as well in the reflected
region Re s ≤ k − c, and the BV assumption from (A) handles the missing strip:
therefore Φ(s) + as0 + Ca 1
k−s is of order one on C. Since sΓ(s) is entire and of order 1,
0

−1
(s − k)φ(s) = (s − k)( 2π s
λ ) Γ(s) Φ(s) is also entire and of order 1 (cf. the end of
Section 2.3). The functional equation
 2s−k
2π Γ(k − s)
φ(s) = C φ(k − s)
λ Γ(s)
shows that
(3.8) |φ(σ + it)| = O(t2c−k ) , σ = k −c < 0
just as in the proof of Lemma 2.1. We conclude from the Phragmen-Lindelöf Princi-
ple (Proposition 2.5) that φ(s) is O(|Im s|K ) for some K, uniformly as |Im s| → ∞
in the strip k − c ≤ Re s ≤ c. Since this growth is at most polynomial, the expo-
nential decay from Stirling’s formula (2.15) gives us that Φ(σ + it) decays faster
than any polynomial in |t| as |t| → ∞, uniformly for σ in the interval [k − c, c].
Thus the integrals
Z c+ir Z c−ir
x−s Φ(s) ds , x−s Φ(s) ds −→ 0 , as r → ∞ ,
k−c+ir k−c−ir
RIEMANN’S ZETA FUNCTION AND BEYOND 15

and the contour shift between (3.5) and (3.6) is valid.


Now, let us resume from (3.6) and apply the functional equation from (A):
Z
−k C
f (ix) − C ao x = x−s Φ(k − s) ds
2πi σ = k − c < 0
Z
C
= xs−k Φ(s) ds (upon s 7→ k − s)
2πi σ = c > k
i
= C x−k (f ( ) − a0 ) by (3.5) ,
x
or
i
f (ix) = C x−k f ( ) ,
x
which is property (ii) of the definition of M (λ, k, C). Properties (i) and (iii) are
immediate from the definition of f (τ ) in (3.4), and we conclude the proof that (B)
follows from (A).
Now suppose (B). We will essentially follow Riemann’s original argument from
Section 2.1, using the integral representation (cf. (2.23))
Z ∞
Φ(s) = ts−1 (f (it) − a0 ) dt .
0

Then

1 ∞ 1
ts
Z Z
i
ts−1 (f (it) − a0 ) dt = t−s−1 f ( ) dt − a0
0 1 t s 0
Z ∞
a0 C a0
= C tk−s−1 (f (it) − a0 ) dt − − .
1 s k−s
Thus
Z ∞
a0 C a0
Φ(s) + + = [ts−1 (f (it) − a0 ) + tk−s−1 C(f (it) − a0 )] dt .
s k−s 1

This expression is clearly EBV, and Φ(s) = C Φ(k − s) (just as in the proof of
Theorem 2.1), whence (A). 

By reducing a question about Dirichlet series to one about modular forms, The-
orem 3.1 represents a great step forward from Riemann’s treatment of ζ. In par-
ticular, it puts his original argument into a very useful and fruitful context. Note
that a specified type of Dirichlet series is connected to any modular form satisfying
 
aτ + b
f = (cτ + d)k f (τ )
cτ + d
for

(3.9)
    
a b a b
∈ SL(2, Z) = a, b, c, d ∈ Z , a d − b c = 1 ,
c d c d
−1
the group of substitutions generated by τ → τ + 1 and τ → τ .
16 STEPHEN S. GELBART AND STEPHEN D. MILLER

3.2. Weil’s Converse Theorem (1967). A. Weil in 1967 completed Hecke’s the-
ory by similarly characterizing modular forms for congruence subgroups, such as
  
a b
(3.10) Γ0 (N ) = ∈ SL(2, Z) c ≡ 0 (mod N ) .
c d
(These subgroups in general have many generators, whereas Hecke’s Theorem deals
with modular forms only for the groups generated by τ 7→ τ + λ and τ 7→ − τ1 .)
Weil’s breakthrough was to twist the series φ(s) by Dirichlet characters. Recall from
Section 2.2 that a Dirichlet character modulo r is a periodic function χ : Z → C
which is completely multiplicative (i.e. χ(nm) = χ(n)χ(m)), and satisfies
χ(1) = 1 , χ(n) = 0, if (n, r) > 1.
Given a Dirichlet character χ modulo r and a proper multiple r0 of r, one may form
a Dirichlet character χ0 modulo r0 by setting
(
χ(n) , (n, r0 ) = 1,
χ0 (n) =
0 , otherwise.
Such a character χ0 obtained this way is termed imprimitive, and one which is not,
primitive. The importance of primitive characters is that their functional equations
are simpler (see [30]). Weil’s converse theorem gives a condition for modularity
under Γ0 (N ) in terms of the functional equations of Dirichlet series twisted by
primitive characters:
Theorem
P∞ 3.2. (Weil [178]) Fix positive integers N and k, and suppose L(s) =
−s
n=1 an n satisfies the following conditions:
• (i) L(s) is absolutely convergent for Re s sufficiently large;
• (ii) for each primitive character χ of modulus r with (r, N ) = 1,

X
Λ(s, χ) = (2π)−s Γ(s) an χ(n) n−s
n=1

continues to an Entire function of s, Bounded in Vertical strips;


• (iii) Each such Λ(s, χ) satisfies the Functional Equation
k
(3.11) Λ(s, χ) = wχ r−1 (r2 N ) 2 −s Λ(k − s, χ̄) ,
where
wχ = ik χ(N ) g(χ) 2
and the Gauss sum
X
g(χ) = χ(n) e 2 π i n/r .
n (mod r)
P∞ 2πinz
Then f (z) = n=1 an e belongs to the space of modular forms for Γ0 (N )
(i.e.
   
az + b a b
f = (c z + d)k f (z) , for all ∈ Γ0 (N ) ,
cz + d c d
and satisfies a holomorphy condition at its “cusps” analogous to property (iii) in
the definition of M (λ, k, C) in Section 3.1).
RIEMANN’S ZETA FUNCTION AND BEYOND 17

Note that the trivial character (with χ(n) ≡ 1) is primitive, so the statement
includes the L-functions used in Theorem 3.1. Property (iii) is certainly satisfied if
L(s) is the L-function of a modular form, as can be shown using a slight variant of
Hecke’s argument used in proving Theorem 3.1. For the obvious reason, we refer
to this Theorem as “Weil’s converse to Hecke Theory”. For a proof, see [10], [68],
or [125].

3.3. Maass Forms (1949). In addition to the holomorphic modular forms on


the complex upper half plane H, there are the non-holomorphic modular forms in-
troduced by Maass [111]. These are equally important, but far more mysterious.
The literature has slight differences in the terminology, but for usa Maass form
d2 d2
will be a non-constant eigenfunction of Laplace operator ∆ = −y 2 dx 2 + dy 2 in
L2 (Γ\H), where Γ is a discontinuous subgroup of SL(2, R), e.g. a congruence sub-
group. The laplacian condition replaces the holomorphy condition here. In contrast
to the holomorphic modular forms, all of which have constructions and geometric
interpretations, the vast majority of Maass forms lack constructions or identifica-
tion. Their mere existence is so subtle that Selberg invented the trace formula [156]
simply to show that they exist for Γ = SL(2, Z)! In fact, deformation results such
as those of Phillips-Sarnak and Wolpert [127–129, 149, 182, 183] demonstrate that
Maass forms are scarce for the generic discrete subgroup Γ ⊂ SL(2, R). For this
reason we shall stick to congruence subgroups Γ for the rest of this exposition.
For now, consider a Maass form φ for Γ = Γ0 (N ) (for simplicity the reader may
take N = 1 and Γ = SL(2, Z)). The Fourier expansion of φ is given by
X √
(3.12) φ(x + iy) = an y Kν ( 2 π |n| y ) e2 π i n x ,
n6=0

where an are coefficients and Kν (t) is the K-Bessel function



π I−s (z) − I s (z) X ( z2 ) s+2m
(3.13) Ks (z) = , Is (z) = .
2 sin πs m=0
m! Γ(s + m + 1)

The parameter ν is related to the Laplace eigenvalue of φ by λ = 1/4 − ν 2 , where


∆φ = λφ. Hecke’s method was extended by Maass to obtainPthe analytic continu-

ation and functional equations of the L-functions L(s, φ) = n=1 an n−s of Maass
R∞
forms on Γ0 (N ) through the integral 0 φ(i y) y s−1/2 dy
y . When Γ = SL(2, Z), for
example, this integral is unchanged by the substitution s 7→ 1 − s. Maass also
proved a converse theorem for his Maass forms for Γ = SL(2, Z); see the comments
at the end of the section.

3.4. Hecke Operators. The Euler product structure of the Riemann ζ-function
has an analog for modular form L-functions through Hecke operators. For any
positive integer n, the Hecke operator
 
1 X X az + b
(3.14) Tn (f )(z) = ak f
n d
ad = n 0≤b<d

preserves the space of modular forms of weight k for Γ0 (N ), so long as n and N


are relatively prime. The same formula applies to Maass forms when k = 0 and the
prefactor n1 is replaced by √1n . A few other operators are used as well, to take into
account symmetries of Γ0 (N ) by which modular forms can be “diagonalized.” In
18 STEPHEN S. GELBART AND STEPHEN D. MILLER

addition to being eigenfunctions of a differential operator (i.e. either the Cauchy-


Riemann operator ∂¯ or the laplacian ∆), a basis of modular forms or Maass forms
can be chosen among eigenfunctions of the Hecke operators as well. As a result,
identities amongst the coefficients can be proven. These are nicely expressed as
factorizations of the L-functions of modular forms. For example, when Γ = SL(2, Z)
the L-series of a holomorphic form of weight k factors as

X Y −1
(3.15) L(s) = an n−s = 1 − ap p−s + p k−1−2s ,
n=1 p

a formula which remains valid for Maass forms if k is taken to be 1.


We end this section with some remarks about the Converse Theorem 3.2. Maass
observed that Hecke’s argument for Theorem 3.1 applies to his Maass forms for
Γ = SL(2, Z) as well, but this method does not prove a converse theorem for Γ0 (N )
for N large. The reason for this is that the group Γ0 (N ) can have many generators,
which are not accounted for by simply one functional equation alone. Interestingly,
Conrey and Farmer [29] have found that by using Hecke operators, a converse
theorem can be proved for a surprisingly large range of N using only a single
functional equation. In another direction, Booker [6] has recently discovered that
the converse theorem requires only a single functional equation when it is specialized
to the L-functions coming from Galois representations, regardless of how large N is.
It is an open question whether or not Weil’s argument applies to Maass forms. A
key point for Weil is that radially symmetric holomorphic functions are necessarily
constant; this is not true in the non-holomorphic case because there are spherical
functions (formed by radially-symmetrizing Im(z)s ), and so Weil’s argument does
directly apply. However, there is nevertheless an applicable converse theorem due
to Jacquet and Langlands, which we will come to in Section 7.1.

4. L-functions from Eisenstein Series (1962-)


In the last section we saw the Mellin transform provided a connection between
holomorphic modular forms, and certain Dirichlet series generalizing ζ(s). Another
quite different connection comes from a family of non-cuspidal modular forms, the
Eisenstein series
X 1
(4.1) Gk (z) = , k even, ≥ 2.
2
(m z + n)k
(m,n) ∈ Z −{0,0}

It is not difficult to show that Gk (z) is a holomorphic modular form of weight k for
SL(2, Z). Via a Poisson summation argument over m one can obtain the Fourier
expansion

2(2π i)k X
(4.2) Gk (z) = 2 ζ(k) + σk−1 (n) e2 π i n z ,
(k − 1)! n=1

where σk−1 (n) is defined in terms of the divisors of n by


X
σt (n) = dt
d|n

(see [160, Section 7.5.5] for details). The appearance of ζ(k) here is the first example
of a very general phenomena, which ultimately leads to the Langlands-Shahidi
RIEMANN’S ZETA FUNCTION AND BEYOND 19

method (Section 8). In the next section, we will describe the generalized non-
holomorphic Eisenstein series considered by Selberg, and their connection to the
analytic properties of the Riemann ζ-function throughout the complex plane – not
just at special integral values alone.

4.1. Selberg’s Analytic Continuation. Selberg’s method [157] can be used to


obtain the analytic continuation and functional equations of the L-functions that
arise in the “constant terms” of Eisenstein series. We shall sketch a form of it in the
classical case of the upper half plane H = {z = x + iy | y > 0}, and the simplest-
possible Eisenstein series. Here we shall summarize the main steps involved; the
details can be found in [7, 91]. We will turn to the general case in Section 8.
Define
1 X ys
E(z, s) =
2 2
|m z + n|2s
(m,n)=Z −{0}
(4.3) gcd(m,n)=1
1 1 X ys
= ,
2 ζ(2 s) |m z + n|2s
(m,n)=Z2 −{0}

for z ∈ H and σ = Re s > 1. This series converges absolutely and uniformly in any
compact subset of the region Re s > 1, and is the first example of a non-holomorphic
Eisenstein series. Very importantly, E(z, s) is unchanged by the substitutions z 7→
az+b
cz+d coming from any matrix in (3.9). Selberg considers the problem of analytically
continuing E(z, s) with respect to s to obtain another functional equation, as we
shall now explain. (Actually, Selberg had several different arguments to do this,
but they mainly appeal to spectral theory to obtain the important properties of
analytic continuation and functional equation of Eisenstein series.)
To motivate the statement of the functional equation, let us first consider the
Fourier expansion of E(z, s). It is given by

X
(4.4) E(z, s) = E(x + i y , s) = am (y, s) e 2 π i m x
m∈Z
2πix
where e(x) = e , and
Z 1
am (y, s) = E(x + iy, s) e−2 π i m x dx .
0
We shall need here only the coefficients a0 and a1 . If one computes directly for
Re s > 1, using the “Bruhat decomposition”4 for SL(2, Z) and recalling (3.13), one
obtains

(4.5) a0 (y, s) = y s + φ(s) y 1−s


 
4The Bruhat decomposition states that all matrices a b
∈ SL(2, Z) with c 6= 0 may
c d
   
1 r α β 1 s
be written as products , where: r and s range over Z; γ over
0 1 γ δ 0 1
Z − {0}; δ over (Z/γZ)∗ ; and α and β are any two integers (which depend on γ and δ, of course)
satisfying αδ − βγ = 1.
20 STEPHEN S. GELBART AND STEPHEN D. MILLER

and

y Ks− 21 (2 π |n| y)
(4.6) an (y, s) = 2 |n|s−1 σ1 − 2 s (n) ,
π −s Γ(s) ζ(2s)
with

Γ(s − 21 ) ζ(2s − 1) ξ(2s − 1)


φ(s) = π 1/2 =
Γ(s) ζ(2s) ξ(2s)
(see [10] for details). In general, φ(s) is called the “constant term” or “scattering”
matrix of E(z, s).
Having described the Eisenstein series E(z, s), we now state and prove Selberg’s
theorem:
Theorem 4.1. (Selberg – see [157]) E(z, s) has a meromorphic continuation to the
whole complex s-plane, and satisfies the functional equation
(4.7) E(z, s) = φ(s) E(z, 1 − s) .

(A Misleading) Proof: Theorem 2.1 and (4.5-4.6) can then be applied to show
that each term in the Fourier expansion
X
E(z, s) = an (y, s) e 2 π i n x
n∈Z

is meromorphic and satisfies the functional equation (4.7). The sum converges
rapidly because Ks (y) decays exponentially as y → ∞. Hence the whole sum is
meromorphic on C, and satisfies (4.7). 

We wrote that the above proof is “misleading” because, although it demonstrates


a connection to Theorem 2.1, in practice it has turned out to be much more fruitful
to reverse the logic – and conclude properties of L-functions from those of Eisenstein
series! Indeed, Theorem 4.1 can be proven using spectral theory, and even in a very
non-arithmetic setting (see [7, 28, 91] for more details). The reader may already
have noticed a similarity between the Fourier expansion of Eisenstein series in (4.4-
4.6), and those of Maass forms in (3.12). In fact, the Eisenstein series E(x + iy, s)
d2 d2
is an eigenfunction of the Laplace operator ∆ = −y 2 dx 2 + dy 2 , with eigenvalue
s(1 − s). A main point in arguing the functional equation is Maass’ lemma, which
ultimately implies that because E(z, s) and E(z, 1−s) share the Laplace eigenvalue
s(1 − s), the two must be multiples of each other. The ratio can be found to be
φ(s) by the inspecting the constant term a0 (y, s), and so the functional equation
(4.7) can be proven without knowing ζ’s functional equation ξ(s) = ξ(1 − s).
Of course, Selberg proved his Theorem 4.1 in much greater generality than we
have stated. Our point is that the analytic continuation and functional equation
for the Eisenstein series furnish an analytic continuation and functional equation
for the Riemann ζ-function. To analytically continue ζ(s), basically “the constant
term” is enough: reading through the spectral proof of the analytic continuation
of φ(s) for E(z, s), one demonstrates that ξ(s) is holomorphic everywhere, save for
simple poles at s = 0 and 1. To get the functional equation, we need to consider
RIEMANN’S ZETA FUNCTION AND BEYOND 21

the non-trivial Fourier coefficient a1 (y, s). Theorem 4.1 yields


(4.8)

2 y Ks−1/2 (2 π y)
= a1 (y, s)
ξ(2s)

ξ(2s − 1) ξ(2s − 1) 2 y K1/2−s (2 π y)
= a1 (y, 1 − s) = ;
ξ(2s) ξ(2s) ξ(2 − 2s)
1+s0
then, using Ks = K−s and setting s = 2 , we have

ξ(s0 ) = ξ(1 − s0 ) ,
exactly the Functional Equation for ζ(s0 ). Incidentally, the same analysis applied
to the general Fourier coefficient an (y) from (4.6) does not give any additional
information (this is because the extra factor |n|s−1 σ1−2s (|n|) already obeys the
functional equation). Boundedness in Vertical strips is another matter, which we
will return to in Section 8.3. Selberg’s work on GL(2) was extended by Langlands
[95, 97] to cover Eisenstein series on general groups, where the analysis is much
more difficult. This forms the basis of the Langlands-Shahidi method, the topic of
Section 8.

5. Generalizations to Adele Groups


In the remaining sections of the paper, we will revisit the techniques and topics
of the earlier sections, but in the expanded setting of automorphic forms on groups
over the adeles. The adeles themselves enter as a language to keep track of the
arithmetic bookkeeping needed for complicated expressions, such as the computa-
tions over general number fields in Section 2.2. They are convenient even in the
simplest examples when the ground field is Q. For instance, we shall see in the
next section how Tate’s thesis naturally produces the Euler product formula for
the Riemann ζ-function:
Y −1 Y
1 − p−s 1 + p−s + p−2s + · · · ,

ζ(s) = =
p p

a formula which itself is a restatement of the unique factorization theorem for


integers. They will be useful in Section 8 for computations involving the Eisenstein
series for SL(2, Z) from Section 4. In general, they are extremely valuable on
general groups, where they give clues for how to structure terms in large sums into
an “Eulerian” form.
The adeles and their notable features are perhaps better explained later on,
within the context of the arguments in which they are used. Nevertheless we give
the basic definitions before proceeding. Given a rational number x, let

(5.1) |x|p := p−ordp (x) ,


where ordp (x) denotes the exponent of p occurring in the unique factorization of
x ∈ Q. This p-adic valuation defines a metric on Q by dp (x, y) = |x − y|p , and its
completion is Qp , the field of p-adic numbers. More concretely, Qp may be viewed
as the formal Laurent series in p
(5.2) x = ck p k + ck+1 p k+1 + · · · , ck 6= 0 , k ∈ Z
22 STEPHEN S. GELBART AND STEPHEN D. MILLER

with integral coefficients 0 ≤ cj < p; alternatively it may be thought of as consisting


of base-p expansions with only finitely many digits to the right of the “decimal”
point, but perhaps infinitely many to the left. Within Qp lies its ring of integers, Zp ,
which is the completion of Z under |·|p . It may instead be viewed as the elements of
Qp as in (5.2) which have k ≥ 0, or those with no digits to the right of the decimal
point in their base-p expansion. The p-adic valuation of course extends to Qp : the
absolute value of x given in (5.2) is p−k , and Zp = {x ∈ Qp | |x|p ≤ 1}. Similarly,
the multiplicative subgroups are Q∗p = Qp − {0} and Z∗p = {x ∈ Qp | |x|p = 1}.
The adeles are formed by piecing together all Qp along with R, which may be
viewed as Q∞ , the completion of Q under the usual archimedean absolute value.
Concretely, the adeles A are the restricted direct product of the Qp with respect to
the Zp ; that means the adeles are infinite-tuples of the form

(5.3) a = (a∞ ; a2 , a3 , a5 , a7 , a11 , . . .) , ap ∈ Qp for all p ≤ ∞

such that all but finitely many ap lie in Zp . Similarly the ideles A∗ are the restricted
direct product of all Q∗p with respect to Z∗p . Addition and multiplication are defined
componentwise in A and A∗ . The rational numbers embed diagonally into the ring A
and play a fundamental role, which will become apparent shortly when it appears in
Tate’s thesis. The adeles, or more properly the ideles, themselves have an absolute
value; its value on a in (5.3) is
Y
|a|A = |ap |p .
p≤∞

Note that this is actually a finite product, because almost all ap have absolute
value equal to one, a theme which underlies many adelic concepts. The diagonally-
embedded Q∗ consists of the ideles with |a|A = 1.
The above construction can be generalized to an arbitrary number field – or even
“global field” – F to obtain its adele ring AF (see [93, 142]). Most constructions in-
volving AQ generalize to AF , though we will mainly focus on F = Q for expositional
ease. Adeles are usually viewed much more algebraically and with much greater
emphasis on their topology (which we have hardly touched); our intention here is
rather to give enough background to illuminate their effectiveness in analysis.

6. Tate’s Thesis (1950)


In his celebrated 1950 Ph.D. thesis [173], J. Tate reinterpreted the methods of
Riemann and Hecke in terms of harmonic analysis on the ideles A∗ of a number field
F . Tate’s method succeeded in precisely isolating and identifying the contribution
to the functional equation from each of the ramified prime ideals P not treated in the
product (2.13), a delicate problem which appeared complicated from the perspective
of Hecke’s classical method. At the same time, Tate’s method is powerful enough
to uniformly reprove the analytic continuation and functional equations of Hecke’s
L-functions. For this local precision, uniformity, and flexibility, Tate’s method
has influenced the many adelic methods at the forefront today. In this section we
explain Tate’s construction and the role of the devices he employs, via a comparison
with Riemann’s argument in Section 2.
RIEMANN’S ZETA FUNCTION AND BEYOND 23

Let us recall Riemann’s integral from Section 2, after a harmless change of vari-
ables:
Z ∞ X 2 2
(6.1) ξ(s) = π −s/2 Γ( 2s ) ζ(s) = xs e−π n x d∗ x .
0 n6=0

Tate instead considers the sum over Z − {0} as an integral over a disconnected
group. In order to keep the flexibility of treating more general sums, he instead
essentially integrates the characteristic function of Z over a much larger set in his
generalized ζ-integral
Z
(6.2) ζ(f, c) = f (a) c(a) d∗ a .
AQ ∗

Here c(a) is any quasi-character of A∗ – that is, a continuous homomorphism from


A∗ to C∗ – which is trivial on Q∗ (for example, we saw before that |a|A is trivial on
Q∗ ); d∗ a is the multiplicative Haar measure on A∗ pieced together as a product of
the local Haar measures d∗ x∞ = |x| dx
and d∗ xp . The latter is normalized so that Z∗p
has measure 1. Finally, the function f is taken to be a product
Y
(6.3) f (a∞ ; a2 , a3 , a5 , . . .) = fp (ap )
p≤∞

of functions fp on Qp , which may depend on the quasi-character c. In the simplest


possibility, which is that c(a) = |a|sA , let us choose
(
1 , |x|p ≤ 1 ,
fp (x) = χZp (x) =
0 , otherwise ,
2
and f∞ (x) = e−π x ; then the integral ζ(f, |·|sA ) actually recovers Riemann’s integral.
This can be seen as follows: first we may “fold” the integral to one over Q∗ \A∗ :
 
Z Z X
(6.4) f (a) |a|sA d∗ a = |a|sA  f (qa) d∗ a .
A∗ Q∗ \A∗ q∈Q∗

The strong approximation principle states that (0, ∞) × Z b ∗ is a fundamental do-


main for Q∗ \A∗ , where Z b∗ = ∗
Q
p<∞ Zp . It is easy to see that f (qa) ≡ 0 on this
fundamental domain unless the rational q is actually an integer, for otherwise, the
p-adic valuation |qa|p = |q|p > 1 for any prime p in the denominator of q. Thus the
role of the fp is to select the integers amongst Q, and (6.4) becomes
 
Z X
(6.5) |a|sA  f (na) d∗ a .
b∗
(0,∞)×Z n6=0

b∗
Now fp ((na)p ) ≡ 1 for all p < ∞, and so the integrand is independent of the Z
factor, which has volume 1 under the Haar measure. Now (6.5) amounts to
Z ∞ X 2 2
(6.6) |a∞ | s e−π n a∞ d∗ a∞ ,
0 n6=0

i.e. (6.1). Thus Tate’s and Riemann’s integrals match for ζ(s).
24 STEPHEN S. GELBART AND STEPHEN D. MILLER

At the same time, the global integral on the lefthand side of (6.4) factors as a
product
Y Z Z  YZ
s ∗ −π|x|2 s dx
(6.7) fp (x) |x|p d xp = e |x| · |xp |sp d∗ xp .
Q∗ R |x| p Zp
p≤∞ p

The integral over R gives π −s/2 Γ( 2s ), and the p-adic integral may actually be broken
up over the “shells” pk Z∗p = {|xp |p = p−k }, k ≥ 0, to give the geometric series
P∞ −k s
k=0 p = (1 − p−s )−1 . This gives the Euler product formula for ζ(s), along
with its natural companion factor π −s/2 Γ( 2s ) for p = ∞ – in other words, the
completed Riemann ξ-function.
We should note that the role of the adelic absolute value (and in particular that
its value is 1 on Q) corresponds to the change of variables x 7→ x/n in the classical
picture. In general for a global field F , we may write the quasi-character c(a) in
the form c0 (a)|a|s , where c0 : A∗ → C∗ is a character of modulus 1. Then c0 (a)
corresponds to χ, a “Hecke character” for F (Section 2.2), and ζ(f, c) differs from
Y −1
LF (s, χ) = 1 − χ(P) (N P)−s
P

(where P now runs over all prime ideals of F ) by only a finite number of factors.
In this idelic setting, Tate uses a Fourier theory and Poisson summation formula
on the ring of adeles A, and proves the elegant functional equation
(6.8) ζ( f , c ) = ζ( fb, b
c),
where fb is the “adelic Fourier transform” of f and b c(a) = c0 (a)|a|1−s . The func-
tional equation for LF (s, χ) may be extracted from this. To illustrate with our
example of the Riemann ζ-function, recall that we had taken c0 to be identically
equal to 1, and in fact our f = fb, so that
s
Y
ζ(f0 , | · |s ) = π − 2 Γ( 2s ) (1 − p−s )−1 = ξ(s) .
p

The functional equation FE is then immediate from (6.8). Tate’s method of course
also yields the Entirety and Boundedness in Vertical strips.

7. Automorphic forms on GL(n)


Thus far we have seen two types of L-functions: The Riemann ζ-function and its
cousins that are treated in the Riemann-Hecke-Tate theory (Section 2), and the L-
functions of modular forms in Hecke’s (other) theory (Section 3.1). We now under-
stand these L-functions to be part of a family, the L-functions of automorphic forms
on GL(n, A). The integrals in Tate’s thesis are over A∗ , which is just GL(1, A),
and the quasi-characters are viewed as automorphic forms on GL(1, Q)\GL(1, A).
We shall now explain how to view the modular forms we saw in Section 3.1 as au-
tomorphic forms on GL(2, Q)\GL(2, A). This leads to two major generalizations:
first to a general number field (or indeed even a global field) F instead of Q, and
second to an arbitrary reductive algebraic group G instead of GL(1) or GL(2).
To recap from Section 3.1, a holomorphic modular form of weight k for Γ =
SL(2, Z) ⊂ SL(2, R) is a holomorphic function on the complex upper half plane H
such that
RIEMANN’S ZETA FUNCTION AND BEYOND 25

 

az+b
 a b
• f cz+d = (cz + d)k f (z) for all ∈Γ
c d
• f (z) has a Fourier expansion f (z) = n≥0 cn e2 π i n z . In addition f is a
P
cusp form if c 0 = 0.
The above definition of course extends to more general groups Γ, such as the con-
gruence subgroups in (3.10). Before considering f as a function on GL(2, A), we
must first explain how to consider f as a function on GL(2, R), or even SL(2, R).
Indeed, there is a correspondence between holomorphic modular forms f of weight k
for Γ\H, and certain functions F on Γ\SL(2, R) defined via the following relations:
  
a b ai + b
F = f (c i + d)−k ,
c d ci + d
(7.1)   √ 
1 x y
f (x + i y) = y −k/2 F √ .
1 1/ y
(We leave matrix entries blank if they are zero.) The key reason for this correspon-
dence is that H is isomorphic to the quotient SL(2, R)/SO(2, R). For more details
and a precise characterization of F , see [10] or [40].
Aside from the holomorphic modular forms, the most significant automorphic
forms on H are the non-holomorphic Maass forms: non-constant, L2 Laplace eigen-
functions on the quotient Γ\H. We described these in Section 3.3. Because of
the identification H ∼= SL(2, R)/SO(2, R), Maass forms can directly be viewed as
functions on Γ\SL(2, R).
Now that we view the holomorphic and Maass modular forms on the group
G = SL(2, R), wide generalizations are possible, and techniques from representa-
tion theory may be applied. The group G acts on L2 (Γ\G) by the right regular
representation, which is translation on the right:
(7.2) [ρ(g)f ](h) = f (hg) .
The study of automorphic forms on Γ\G now becomes understanding the decom-
position of the very large (and highly reducible) representation ρ into irreducible
components. This is the starting point for the notion of “automorphic representa-
tion,” but for that we first need to delve more into the arithmetic nature of Γ, and
consider G adelically.
In addition to the action on the right, left-translation by rational matrices is
very important in many constructions in automorphic forms. We have, therefore,
also the left regular representation:
(7.3) [λ(g)f ](h) = f (g −1 h) ,
which maps L2 (Γ\G) to L2 (gΓg −1 \G). In general this moves automorphic forms for
one congruence subgroup Γ to those on a conjugate, which may be wildly different.
For this reason it is natural to act on the left only by rational matrices g, so that
the conjugate of Γ is still closely related to a congruence subgroup.5 In fact, many
fundamental constructions (such as Hecke operators) require action by rational
matrices g which lie in GL(2, Q), but not SL(2, Q). It is for this reason that
we will consider adelic automorphic forms on GL(2, A), not SL(2, A), though a
theory exists for that group as well. Because GL(2, R) is one dimension larger than
SL(2, R), we technically need to consider L2ω (ZΓ\GL(2, R)) where Z is the center of
5See [113] for a thorough explanation.
26 STEPHEN S. GELBART AND STEPHEN D. MILLER

GL(2, R) (=scalar multiples of the identity matrix). Here ω is a central character


(that is, a character of Z) and this L2 space consists of functions on G which
transform by Z according to ω, but which are otherwise square-integrable on the
quotient ZΓ\GL(2, R). As a minor technicality, we will now consider Γ = GL(2, Z)
instead of SL(2, Z) to make the picture more uniform. The setup of this paragraph
works equally well for Γ = GL(n, Z) and G = GL(n, R).
Finally we now come to the adeles. The adelic group GL(n, A) is the product of
GL(n, R) with GL(n, Af ), the direct product of all GL(n, Qp ) with respect to their
integral subgroups GL(n, Zp ). We have already seen – at least in the case n = 2 –
that the first factor, GL(n, R), acts on automorphic functions on ZΓ\G on the right,
and that rational matrices act on the left. Just as with A∗ = GL(1, A) in Tate’s
thesis, there is a version of the strong approximation theoremQ for GL(n, A). It
states that GL(n, A) = GL(n, Q)GL(n, R)Kf , where Kf = p<∞ GL(n, Zp ). We
now define an action of GL(n, Af ) on the left that extends the action of GL(n, Q):

(7.4) [λ(gf )F ](h) = [λ(γ)F ](h) = F (γ −1 h) ,


where γ ∈ GL(n, Q) is the factor guaranteed by the strong approximation theorem
in writing gf ∈ GL(n, Af ) ⊂ GL(n, A) as a product. This definition is well defined,
because any two close “approximants” γ must be related by a multiple of an integral
matrix, and F is presumed to be invariant under GL(n, Z). Roughly speaking, the
topology on the adeles is given in terms of a basis of products of GL(n, Zp ) and
finite index subgroups, which are related to congruence groups. Thus the adelic
topology aligns with the invariance of F under matrices in a congruence subgroup.
Unifying these actions leads to the notion of adelic representations and adelized
automorphic forms, where F , instead of being a function on GL(n, R) alone, is
padded with extra variables. Namely, we have an adelic function
(7.5) FA (g∞ ; g2 , g3 , g5 . . .)
such that almost all gp lie in GL(n, Zp ), and
(7.6) FA (g∞ ; g2 , g3 , g5 . . .) = [λ(g2 ) λ(g3 ) λ(g5 ) · · · λ(gp ) · · · F ](g∞ ) ,
where the number of λ(gp )’s that act in an nontrivial way is finite, and their actions
for various p’s commute with each other. The adelized function F has the properties

(7.7) F (g∞ ) = FA (g∞ ; 1, 1, 1, · · · ) and FA (γ g) = FA (g) ,


for any diagonally embedded rational matrix γ. The center Z and central charac-
ter ω have analogous adelic versions, which are related to automorphic forms on
GL(1, A), in fact. The right regular representation now acts on GL(n, A) by the
formula (7.2), but note that this right action of the factor GL(n, Af ) is really a left
action on GL(n, R).
This leads us to our final version of automorphic representation: an irreducible
subrepresentation of the action of the right regular representation ρ on GL(n, A)
on L2ω (ZA GL(n, Q)\GL(n, A)). The constituents of these subspaces are general-
izations of the automorphic forms we encountered previously. All have classical
counterparts as functions with transformation properties for various congruence
subgroups Γ of GL(n, Z), but the adelic version provides a uniform framework.
The role of Γ itself is replaced by right-invariance under finite index subgroups Kf0
RIEMANN’S ZETA FUNCTION AND BEYOND 27

of Kf = p<∞ GL(n, Zp ); here Γ = {γR | γ ∈ GL(n, Q), γf ∈ Kf0 }, where γR and


Q

γf denote the projections of γ ∈ GL(n, A) to the factors GL(n, R) and GL(n, Af ),


respectively. We note that forms for various conjugate subgroups all fall into the
same irreducible adelic automorphic representation, as do “newforms.” More im-
portantly, the theory of Hecke operators (Section 3.4) for powers of a prime p (e.g.
the Tpk from Section 3.4) can be recast as the study of the action of GL(n, Qp ),
which allows the powerful representation theory of this p-adic group to be used.
In general, the adelic framework unifies many constructions in automorphic forms,
and explains their effectiveness. Better yet, it provides insight for new constructions
which would seem very difficult to uncover using only the classical perspective.
7.1. Jacquet-Langlands (1970). In 1970, a remarkable book was published:
“Automorphic Forms on GL(2)”, by H. Jacquet and R. Langlands [72]. The irre-
ducible unitary representations π of GL(n, A) discussed above factor into restricted
tensor products π ∼ = ⊗p≤∞ πp , where πp is a “local representation” of GL(n, Qp ).
One can treat the case of a number field, or even an arbitrary global field in a
similar way. For n = 2, Jacquet and Langlands rephrase Hecke’s theory from
Section 3.1 using adelic machinery, much in the way Tate reworked Riemann and
Hecke’s classical arguments. In particular, they attach a global L-function Λ(s, π)
(a Dirichlet series times a product of gamma factors, such as the π −s/2 Γ(s/2) that
differentiates ξ(s) from ζ(s)) to each automorphic representation of GL(2). They
prove that Λ(s, π) is “nice,” meaning that it satisfies the standard properties of
Entirety, Boundedness in Vertical strips, and Functional Equation that Hecke’s
method yields. Secondly they give a criteria for any “nice” L-function of this type
to come from an automorphic representation; that is, a converse theorem.
Although the methods of group representations are new, the underlying tech-
nique of Jacquet-Langlands is fundamentally Hecke’s method, as we shall briefly
describe. However, neither the statement nor proof of their converse theorem is
really Weil’s Theorem 3.2. For example, let us return to the discussion concluding
Section 3. Weil’s proof of his converse theorem demonstrates that only a finite
number of Dirichlet characters are required in his twisting hypothesis (ii). In fact,
Piatetski-Shapiro [130], carefully examining this point, discovered an important
simplifying feature in the early 1970s which has become one of the most important
technical devices in today’s applications. He found that Jacquet-Langlands’ proof
also requires only a finite number of twists by characters – but a completely disjoint
set of characters from the ones Weil needed! For a classical treatment, see [144].
Recall how in Section 3.1 we considered the L-functions of modular forms for
SL(2, Z). The first example of a modular form whose L-function is entire is Ra-
manujan’s ∆ form
Y
(7.8) ∆(z) = e2 π i z (1 − e2 π i n z ) 24 ,
n≥1

which has weight k = 12. (See [160] for a beautiful


P exposition of ∆ in the context of
Hecke theory.) Expand the product as ∆(z) = n≥1 τ (n)e 2 π i n z and normalize the
coefficients by setting an = nτ11/2
(n)
; in this normalization, the Ramanujan conjecture
(established by Deligne [31]) can be stated uniformly as |ap | ≤ 2 for all primes p.
Ramanujan also conjectured6 that the “standard” (i.e. Hecke) L-series associated

6though not in the language of L-functions.


28 STEPHEN S. GELBART AND STEPHEN D. MILLER

to ∆ has an Euler product over primes, much like ζ:


X Y
(7.9) L(s, ∆) = an n−s = (1 − ap p−s + p−2s )−1 .
n≥1 p

This was proven by Mordell [119], and nowadays we understand the factorization as
being equivalent to the assertion that ∆ is an eigenfunction of the Hecke operators
(3.14) from Section 3.4 – in particular, this is (3.15).
Let us now explain the connection between the arguments of Jacquet-Langlands
and of Hecke. Our starting point is the Fourier expansion of a modular or Maass
form φ(x + iy) in the variable x, in which it is periodic (with period 1 in the case
of SL(2, Z), as we shall now consider). Recall that if φ is a holomorphic cusp form
of weight k,

X
(7.10) φ(z) = cn e 2 π i n z .
n=1

Similarly for a Maass form we have the Fourier expansion (3.12). Up to constants
(and a factor of y k/2 in the holomorphic case, like in (7.1)), we may write these
expansions as
X an
(7.11) 1/2
W (2 π n y) e2 π i n x ,
n6=0
|n|
cn
where as before an = in the holomorphic case, and
n(k−1)/2
 k/2 −y
y e , φ holomorphic,
(7.12) W (y) = p
|y| Kν (|y|) , φ a Maass form.
In Section 7 we saw how both holomorphic and Maass forms can be viewed as
functions on GL(2, R). With this point of view we can write the corresponding
function, up to constants, as
X an   
n
(7.13) F (g) = W g ,
|n|1/2 1
n6=0

where
   
1 x y1
(7.14) W k = e2 π i n x W (2 π y1 /y2 ) .
1 y2
(Here the matrix k on the lefthand side is orthogonal; all matrices in GL(2) can be
written in that form according to the Iwasawa decomposition.) This W (g) is called
a “Whittaker” function in connection with the special functions it is related to. It
satisfies a transformation law on the left:
  
1 u
(7.15) W g = e2 π i u W (g) ,
1
and thus can be obtained from the integral
Z 1   
1 u
(7.16) W (g) = F g e−2 π i u du .
0 1
The adelic method of Jacquet and Langlands involves incorporating the coeffi-
cient an into a cognate Whittaker function which generalizes the properties (7.15)
RIEMANN’S ZETA FUNCTION AND BEYOND 29

and (7.16). Consider now the adelized version FA of F defined in (7.5-7.7), and
define its adelic Whittaker function
Z   
1 u
(7.17) WA (gA ) = FA gA ψ(−u) du , gA ∈ GL(2, A) ,
Q\A 1
where ψ is a non-trivial character of A that is trivial on the subgroup Q (which
we recall is diagonally embedded into A). The measure du is normalized to give
Q\A measure 1. The definition depends on the precise choice of character, but all

non-trivial characters can be written  for some q ∈ Q , and this q can
 as ψ(qu)
q
be absorbed into gA via the matrix ; changing variables does not affect
1
the measure since the adelic absolute value |q|A = 1. The result is that FA can be
reconstructed from WA via the succinct formula
  
X q
(7.18) FA (gA ) = WA gA ,

1
q∈Q

much like (7.13).


Now we shall make a tacit assumption that our original modular form is a Hecke
eigenform (see Section 3.4). Our Whittaker function here, like many adelic func-
tions, can be expressed as a product of local Whittaker functions Wp on GL(2, Qp ):
Y
(7.19) WA (gA ) = Wp (gp ) , gA = (g∞ ; g2 , g3 , g5 , g7 . . .) ,
p≤∞

where each Wp obeys a transformation law similar to (7.15). In fact, just as with
the Iwasawa decomposition in (7.14), the local Whittaker functions depend only on
diagonal matrices, and actually their value there is related to the original Fourier
pk

coefficient by W = apk . This last fact underlies the connection between
1
(7.11) and (7.18): the extra adelic variables encode the value of the Fourier coeffi-
cients; these are very often zero, notably when k < 0 and the subscript is no longer
an integer. This is why the sum over Q, which appears to be much larger, actually
corresponds to the sum over Z.
Jacquet and Langlands use this theory beautifully to write the global L-function
as
Z  
a s−1/2 ∗
(7.20) Λ(s) = FA |a|A d a.
Q∗ \A∗ 1
This
 has afunctional equation s 7→ 1 − s, owing to the invariance of W under
−1
, just as in Hecke’s argument. Now by substituting (7.18) and collapsing
1
the common Q∗ from the quotient and sum together (“unfolding”), the integral
(7.21) Z    
a s−1/2 ∗
Y Z ap
Λ(s) = WA |a|A d a = Wp |ap |ps−1/2 d∗ ap
A ∗ 1 Q∗ 1
p≤∞ p

splits as a product of local integrals. The ones for p < ∞ separately give the local
factors of an Euler product which represents the Dirichlet series for Hecke’s L-
function L(s), and the integral for p = ∞ gives the corresponding Γ-functions L(s)
must be multiplied by in order to have a clean functional equation. This is entirely
30 STEPHEN S. GELBART AND STEPHEN D. MILLER

analogous to the situation in Tate’s thesis after (6.7), and the computations are
deep down identically those needed for the classical treatment in Section 3. Details
can be found in [10, 48, 72].
We should emphasize that the method is far more general and has strong ad-
vantages in its local precision, in that it gives a very satisfactory treatment of the
contribution to the functional equation by each prime. Also the technique works
for congruence subgroups, as well as over general global fields. Just as Tate’s thesis
understood Riemann’s ζ-function in terms of A∗ = GL(1, A), Jacquet-Langlands
subsumed the theory of modular forms and their L-functions through GL(2, A).
Subsequently, efforts were underway to provide a similar theory for general groups,
most notably GL(n, A).

7.2. Godement-Jacquet (1972). Tate (see Section 6) redid Hecke (Section 2.2)
by using adeles, developing a Poisson summation formula, and working with
Z
ζ(f, c) = f (a) c(a) d∗ a .
A∗
R. Godement and H. Jacquet [49] generalized Tate by working with GL(n) for
arbitrary n instead of GL(1). In particular, they proved that the global, completed
L-functions of automorphic forms on GL(n) satisfy properties E, BV and FE of
Section 1 (actually their integral representation for GL(2) is completely different
than Jacquet-Langlands and Hecke).
We shall not pursue this avenue here, but will briefly describe the L-functions
of cusp forms on GL(n). Recall how after (7.8) we renormalized the coefficients
of Ramanujan’s ∆-form by a factor of n(k−1)/2 . This can be carried out for any
holomorphic cusp form f of weight k for SL(2, Z), resulting in an Euler product of
the same form as (7.9). Writing ap = αp + αp−1 , the Euler product factors further
as
Y
(7.22) L(s, f ) = (1 − αp p−s )−1 (1 − αp−1 p−s )−1 .
p

The preceding expression is called a degree two Euler product because of its two
p−s )−1 .
Q
factors, in comparison with the degree one Euler product ζ(s) = p (1 −
The L-functions of cusp forms φ for GL(n, AQ ) are Euler products of degree n,
n
YY
(7.23) L(s, φ) = (1 − αp,j p−s )−1 .
p j=1

To form their global, completed L-functions, they must be multiplied by a product


of n Γ-factors,
n
Y
(7.24) L∞ (s, φ) = ΓR (s + µj ) ,
j=1

where the µj are special complex parameters related to π (for example, the ν
from Maass forms in (3.12)), and ΓR (s) = π −s/2 Γ(s/2) is again the factor which
distinguishes ξ(s) from ζ(s). The completed L-function,
(7.25) Λ(s, φ) = L∞ (s, φ) L(s, φ)
RIEMANN’S ZETA FUNCTION AND BEYOND 31

is Entire (unless n = 1 and Λ(s, φ) = ζ(s)), Bounded in Vertical strips, and satisfies
the Functional Equation
(7.26) Λ(s, φ) = w Q1/2 − s Λ(1 − s, φ̃) .
Here w is a complex number of modulus one, the “conductor” Q is a positive integer
(related to the congruence subgroup φ comes from), and φ̃ is the “contragredient”
automorphic form to φ (coming from the automorphic representation dual to φ’s).
The notion of contragredient does not really rear its head in the previous topics we
have covered, but is a feature of the more general functional equations.

7.3. Jacquet-Piatetski-Shapiro-Shalika (1979). Another proof of the analytic


properties of the standard L-functions of cusp forms on GL(n) is a generalization
of Hecke’s method (Section 3.1). In the 1970s Piatetski-Shapiro and Shalika [131,
169] independently developed their “Whittaker” expansions on GL(n) in order to
generalize the expansion (7.18) of Jacquet-Langlands. The Whittaker function on
GL(n, AQ ) is given by the integral
(7.27)  1 u12 u13 ··· u1n  
Z 1 u23 ··· u2n
WA (gA ) = FA  .. .. .  
. . ..  gA  ψ(u12 + u23 + u n−1 n ) du ,

N (Q)\N (A) 1 u n−1 n
1
with the integration over the subgroup N of unit upper triangular matrices. The
expansion of FA in terms of WA is given by
X
(7.28) FA (gA ) = WA (γ g),
γ∈N (Q)\P (Q)

where P is the subgroup of G consisting of matrices whose bottom row is (0 0 · · · 0 1).


This also has an explicit, classical version (“neoclassical,” in the terminology of
Jacquet), which can be found in [9, 71] – both of which are excellent references for
this section.
In a series of papers, Jacquet, Piatetski-Shapiro, and Shalika used these ex-
pansions to generalize Hecke’s construction to L-functions of automorphic forms
on GL(n) (including properties E, BV, and FE), and prove a converse theorem
for GL(3) (see [8, 9, 71, 73, 74], and [115, 117] for a different treatment). This is
a big advantage over the method of Godement-Jacquet, whose integral is over the
large group GL(n). The integrals of Jacquet, Piatetski-Shapiro, and Shalika instead
involve integration over one dimensional subgroups, matching the one complex vari-
able of the L-functions L(s). In later papers of Cogdell and Piatetski-Shapiro, a
powerful converse theorem has been established for GL(n) (see [27] and Section 9).
These techniques lie close to the heart of the “Rankin-Selberg” method, which uses
integral representations to generate a wide variety of the Langlands L-functions we
will come to in Sections 8 and 9 (see [9] for a thorough, though slightly out of date,
survey). While the statement of the converse theorem is quite technical, it is simi-
lar in form to Weil’s Theorem 3.2, in that it involves the assumptions of Entirety,
Boundedness in Vertical strips, and Functional Equation; as in the above proof
of Hecke’s theorem Theorem 3.1, these are used to shift a contour integral which
reconstructs an automorphic form using Mellin inversion. However, an important
difference is that their converse theorem typically involves twisting by automorphic
forms on GL(m), not merely Dirichlet characters (which, we have seen, correspond
32 STEPHEN S. GELBART AND STEPHEN D. MILLER

to automorphic forms on GL(1)). This is an important topic in Section 9, where


we state a typical version in Theorem 9.1; a general account can be found in [21].

8. Langlands-Shahidi (1967-)
This section is meant for readers having some familiarity with Lie groups, but
it can be skipped without loss of continuity. References include [3, 7, 44, 56, 118,
166, 167]. Our purpose here is to describe the general method of obtaining analytic
properties of L-functions from Eisenstein series, generalizing Selberg’s method for
ζ(s) in Section 4.1. Many of the applications in Section 9 are based upon properties
yielded by the Langlands-Shahidi method.
The theory of Eisenstein series was widened by Langlands to more general Lie
groups in [95, 97]; in particular Langlands proved the analytic continuation and
functional equations that were useful in Selberg’s proof of the analytic properties
of ζ(s). In his Yale monograph [96], Langlands considered the constant terms of
the completely general Eisenstein series. This time, a wide variety of (generalized)
L-functions appeared; his analysis gives their meromorphic continuation. The cal-
culations involved are quite complicated and are performed adelically; they led
Langlands to define the L-group and ultimately to the formulation of his functori-
ality conjectures.
Recall the example of GL(2) from Section 4.1 (which we will reconsider through
group representations and adeles in Section 8.2). There, analysis of the constant
term and first Fourier coefficient already sufficed for the analytic continuation and
functional equation of ζ(s) via Selberg’s method. Langlands proposed studying the
non-trivial Fourier coefficients in general, and Shahidi has now worked that the-
ory out ([161–168]) along with Kim and others. In general it has been a difficult
challenge to prove the L-functions arising in the constant terms and Fourier coeffi-
cients are entire. The analytic continuation of Eisenstein series typically gives the
meromorphic continuation to C, except for a finite number of poles on the real axis
between 0 and 1; these come from points where the Eisenstein series themselves are
not known to be holomorphic. A recent breakthrough came with a clever observa-
tion of H. Kim: the residues of the Eisenstein series at these potential singularities
are L2 , non-cuspidal automorphic forms, and – as in Section 7 – give rise to unitary
representations that can be explicitly described by the Eisenstein series they came
from. Kim remarked that results about the classification of irreducible unitary rep-
resentations show that many of these potential representations do not exist, thus
allowing one to conclude the holomorphy of the Eisenstein series at these points in
question! When combined with [162, 164], this has recently led to new examples of
entire L-functions (more on this in Section 9).

8.1. An Outline of the Method. The following is a brief sketch of the main
points of the method; a fuller introduction with more definitions and detailed ex-
amples can be found in [166]. Detailed examples of constant term calculations can
be found in many places, e.g. [44, 96, 100, 114]. Though it is possible to describe the
method without adeles (as was done in Section 4.1), their use is key in higher rank
for factoring infinite sums and product expansions into L-functions. Because the
Langlands-Shahidi method utilizes various algebraic groups, we will have to assume
some familiarity with the basic concepts. For this reason we include an example of
the GL(2) case in Section 8.2.
RIEMANN’S ZETA FUNCTION AND BEYOND 33

Let F be a global field, A = AF its ring of adeles, and G a split algebraic group
over F . Much carries over to quasi-split case as well, and we will highlight the
technical changes needed for this at the end. Fix a Borel (= a maximal connected
solvable) subgroup B ⊂ G, and a standard maximal parabolic P ⊃ B defined over
F .7 Decompose B = T U, where T is a maximal torus. The parabolic can be also
decomposed as P = M N , where the unipotent radical N ⊂ U , and M is the unique
Levi component containing T . Denote by L G,L M,L N, etc. the Langlands dual
L-groups (see [166] for definitions).
One of the key aspects of this method is that it uses many possibilities of parabol-
ics of different groups G, especially exceptional groups. This is simultaneously a
strength (in that there is a wide range of exotic possibilities) and a limitation (in
that there are only finitely many exceptional groups).
8.1.1. Cuspidal Eisenstein Series. Recall that an automorphic form in L2 (Γ\G)
is associated to a (unitary) automorphic representation of G. Let π = ⊗v πv be
a cuspidal automorphic representation of M (A); we may assume that almost all
components πv are spherical unitary representations (meaning that they have a
vector fixed by G(Ov ), where Ov is the ring of integers of the local field Fv ). For
these places v the equivalence class of the unitary representation πv is determined
by a semisimple conjugacy class tv ∈ L G, the L-group. This conjugacy class is used
to define the L-functions below in (8.3). The finite number of exceptional places
are where π ramifies.
A maximal parabolic subgroup P has a modulus character δP , which is the ratio
of the Haar measures on M ·N and N ·M . It is related to the simple root of G which
does not identically vanish on P . For any automorphic form φ in the representation
space of π, we can define the Eisenstein series
X
(8.1) E(s, g, φ) = φ(γg) δP (γg)s
γ∈P (F )\G(F )

and their constant terms


Z
(8.2) c(s, g, φ) = E(s, ng, φ) dn ,
N 0 (F )\N 0 (A)

where N 0 is the unipotent radical of the opposite parabolic P 0 to P (it is related


by the longest element in the Weyl group). One can view the constant term as an
automorphic form on M , and we will shortly relate it to φ and π. The measure dn
is normalized to give the quotient N 0 (F )\N 0 (A) volume 1. The notion of constant
term applies to any parabolic, but P 0 is the most useful one for our purposes.
8.1.2. Langlands L-functions. If ρ is a finite-dimensional complex representation of
L
M , and S is a finite set including the archimedean and ramified places of F and
π, then the partial Langlands L-function is
Y
(8.3) LS (s, π, ρ) = det(I − ρ(tv ) qv−s )−1 .
v ∈S
/

7The theory has an extension to non-maximal parabolic subgroups, but this does not yield any
extra information about L-functions. This matches the fact that the Eisenstein series for maximal
parabolic subgroups depend on one complex variable, as do L-functions.
34 STEPHEN S. GELBART AND STEPHEN D. MILLER

Here qv is the cardinality of the residue field of Fv , a prime power. The full, com-
pleted, L-function involves extra factors for the places in S, whose definition is
technical and in general difficult. This is connected to the local Langlands cor-
respondence, proven recently by Harris and Taylor for GL(n) and by Jiang and
Soudry for SO(2n + 1) (see [14, 57, 59, 62, 63, 65, 78, 101]). When ρ is the standard
representation of L GL(n) = GL(n) and F = Q, the Euler factors in (8.3) agree
with those in (7.23); in general the degree of LS (s, π, ρ) equals the dimension of ρ.
8.1.3. The Constant Term Formula. The constant term formula involves the sum
of two terms. The first, which only occurs when the parabolic P is its own opposite
P 0 , is φ(g)δP (g)s – simply the term in (8.1) for γ = the identity matrix. Langlands
showed that the map from φ to the second term is described by an operator
 
m
Y L(a j s , π̃ , r j )  ⊗v∈S A(s, πv ) ,
(8.4) M (s, π) = 
j=1
L(1 + aj s , π̃ , r j )

where the A(s, πv ) are a finite collection of operators, r the adjoint action of L M
on the lie algebra of L N , r1 , . . . , rm the irreducible representations it decomposes
into, and aj integers which are multiples of each other (coming from roots related
to the rj ). The variety of decompositions of r is what gives this method much of its
power for treating complicated L-functions. See [166] for a fuller discussion, along
with an example for the Lie group G2 and the symmetric cube L-function. Tables
listing Lie groups and the representations rj occurring for them can be found in
[96] and [163], for example.
8.1.4. The Non-Constant Term: Local Coefficients. We must now make a further
restriction on the choice of π involved, namely that it be generic, i.e. have a
Whittaker model. This means that if ψ is a generic unitary character of U (F )\U (A),
we need to require
Z
W (g, ψ) = φ(ng) ψ(n) dn 6= 0 , UM = U ∩ M
UM (F )\UM (A)

for some φ and g (we have already seen this notion in (7.17) and (7.27)).
Shahidi’s formula uses the Casselman-Shalika formula for Whittaker functions
(see [18, 171]) to express the following non-constant term at the identity g = e as

Z m
Y 1 Y
(8.5) E(s, ne, φ) ψ(n) dn = · Wv (e) ,
N 0 (F )\N 0 (A) j=1
L(1 + aj s , π̃ , rj )
v∈S

for a certain choice of φ. Applying the functional equation of the Eisenstein se-
ries (which has the constant-term ratio involved), one gets the “crude” functional
equation for the product of m L-functions

m
Y m
Y Y
(8.6) LS (aj s , π̃ , rj ) = LS (1 − aj s , π , rj ) · (local factors) .
j=1 j=1 v∈S

Shahidi’s papers [162] and [164] match all the local factors above to the desired
L-functions (cf. the remark after (8.3)). This gives the full functional equation for
these m L-functions, but only when multiplied together. His 1990 paper [164] uses
RIEMANN’S ZETA FUNCTION AND BEYOND 35

an induction argument to isolate the functional equation of each of the above m


factors separately.

8.1.5. Analytic Properties and the Quasi-Split Case. It still remains to prove that
the L-functions are entire, except perhaps at s = 0 and 1 (where the order of the
poles is understood, like for ξ(s)). The theory of Eisenstein series provides this
full analyticity for the L-functions arising in the constant term unless π satisfies a
self-duality condition; even in this case, it can be shown that the L-functions have
only a finite number of poles, all lying on the real axis between 0 and 1. Kim’s
observation of using the unitary dual has worked in many cases to eliminate this
possibility. It is also always possible to remove the potential poles by twisting by
a highly-ramified GL(1) character of AF ; this has been crucial for applications to
functoriality through the converse theorem [25–27], which we come to in Section 9.
The main difference in the quasi-split case is that the action of the Galois group
GF is no longer trivial. The L-groups are potentially disconnected, inasmuch as
they are semi-direct products of a connected component with GF . Also, the repre-
sentation ρ used to define the Langlands L-functions in (8.3) may also depend on
the place v.

8.2. GL(2) Example. Here we reconsider the ζ-function example from Section 4.1,
but in the framework of the Langlands-Shahidi method. In this setting, the Eisen-
stein series on G = GL(2) is defined by
X
(8.7) E(s, g, f ) = f (γg) ,
γ∈B(Q)\G(Q)
 
x a
where M = P = P 0 = B = ⊂ G is the Borel subgroup/minimal
b
parabolic (GL(2) is too small a group to afford other interesting choices). The
Eisenstein series formed from π are related to the representations induced from π,
from M (A) to G(A). In (8.7) we may absorb the factor δPs into f by taking a
G(A)
vector in the induced representation I(s) = IndB(A) |a|s , which roughly speaking is
the space of functions
    
a x s+1
f : G(A) → C f g = |a| f (g) .
a−1
The Eisenstein series (8.7) converges for Re(s) sufficiently large. In fact, we may
choose our f ∈ I(s) so that E(s, g, f ) reduces to just the classical Eisenstein series
E(z, 1+s
2 ) considered in Section 4.1. To do this, we take f to be identically 1 on
K = O(2, R) × p<∞ GL(2, Zp ), and use the fact B(Q)\G(Q) ' B(Z)\G(Z)8;
b Q

this, in view of the Iwasawa decomposition G = B K, b is the simplest choice. It


corresponds to (8.7), taking π to be the trivial representation of M (A). In general,
Eisenstein series are always induced from automorphic forms on smaller groups,
which in the example here is just the constant function on the factors M (A) =
GL(1, A) × GL(1, A) ⊂ B.

8Classically speaking, this isomorphism comes from the decomposition of any rational matrix
g ∈ GL(2, Q) as g = bu, b ∈ B(Q), u ∈ GL(2, Z). The resulting indexing of B(Z)\G(Z) via
rational matrices gives powerful insight into how to arrange the summands of the Eisenstein series
into a computationally-useful form.
36 STEPHEN S. GELBART AND STEPHEN D. MILLER

To compute the constant term, we appeal to the Bruhat decomposition


G = B t BwB = B t BwN ,
   
(8.8) −1 1 ?
w= , N= ⊂G
1 1
which is valid over any field. When applied to γ ∈ B(Q)\G(Q), it allows us to
compute the constant term integral over N :
(8.9) Z X Z
c(s, g, f ) = E(s, ng, f ) dn = f (g) + f (w γ n g) dn
N (Q)\N (A) γ∈N (Q) N (Q)\N (A)

= f (g) + [M (s)f ](g) ,


where M (s) is the intertwining operator
Z
(8.10) M (s)f (g) = f (w n g) dn
N (A)
Q
from I(s) to I(−s). If f is chosen to be a product f (g) = p fp (gp ), then the
integral (8.10) factors further to give an Euler product, in analogy to (6.4) and
(6.7). For the details of this example, see Langlands’ article [100]; in general, his
constant term method [96] gives similar integrals for constant terms over general
groups. In any event, for our example here where f is trivial on K b (i.e. so that
ξ(s)
E(s, f, g) recovers the classical Eisenstein series), [M (s)f ](e) = ξ(s+1) .
To complete our discussion let us again compute the ψ-th Fourier coefficient of
E(s, g, f ), where ψ is a non-trivial additive character of N (A) trivial on N (Q) (or,
equivalently, a non-trivial additive character of A trivial on Q). Afterwards we will
use the functional equation of E(s, g, f ) to get the functional equation of ζ(s).
We find
Z
Eψ (s, e, f ) := E(s, n, f ) ψ(n) dn
N (Q)\N (A)
(8.11)
c(s)
= ,
ξ(s + 1)
and
ξ(s) c(−s)
Eψ ( −s , e , M (s)f ) = ,
ξ(s + 1) ξ(1 − s)
where c(s) = c(−s) is non-zero (it is related to the K-Bessel function appearing in
(4.6)). So using the functional equation

E(s, e, f ) = E(−s, e , M (s)f ) ,


it follows that
ξ( s ) = ξ( 1 − s ) .
8.3. Boundedness in Vertical Strips and Non-vanishing. We have just seen
how the functional equation and several analytic properties of L-functions can be
obtained via Langlands’ analytic continuation of Eisenstein series, which itself relies
on spectral theory. Through (8.11) and the known holomorphy of E(s, f, g) on the
line Re s = 0 (which follows from the general spectral analysis), one also obtains a
new proof of the famous result that ζ(s) never vanishes along the line Re s = 1 (see
[75, 120, 153, 161]). Among other things, this result is the key to the standard proof
RIEMANN’S ZETA FUNCTION AND BEYOND 37

of the Prime Number Theorem – as was originally outlined by Riemann himself in


[145]! In fact, this proof of the non-vanishing of L-functions on the line Re s = 1
using Eisenstein series turns out to be the most general method available at present,
in some cases working far inside the known range of absolute convergence of certain
L-functions.
Intriguingly, it is possible to prove Boundedness in Vertical strips using the
Langlands-Shahidi method. This is striking, because our other examples (Riemann,
Hecke, and Tate, e.g. Theorem 2.1) all acquire BV through an integral represen-
tation of an L-function; here the L-function’s analytic properties are obtained very
indirectly. Not surprisingly, the argument is more round-about and subtle, but pays
off in that it turns out – again – to extend to more general L-functions than treated
by other methods alone (see [45, 153]). This has been very important for applica-
tions to the Langlands functoriality conjectures through the converse theorem (see
Section 9 for further discussion).
Recall that our BV assertion is that s(s − 1)ξ(s) is bounded for s in any vertical
strip. In our situation, the fact that
ξ(s)
r(s) =
ξ(s + 1)
ρ
satisfies the finite order estimate O(e|s| ) in Re s ≥ 12 is possible to prove using
spectral theory. We recall that ξ(s) satisfies P that finite order inequality in the
∞ −3/2
region Re s ≥ 3/2; this is because |ζ(s)| ≤ n=1 n = ζ(3/2) is bounded
1+ε
there, and Stirling’s formula (2.21) shows that Γ(s/2) = O(e|s| ) there. Thus
ξ(s) = r(s)ξ(s+1) obeys the finite order inequality in Re s ≥ 1/2, and the functional
equation shows this is true for Re s ≤ 1/2 as well – giving a new proof that ξ(s)
is of finite order. For ζ and the general L-functions considered in [45], additional
Eisenstein series and group representations, along with some results of [56] and
[123], are required. In particular, more is needed than mere meromorphicity of the
Eisenstein series alone. One also needs an estimate on the growth rate of Eisenstein
series in vertical strips going beyond the original work of Langlands, which here is
not being used as simply a “black box.” For a different, though related, method
of obtaining BV and the non-vanishing of ξ(s) on the line Re(s) = 1 through
Eisenstein series, see [151].

9. The Langlands Program (1970-)


Many articles have been addressed to the “Langlands program” (e.g. [1, 42]
and the references in the introduction), and it is not our desire to add to these.
However, one part of the program is closely connected to our discussion: namely, it
“explains” why the analytic continuation and functional equation of the L-functions
of automorphic forms on GL(n) probably suffice to ensure that any L-series in
arithmetic has an analytic continuation and functional equation!

9.1. The Converse Theorem of Cogdell-Piatetski-Shapiro (1999). Let’s see


why: many arithmetic objects, such as elliptic curves, have an L-series attached to
them which are conjectured to be entire, and have functional equations similar to
those possessed by our L-functions. Some very interesting examples, which we will
not touch on here directly, involve the Artin conjecture (which involves L-series of
Galois representations); see [80, 147].
38 STEPHEN S. GELBART AND STEPHEN D. MILLER

For an elliptic curve E defined over the rational numbers, this L-series, called
the Hasse-Weil L-function L(s, E), is defined by counting points on E over varying
finite fields. Here 1 + p − ap is the number of points on the reduced curve modulo
p, and L(s, E) is defined by the Euler product in (3.15) with k = 2, except for
a finite number of exceptional prime factors (see [172]). The resemblance to the
L-functions of holomorphic modular forms of weight 2 is the springboard for the
celebrated “Modularity Conjecture” of Taniyama, Shimura, and Weil, which was
recently proven in [181], [175], and [12]. It asserts that L(s, E) = L(s, f ), the L-
function of some holomorphic cusp form f of weight 2 on Γ0 (N ), where N is a subtle
invariant (the “conductor”) calculable from the arithmetic of the curve.9 Since
the L-functions of modular forms are known to be entire through Hecke’s theory
(Section 3.1), we therefore now know that the Hasse-Weil L-functions of rational
elliptic curves are entire. Among other things, this gives a definition of L(s, E) at
the center of its critical strip, where the Birch-Swinnerton-Dyer conjecture asserts
deep relations with arithmetic ([5, 54, 90]).
One might ask if the modularity of an elliptic curve might be proved using Weil’s
Converse Theorem 3.2. Unfortunately, this route requires one to know that the
Hasse-Weil L-functions are entire beforehand, which at present seems far beyond
reach. Thus the prospect of proving entirety and applying the Converse Theorem
to these arithmetic L-functions seems to be begging the question. However, it is
an interesting fact the Converse Theorem on GL(3) (proven in 1979 by Jacquet,
Piatetski-Shapiro, and Shalika) [73] does enter into the proof of the Modularity
Conjecture, as it had been earlier used to establish a key step: the Langlands-
Tunnell Theorem [99, 177].
Furthermore the Converse Theorem (in the form developed by Piatetski-Shapiro
and Cogdell, e.g. [27], Theorem 9.1) has had remarkable success towards the Lang-
lands Program in a different aspect. Roughly speaking, the Langlands conjectures
assert correspondences between automorphic forms on different groups. When
starting with an automorphic form, it is often possible to prove the Entirety,
Boundedness in Vertical strips, and Functional Equations of the (twisted) L-
functions the Converse Theorem requires. This has led to very significant progress,
especially on liftings of automorphic forms on GL(n) to GL(m), n < m. In particu-
lar, the recent breakthroughs of Kim and Shahidi (cf. [23, 82, 85–87] and Section 8),
and also of Lafforgue ([37, 92, 105]), have proven many important new examples
of Langlands “Functoriality”, by showing certain L-functions are entire and then
appealing to the Converse Theorem of Cogdell and Piatetski-Shapiro ([27]). Unfor-
tunately the precise statements connected to the use of the Converse Theorem are
quite technical and complicated, and so we will just make do with a less-technical
(but still very useful) version of the Converse Theorem in Theorem 9.1. We will
then summarize the main applications in Section 9.3. In short the basic idea, which
can also been see through the standard functoriality conjectures of Langlands, is
the following: any Langlands L-function (8.3) – of any automorphic form, on any
group, over any field – should itself be the L-function of an automorphic form on
GL(n, AQ ). So GL(n, AQ ) is speculated to the mother of all automorphic forms,

9Weil’s contribution [178] to the conjecture is closely related to Theorem 3.2: in fact his
prediction of modularity on Γ0 (N ), N being the conductor of E, comes from a comparison of
the expected functional equations of Hasse-Weil L-functions with the exact form of (3.11) in
Theorem 3.2.
RIEMANN’S ZETA FUNCTION AND BEYOND 39

and its offspring L-functions are already known to have an analytic continuation
and functional equation.
9.2. Examples of Langlands L-functions: Symmetric Powers. To get the
full statements of the Langlands conjectures, one requires even more than the def-
initions of the Langlands L-functions from (8.3). For simplicity and the benefit of
readers who have skipped Section 8, we will explain the relevant L-functions here
in the everywhere-unramified case, which corresponds to cusp forms on GL(n, R)
invariant under GL(n, Z).
Recall the Euler product for the L-function of a modular form f from (7.22); this
formula is also valid for the Maass forms from Section 3.3, of course provided they
are eigenforms of all the Hecke operators Tp . To better highlight the symmetries
involved (as well as to allow for more generality), let us introduce the parameter
βp = αp−1 and rewrite (7.22) as
Y
(9.1) L(s, f ) = (1 − αp p−s )−1 (1 − βp p−s )−1 .
p

Langlands has defined higher degree Euler products from L(s, f ) called the sym-
metric k-th power L-functions:

k
YY
(9.2) L(s, Symk f ) = (1 − αpj βpk−j p−s )−1 .
p j=0

The definition of the symmetric power L-functions is a general example of Lang-


lands’ method of creating new Dirichlet series out of Euler products. The major
challenge, as we shall see, is to derive important analytic properties of these Dirich-
let series, and thereby put them on the same footing as the other L-functions we
have come across. His formulation is in terms of finite dimensional representations,
which in this case of GL(2) is the k + 1-dimensional representation on homogeneous
polynomials of degree k. Other examples give rise to Euler products which are very
symmetric, like this one on the righthand side of (9.2). In general, one starts by
factoring the L-function of an automorphic form on GL(n) as
n
YY
(9.3) L(s, π) = (1 − αp,j p−s )−1 .
p j=1

New Euler products may be taken using symmetric combinations of the αp,j above
(the individual αp,j chiefly have meaning only in the context of their aggregate
{αp,j }1≤j≤n ). In addition to the symmetric powers for GL(2), there are symmetric
and exterior powers for GL(n):
Y Y
(9.4) L(s, π, Symk ) = (1 − αp,i1 αp,i2 · · · αp,ik p−s )−1
p i1 ≤i2 ≤···≤ik

Y Y
(9.5) L(s, π, Extk ) = (1 − αp,i1 αp,i2 · · · αp,ik p−s )−1 .
p i1 <i2 <···<ik

Given another L-function on GL(m)


m
YY
0
(9.6) L(s, π ) = (1 − βp,k p−s )−1 ,
p k=1
40 STEPHEN S. GELBART AND STEPHEN D. MILLER

Langlands forms the “Rankin-Selberg” tensor product


YY n Y m
(9.7) L(s, π ⊗ π 0 ) = (1 − αp,j βp,k p−s )−1 ,
p j=1 k=1

in analogy with the classical constructions [143, 155] for GL(2) (see [9]). There is a
complementary theory for Γ-factors and completed, global Langlands L-functions
as well. The general Langlands construction is in terms of finite dimensional repre-
sentations of L-groups (Section 8.1.2); in particular, they can be repeated in various
configurations. Now, thanks to the recent proof of the local Langlands correspon-
dence by Harris and Taylor for GL(n) [14, 57, 59, 62, 63, 65, 101], the definitions at
the ramified places can be made also. Langlands’ deep conjectures, in these cases
here, assert that each of the L-functions defined above is in fact the L-function of
some automorphic form on some GL(d), where d is the degree of the Euler product
in each case (i.e. the number of factors occurring for each prime). Or, in other
words, if his symmetric-looking Euler products look like the Euler product of an
automorphic form as in (9.3), they probably are!
9.3. Recent Examples of Langlands Functoriality (2000-). To continue, we
now wish to focus on the examples of Langlands’ lifting mentioned above. We will
describe various lifts which start with automorphic forms on GL(n), and create
automorphic forms on some GL(m), m > n. Though many examples of Lang-
lands functoriality are known in various types of cases, this class is very analytic
and has largely been unapproachable without using the types of analytic proper-
ties of L-functions that we have come across in this paper. When considered as
correspondences between eigenfunctions one space and another, the lifts below are
quite stunning theorems in harmonic analysis, made possible by a deep use of the
arithmetic of GL(n, Z).
Having explained the tensor product L-function (9.7), we can now state a version
of the converse theorem (in practice, slightly weaker assumptions are often used, as
well as analogs over different number fields):
Theorem 9.1. (GL(n) × GL(n − 2) Converse Theorem – [27])
Consider the Euler product L(s, π) (9.3), and assume that it is convergent for
Re s sufficiently large. Suppose that L(s, π) along with all possible tensor product L-
functions L(s, π ⊗ τ ), for τ an arbitrary cuspidal automorphic form on GL(m, AQ ),
1 ≤ m ≤ n−2, satisfy Entirety, Boundedness in Vertical strips, and the Functional
Equation. Then L(s, π) is in fact the L-function of a cuspidal automorphic form
on GL(n, AQ ).
Of course, in this statement we have not described the global L-function (e.g.
Γ-factors) whose analytic properties we are describing, but it is similar to the ones
from Section 3.1. Needless to say, Theorem 9.1 is a generalization of Theorem 3.2.
When n = 3, it is an earlier theorem of Jacquet-Piatetski-Shapiro-Shalika [73]. To
use the Converse Theorem to establish lifting to GL(n), one still needs to show that
various tensor product L-functions obey the analytic conditions it requires. Such
properties are themselves very difficult assertions in their own right, and progress
has been hard won. We shall now describe the established lifts from GL(n) to
GL(m) that were mentioned at the end of the last subsection.
The first such example is the symmetric square lift Sym2 : GL(2) → GL(3), the
so-called Gelbart-Jacquet lift [41]. Because it is the simplest of these to explain,
RIEMANN’S ZETA FUNCTION AND BEYOND 41

we will spend a moment to go over how it is proved. A central role is played by


Shimura’s integral representation of the symmetric square L-function [170]; one
obtains the necessary analytic properties of L(s, Sym2 π ⊗ χ), where χ is a Dirichlet
character (recall that these are automorphic forms on GL(1)). Then the Converse
Theorem of [73] (i.e. Theorem 9.1 with n = 3) implies the existence of a cuspidal
automorphic representation Π whose L-function L(s, Π) = L(s, Sym2 π) – i.e., the
Langlands functorial symmetric square lift from GL(2) to GL(3).
Examples on GL(n) for n ≥ 4 require the Cogdell-Piatetski-Shapiro versions of
the Converse theorem, and are quite technical, even in description. Fortunately,
many have been achieved in the last few years, mainly as a consequence of new
analytic properties from the Langlands-Shahidi method (Section 8), mined from
various configurations of parabolic subgroups in exceptional groups such as E8 .
Here is a summary of the recent lifts:

Theorem 9.2. The following instances of Langlands functoriality are known. That
is, in each case there are automorphic forms on the target GL(n) whose standard
L-functions agree with the Langlands L-functions on the source group (cf. Sec-
tion 9.2):
• Gelbart-Jacquet [41]. Sym2 : GL(2) → GL(3).
• Ramakrishnan [140]. Tensor Product: GL(2) × GL(2) → GL(4).
• Kim-Shahidi [85, 86]. Tensor product: GL(2) × GL(3) → GL(6).
• Kim-Shahidi [85, 86]. Sym3 : GL(2) → GL(4).
• Kim [82, 85]. Ext2 : GL(4) → GL(6) weakly automorphic (bad at 2 and 3).
• Kim [82, 85]. Sym4 : GL(2) → GL(5).
• Cogdell-Kim-Piatetski-Shapiro-Shahidi [23,24]: Weak functoriality to GL(n)
for generic cusp forms on split classical groups.

The notion of “weak” automorphy means that an automorphic form on the target
GL(n) exists whose L-function matches the desired Euler product – but except
perhaps for a finite number of factors. Much more about these results can be found
in these references, and also the ICM lectures [22, 168]. Ramakrishnan’s result
used an integral representation for a triple-product L-function ([39, 58, 66, 134]),
but can also now be proven using the Langlands-Shahidi method ([82]). The last
example mentioned here is a lift from generic cuspidal automorphic forms on SO(n)
or Sp(2n) to some GL(m) (see [22–24, 168]). A differing “descent method” (i.e.
studying the opposite direction of the lift) of Ginzburg, Rallis, and Soudry [47] can
be used to establish the lifts of [23, 24] in strong form; in other words, the adjective
“weak” can be removed from the last assertion of Theorem 9.2.
Of course Langlands’ conjectures go far beyond these examples involving only
GL(n) over a number field. Other routes, through theta liftings (see [2, 106]) and
the Arthur-Selberg trace formula (see [1,43,89]), have also provided many instances
of Langlands Functoriality. In particular, the trace formula is in some sense the
most successful when successful, in that it gives a very complete description and
characterization of the lifts it treats. Nevertheless, the full force of Langlands’
Conjectures seem absolutely beyond current technology (see [104] for intriguing
comments by Langlands on the limitations of the trace formula). We shall not
describe these nor the exact formulations of the Converse Theorem here, though
we hope we have transmitted the flavor of the arguments and technical analytic
properties such as EBV which have put these recent results in grasp.
42 STEPHEN S. GELBART AND STEPHEN D. MILLER

9.4. Applications to Number Theory (2001-). The coefficients of modular


forms on the complex upper half plane H play a fundamental role in many problems
in numberP theory. For example, the coefficients of holomorphic modular forms
φ(z) = n≥0 cn e 2 π i n z can be related to various counting problems, such as the
number of ways to represent a number as a sum of squares, or the number of
points on an elliptic curve (Section 9.1). The coefficients an of the non-holomorphic
Maass forms in (3.12) are also related to number theory as well, ranging for example
ax+bx̄
from Galois theory to the properties of Kloosterman sums xx̄≡1 (mod p) e2 π i p
P

[50, 69, 70, 148, 158]. The sizes of the an and eigenvalue parameter ν, along with
their distributions, are very important in many instances; in the remainder of this
section, we will describe the role of the analytic properties of L-functions in gleaning
some of this information.

9.4.1. Progress towards the Ramanujan and Selberg conjectures. Recall Ramanu-
jan’s ∆ form, defined in (7.8). We mentioned that Ramanujan conjectured a bound
on the normalized coefficients an of his ∆ form, a bound which has a natural gen-
eralization to the coefficients of modular forms of any weight, and to Maass forms
as well.
Conjecture 9.3. (Ramanujan Conjecture) Let φ(x + iy) be either a holomorphic
cusp form of weight k with Fourier coefficients cn = an n(k−1)/2 as in (7.10), or
a Maass form with Fourier coefficients an as in (3.12). Then an = O(|n|ε ) for
any ε > 0 (of course the implied constant in the O-notation here may depend on
ε). When φ is a Hecke eigenform and a1 is normalized to be 1, then equivalently
|ap | ≤ 2.
This conjecture was proven in the holomorphic case by Deligne [31] in 1974, but
remains open for Maass forms.
Years later after Ramanujan, Selberg made a separate conjecture about the size
of the parameter ν that enters into the Fourier expansion of Maass forms. It is
related to the Laplace eigenvalue by λ = 1/4 − ν 2 . Selberg conjectured
Conjecture 9.4. (Selberg, 1965 [158]) Let λ > 0 be the Laplace eigenvalue of a
Maass form for Γ\H, where Γ is a congruence subgroup of SL(2, Z). Then λ ≥ 41
(equivalently, ν is purely imaginary).
Selberg was originally motivated by questions involving cancellation in sums of
Kloosterman sums, but his question is a deep one about the nature of the Rie-
mann surfaces Γ\H. Their volumes grow to infinity as their index increases, and
one would naively expect an accumulation of small Laplace eigenvalues. However,
Selberg predicts a barrier at λ = 41 . This has implications for the geometry of
Γ\H: intuitively, small eigenvalues are a measure of how close a surface is to being
disconnected, since, after all, the multiplicity of the eigenvalue λ = 0 is the number
of disconnected components. These ideas have played a crucial role in the devel-
opment of expander graphs: discrete combinatorial networks which have relatively
few edges connecting their vertices, but which are extremely difficult to disconnect
by removing only a moderate number of edges. See [107, 108, 112, 124, 148, 150]. By
the way, Maass forms with eigenvalue exactly equal to 14 are known to exist, and
in fact they will play a role later at the end of this section. So Selberg’s conjecture,
if true, is sharp!
RIEMANN’S ZETA FUNCTION AND BEYOND 43

Not long after Selberg’s conjecture, Satake observed a unifying reformulation


of both the Ramanujan and Selberg conjectures, in terms of representation theory
(more specifically, tempered representations). Suppose φ is a Hecke eigenform. A
key idea was the parametrization of the Hecke eigenvalues (which are also Fourier
coefficients) ap , for p prime, as ap = αp + αp−1 , αp ∈ C (cf. (7.22)). This has
meaning from the representation theory of the group GL(2, Qp ), and is analogous
to the convention of writing the Laplace eigenvalue λ = 14 − ν 2 . The statement that
|ap | ≤ 2 is equivalent to proscribing that the complex modulus satisfy |αp | = 1.
Writing αp as pµp , the connection between Ramanujan’s and Selberg’s conjectures
becomes even more clear: both ν and all µp = logp (αp ) should be purely imaginary.
The generalized Ramanujan-Selberg conjecture asserts this phenomenon holds
for GL(n):
Conjecture 9.5. If π is a cusp form on GL(n) which is unramified at the prime
p, the quantities αp,j appearing in the Euler product (9.3) obey |αp,j | = 1.
All but a finite number of primes are ramified for π, and none of them are
when Γ = GL(n, Z). A similar statement for the archimedean case p = ∞ is
conjectured to be true for the parameters µ∞,j , generalizing ν, that appear in the
Γ-factors that multiply the L-function in its global, completed form (see (7.24)). For
the cognoscenti we will note that the Ramanujan conjecture 9.5 has a statement
in terms of representation theory which covers the ramified places as well: the
associated local representations πp of GL(n, Qp ) should all be tempered.
Though the generalized conjecture for GL(n) is of course no easier than it was
for GL(2), this added perspective has been crucial for two reasons. The first is
that we know a “trivial” or “local bound” coming from representation theory [76]
that |Re µp,j | < 1/2 for all places p ≤ ∞. For GL(2) this is quite trivial indeed:
it states, for example, that the Laplace eigenvalue is merely positive, and that the
corresponding bound on the Hecke eigenvalue ap comes directly from the bound-
edness of a cusp form. However, for GL(n), n > 2, this bound actually becomes
quite deep, due to a separation feature between the trivial and non-trivial unitary
irreducible representations of GL(n).
The second reason is that the Langlands program connects automorphic forms
on different GL(n)’s, for example through symmetric powers. Notably, the factors
(1 − αpn−1 p−s )−1 and (1 − αp1−n p−s )−1 occur in the Euler product for the n − 1-st
symmetric power from GL(2) to GL(n) (formula (9.2)). If this symmetric power L-
function was indeed the L-function of a cusp form, we would conclude that p−1/2 <
|αpn−1 | < p1/2 from the “trivial bounds” above. This gives an improved bound
towards the Ramanujan-Selberg conjectures for any n for which the symmetric
power lifting can be established – a bound which approaches the conjecture |αp | = 1
itself as n → ∞. A similar magnification can be performed with the archimedean
parameters µ∞,j .
Thus the Langlands program (in particular, the symmetric power functorial lift-
ings) implies both the Ramanujan and Selberg conjectures, and their generalizations
to GL(n)! It should be noted that this strategy is different from Deligne’s and other
arguments coming from algebraic geometry – which themselves have been success-
ful for certain cohomological forms, but do not apply to Maass forms, for example.
(Actually Deligne’s argument uses the “magnification” mechanism of the previous
paragraph, but in a different context.) Anytime a new lift is proven or a new bound
on the |αp,j | of cusp forms on GL(n) is established, it results in a bound towards the
44 STEPHEN S. GELBART AND STEPHEN D. MILLER

Ramanujan and Selberg conjectures. Using results of [109], an analytic technique of


[35], and the recent progress of Kim-Shahidi described in the previous subsection,
the following bounds have been proven:
Theorem 9.6. (Kim-Sarnak [82, Appendix 2]) If π is a cusp form on GL(2, AQ )
then
(9.8) p−7/64 ≤ |αp | ≤ p 7/64 , if π is unramified at p,
and
(9.9)
 2
975 1 7
λ ≥ = − ≈ .238037 , if π comes from a Maass form.
4096 4 64
This theorem is for Q, but results are also possible over general number fields.
7
A weaker estimate (with 64 replaced by 19 ) is established by Kim-Shahidi in [87],
using their recent progress and techniques from [163].

9.4.2. The distribution of the Hecke eigenvalues, Sato-Tate. Having seen that the
Hecke eigenvalue parameters αp for a GL(2) modular, Hecke eigenform should lie
on the unit circle in the complex plane, we now turn to their distribution over
this circle as p varies. The question has its origin in conjectures and investigations
made independently by Sato and Tate [174] for the ap of rational elliptic curves
(Section 9.1). Namely, if we consider the phase of αp , i.e. the angle θp such that
ap = 2 cos θp , the θp ∈ [0, π] should be equidistributed with respect to the measure
2 2
π sin θ dθ. This means that
Z β  
#{ α < θp < β | p ≤ X} 2 2
(9.10) lim = sin θ dθ ;
X→∞ #{ p ≤ X} α π
when viewed in terms of the ap themselves, √ the conjecture states that a histogram
1
of the ap is governed by the distribution 2π 4 − x2 , which looks like a semi-circle
(really, semi-ellipse) between −2 and 2 of area 1. The Sato-Tate semi-circle measure
occurs in many contexts; here it is related to the Weyl integration formula, which
weighs the relative sizes of conjugacy classes in SL(2, R). This conjecture is not
meant to be valid for all elliptic curves (nor, by extension, to all modular forms
via Wiles et al), but instead only for the “typical” (i.e. non-CM) elliptic curve.
In the other cases, the distribution is much simpler and the desired results are
known (see [159]). Regardless, the Sato-Tate conjecture is expected to also hold
for most modular and Maass forms, as we shall see shortly. See [98, p.210] for a
generalization to GL(n).
As in nearly all distributional questions in number theory, an equivalent formu-
lation of the Sato-Tate conjecture can be made in terms of the moments
X
(9.11) Sm (X) := am
p .
p≤X

Conjecturally Sm (X)/π(X) should tend to the constant


Z 2
Sm (X) 1 p
(9.12) lim = xm 4 − x2 dx
X→∞ π(X) 2 π −2
(π(X), as in the introduction, refers to the number of primes p ≤ X). In fact, the
truth of (9.12) for all m ≥ 0 implies the Sato-Tate conjecture (9.10).
RIEMANN’S ZETA FUNCTION AND BEYOND 45

In view of the connection with the powers of ap = αp + αp−1 , and symmetric


power L-functions in the previous subsection, it is not surprising that symmetric
power L-functions play a role in the Sato-Tate conjecture as well. In fact, the non-
vanishing and holomorphy of the m-th symmetric power L-function L(s, Symm π)
in the region Re s ≥ 1 implies the m-th moment (9.12) (see [159,165], and also Ogg’s
paper [126], which shows the holomorphy actually implies non-vanishing). Kim and
Shahidi have now established this for m ≤ 9 ([87]). Actually the nonvanishing of
ζ(s) in the region Re s ≥ 1Pis essentially what Riemann observed implies the prime
number theorem π(X) = p≤X 1 ∼ X/ log X, so it is natural to see this analytic
condition appear in a counting problem. This is a typical way exotic L-functions
enter into analytic number theory.
Though the formulation of (9.10) here implicitly assumed the Ramanujan con-
jectures, (9.12) is more general. It can be viewed as saying that the Ramanujan
conjecture is true on average – and much more. We note in passing that various on-
average results can be proven using the theory of L-functions. The Rankin-Selberg
method [143, 155] has its origin in this issue for GL(2); the generalization of the
Rankin-Selberg method to GL(n) ([74, 162]) also gives a weaker on-average version
of Ramanujan. One can also give a relatively large lower bound on the percentage
of primes p such that the Ramanujan conjecture holds for p ([87, 139, 141]).
Finally, we conclude by describing a result of Sarnak [152]. We had mentioned be-
fore that the Maass forms for SL(2, Z)\H – the non-holomorphic L2 -eigenfunctions
of the Laplacian – are quite mysterious in nature, and none has been explicitly
described. However Maass, in his original paper [111], constructed some examples
for Γ\H, where Γ is a congruence subgroup of SL(2, Z). These, and generalizations
coming from Galois theory through the Artin conjecture, are very special types of
Maass forms, and come from algebraic constructions. In particular, several give fas-
cinating examples of Maass forms (3.12) whose coefficients an are relatively small
integers – bounded in absolute value by the number of divisors of n. This is remark-
able because of the discreteness and limitation of the possible coefficients. In [152]
Sarnak considers hypothetical Maass forms with integral coefficients that are not
examples of the known constructions from Galois theory. In these cases, the results
of Kim and Shahidi [87] on the non-vanishing and holomorphy of L(s, Symm π) on
the line Re s = 1 give the asymptotics of the m-th moment, i.e. (9.12), for m ≤ 9.
If the coefficients are indeed integral, the Ramanujan conjecture asserts that the ap
should only assume one of the five values {−2, −1, 0, 1, 2}. This constraint makes
it difficult to match the predicted moments, and in fact with m = 6 it is possible
to show the impossibility of all the ap being integral. Indeed, even without the
Ramanujan conjecture, the assumption that all ap ∈ Z can be ruled out simply by
taking linear combinations of (9.12), and concluding that

Z 2
1 X 1 p
(9.13) lim P (ap ) = P (x) 4 − x2 dx = 1 .
X→∞ π(X) 2π −2
p≤X

Here P (x) = x2 (4 − x2 )(x2 − 1), a sixth degree polynomial which vanishes at the
integers {−2, −1, 0, 1, 2}, and is negative at all others; a contradiction arises because
the righthand side is positive. As a result, one obtains the first algebraicity result
in the subject of Maass forms:
46 STEPHEN S. GELBART AND STEPHEN D. MILLER

Theorem 9.7. (Sarnak [152]) Let φ be a Maass form for Γ\H as in (3.12) with
integral coefficients. Assume φ is a Hecke eigenform. Then φ arises from a Galois
representation, and in particular the Laplace eigenvalue of φ is 14 (i.e. ν = 0).

A generalization has been established by Brumley [13]. We leave the reader


with a some open conjectures – both widely believed to be true and supported
by numerical evidence – on which the ideas of functoriality and L-functions have
brought an interesting perspective.

Conjecture 9.8. (See [17]) Let φ be a Maass form which has Laplace eigenvalue
1
4 . Does φ necessarily arise from a Galois representation?

Conjecture 9.9. (Cartier, [16]) Is the Laplace spectrum of Maass forms for SL(2, Z)\H
simple? In other words, can there be two linearly-independent Maass forms on
SL(2, Z)\H sharing the same eigenvalue?

References
[1] James Arthur, The principle of functoriality, Bull. Amer. Math. Soc. (N.S.) 40 (2002), 39–53
(electronic), Mathematical challenges of the 21st century (Los Angeles, CA, 2000). 1 943 132
[2] James Arthur and Stephen Gelbart, Lectures on automorphic L-functions, L-Functions and
Arithmetic (Durham, 1989), London Math. Soc. Lecture Note Ser., vol. 153, Cambridge Univ.
Press, Cambridge, 1991, pp. 1–59. MR 92j:11045
[3] Joseph Bernstein, Meromorphic Continuation of Eisenstein Series, IAS/Park City Lecture
Notes, Park City, Utah, 2002.
[4] Joseph Bernstein and Stephen Gelbart (eds.), An Introduction to the Langlands Program,
Birkhauser, 2003.
[5] B. J. Birch and H. P. F. Swinnerton-Dyer, Notes on elliptic curves. II , J. Reine Angew.
Math. 218 (1965), 79–108. MR 31 #3419
[6] Andrew Booker, Poles of Artin L-functions and the strong Artin conjecture, Ann. of Math.
(2), to appear.
http://www.math.princeton.edu/~arbooker/papers.
[7] Armand Borel, Automorphic forms on SL2 (R), Cambridge Tracts in Mathematics, vol. 130,
Cambridge University Press, Cambridge, 1997, ISBN 0-521-58049-8. MR 98j:11028
[8] Daniel Bump, Automorphic forms on GL(3, R), Lecture Notes in Mathematics, vol. 1083,
Springer-Verlag, Berlin, 1984, ISBN 3-540-13864-1. MR 86g:11028
[9] , The Rankin-Selberg method: a survey, Number Theory, Trace Formulas and Discrete
Groups (Oslo, 1987), Academic Press, Boston, MA, 1989, pp. 49–109. MR 90m:11079
[10] , Automorphic forms and representations, Cambridge Studies in Advanced Mathe-
matics, vol. 55, Cambridge University Press, Cambridge, 1997, ISBN 0-521-55098-X. MR
97k:11080
[11] Daniel Bump and David Ginzburg, Symmetric square L-functions on GL(r), Ann. of Math.
(2) 136 (1992), 137–205. MR 93i:11058
[12] Christophe Breuil, Brian Conrad, Fred Diamond, and Richard Taylor, On the modularity
of elliptic curves over Q: wild 3-adic exercises, J. Amer. Math. Soc. 14 (2001), 843–939
(electronic). MR 2002d:11058
[13] Farrell Brumley, Maass cusp forms with quadratic integer coefficients, Int. Math. Res. Not.
(2003), 983–997. 1 962 012
[14] Henri Carayol, Preuve de la conjecture de Langlands locale pour GLn : travaux de Harris-
Taylor et Henniart, Astérisque (2000), Exp. No. 857, 4, 191–243, Séminaire Bourbaki, Vol.
1998/99. MR 2001i:11136 (French, with French summary)
[15] Pierre Cartier, Des nombres premiers à la géométrie algébrique (une brève histoire de la
fonction zéta), Analyse diophantienne et géométrie algébrique, Cahiers Sém. Hist. Math.
Sér. 2, vol. 3, Univ. Paris VI, Paris, 1993, pp. 51–77. MR 94i:11060 (French)
[16] P. Cartier, Some numerical computations relating to automorphic functions. (English), Com-
puters in Number Theory, Proc. Atlas Sympos. No.2, 1971, pp. 37–48.
RIEMANN’S ZETA FUNCTION AND BEYOND 47

[17] W. Casselman, GLn , Algebraic Number Fields: L-Functions and Galois Properties (Proc.
Sympos., Univ. Durham, Durham, 1975), Academic Press, London, 1977, pp. 663–704. MR
58 #27777
[18] W. Casselman and J. Shalika, The unramified principal series of p-adic groups. II. The
Whittaker function, Compositio Math. 41 (1980), 207–231. MR 83i:22027
[19] J. Brian Conrey, The Riemann hypothesis, Notices Amer. Math. Soc. 50 (2003), 341–353. 1
954 010
[20] Laurent Clozel, Spectral Theory of Automorphic Forms, IAS/Park City Lecture Notes, Park
City, Utah, 2002.
[21] J. W. Cogdell, L-functions and Converse Theorems for GL(n), IAS/Park City Lecture Notes,
Park City, Utah, 2002,
http://www.math.okstate.edu/~cogdell/.
[22] J. W. Cogdell and I. I. Piatetski-Shapiro, Converse theorems, functoriality, and applica-
tions to number theory, Proceedings of the International Congress of Mathematicians, Vol.
II (Beijing, 2002), Higher Ed. Press, Beijing, 2002, pp. 119–128. 1 957 026
[23] J. W. Cogdell, H. H. Kim, I. I. Piatetski-Shapiro, and F. Shahidi, On lifting from classical
groups to GLN , Publ. Math. Inst. Hautes Études Sci. (2001), 5–30. MR 2002i:11048
[24] , Functoriality for the classical groups, preprint.
[25] J. W. Cogdell and I. I. Piatetski-Shapiro, Converse theorems for GLn , Inst. Hautes Études
Sci. Publ. Math. (1994), 157–214. MR 95m:22009
[26] , A converse theorem for GL4 , Math. Res. Lett. 3 (1996), 67–76. MR 97d:22019
[27] , Converse theorems for GLn . II , J. Reine Angew. Math. 507 (1999), 165–188. MR
2000a:22029
[28] Paul Cohen and Peter Sarnak, Notes on Eisenstein Series, Stanford University, 1980.
[29] J. B. Conrey and D. W. Farmer, An extension of Hecke’s converse theorem, Internat. Math.
Res. Notices (1995), 445–463. MR 96i:11101
[30] Harold Davenport, Multiplicative number theory, 3rd ed., Graduate Texts in Mathematics,
vol. 74, Springer-Verlag, New York, 2000, ISBN 0-387-95097-4, Revised and with a preface
by Hugh L. Montgomery. MR 2001f:11001
[31] Pierre Deligne, La conjecture de Weil. I , Inst. Hautes Études Sci. Publ. Math. (1974), 273–
307. MR 49 #5013 (French)
[32] W. Duke, Hyperbolic distribution problems and half-integral weight Maass forms, Invent.
Math. 92 (1988), 73–90. MR 89d:11033
[33] , Rational points on the sphere, Ramanujan Journal, Rankin Volume, to appear.
http://www.math.ucla.edu/~duke.
[34] W. Duke, J. Friedlander, and H. Iwaniec, Bounds for automorphic L-functions, Invent. Math.
112 (1993), 1–8. MR 94c:11043
[35] W. Duke and H. Iwaniec, Estimates for coefficients of L-functions. I , Automorphic Forms
and Analytic Number Theory (Montreal, PQ, 1989), Univ. Montréal, Montreal, QC, 1990,
pp. 43–47. MR 92f:11068
[36] H. M. Edwards, Riemann’s zeta function, Dover Publications Inc., Mineola, NY, 2001, ISBN
0-486-41740-9, Reprint of the 1974 original [Academic Press, New York; MR 57 #5922]. MR
2002g:11129
[37] Edward Frenkel, Recent advances in the Langlands program, http://arxiv.org/PS_cache/
math/pdf/0303/0303074.pdf.
[38] Wee Teck Gan, Benedict Gross, and Gordan Savin, Fourier coefficients of modular forms on
G2 , Duke Math. J. 115 (2002), 105–169. 1 932 327
[39] Paul B. Garrett, Decomposition of Eisenstein series: Rankin triple products, Ann. of Math.
(2) 125 (1987), 209–235. MR 88m:11033
[40] Stephen S. Gelbart, Automorphic forms on adèle groups, Princeton University Press, Prince-
ton, N.J., 1975, Annals of Mathematics Studies, No. 83. MR 52 #280
[41] Stephen Gelbart and Hervé Jacquet, A relation between automorphic representations of
GL(2) and GL(3), Ann. Sci. École Norm. Sup. (4) 11 (1978), 471–542. MR 81e:10025
[42] Stephen Gelbart, An elementary introduction to the Langlands program, Bull. Amer. Math.
Soc. (N.S.) 10 (1984), 177–219. MR 85e:11094
[43] , Lectures on the Arthur-Selberg trace formula, University Lecture Series, vol. 9, Amer-
ican Mathematical Society, Providence, RI, 1996, ISBN 0-8218-0571-1. MR 98d:22017
48 STEPHEN S. GELBART AND STEPHEN D. MILLER

[44] Stephen Gelbart and Freydoon Shahidi, Analytic properties of automorphic L-functions,
Perspectives in Mathematics, vol. 6, Academic Press Inc., Boston, MA, 1988, ISBN 0-12-
279175-4. MR 89f:11077
[45] , Boundedness of automorphic L-functions in vertical strips, J. Amer. Math. Soc. 14
(2001), 79–107 (electronic). MR 2003a:11056
[46] David Ginzburg and Stephen Rallis, The exterior cube L-function for GL(6), Compositio
Math. 123 (2000), 243–272. MR 2001j:11025
[47] David Ginzburg, Stephen Rallis, and David Soudry, Generic automorphic forms on SO(2n +
1): functorial lift to GL(2n), endoscopy, and base change, Internat. Math. Res. Notices
(2001), 729–764. MR 2002g:11065
[48] Roger Godement, Notes on Jacquet-Langlands’ theory, The Institute for Advanced Study,
1970, (mimeographed notes).
[49] Roger Godement and Hervé Jacquet, Zeta functions of simple algebras, Springer-Verlag,
Berlin, 1972, Lecture Notes in Mathematics, Vol. 260. MR 49 #7241
[50] D. Goldfeld and P. Sarnak, Sums of Kloosterman sums, Invent. Math. 71 (1983), 243–250.
MR 84e:10037
[51] I. S. Gradshteyn and I. M. Ryzhik, Table of integrals, series, and products, 5th ed., Academic
Press Inc., Boston, MA, 1994, ISBN 0-12-294755-X, Translation edited and with a preface by
Alan Jeffrey. MR 94g:00008
[52] Benedict H. Gross, Algebraic modular forms, Israel J. Math. 113 (1999), 61–93. MR
2001b:11037
[53] Benedict H. Gross and Stephen S. Kudla, Heights and the central critical values of triple
product L-functions, Compositio Math. 81 (1992), 143–209. MR 93g:11047
[54] Benedict H. Gross and Don B. Zagier, Heegner points and derivatives of L-series, Invent.
Math. 84 (1986), 225–320. MR 87j:11057
[55] H. Hamburger, Über die Funktionalgleichung der ζ-Funktion, Math. Z., 10 (1921), 240 –258,
11 (1921), 224 –245, 13 (1922), 283 –311.
[56] Harish-Chandra, Automorphic forms on semisimple Lie groups, Notes by J. G. M. Mars.
Lecture Notes in Mathematics, No. 62, Springer-Verlag, Berlin, 1968. MR 38 #1216
[57] Michael Harris, On the local Langlands correspondence, Proceedings of the International
Congress of Mathematicians, Vol. II (Beijing, 2002), Higher Ed. Press, Beijing, 2002, pp. 583–
597. 1 957 067
[58] Michael Harris and Stephen S. Kudla, The central critical value of a triple product L-function,
Ann. of Math. (2) 133 (1991), 605–672. MR 93a:11043
[59] Michael Harris and Richard Taylor, The geometry and cohomology of some simple Shimura
varieties, Annals of Mathematics Studies, vol. 151, Princeton University Press, Princeton, NJ,
2001, ISBN 0-691-09090-4, With an appendix by Vladimir G. Berkovich. MR 2002m:11050
[60] Erich Hecke, Mathematische Werke, 3rd ed., Vandenhoeck & Ruprecht, Göttingen, 1983,
ISBN 3-525-40729-7, With introductory material by B. Schoeneberg, C. L. Siegel and J.
Nielsen. MR 86a:01049 (German)
[61] , Lectures on Dirichlet series, modular functions and quadratic forms, Vandenhoeck
& Ruprecht, Göttingen, 1983, ISBN 3-525-40727-0, Edited by Bruno Schoeneberg; With the
collaboration of Wilhelm Maak. MR 85c:11042
[62] Guy Henniart, Une preuve simple des conjectures de Langlands pour GL(n) sur un corps p-
adique, Invent. Math. 139 (2000), 439–455. MR 2001e:11052 (French, with English summary)
[63] , Sur la conjecture de Langlands locale pour GLn , J. Théor. Nombres Bordeaux 13
(2001), 167–187, 21st Journées Arithmétiques (Rome, 2001). MR 2002f:11178 (French)
[64] G. Henniart, Progrés rècents en fonctorialitè de Langlands, Seminaire Bourbaki Exposé
890 (Juin 2001), 890-1 to 890-21.
[65] Guy Henniart, Une caractérisation de la correspondance de Langlands locale pour GL(n),
Bull. Soc. Math. France 130 (2002), 587–602. 1 947 454 (French)
[66] Tamotsu Ikeda, On the gamma factor of the triple L-function. I , Duke Math. J. 97 (1999),
301–318. MR 2000a:11080
[67] Henryk Iwaniec, Fourier coefficients of modular forms of half-integral weight, Invent. Math.
87 (1987), 385–401. MR 88b:11024
[68] , Topics in classical automorphic forms, Graduate Studies in Mathematics, vol. 17,
American Mathematical Society, Providence, RI, 1997, ISBN 0-8218-0777-3. MR 98e:11051
RIEMANN’S ZETA FUNCTION AND BEYOND 49

[69] , Spectral methods of automorphic forms, 2nd ed., Graduate Studies in Mathematics,
vol. 53, American Mathematical Society, Providence, RI, 2002, ISBN 0-8218-3160-7. 1 942
691
[70] H. Iwaniec and P. Sarnak, Perspectives on the analytic theory of L-functions, Geom. Funct.
Anal. (2000), 705–741, GAFA 2000 (Tel Aviv, 1999). MR 2002b:11117
[71] Hervé Jacquet, Dirichlet series for the group GL(n), Automorphic Forms, Representation
Theory and Arithmetic (Bombay, 1979), Tata Inst. Fund. Res. Studies in Math., vol. 10,
Tata Inst. Fundamental Res., Bombay, 1981, pp. 155–163. MR 83j:10032
[72] H. Jacquet and R. P. Langlands, Automorphic forms on GL(2), Springer-Verlag, Berlin, 1970,
http://sunsite.ubc.ca/DigitalMathArchive/Langlands/JL.html. MR 53 #5481
[73] Hervé Jacquet, Ilja Iosifovitch Piatetski-Shapiro, and Joseph Shalika, Automorphic forms on
GL(3)., Ann. of Math. (2) 109 (1979), 169–258. MR 80i:10034b
[74] H. Jacquet, I. I. Piatetskii-Shapiro, and J. A. Shalika, Rankin-Selberg convolutions, Amer. J.
Math. 105 (1983), 367–464. MR 85g:11044
[75] Hervé Jacquet and Joseph A. Shalika, A non-vanishing theorem for zeta functions of GLn ,
Invent. Math. 38 (1976/77), 1–16. MR 55 #5583
[76] H. Jacquet and J. A. Shalika, On Euler products and the classification of automorphic rep-
resentations. I , Amer. J. Math. 103 (1981), 499–558,
http://www.jstor.org/view/00029327/di994444/99p0113q/0. MR 82m:10050a
[77] Hervé Jacquet and Joseph Shalika, Exterior square L-functions, Automorphic Forms,
Shimura Varieties, and L-Functions, Vol. II (Ann Arbor, MI, 1988), Perspect. Math., vol. 11,
Academic Press, Boston, MA, 1990, pp. 143–226. MR 91g:11050
[78] D. Jiang and D. Soudry, Generic representations and local Langlands reciprocity law for
p-adic SO2n+1 ( 2001), preprint.
[79] N. M. Katz, The work of Pierre Deligne, Proceedings of the International Congress of Math-
ematicians (Helsinki, 1978), Acad. Sci. Fennica, Helsinki, 1980, pp. 47–52. MR 82g:01060
[80] D. Kazhdan, B. Mazur, and C.-G. Schmidt, Relative modular symbols and Rankin-Selberg
convolutions, J. Reine Angew. Math. 519 (2000), 97–141. MR 2001j:11026
[81] Henry H. Kim, Langlands-Shahidi method and poles of automorphic L-functions: application
to exterior square L-functions, Canad. J. Math. 51 (1999), 835–849. MR 2000f:11058
[82] , Functoriality for the exterior square of GL4 and the symmetric fourth of GL2 , J.
Amer. Math. Soc. 16 (2003), 139–183 (electronic),
http://www.ams.org/jourcgi/jour-getitem?pii=S0894034702004101. 1 937 203
[83] , Applications of Langlands’ functorial lift of odd orthogonal groups, Trans. Amer.
Math. Soc. 354 (2002), 2775–2796 (electronic). MR 2003c:22025
[84] Henry H. Kim and Freydoon Shahidi, Symmetric cube L-functions for GL2 are entire, Ann.
of Math. (2) 150 (1999), 645–662. MR 2000k:11065
[85] , Functorial products for GL2 × GL3 and functorial symmetric cube for GL2 , C. R.
Acad. Sci. Paris Sér. I Math. 331 (2000), 599–604. MR 2002i:11049 (English, with English
and French summaries)
[86] , Functorial products for GL2 × GL3 and the symmetric cube for GL2 , Ann. of Math.
(2) 155 (2002), 837–893, With an appendix by Colin J. Bushnell and Guy Henniart. 1 923
967
[87] , Cuspidality of symmetric powers with applications, Duke Math. J. 112 (2002), 177–
197. MR 2003a:11057
[88] A. W. Knapp, Introduction to the Langlands program, Representation Theory and Auto-
morphic Forms (Edinburgh, 1996), Proc. Sympos. Pure Math., vol. 61, Amer. Math. Soc.,
Providence, RI, 1997, pp. 245–302. MR 99d:11123
[89] A. W. Knapp and J. D. Rogawski, Applications of the trace formula, Representation Theory
and Automorphic Forms (Edinburgh, 1996), Proc. Sympos. Pure Math., vol. 61, Amer. Math.
Soc., Providence, RI, 1997, pp. 413–431. MR 98j:11105
[90] V. A. Kolyvagin and D. Yu. Logachëv, Finiteness of the Shafarevich-Tate group and the group
of rational points for some modular abelian varieties, Algebra i Analiz 1 (1989), 171–196.
MR 91c:11032 (Russian)
[91] Tomio Kubota, Elementary theory of Eisenstein series, Kodansha Ltd., Tokyo, 1973. MR
55 #2759
[92] Laurent Lafforgue, Chtoucas de Drinfeld et correspondance de Langlands, Invent. Math. 147
(2002), 1–241. MR 2002m:11039 (French)
50 STEPHEN S. GELBART AND STEPHEN D. MILLER

[93] Serge Lang, Algebraic number theory, 2nd ed., Graduate Texts in Mathematics, vol. 110,
Springer-Verlag, New York, 1994, ISBN 0-387-94225-4. MR 95f:11085
[94] , Complex analysis, 4th ed., Graduate Texts in Mathematics, vol. 103, Springer-Verlag,
New York, 1999, ISBN 0-387-98592-1. MR 99i:30001
[95] R. P. Langlands, Eisenstein series, Algebraic Groups and Discontinuous Subgroups (Proc.
Sympos. Pure Math., Boulder, Colo., 1965), Amer. Math. Soc., Providence, R.I., 1966,
pp. 235–252,
http://sunsite.ubc.ca/DigitalMathArchive/Langlands/automorphic.html#esboulder.
MR 40 #2784
[96] Robert P. Langlands, Euler products, Yale University Press, New Haven, Conn., 1971,
http://sunsite.ubc.ca/DigitalMathArchive/Langlands/automorphic.html#
eulerproducts. MR 54 #7387
[97] , On the functional equations satisfied by Eisenstein series, Springer-Verlag, Berlin,
1976,
http://sunsite.ubc.ca/DigitalMathArchive/Langlands/automorphic.html#esspringer.
MR 58 #28319
[98] R. P. Langlands, Automorphic representations, Shimura varieties, and motives. Ein
Märchen, Automorphic Forms, Representations and L-Functions (Proc. Sympos. Pure Math.,
Oregon State Univ., Corvallis, Ore., 1977), Part 2, Proc. Sympos. Pure Math., XXXIII, Amer.
Math. Soc., Providence, R.I., 1979, pp. 205–246. MR 83f:12010
[99] Robert P. Langlands, Base change for GL(2), Annals of Mathematics Studies, vol. 96,
Princeton University Press, Princeton, N.J., 1980, ISBN 0-691-08263-4, 0-691-08272-3,
http://sunsite.ubc.ca/DigitalMathArchive/Langlands/basechange.html#book. MR
82a:10032
[100] R. P. Langlands, Eisenstein series, the trace formula, and the modern theory of automor-
phic forms, Number Theory, Trace Formulas and Discrete Groups (Oslo, 1987), Academic
Press, Boston, MA, 1989, pp. 125–155,
http://sunsite.ubc.ca/DigitalMathArchive/Langlands/endoscopy.html#oslo. MR
90e:11077
[101] , On the classification of irreducible representations of real algebraic groups, Repre-
sentation Theory and Harmonic Analysis on Semisimple Lie Groups, Math. Surveys Monogr.,
vol. 31, Amer. Math. Soc., Providence, RI, 1989, pp. 101–170. MR 91e:22017
[102] Robert P. Langlands, Where stands functoriality today? , Representation Theory and Au-
tomorphic Forms (Edinburgh, 1996), Proc. Sympos. Pure Math., vol. 61, Amer. Math. Soc.,
Providence, RI, 1997, pp. 457–471. MR 99c:11140
[103] , The trace formula and its applications: an introduction to the work of James
Arthur , Canad. Math. Bull. 44 (2001), 160–209. MR 2002e:22024
[104] Robert Langlands, Beyond endoscopy, Contributions to Automorphic Forms, Geometry
and Number Theory: Shalika Fest 2002 ( Haruzo Hida Dinakar Ramakrishnan Freydoon
Shahidi, eds.), to appear.
[105] Gérard Laumon, La correspondance de Langlands sur les corps de fonctions (d’après Lau-
rent Lafforgue), Astérisque (2002), 207–265, Séminaire Bourbaki, Vol. 1999/2000. MR
2003b:11052 (French)
[106] Jian-Shu Li and Joachim Schwermer, Automorphic representations and cohomology of arith-
metic groups, Challenges for the 21st Century (Singapore, 2000), World Sci. Publishing, River
Edge, NJ, 2001, pp. 102–137. MR 2003c:11051
[107] Alexander Lubotzky, Discrete groups, expanding graphs and invariant measures, Progress
in Mathematics, vol. 125, Birkhäuser Verlag, Basel, 1994, ISBN 3-7643-5075-X, With an
appendix by Jonathan D. Rogawski. MR 96g:22018
[108] A. Lubotzky, R. Phillips, and P. Sarnak, Ramanujan graphs, Combinatorica 8 (1988), 261–
277. MR 89m:05099
[109] W. Luo, Z. Rudnick, and P. Sarnak, On Selberg’s eigenvalue conjecture, Geom. Funct. Anal.
5 (1995), 387–401. MR 96h:11045
[110] Wenzhi Luo, Zeév Rudnick, and Peter Sarnak, On the generalized Ramanujan conjecture
for GL(n), Automorphic Forms, Automorphic Representations, and Arithmetic (Fort Worth,
TX, 1996), Proc. Sympos. Pure Math., vol. 66, Amer. Math. Soc., Providence, RI, 1999,
pp. 301–310. MR 2000e:11072
RIEMANN’S ZETA FUNCTION AND BEYOND 51

[111] Hans Maass, Über eine neue Art von nichtanalytischen automorphen Funktionen und die
Bestimmung Dirichletscher Reihen durch Funktionalgleichungen, Math. Ann. 121 (1949),
141–183. MR 11,163c (German)
[112] G. A. Margulis, Explicit constructions of expanders, Problemy Peredači Informacii 9 (1973),
71–80. MR 58 #4643 (Russian)
[113] , Discrete subgroups of semisimple Lie groups, Ergebnisse der Mathematik und ihrer
Grenzgebiete (3) [Results in Mathematics and Related Areas (3)], vol. 17, Springer-Verlag,
Berlin, 1991, ISBN 3-540-12179-X. MR 92h:22021
[114] Stephen D. Miller, Cusp Forms on SL3 (Z)\SL3 (R)/SO3 (R), Ph.D. dissertation, Princeton
University, 1997
http://www.math.rutgers.edu/~sdmiller/thesis.html/.
[115] Stephen D. Miller and Wilfried Schmid, Summation Formulas, from Poisson and Voronoi
to the Present, Noncommutative Analysis, in Honor of Jacques Carmona, Progress in Math-
ematics, vol. 220, Birkhäuser, 2003, http://www.math.rutgers.edu/~sdmiller/voronoi.
[116] , Distributions and Analytic Continuation of Dirichlet Series,
http://www.math.rutgers.edu/~sdmiller/voronoi.
[117] , Automorphic Distributions, L-functions, and Voronoi Summation for GL(3),
http://www.math.rutgers.edu/~sdmiller/voronoi.
[118] C. Mœglin and J.-L. Waldspurger, Spectral decomposition and Eisenstein series, Cambridge
Tracts in Mathematics, vol. 113, Cambridge University Press, Cambridge, 1995, ISBN 0-521-
41893-3, Une paraphrase de l’Écriture [A paraphrase of Scripture]. MR 97d:11083
[119] L. J. Mordell, On Mr. Ramanujan’s Empirical Expansions of Modular Functions, Proc.
Cambridge Phil. Soc. 19 (1917), 117 –124.
[120] Carlos J. Moreno, Analytic proof of the strong multiplicity one theorem, Amer. J. Math.
107 (1985), 163–206. MR 86m:22027
[121] Goran Muić, A proof of Casselman-Shahidi’s conjecture for quasi-split classical groups,
Canad. Math. Bull. 44 (2001), 298–312. MR 2002f:22015
[122] , Some results on square integrable representations; irreducibility of standard repre-
sentations, Internat. Math. Res. Notices (1998), 705–726. MR 99f:22031
[123] Werner Müller, The trace class conjecture in the theory of automorphic forms, Ann. of
Math. (2) 130 (1989), 473–529. MR 90m:11083
[124] M. Ram Murty, Ramanujan Graphs, Journal of the Ramanujan Mathematical Society, to
appear.
[125] Andrew Ogg, Modular forms and Dirichlet series, W. A. Benjamin, Inc., New York-
Amsterdam, 1969. MR 41 #1648
[126] A. P. Ogg, A remark on the Sato-Tate conjecture, Invent. Math. 9 (1969/1970), 198–200.
MR 41 #3481
[127] R. S. Phillips and P. Sarnak, On cusp forms for co-finite subgroups of PSL(2, R), Invent.
Math. 80 (1985), 339–364. MR 86m:11037
[128] , The Weyl theorem and the deformation of discrete groups, Comm. Pure Appl.
Math. 38 (1985), 853–866. MR 87f:11035
[129] R. Phillips and P. Sarnak, Perturbation theory for the Laplacian on automorphic functions,
J. Amer. Math. Soc. 5 (1992), 1–32. MR 92g:11056
[130] I. I. Pjateckij-Šapiro, On the Weil-Jacquet-Langlands theorem, Lie Groups and Their Rep-
resentations (Proc. Summer School, Bolyai János Math. Soc., Budapest, 1971), Halsted, New
York, 1975, pp. 583–595. MR 53 #10719
[131] I. I. Piatetski-Shapiro, Multiplicity one theorems, Automorphic Forms, Representations and
L-Functions (Proc. Sympos. Pure Math., Oregon State Univ., Corvallis, Ore., 1977), Part 1,
Proc. Sympos. Pure Math., XXXIII, Amer. Math. Soc., Providence, R.I., 1979, pp. 209–212.
MR 81m:22027
[132] S. J. Patterson and I. I. Piatetski-Shapiro, The symmetric-square L-function attached to
a cuspidal automorphic representation of GL3 , Math. Ann. 283 (1989), 551–572. MR
90d:11070
[133] I. Piatetski-Shapiro and Ravi Raghunathan, On Hamburger’s theorem, Lie Groups and Lie
Algebras: E. B. Dynkin’s Seminar, Amer. Math. Soc. Transl. Ser. 2, vol. 169, Amer. Math.
Soc., Providence, RI, 1995, pp. 109–120. MR 97a:11134
[134] I. Piatetski-Shapiro and Stephen Rallis, Rankin triple L functions, Compositio Math. 64
(1987), 31–115. MR 89k:11037
52 STEPHEN S. GELBART AND STEPHEN D. MILLER

[135] I. Piatetski-Shapiro and S. Rallis, A new way to get Euler products, J. Reine Angew. Math.
392 (1988), 110–124. MR 90c:11032
[136] I. Piatetski-Shapiro, S. Rallis, and G. Schiffmann, L functions for the group G2 , Bull. Amer.
Math. Soc. (N.S.) 23 (1990), 389–399. MR 91a:11023
[137] , Rankin-Selberg integrals for the group G2 , Amer. J. Math. 114 (1992), 1269–1315.
MR 93m:22022
[138] Ravi Raghunathan, A converse theorem for Dirichlet series with poles, C. R. Acad. Sci.
Paris Sér. I Math. 327 (1998), 231–235. MR 99i:11034 (English, with English and French
summaries)
[139] Dinakar Ramakrishnan, On the coefficients of cusp forms, Math. Res. Lett. 4 (1997), 295–
307. MR 98e:11064
[140] , Modularity of the Rankin-Selberg L-series, and multiplicity one for SL(2), Ann. of
Math. (2) 152 (2000), 45–111. MR 2001g:11077
[141] , Existence of Ramanujan primes for GL(3), Contributions to Automorphic Forms,
Geometry and Number Theory: Shalika Fest 2002 ( Haruzo Hida Dinakar Ramakrishnan
Freydoon Shahidi, eds.), to appear.
[142] Dinakar Ramakrishnan and Robert J. Valenza, Fourier analysis on number fields, Graduate
Texts in Mathematics, vol. 186, Springer-Verlag, New York, 1999, ISBN 0-387-98436-4. MR
2000d:11002
[143] R. A. Rankin, Contributions to the theory of Ramanujan’s function τ (n) and similar arith-
metical functions. I. The zeros of the function ∞ s
P
n=1 τ (n)/n on the line Rs = 13/2. II. The
order of the Fourier coefficients of integral modular forms, Proc. Cambridge Philos. Soc. 35
(1939), 351–372. MR 1,69d
[144] Michael J. Razar, Modular forms for G0 (N ) and Dirichlet series, Trans. Amer. Math. Soc.
231 (1977), 489–495. MR 56 #2926
[145] B. Riemann, Über die Anzahl der Primzahlen unter einer gegebenen Grösse, Mon. Not.
Berlin Akad (Nov. 1859), 671–680,
See also http://www.maths.tcd.ie/pub/HistMath/People/Riemann/Zeta/.
[146] Jonathan D. Rogawski, Automorphic representations of unitary groups in three variables,
Annals of Mathematics Studies, vol. 123, Princeton University Press, Princeton, NJ, 1990,
ISBN 0-691-08586-2, 0-691-08587-0. MR 91k:22037
[147] , Functoriality and the Artin conjecture, Representation Theory and Automorphic
Forms (Edinburgh, 1996), Proc. Sympos. Pure Math., vol. 61, Amer. Math. Soc., Providence,
RI, 1997, pp. 331–353. MR 98j:11106
[148] Peter Sarnak, Some applications of modular forms, Cambridge Tracts in Mathematics,
vol. 99, Cambridge University Press, Cambridge, 1990, ISBN 0-521-40245-6. MR 92k:11045
[149] P. Sarnak, On cusp forms. II , Festschrift in Honor of I. I. Piatetski-Shapiro on the Occa-
sion of His Sixtieth Birthday, Part II (Ramat Aviv, 1989), Israel Math. Conf. Proc., vol. 3,
Weizmann, Jerusalem, 1990, pp. 237–250. MR 93e:11068
[150] Peter Sarnak, Selberg’s eigenvalue conjecture, Notices Amer. Math. Soc. 42 (1995), 1272–
1277. MR 97c:11059
[151] , Letter to Stephen Gelbart and Freydoon Shahidi (2001).
[152] , Maass cusp forms with integer coefficients, A Panorama of Number Theory or The
View from Baker’s Garden ( G. Wüstholz, ed.), Cambridge University Press, 2002, pp. 121–
128.
[153] , Nonvanishing of L-functions on Re(s)=1 , Contributions to Automorphic Forms,
Geometry and Number Theory: Shalika Fest 2002 ( Haruzo Hida Dinakar Ramakrishnan
Freydoon Shahidi, eds.), to appear.
[154] , Spectra of hyperbolic surfaces, Bull. Amer. Math. Soc. (N.S.), to appear.
http://www.math.princeton.edu/~sarnak.
[155] Atle Selberg, Bemerkungen über eine Dirichletsche Reihe, die mit der Theorie der Modul-
formen nahe verbunden ist, Arch. Math. Naturvid. 43 (1940), 47–50. MR 2,88a (German)
[156] A. Selberg, Harmonic analysis and discontinuous groups in weakly symmetric Riemannian
spaces with applications to Dirichlet series, J. Indian Math. Soc. (N.S.) 20 (1956), 47–87.
MR 19,531g
[157] Atle Selberg, Discontinuous groups and harmonic analysis, Proc. Internat. Congr. Math-
ematicians (Stockholm, 1962), Inst. Mittag-Leffler, Djursholm, 1963, pp. 177–189. MR 31
#372
RIEMANN’S ZETA FUNCTION AND BEYOND 53

[158] , On the estimation of Fourier coefficients of modular forms, Proc. Sympos. Pure
Math., Vol. VIII, Amer. Math. Soc., Providence, R.I., 1965, pp. 1–15. MR 32 #93
[159] Jean-Pierre Serre, Abelian l-adic representations and elliptic curves, McGill University lec-
ture notes written with the collaboration of Willem Kuyk and John Labute, W. A. Benjamin,
Inc., New York-Amsterdam, 1968. MR 41 #8422
[160] J.-P. Serre, A course in arithmetic, Springer-Verlag, New York, 1973, Translated from the
French; Graduate Texts in Mathematics, No. 7. MR 49 #8956
[161] Freydoon Shahidi, On certain L-functions, Amer. J. Math. 103 (1981), 297–355. MR
82i:10030
[162] , Local coefficients as Artin factors for real groups, Duke Math. J. 52 (1985), 973–
1007. MR 87m:11049
[163] , On the Ramanujan conjecture and finiteness of poles for certain L-functions, Ann.
of Math. (2) 127 (1988), 547–584. MR 89h:11021
[164] , A proof of Langlands’ conjecture on Plancherel measures; complementary series
for p-adic groups, Ann. of Math. (2) 132 (1990), 273–330. MR 91m:11095
[165] , Symmetric power L-functions for GL(2), Elliptic Curves and Related Topics, CRM
Proc. Lecture Notes, vol. 4, Amer. Math. Soc., Providence, RI, 1994, pp. 159–182. MR
95c:11066
[166] , Intertwining Operators, L-functions, and Representation Theory, Lecture Notes
of the Elevent KAIST Mathematics Workshop ( Ja Kyung Koo, ed.), 1996, pp. 1 –63,
http://www.math.rutgers.edu/~sdmiller/L-functions/.
[167] , Langlands-Shahidi Method and Converse Theorems, IAS/Park City Lecture Notes,
Park City, Utah, 2002.
[168] , Automorphic L-functions and functoriality, Proceedings of the International Con-
gress of Mathematicians, Vol. II (Beijing, 2002), Higher Ed. Press, Beijing, 2002, pp. 655–666.
1 957 073
[169] J. A. Shalika, The multiplicity one theorem for GLn , Ann. of Math. (2) 100 (1974), 171–193.
MR 50 #545
[170] Goro Shimura, On the holomorphy of certain Dirichlet series, Proc. London Math. Soc. (3)
31 (1975), 79–98. MR 52 #3064
[171] Takuro Shintani, On an explicit formula for class-1 “Whittaker functions” on GLn over
P -adic fields, Proc. Japan Acad. 52 (1976), 180–182. MR 53 #10991
[172] Joseph H. Silverman, The arithmetic of elliptic curves, Graduate Texts in Mathematics,
vol. 106, Springer-Verlag, New York, 199?, ISBN 0-387-96203-4, Corrected reprint of the
1986 original. MR 95m:11054
[173] J. T. Tate, Fourier analysis in number fields, and Hecke’s zeta-functions, Algebraic Number
Theory (Proc. Instructional Conf., Brighton, 1965), Thompson, Washington, D.C., 1967,
pp. 305–347. MR 36 #121
[174] John T. Tate, Algebraic cycles and poles of zeta functions, Arithmetical Algebraic Geometry
(Proc. Conf. Purdue Univ., 1963), Harper & Row, New York, 1965, pp. 93–110. MR 37 #1371
[175] Richard Taylor and Andrew Wiles, Ring-theoretic properties of certain Hecke algebras, Ann.
of Math. (2) 141 (1995), 553–572. MR 96d:11072
[176] E. C. Titchmarsh, The theory of the Riemann zeta-function, 2nd ed., The Clarendon Press
Oxford University Press, New York, 1986, ISBN 0-19-853369-1, Edited and with a preface by
D. R. Heath-Brown. MR 88c:11049
[177] Jerrold Tunnell, Artin’s conjecture for representations of octahedral type, Bull. Amer. Math.
Soc. (N.S.) 5 (1981), 173–175. MR 82j:12015
[178] André Weil, Über die Bestimmung Dirichletscher Reihen durch Funktionalgleichungen,
Math. Ann. 168 (1967), 149–156. MR 34 #7473 (German)
[179] , Prehistory of the zeta-function, Number Theory, Trace Formulas and Discrete
Groups (Oslo, 1987), Academic Press, Boston, MA, 1989, pp. 1–9. MR 90e:01029
[180] , On Eisenstein’s copy of the Disquisitiones, Algebraic Number Theory, Adv. Stud.
Pure Math., vol. 17, Academic Press, Boston, MA, 1989, pp. 463–469. MR 91m:01016
[181] Andrew Wiles, Modular elliptic curves and Fermat’s last theorem, Ann. of Math. (2) 141
(1995), 443–551. MR 96d:11071
[182] Scott A. Wolpert, Spectral limits for hyperbolic surfaces. I, II , Invent. Math. 108 (1992),
67–89, 91–129. MR 93b:58160
54 STEPHEN S. GELBART AND STEPHEN D. MILLER

[183] , Disappearance of cusp forms in special families, Ann. of Math. (2) 139 (1994),
239–291. MR 95e:11062

Faculty of Mathematics and Computer Science, Nicki and J. Ira Harris Professorial
Chair, The Weizmann Institute of Science, Rehovot 76100, Israel
E-mail address: gelbar@wisdom.weizmann.ac.il

Department of Mathematics, Hill Center-Busch Campus, Rutgers University, 110


Frelinghuysen Rd, Piscataway, NJ 08854-8109
E-mail address: miller@math.rutgers.edu

You might also like