Proceedings of Symposia in Pure Mathematics

Volume 61 (1997), pp. 245–302

Introduction to the Langlands Program

A. W. Knapp

This article is an introduction to automorphic forms on the adeles of a linear

reductive group over a number field. The first half is a summary of aspects of local
and global class field theory, with emphasis on the local Weil group, the L functions
of Artin and Hecke, and the role of Artin reciprocity in relating the two kinds of L
functions. The first half serves as background for the second half, which discusses
some structure theory for reductive groups, the definitions of automorphic and cusp
forms, the Langlands L group, L functions, functoriality, and some conjectures.
Much of the material in the second half may be regarded as a brief introduction to
the Langlands program. There are ten sections:
1. Local Fields and Their Weil Groups
2. Local Class Field Theory
3. Adeles and Ideles
4. Artin Reciprocity
5. Artin L Functions
6. Linear Reductive Algebraic Groups
7. Automorphic Forms
8. Langlands Theory for GLn
9. L Groups and General Langlands L Functions
10. Functoriality

1. Local Fields and Their Weil Groups

This section contains a summary of information about local fields and their Weil
groups. Four general references for this material are [Frö], [La], [Ta3], and [We4].
By a local field is meant any nondiscrete locally compact topological field. Let
F be a local field. If α is a nonzero element of F , then multiplication by α is an
automorphism of the additive group of F and hence carries additive Haar measure
to a multiple of itself. This multiple is denoted |α|F , and it satisfies




−1
h(α x) dx = h(x) d(αx) = |α|F h(x) dx,
F F F

1991 Mathematics Subject Classification. Primary 11F70, 11R39, 11S37, 22E55.

This article is based partly on lectures by Don Blasius in Edinburgh. The author is grateful to
Jonathan Rogawski and David Vogan for offering a number of suggestions about the exposition,
to Dinakar Ramakrishnan for supplying a proof of Theorem 8.8 and giving permission to include
it here, and to Hervé Jacquet for answering many questions.

1997
c American Mathematical Society
0082-0717/97 $1.00 + $.25 per page

245
246 A. W. KNAPP

where dx is an additive Haar measure. We refer to |α|F as the module of α.

By convention, |0|F = 0. The function α → |α|F is continuous on F and is a
homomorphism of the multiplicative group F × into the multiplicative group R× + of
positive reals.
The local field F is said to be nonarchimedean if |α|F satisfies the ultrametric
inequality
|α + β| ≤ max{|α|, |β|}.
Otherwise F is archimedean.
A classification of local fields appears in [We4]. There are only two archimedean
local fields, R and C. In the nonarchimedean case, the set of nonzero values of
| · |F is a discrete subgroup of R× + . The nonarchimedean local fields divide into two
kinds. Those of characteristic 0 turn out to be the p-adic fields, namely the finite
extensions of the fields Qp of p-adic numbers for each prime number p. Those of
characteristic not 0 turn out to be the fields of Laurent series in one variable (finite
in negative powers) over the various finite fields. In this article we consider only
local fields of characteristic 0.
Let us review various constructions of Qp . One way to define Qp is as the
completion of Q in the metric d(x, y) = |x − y|p , where |apn /b|p = p−n if a and
b are integers prime to p. The metric extends to Qp , giving it a locally compact
topology, and the field operations extend as well. It is easy to see that |α|Qp , which
we abbreviate as |α|p , is just the distance between α and 0 in the completed metric.
Another construction begins with the definition of the maximal compact ring Zp
of p-adic integers, which is taken to be the inverse limit
Zp = lim
←
Z/(pn ).
n

The ring Zp is an integral domain having a unique maximal ideal, namely pZp . The
ideals pn Zp form a neighborhood basis of 0, and {α +pn Zp } is a neighborhood basis
about α ∈ Zp . Then Qp may be  described algebraically as the field of quotients of
Zp or as Zp ⊗ Q or as Z[p−1 ] = n≥1 p−n Zp . The last of these descriptions provides
the topology; each p−n Zp is to be open and homeomorphic with Zp . The inclusion
Zp ⊂ Qp makes Zp a compact subring of the local field Qp . Here Zp is precisely the
set of all α ∈ Qp with |α|p ≤ 1, and the maximal ideal pZp is the set of all α ∈ Qp
with |α|p < 1.
Let K be a finite extension of Qp with [K : Qp ] = n. If K is decomposed
as a direct sum of n one-dimensional Qp vector spaces, then the decomposition
automatically respects the topology. The set

OK = {α ∈ K  |α|K ≤ 1}
is a compact open subring of K called the ring of integers of K. It is the unique
maximal compact subring of K and is equal to the integral closure of Zp in K. The
×
group of units OK of OK is the set of elements α with |α|K = 1. The ring OK has
the following properties:
1) OK has a unique maximal ideal mK , namely the set of elements α with
|α|K < 1. The ideal mK is principal, having any element of maximal module
as generator. Such an element will typically be denoted K and is a prime
element. Every nonzero ideal of OK is principal and is a power of mK (with
the corresponding power of K as generator).
2) OK is the inverse limit of OK /(K
n
) on n.
 INTRODUCTION TO THE LANGLANDS PROGRAM 247

3) kK = OK /mK is a finite field kK , the residue field. The inclusion of Zp into

OK induces a map of Zp /pZp into kK , and thus Zp /pZp may be regarded as
the prime field of kK . We write the number of elements in kK as q = pf and
call f the residue degree of K over Qp . The module of K is q −1 .
4) By (1), the ideal pOK of OK is of the form meK = (K e
) for some integer e;
e is called the ramification degree of K over Qp .
5) [K : Qp ] = ef since p−n = |p|K = |K |eK = q −e .
Now let K ⊃ F be two finite extensions of Qp , and write qK , fK , eK and
qF , fF , eF for the respective integers q, f , e. Also let kK and kF be the residue
fields. Then we have an inclusion kF ⊂ kK . We define f = [kK : kF ], and we let
e be the integer such that F OK = meK . The integers f and e are the residue
degree and ramification degree of K over F . Then
f = fK /fF , e = eK /eF , and [K : F ] = ef.
Fix finite extensions K ⊃ F of Qp , and suppose that K/F is Galois. Any F
automorphism of K is module-preserving and hence maps OK to itself and mK to
itself. It therefore induces a automorphism of the quotient, which is the residue
field kK , and this automorphism fixes the residue field kF of F . The result is
a homomorphism Gal(K/F ) → Gal(kK /kF ). Since Gal(kK /kF ) is cyclic with
generator x → xqF , the image of a member of Gal(K/F ) is necessarily of the
n
form x → xqF for some integer n.
Theorem 1.1. If K/F is a Galois extension of finite extensions of Qp and if
kK and kF are the respective residue fields, then Gal(K/F ) maps onto Gal(kK /kF ).
Reference. See [La, p. 15]. For a formulation of this result without the as-
sumption that K/F is Galois, see [Frö, p. 26].

In the setting of Theorem 1.1, we obtain an exact sequence

1 −−−→ IK/F −−−→ Gal(K/F ) −−−→ Gal(kK /kF ) −−−→ 0, (1.1)
where IK/F is the kernel, which is called the inertia group of K over F . We shall
be interested in the effect on (1.1) of letting K swell to F .
For the moment let us drop the assumption that K/F is Galois. Let f and e be
the residue and ramification degrees of K over F . We say that the extension K/F
is unramified if e = 1. In the Galois case, the group Gal(kK /kF ) has order f , and
the exactness of (1.1) implies that IK/F has order e; thus K/F is unramified if and
only if Gal(K/F ) → Gal(kK /kF ) is an isomorphism.
Theorem 1.2. Let F be a finite extension of Qp , and let k be a finite extension
of the residue field kF . Then there exists an unramified finite extension K of F
with residue field kK ∼= k. Such a field K is unique up to F isomorphism and is
Galois over F .
Reference. [Frö, p. 26] or [Se2, p. 54].

Consequently for each f ≥ 1, there is, up to F isomorphism, a unique unram-

ified extension K = Kf of F of degree f , and K/F is Galois. By Theorem 1.1,
Gal(K/F ) ∼ = Gal(kK /kF ) is a cyclic group of order q = pf whose generator Fr is
the lift of the generator x → x|kF | of Gal(kK /kF ). The element Fr of Gal(K/F ) is
248 A. W. KNAPP

called the Frobenius element and is characterized among members of Gal(K/F )

by the congruence Fr(x) ≡ x|kF | mod mK .
Fix an algebraic closure F of F , and regard each K = Kf as contained in
F . When two residue degrees f and f  have the property that f divides f  , the
multiplicative property of residue degrees and ramification degrees implies that
Kf ⊂ Kf
. The result, as f varies, is a directed system of subfields of F . The
union of these subfields is called the maximal unramified extension of F and
is denoted Fur .
Every field map over F of a subfield of F into F extends to an automorphism
of F , and it follows that every member of Gal(Fur /F ) extends to a member of
Gal(F /F ). In other words, there is an exact sequence

1 −−−→ IF −−−→ Gal(F /F ) −−−→ Gal(Fur /F ) −−−→ 0 (1.2)

in which IF is the kernel of π. The group IF is called the inertia group of F (or
of F over F ).
Now let us take the inverse limit of (1.1), letting K swell to F . Then we obtain
a homomorphism of Gal(F /F ) into Gal(k F /kF ), where k F is the algebraic closure
of kF . Then we have the following result.
Theorem 1.3. Every finite extension of F in Fur is unramified. The natural
homomorphism of Gal(F /F ) into Gal(k F /kF ) descends via (1.2) to a topological
isomorphism of Gal(Fur /F ) onto Gal(k F /kF ).
Reference. [Frö, p. 28] or [Se2, p. 54–55].

Let us identify Gal(k F /kF ), so that we can make (1.2) more explicit. The field
k F is the union of its finite subfields. These form a directed system under divisibility
of degree, and the degree can be any positive integer. Therefore Gal(k F /kF ) is the
inverse limit of cyclic groups Z/nZ, indexed according to divisibility of the indices
n. The resulting compact group is denoted Z  and is isomorphic to  Zp . We
p
regard the Frobenius element x → x|kF | in Gal(k F /kF ) as the integer +1, and the
subgroup Z of Z  generated by +1 is dense in Z.  The isomorphism of Theorem
∼ 
1.3 tells us that Gal(Fur /F ) = Z, and we let Fr be the (Frobenius) element of
Gal(Fur /F ) that corresponds to +1 in Z.  The rewritten form of (1.2) is then

π  −−−→ 0
1 −−−→ IF −−−→ Gal(F /F ) −−−→ Z (1.3)

where π is defined to be restriction from F to Fur composed with the isomorphism


to Z.
The Langlands theory of L groups and L functions makes extensive use of the
Weil group of a local field. Let us define this group, sketch some of its properties,
and formulate the theorems of local class field theory in terms of it. A reference for
Weil groups is [Ta3].
We continue with F as a finite extension of Qp and with F as an algebraic
closure of F ; the archimedean case is postponed to the next section. Let GF =
Gal(F /F ). We write Z = Gal(Fur /F ), and we let Z be the infinite cyclic subgroup

of Z generated by the Frobenius element Fr. With π as in (1.3), the Weil group
of F is defined as an abstract group by WF = π −1 (Z) ⊂ Gal(F /F ). The Weil form
 INTRODUCTION TO THE LANGLANDS PROGRAM 249

of the exact sequence (1.3) is then

π
1 −−−→ IF −−−→ WF −−−→ Z −−−→ 0. (1.4)

The relative topology from (1.3) gives Z an unusual topology, but we change matters
to give Z the discrete topology. Correspondingly we retopologize WF so that π is
continuous and IF is homeomorphic with π −1 ({0}).
To understand π and WF better, let K/F be a finite Galois extension. Then
GF maps onto Gal(K/F ) with kernel GK , and in turn Gal(K/F ) maps onto
Gal(kK /kF ) by Theorem 1.1. The effect is to associate to any member σ of GF
n
an integer n and the automorphism x → xqF , with n depending on K and defined
modulo the residue degree of K/F . The inverse limit of the resulting tuple of n’s,
as K varies, defines the member π(σ) of Z.  For π(σ) to be in Z, the condition is
that the inverse limit can be regarded as a single integer n. That is, the members of
the Weil group WF are those members of GF that induce on the algebraic closure
n
k F the automorphism x → xqF for some integer n.
If E/F is a finite extension, then Galois theory says that GE is the subgroup
of GF fixing E. Hence WE is a subgroup of GF . The next theorem identifies this
subgroup.
Theorem 1.4. If E/F is a finite extension, then WE = GE ∩ WF .
Proof. We may regard kE as containing kF with k E = k F . The subgroup
GE ∩ WF consists of the members of GF that induce on k F the automorphism
n n
x → xqF for some integer n and that fix kE . Then xqF = x for all x ∈ kE , and it
[k :k ]a
follows that [kE : kF ] divides n. If we let a = n/[kE : kF ], then qFn = qF E F = qE
a
.
Hence GE ∩ WF ⊂ WE . The reverse inclusion is trivial, and the theorem follows.

 it follows that WF is dense in GF . If E/F is a finite

Since Z is dense in Z,
extension, then GF /GE is a finite set, and the image of WF in it is dense. Therefore
WF maps onto GF /GE . Because of Theorem 1.4, we obtain a bijection
∼
WF /WE → GF /GE ∼
= HomF (E, F ), (1.5)

where HomF (E, F ) is the set of indicated field maps. If E/F is Galois, then
HomF (E, F ) = Gal(E/F ) and (1.5) is a group isomorphism.
Let us mention an alternate definition of WF . If G is a topological group, we let
Gc be the closure of the commutator subgroup and we define Gab = G/Gc . The
closed subgroup GcF of GF = Gal(F /F ) corresponds to a subfield F ab of F called
the maximal abelian extension of F . Its Galois group Gal(F ab /F ) is just Gab F .
Let K be a finite Galois extension of F lying in F , and form K ab . Since
 is abelian, we have K ab ⊃ F ab ⊃ Fur . Therefore we have surjective
Gal(Fur /F ) = Z
maps induced by restriction:

Gal(F /F ) → Gal(K ab /F ) → Gal(Fur /F ) = Z. (1.6)

The inverse image of Z ⊂ Z  in Gal(F /F ) was defined to be WF , and we define

the intermediate inverse image of Z in Gal(K ab /F ) to be WK/F . This construction
carries with it surjective maps WF → WK/F , and these are compatible as K varies.
In addition, any element of F is in some finite Galois extension K of F and therefore
250 A. W. KNAPP

is also in K ab . Hence no nontrivial element of WF can restrict to the identity on

every WK/F , and it follows that WF is the inverse limit
WF = lim WK/F . (1.7)
K
←

2. Local Class Field Theory

In this section we shall state the main results from local class field theory and
translate them into statements about Weil groups. For most of this section, we let
F be a finite extension of Qp and F be an algebraic closure of F . References for
the material in this section are [Ne], [Se1], [Se2], and [Ta3].
Let K be a finite Galois extension of F lying in F and having [K : F ] = n, and
let GK/F = Gal(K/F ). It is known that H 2 (GK/F , K × ) is cyclic of order n with
a canonical generator uK/F . This can be proved rather quickly with the aid of the
theory of the Brauer group [Se1, p. 137], and also a direct cohomological proof is
possible [Se1, p. 130].
Theorem 2.1. If K/F is a finite Galois extension, then the canonical generator
uK/F of H 2 (GK/F , K × ) defines (by means of “cup product”) an isomorphism of
× ×
K/F onto F /NK/F (K ), where NK/F ( · ) denotes the norm map.
Gab
Reference. [Se1, p. 140].

The inverse θK/F : F × /NK/F (K × ) → Gab

K/F of the isomorphism in Theorem 2.1
is called the local reciprocity map of K/F . When GK/F is abelian, then Gab K/F
equals GK/F and θK/F is an isomorphism of F × /NK/F (K × ) onto GK/F .
If x ∈ F × lies in the coset x̄ of F × /NK/F (K × ), then we write θK/F (x̄) =
(x, K/F ).
The symbols (x, K/F ) define homomorphisms F × → GK/F that are compatible
[Se1, p. 140] when K  ⊃ K ⊃ F and K  /F is finite abelian. Taking the inverse
limit, we obtain a homomorphism θF : F × → GF ab /F or θF : F × → Gab F .
In the exact sequence (1.3), the homomorphism π has an abelian image and
therefore descends to a homomorphism π ab defined on Gab F = GF ab /F . Accordingly
we replace (1.3) by
ab
1 −−−→ IF ab /F −−−→ Gab
π 
F −−−→ Z −−−→ 0, (2.1)

where Z = GF /F and where IF ab /F is the kernel of π ab . By Galois theory we may

ur
interpret IF ab /F as GF ab /Fur .
We shall compute π ab ◦ θF : F × → Z. For any x ∈ F × , the number |x|F is a
−1
power of q , and we define v(x) to be that power.
Theorem 2.2. If K/F is a finite unramified extension and if Fr in GK/F
denotes the Frobenius element, then (x, K/F ) = Frv(x) for all x ∈ F × .
Reference. [Se1, p. 141].


Corollary 2.3. For any x ∈ F × , π ab (θF (x)) = v(x) as a member of Z.
Any inverse image in GF of the element θF (x) of Gab
F therefore lies in the Weil
group WF .
 INTRODUCTION TO THE LANGLANDS PROGRAM 251

We shall prove below in Lemma 2 that the homomorphism WFab → Gab F induced
by WF → GF is one-one, and then we may regard (π ab )−1 (Z) as WFab . Thus (2.1)
gives us an exact sequence
π ab
1 −−−→ GF ab /Fur −−−→ WFab −−−→ Z −−−→ 0. (2.2)

Lemma 1. WFc = GcF .

Proof. Certainly WFc ⊂ GcF , and it is GcF ⊂ WFc that needs proof. Let x
be a member of WF with π(x) = 1 ∈ Z. Then WF is the semidirect product of
{xn }∞ c
n=−∞ with IF . So WF is the closure of the subgroup of IF generated by all
commutators of IF and all elements xn ix−n i−1 with i ∈ IF .
Let Y be the smallest closed subgroup of GF containing x. Since Y is compact
abelian and Z is dense in Z, π(Y ) = Z. If g ∈ GF is given, choose y ∈ Y with
π(g) = π(y). Then π(gy −1 ) = 0 shows that gy −1 is in IF , and the identity g =
(gy −1 )y therefore shows that every element of GF is the product of an element of
IF and an element of Y . So GcF is the closure of the subgroup of IF generated
by all commutators of IF and all elements yiy −1 i−1 with y ∈ Y and i ∈ IF . The
commutators of IF are in WFc , and the element yiy −1 i−1 is the limit of elements
xn ix−n i−1 with n varying through a suitable sequence. Hence yiy −1 i−1 is in WFc ,
and GcF ⊂ WFc .

Lemma 2. The homomorphism WFab → Gab

F induced by WF → GF is one-one.

Proof. We need to prove that WF ∩ GcF = WFc . Since GcF ⊂ IF ⊂ WF , we

need GcF = WFc . But this is just what Lemma 1 gives.

With Lemma 2 proved, (2.2) now follows. Corollary 2.3 implies that the restric-
tion of θF to OF× is in the kernel of π ab on WF . Hence it is in the image of GF ab /Fur .
ab

We can put this information and the full strength of Corollary 2.3 together in a
diagram with exact rows and commutative squares

OF×
v
1 −−−−→ −−−−→ F × −−−−→ Z −−−−→ 0
  
θ θ 
F F 1 (2.3)
ab
π
1 −−−−→ GF ab /Fur −−−−→ WFab −−−−→ Z −−−−→ 0.

Let us now state in its classical form the Existence Theorem of local class field
theory.
Theorem 2.4. The map K → F × /NK/F (K × ) is a bijection of the set of finite
abelian Galois extensions K of F onto the set of open subgroups of F × of finite
index.
Reference. [Se1, p. 143].

Use of the Weil group allows us to restate this result more simply.
Corollary 2.5. The local reciprocity map θF is a topological isomorphism of
F × onto WFab .
252 A. W. KNAPP

Proof. If K/F is a finite abelian extension within F ab , then Theorem 2.1 shows
×
that the composition of θF followed by the quotient map Gab F → GK/F carries F
×
onto GK/F . Letting K vary, we see that θF carries F onto a dense subgroup of
Gab . Since OF× is compact, it follows from (2.3) that θF carries OF× onto GF ab /Fur .
A second
 application of (2.3) shows that θF carries F × onto WFab . The kernel of
×
θF is K NK/F (K ), the intersection being taken over all finite extensions K of F
lying in F ab . If i and j are any integers ≥ 0, then the set {(Fi )n (1 + OFj ) | n ∈ Z}
is an open subgroup of F × of finite index. By Theorem 2.4 it is NK/F (K × ) for
some finite abelian K/F . Hence

NK/F (K × ) ⊂ {(Fi )n (1 + OFj ) | n ∈ Z} = {1},
K i,j

and θF is one-one. Consequently θF : F × → WFab is a group isomorphism. Since

θF : OF× → GF ab /Fur is continuous and OF× is compact, θF : OF× → GF ab /Fur is a
homeomorphism. Then it follows that θF : F × → WFab is a homeomorphism.

Remarks.
1) In Corollary 2.5, θF carries a prime element of OF to an element of WFab that
acts as a Frobenius automorphism in every unramified extension. Thus θF (x)(y) =
−1
y |x|F for y in any unramified extension. Some authors adjust a sign somewhere
to make θF (x)(y) = y |x|F ; see [Ta3, p. 6] for a discussion of this point. For these
authors the later definitions of L functions are likely to be what we, with our
traditional definitions, would call the L function of the contragredient.
2) Theorem 2.4, which is the difficult result in local class field theory, is essentially
equivalent with Corollary 2.5. A proof that Corollary 2.5 implies Theorem 2.4 may
be based on [Se1, p. 144].
3) Corollary 2.5 implies that the (continuous) one-dimensional representations
of F × are parametrized by the continuous homomorphisms of WF into C× . This is
a point of departure for conjectures of Langlands about parametrizing irreducible
representations of linear reductive groups over F . We return to this matter in §8.

Let us return to the group WK/F defined from (1.6), where K is a finite Galois
extension of F lying in F . The kernel of the map WF → WK/F in (1.7), in view
of (1.6), is the set of all w ∈ WF that act as Galois elements by 1 on K ab . They
are in particular members of GF . Being 1 on K, they are in GK . Being 1 on K ab ,
they are in GcK , which equals WK c
by Lemma 1. As a result we have

WK/F ∼ c
= WF /WK . (2.4)

We form the exact sequence

1 −−−→ WK /WK
c
−−−→ WF /WK
c
−−−→ WF /WK −−−→ 0. (2.5)
c
The quotient WK /WK is just WK/K by (2.4), and this is by definition the subgroup
ab
of GK inducing an integral power of the Frobenius. By Lemma 2 and the derivation
ab
of (2.2), we can identify this subgroup with WK . Applying Corollary 2.5 and
substituting into (2.5) from (1.5) amd (2.4), we obtain an exact sequence

1 −−−→ K × −−−→ WK/F −−−→ GK/F −−−→ 1. (2.6)

 INTRODUCTION TO THE LANGLANDS PROGRAM 253

Such an exact sequence yields by standard cohomology of groups a member of

H 2 (GK/F , K × ). Tracking down the isomorphisms that led to (2.6) allows one to
identify this cohomology element.
Theorem 2.6. The cohomology class of the exact sequence (2.6) in the group
H 2 (GK/F , K × ) is exactly the canonical generator uK/F .
Reference. [Ta3, pp. 4–5].

Theorem 2.6 allows a fairly explicit understanding of WK/F for nonarchimedean

local fields.
For the archimedean local fields R and C, we turn Theorem 2.6 around and use
it as a definition of the Weil group. In the case of R, H 2 (GC/R , C× ) is cyclic of
order 2, and WC/R is defined correspondingly to be a group that fits into a nonsplit
exact sequence
1 −−−→ C× −−−→ WC/R −−−→ GC/R −−−→ 1.
Specifically we take WC/R = C× ∪ jC× , where j acts on C× by complex conjugation
and where j 2 = −1 ∈ C× . In view of (1.7), we make the definition
WR = WC/R = C× ∪ jC× .
Similarly we are led to define WC = WC/C = C× . With these definitions we readily
check that Theorem 2.1 and Corollary 2.5 remain valid for R and C.

3. Adeles and Ideles

Adeles occur in the study of “global fields,” which are of two kinds. The global
fields of characteristic 0 are the number fields, the finite extensions of Q. The
global fields of characteristic nonzero are the finite extensions of the formal rational
functions over a finite field. We shall limit our discussion to number fields. General
references for adeles are [Cas], [La], and [We4].
The idea with adeles is to study number-theoretic questions about a number field
by first studying congruences. For example, to study the factorization of a monic
polynomial with integer coefficients, we first study the factorization modulo each
prime. In addition, we consider any limitations imposed by treating the polynomial
as having real coefficients. Thus, in the case of Q, we use a structure that incor-
porates congruences modulo each prime (as well as powers of the prime), together
with information about R. The ring of adeles AQ is the structure in question. We
defer to §5 the way it carries information about factorization of polynomials.
In the case of Q, let P = {∞} ∪ {primes}. For v ∈ P , the field Qv is to be the
field of p-adic numbers if v is a prime p, and it is to be R if v = ∞. Let S ⊂ P be
a finite set containing ∞, and define

 
 
AQ (S) = Qv × Zv .
v∈S v ∈S
/

With the product topology, AQ (S) is a locally compact commutative topological

ring. If S1 ⊂ S2 , then AQ (S1 ) ⊂ AQ (S2 ). The directed system of inclusions allows
us to define AQ as the direct limit

AQ = lim
→
A Q (S) = AQ (S).
S S
254 A. W. KNAPP

The direct-limit topology makes each AQ (S) be open in AQ , the relative topology
being the locally compact topology above. Then AQ is a locally compact commu-
tative topological ring known as the adeles of Q.
Elements of AQ may be regarded as tuples

x = (x∞ , x2 , x3 , x5 , . . . , xv , . . . ) = xv
v

with almost all (i.e., all but finitely many) xv having |xv |v ≤ 1. (Here | · |∞
denotes the module for the local field R, which is just the usual absolute value.)
Often one writes simply x = (xv ).
The adeles are the result of a construction called restricted direct product. Sup-
pose that I is a nonempty index set, that Xi is a locally compact Hausdorff space
for each i ∈ I, and that a compact open subset Ki of Xi is specified for all i outside
a finite subset S∞ of I. If S is any finite subset of I containing S∞ , we can define

 
 
A(S) = Xi × Ki , (3.1)
i∈S i∈S
/

and A(S) will be locally compact Hausdorff. The direct limit A of the A(S)’s as
S increases is called the restricted direct product of the Xi relative to the Ki .
Thespace A is locally compact Hausdorff,  and each A(S) is open in it. An element
of i∈I xi of the Cartesian product i∈I Xi is in A if and only if xi is in Ki for
almost all i.
In practice, Xi is usually a locally compact group and Ki is a compact open
subgroup. Then A is a locally compact group. In the case of AQ , the finite set S∞
is {∞}, each Xi is a locally compact ring (namely Qp or R), and Ki is a compact
open subring (namely Zp ); thus AQ is a locally compact ring.
For a general number field F (possibly Q itself), we construct the ring AF of
adeles of F as follows. A completion of F is a pair (λ, K), where K is a local
field and λ : F → K is a field map with dense image. Two completions (λ, K) and
(λ , K  ) are equivalent is there is a topological isomorphism ρ : K → K  such that
ρ ◦ λ = λ . A place is an equivalence class of completions. Places are typically
denoted v, and a representative of the corresponding local field is denoted Fv . When
F = Q, the only places are those coming from embedding Q in R and in each Qp
for p prime.
An isomorphism that exhibits two completions as corresponding to the same
place preserves the module. Consequently restriction of the module to F gives a
well defined function | · |v on F .
Suppose that F  /F is an extension of number fields, and let w be a place of F  .
Regard w as a field map w : F  → Fw . It is not hard to see that the closure w(F )
of w(F ) is a local field, that Fw is a finite algebraic extension of w(F ), and that
the restriction of w to F determines a place v of F . In this case we say that v is
the place of F that lies below w, and that w lies above v. We write w | v.
Theorem 3.1. Let F  /F be an extension of number fields, and let v be a place
of k. Then there exists a place of F  lying above v, and there are only finitely many
such places.
Reference. [We4, p. 45].
 INTRODUCTION TO THE LANGLANDS PROGRAM 255

We can apply Theorem 3.1 to construct AF . The Xi are the various Fv . All
of these are p-adic fields except those with v lying above the place ∞ of Q, and
Theorem 3.1 says that there are only finitely many such places. We take S∞ to be
the set of places lying above ∞, and we let the Ki ’s be the rings of integers Ov in
Fv . Then AF is the restricted direct product of the Fv relative to the Kv , and it is
a locally compact commutative ring.
We can get some insight into the places of a number field F by treating F
as an extension of Q and considering all places lying over a place of Q (a prime
or ∞). By the theory of semisimple algebras, the algebra F ⊗ Qv over Qv is
a finite direct sum of fields, each of which is a finite extension of Qv . Fix an
algebraic closure Qv , and consider the set of field maps HomQ (F, Qv ). The group
Dv = Gal(Qv /Qv ) acts on this set of maps by acting on the values of each map.
Let Pv (F ) = Dv \HomQ (F, Qv ). Then

F ⊗Q Qv ∼ = Fw ,
w∈Pv (F )

with the right side involving each place lying over v just once.
Let us consider this decomposition for the infinite places, those lying above ∞
of Q. We shall write F∞ for the algebra F ⊗Q R. This R algebra is of the form
Rr1 × Cr2 for some integers r1 ≥ 0 and r2 ≥ 0 satisfying r1 + 2r2 = [F : Q]. Since
D∞ = Gal(C/R) consists of 1 and complex conjugation, P∞ (F ) consists of the set
HomQ (F, C) of embeddings of F into C, with two embeddings identified when they
are complex conjugates of one another. Thus r1 is the number of embeddings into
R, and r2 is the number of complex-conjugate pairs of nonreal embeddings of F
into C.
Next let us consider the finite places. Recall that the ring of integers O in F
consists of all elements satisfying a monic polynomial equation with Z coefficients.
The places lying over the primes of Q are related to the nontrivial prime ideals
of O. Let v be a place of F lying above a prime p for Q. It is easy to see that
the mapping F → Fv carries O into the ring of integers Ov of Fv . Let mv be the
maximal ideal of Ov , and let Pv be the inverse image in O of mv . Then Pv is a
prime ideal of O. That is, every finite place of F leads to a prime ideal of O.
This fact admits a converse. Before stating the converse, we recall that the
nonzero ideals of O admit unique factorizations as products of prime ideals, the
exponents of the prime ideals being integers ≥ 0. The notion of ideal can be
extended to fractional ideal; a fractional ideal is just a set of the form n−1 I
for an ideal I and some nonzero n ∈ Z. It is not hard to see that the nonzero
fractional ideals form a group. Consequently the nonzero fractional ideals admit
unique factorizations as products of prime ideals, the exponents of the prime ideals
being integers that are not necessarily ≥ 0.
Theorem 3.2. Let P be a nontrivial prime ideal of O, and let q = |O/P |. For
each x ∈ F × , let v(x) be the power of P that appears in the factorization of the
principal fractional ideal (x) = xO, and define |x|P = q −v(x) . Then | · |P defines a
metric on F , and the completion of F in this metric is a local field whose module is
the continuous extension of | · |P . The result is a place of F , and the prime ideal
of O associated to this place is just P .
Since the factorization of (x) in the theorem is finite, it follows that |x|v is
different from 1 for only finitely many places v. One consequence of this fact is
256 A. W. KNAPP

that any x ∈ F embeds diagonally as an element of AF . We shall make constant

use of this embedding. It is tempting to write diag F for the image, but this
notation soon becomes unwieldy and it is customary to denote the image simply
by F . Briefly the subset F of AF always means the diagonally embedded version of
F unless the contrary is stated.
 A second consequence of the fact that |x|v = 1 for only finitely many v is that
×
v |x|v is well defined for each x ∈ F . The next theorem tells the value of this
product.
 ×
Theorem 3.3 (Artin product formula). v |x|v = 1 for all x ∈ F .

Reference. [La, p. 99].

Theorem 3.4. The image of F in AF is discrete, and the quotient group AF /F

is compact.
Reference. [La, p. 139].

The construction of the ideles of the number field F is a second use of the notion
of a restricted direct product. The index set I is the set {v} of places, the factors
Xi are the multiplicative groups Fv× , the subset S∞ is the set of infinite places,
and the Ki are the groups of units Ov× . The restricted direct product is a locally
compact abelian group denoted A× F and called the ideles of F .
If x is in F × , then the tuple consisting of x in every place is an idele as a
consequence of the discussion before Theorem 3.3. Hence F × embeds diagonally
in A× ×
F . We denote the image simply by F , understanding that F
×
is diagonally
embedded unless the contrary is stated.
If one tracks down the definitions, the topology on A×
F is not the relative topology
×
from AF but is finer. Actually AF gets the relative topology from

{(x, y) ∈ AF × AF | xy = 1}.

Since F × is discrete in AF (Theorem 3.4) and since the topology on A× F is finer

than the relative topology on AF , F × is discrete in A×
F.
Being a locally compact abelian group, AF has a Haar measure µAF , and this can
evidently be taken to be the product of the Haar measure on each Fv if the latter
are normalized at almost every place to assign mass one to Ov . Multiplication by
any r ∈ A×F is an automorphism of AF and hence carries µAF to a multiple of itself.
We write
dµAF (rx) = |r|AF dµAF (x)
and call |r|AF the module of r. Arguing first with r equal to 1 in all but one place
and then passing to general r, we see that |r|AF = v |rv |v . Theorem 3.3 therefore
implies that any element x of F × has |x|AF = 1.
Let (A× ×
F ) be the subgroup of elements of AF of module 1, i.e., the kernel of the
1
× ×
homomorphism | · |AF : AF → R+ . We have just seen that F × lies in this subgroup,
and we saw above that F × is discrete. This proves the easy half of the following
theorem.
Theorem 3.5. The diagonal embedding F × → A× F carries F
×
to a discrete
× 1 × 1 ×
subgroup of (AF ) , and the quotient (AF ) /F is compact.
 INTRODUCTION TO THE LANGLANDS PROGRAM 257

Remarks. The compactness of the quotient is closely related to the Dirichlet

Unit Theorem. For a proof of the result and a discussion of the connection, see
[Cas, pp. 70–73] and [La, pp. 142–146]. For a direct argument in a more general
context, see [We4, p. 76].

The relationship between A× × 1

F and (AF ) is that

A× × 1
F = (AF ) × R (3.2)
for a noncanonical subgroup R isomorphic to R× + . In fact, fix an infinite place v.
Then Fv× contains a subgroup R× + , and we let R be the image of this subgroup in
A×F under inclusion into the v th
place. It is clear that the module map | · |AF carries
R one-one onto R× + . Thus (3.2) follows.
From (3.2) and Theorem 3.5, it follows that
A×
F /H is compact (3.3)
for any subgroup H of A× ×
F containing RF . Recall that F∞ = F ⊗Q R gives
×
the archimedean component of AF (with 1 in all finite places). Then F∞ gives
× ×
the archimedean component of AF . Since F∞ contains R, (3.3) implies that

A× × × ×
∞ F ) is× compact. Let U1 be the ring v finite Ov , so that K1 = U1 is
F /(F
just v finite Ov .
Corollary 3.6. Suppose that K is any open compact subgroup of K1 . Then the
set XK = F × \A× ×
F /(F∞ K) is finite.

Remark. When K = K1 , use of the relationship discussed below between ideles

and fractional ideals allows one to identify XK1 with the ideal class group of F .
Thus the corollary gives an adelic proof of the finiteness of the class number.

Proof. The above considerations show that A× × ×

F /(F∞ F ) is compact, and hence
XK is compact. On the other hand, F∞ K is open in A×
× × ×
F . Thus AF /(F∞ K) is
discrete, and consequently XK is discrete.

Historically, ideles were introduced before adeles. Chevalley’s purpose in intro-

ducing ideles was to extend class field theory to infinite abelian extensions. But, as
is indicated in the introduction of [Ch], the theory of ideles served also the purpose
of reinterpreting results about fractional ideals and related notions. We give some
details about this point in order to prepare for Artin reciprocity, which will be
discussed in the next section. For each x = (xv ) ∈ A× F , we create a fractional
ideal as follows. Each finite place v corresponds to some prime ideal ℘ in O. With
−v(x ) 
qv = |Ov /mv | and |xv |v = qv v , associate to x the fractional ideal ℘ ℘v(xv ) . If
x = (x0 , x0 , . . . ) is in F × , then this definition reproduces the fractional ideal (x0 )
in its factored form; hence there is no ambiguity if we refer to this fractional ideal
as (x) in every case. is a finite set of places containing S∞ and if x is an idele,
 If S v(x
we define (x)S = ℘∈S / ℘ v)
.
Artin reciprocity initially involves a homomorphism of fractional ideals into an
abelian Galois group, and we shall want to lift this homomorphism to a homomor-
phism of the group of ideles. We shall impose a continuity condition. Let S be a
finite set of places containing S∞ , and let I S be the subgroup of fractional ideals
whose ℘ factor is (1) for every finite ℘ ∈ S. If G is an abelian topological group, we
258 A. W. KNAPP

say that a homomorphism ϕ : I S → G is admissible if, for each neighborhood N

of 1 in G, there exists δ > 0 such that ϕ((a)S ) is in N whenever a is a member of
F × such that |a − 1|v < δ for all v ∈ S (including all infinite places). To reinterpret
admissible ϕ’s in terms of ideles, we use the following Weak Approximation
Theorem.
Theorem 3.7. Let | · |n , 1 ≤ n ≤ N be distinct places of F , finite or infinite.
If xn , 1 ≤ n ≤ N , are members of F and if 7 > 0 is given, then there exists ξ ∈ F ×
with |ξ − xn |n < 7 for 1 ≤ n ≤ N .
Reference. [Cas, p. 48].

Theorem 3.8. Let S be a finite set of places containing S∞ , and let G be a

compact abelian group (usually finite abelian or S 1 ). If ϕ : I S → G is an admissible
homomorphism, then there exists a unique homomorphism ϕ̃ : A× F → G such that
(a) ϕ̃ is continuous
(b) ϕ̃ is 1 on F ×
(c) ϕ̃(x) = ϕ((x)S ) for all x = (xv ) ∈ A×
F such that xv = 1 for all v ∈ S
(including all infinite places).

Reference. [Ta2, pp. 169–170].

Proof of uniqueness. Given x ∈ A× F , choose, by Theorem 3.7, a sequence of

elements an of F × such that an → x−1 at the places of S. For y ∈ A× S
F , let y be
the idele with all entries in places of S set equal to 1, and let yS be the idele with
all entries in places outside S set equal to 1. For every n, we have

ϕ̃(x) = ϕ̃(an x) by (b)

S
= ϕ̃((an x)S )ϕ̃((an x) ) since ϕ̃ is a homomorphism
S
= ϕ̃((an x)S )ϕ((an x) ) by (c).

The first factor on the right side tends to 1 by (a). Thus

ϕ̃(x) = lim ϕ((an x)S ).

n→∞

4. Artin Reciprocity
A general reference for Artin reciprocity is [Ta2]. Let K/F be a Galois extension
of number fields, and let G = Gal(K/F ) be the Galois group.
Let v be a finite place of F , and let ℘ be the corresponding prime ideal of OF .
The ideal ℘OK of OK has a factorization into prime ideals of OK , say

℘OK = P1e1 · · · Pgeg

with the Pj distinct and with all ej > 0. Here the Pj ’s are exactly the prime ideals
of OK that contain ℘, and each Pj has Pj ∩ OF = ℘. We say that the Pj lie above
℘. From the correspondence of finite places to prime ideals, it is easy to see that
the places wj corresponding to the Pj ’s are exactly the places that lie above v.
 INTRODUCTION TO THE LANGLANDS PROGRAM 259

e
Theorem 4.1. In the factorization ℘OK = P1e1 · · · Pg g into prime ideals in
OK , the Galois group G permutes P1 , . . . , Pg , and the action of G on the g-element
set is transitive. Consequently e1 = · · · = eg . Moreover, if |OF /℘| = q and
|OK /Pj | = q fj , then f1 = · · · = fg . If e denotes the common value of the ej and if
f denotes the common value of the fj , then
ef g = [K : F ].
The prime ideal ℘ of OF is said to ramify in K if e > 1. Ramification is an
exceptional occurrence: If v lies above the place p of Q, then ramification of ℘ in K
implies that p divides the absolute discriminant of K. In particular, only finitely
many prime ideals of OF ramify in K.
In the situation of the theorem, let P be one of P1 , . . . , Pg . Let w and v be the
places of K and F corresponding to P and ℘, so that w lies above v. Write KP
and F℘ for the completions. Define
GP = {σ ∈ G | σ(P ) = P }.
This group is called the decomposition group relative to P . The members of
GP acts as isometries of K in the norm | · |P . Consequently σ extends to an
automorphism of KP , and we see that we can think in terms of an inclusion
GP 9→ Gal(KP /F℘ ). (4.1)
Theorem 4.2. The embedding of GP in (4.1) is onto Gal(KP /F℘ ), and KP /F℘
is a Galois extension of local fields.
We observed before Theorem 1.1 that each member of Gal(KP /F℘ ) acts by
an isometry and consequently induces an automorphism of Gal(KP /F℘ ) into
Gal(kP /k℘ ), where kP and k℘ are the respective residue fields. Theorem 1.1 says
that the resulting homomorphism is onto. Thus we can rewrite the exact sequence
(1.1) in this context as
1 −−−→ IP −−−→ GP −−−→ Gal(kP /k℘ ) −−−→ 0, (4.2)
where the inertia group IP is defined to be the kernel. With e, f , and g as in
Theorem 4.1, we know that ef g = [K : F ]. Since G acts transitively on P1 , . . . , Pg
by Theorem 4.1, the isotropy subgroup GP at P has ef g/g = ef elements. By
Theorem 4.2, |Gal(KP /F℘ )| = ef . Our definitions make |Gal(kP /k℘ )| = f , and
therefore |IP | = e.
In other words, if ℘ is unramified in K (as is the case for almost all prime ideals of
F ), the extension KP /F℘ is unramified in the sense of §1. In this case, Gal(KP /F℘ )
contains a well defined Frobenius element, as in the definition following Theorem
1.2. If P is replaced by another prime ideal P  lying over ℘, then GP is conjugate to
GP
by an element of G carrying P to P  , and this conjugacy carries the Frobenius
element to the Frobenius element. In terms of ℘ as a given piece of data, the
Frobenius element is then any element of a certain conjugacy class of G.
Artin reciprocity deals with the situation that G = Gal(K/F ) is abelian. In this
case when ℘ is a prime ideal of OF that is unramified in K, the conjugacy class of
Frobenius elements reduces to a single element, and we can unambiguously denote
the Frobenius element by the notation
 
K/F
∈ G.
℘
260 A. W. KNAPP

Let S be the finite

 set of all infinite places of F and all finite places of F that ramify
n
in K. If X = j ℘j j is the factorization of a fractional ideal of F into primes, we
 nj
recall that X S = j ∈S
/ ℘j . Then we can define
   n
K/F K/F j
= .
XS ℘j
j ∈S
/

The resulting homomorphism of I S into G is called the Artin symbol of K/F .

Theorem 4.3 (Artin reciprocity, first form). Let K/F be a finite abelian Galois
extension of degree n, and let S be the finite set of all infinite places of F and all
finite places of F that ramify in K. If a ∈ F × is such that a is in (Fv× )n for every
v ∈ S and if (a) denotes the fractional ideal aOF , then
 
K/F
= 1.
(a)S

Reference. [Ta2, p. 167].

This first form of Artin reciprocity is the weakest of three forms that we shall
consider. However, it is already strong enough so that with a little computation it
implies quadratic reciprocity [Cas-Fr, pp. 348–350]. It also implies a more general
mth power reciprocity theorem due to Kummer.
We shall now sharpen the statement of Artin reciprocity so as to be able to bring
Theorem 3.8 to bear.
Theorem 4.4 (Artin reciprocity, second form). Let K/F be a finite abelian
Galois extension, and let S be the finite set of all infinite places of F and all finite
places of F that ramify in K. There exists δ > 0 such that whenever a ∈ F × has
|a − 1|v < δ for all v ∈ S, then
 
K/F
= 1.
(a)S

Reference. [Ta2, p. 167].

Theorem 4.4 implies Theorem 4.3 by a simple argument [Ta2, p. 167] using the
Weak Approximation Theorem (Theorem 3.7).
In the terminology at the end of §3, Theorem 4.4 says that the homomorphism
I S → G defined by the Artin symbol is admissible. By Theorem 3.8, the Artin
symbol lifts uniquely to a continuous homomorphism of the idele class group of
F,
CF = A× ×
F /F , (4.3)
into G. Let us call this homomorphism the Artin map of K/F and denote it by
θK/F : CF → G.
Theorem 4.5. Let K/F be a finite abelian Galois extension, let v be a place of
F , and let w be a place of K lying above v. If iv : Fv× → CF denotes the composition
of inclusion of Fv into the v th place of A×
F followed by the quotient map to CF , then
the Artin map and the local reciprocity map are related by θK/F ◦ iv = θKw /Fv as
homomorphisms Fv× → G.
 INTRODUCTION TO THE LANGLANDS PROGRAM 261

Reference. [Ta2, p. 175].

Theorems 4.4 and 4.5 are the main facts about the Artin symbol and Artin
map that we need in the next section. For completeness we include a little more
information at this time. It is apparent from Theorem 4.5 and the precise statement
of local class field theory in Theorem 2.4 that there has to be a sharper statement
of Artin reciprocity than in Theorem 4.4. Here is such a result.
Theorem 4.6 (Artin reciprocity, third form). Let K/F be a finite abelian Galois
extension, and let S be the finite set of all infinite places of F and all finite places
of F that ramify in K. If an element a ∈ F × is a norm from Kw for all w lying
over places of S, then  
K/F
= 1.
(a)S
Reference. [Ta2, p. 176].

Now we think of K as varying. Namely we fix an algebraic closure F of F and

consider finite abelian Galois extensions K of F lying in F . Let GF be the Galois
c
group of F over F , and let Gab
F = GF /G be the Galois group of the maximal abelian
extension of F . If K is a finite abelian extension of F , then the Artin map θK/F is a
continuous homomorphism of CF into GK/F = Gal(K/F ). These homomorphisms
have a compatibility property that allows us to lift them to a single continuous
homomorphism with values in the inverse limit, namely θF : CF → Gab F . We call
θF the Artin map of F . It follows from the various compatibility properties that
Theorem 4.5 can be restated in this notation as
θF ◦ iv = θFv (4.4)
as homomorphisms Fv× → Gab
F .

Lemma. The Artin map θF carries CF onto Gab

F .

Sketch of proof. By (3.2), A× × 1

F = R × (AF ) . Since F
×
lies in (A× 1
F ) , we
∼ × 1 ×
obtain CF = R × CF , where CF = (AF ) /F . Since R is connected and Gab
1 1
F is
totally disconnected, θF (R) = 1. Thus θF (CF ) = θF (CF1 ), and this is compact
by Theorem 3.5. On the other hand, one shows that the composition of θF and
passage to any finite quotient of Gab
F carries CF onto the finite quotient. Hence
θF (CF ) is dense in Gab
F .

In practice, Artin reciprocity is proved at the same time as the Existence The-
orem of global class field theory, whose statement is given in Theorem 4.7 below.
If v is a place of F and w is a place of K lying above v, then the norm map
NKw /Fv : Kw ×
→ Fv× is well defined. We set NK/F : A× ×
K → AF equal to the
 ×
coordinate-by-coordinate product NK/F = w NKw /Fv . It is clear that NK/F
carries diagonally embedded K × to diagonally embedded F × and therefore descends
to a homomorphism NK/F : CK → CF .
Theorem 4.7. The map K → CF /NK/F (CK ) is a bijection of the set of finite
abelian Galois extensions K of F onto the set of open subgroups of CF of finite
index. The field corresponding to a subgroup B is the fixed field of the subgroup
θF (B) of Gab
F .
262 A. W. KNAPP

References. [Ta2, p. 172] and [Ar-Ta, p. 70].

Corollary 4.8. The kernel of the Artin map θF of CF onto Gab

F is the identity
component (CF )0 of CF .
Reference. [Ta2, p. 173].

We can summarize Corollary 4.8 as saying that the sequence

θ
1 −−−→ (CF )0 −−−→ CF −−−
F
→ Gab
F −−−→ 0

is exact.

5. Artin L Functions
At the beginning of §3, we mentioned that the ring of adeles carries information
about the factorization of polynomials, and we shall elaborate on this assertion now.
The Artin L functions to be introduced in this section encode this information as
explicit functions of a complex variable given by product formulas. Artin reciprocity
enables one to recognize certain Artin L functions as arising in another way that
shows that they have nice analytic properties.
Example 1. For the polynomial R(X) = X 2 + 1, we ask how R(X) reduces
modulo p for primes p = 2. Before giving the well known answer, let us encode the
problem in a generating function. Put
  
−1 +1 if X 2 + 1 factors completely modulo p
=
p −1 if X 2 + 1 is irreducible modulo p,
and define
 1
L(s) =   .
−1
p=2 1− p p−s
This certainly converges for Re s > 1. The well known answer to our question
amounts to giving the pattern for −1p , which is
  
−1 +1 if p = 4k + 1
=
p −1 if p = 4k − 1.
This is the simplest case of quadratic reciprocity. The point to observe is that the
pattern is described by finitely many linear congruences. If we define


 +1 if n ≡ 1 mod 4
χ(n) = −1 if n ≡ 3 mod 4


0 if n even,
then
 1
L(s) = .
p
1 − χ(p)p−s
From the property χ(mm ) = χ(m)χ(m ), we obtain
∞
χ(n)
L(s) = .
n=1
ns
 INTRODUCTION TO THE LANGLANDS PROGRAM 263

In this form, L(s) becomes more manageable. This series is absolutely convergent
for Re s > 1. Use of summation by parts shows that L(s) converges for Re s > 0,
and an elementary argument shows that L(s) continues to an entire function. It is
not hard to see that L(1) = 0, and from this fact it follows that there are infinitely
many primes p = 4k + 1 and infinitely many primes p = 4k − 1. In other words,
interesting information about primes has been encoded in L(s) at a spot on the
boundary of the region where L(s) converges absolutely. Finding the pattern for
1
  enables us to extract this information.
1 − −1 p

Example 2 ([Buhl] and [Lgl6]). Let us consider the polynomial

R(X) = X 5 + 10X 3 − 10X 2 + 35X − 18.

This has discriminant 26 58 112 , and the question is to find the pattern of how R(X)
reduces modulo p for p = 2, 5, 11. For example, we can readily find by computer
that R(X) is irreducible modulo p for p = 7, 13, 19, 29, 43, 47, 59, . . . . Similarly we
find that R(X) splits completely for p = 2063, 2213, 2953, 3631, . . . . What is the
pattern? These sequences of primes are not related to linear congruences, and the
Langlands theory gives conjectures that describe the pattern. Let F be the splitting
field of R(X) over Q, and let G be the Galois group. Since the discriminant is a
square, G ⊂ A5 . The group G contains a Frobenius element Frp for each p = 2, 5, 11,
and this element is the lift to Gp of a generator of the Galois group of R(X) mod p.
Modulo p = 7, R(X) is irreducible; so G has an element of order 5. Modulo p = 3,
R(X) is the product of two linear factors and an irreducible cubic; so G has an
element of order 3. Since A5 is generated by any two elements of respective orders
5 and 3, we conclude that G = A5 .
For any p = 2, 5, 11, the Galois group of R(X) mod p tells us a great deal
about the factorization of R(X) mod p. The generator of this group is a Frobenius
element, which can be any element in a particular conjugacy class of G. In the
case of A5 , the order of an element determines the conjugacy class of the element
in A5 unless the order is 5, for which there are two conjugacy classes. The order
of the Frobenius element is f . Order 4 does not occur in A5 , and thus f = 1, 2,
3, or 5. If f = 5, R(X) mod p is irreducible. If f = 3, R(X) mod p is the product
of two distinct linear factors and an irreducible cubic. If f = 2, the element of G
has to be the product of two 2-cycles; thus R(X) mod p has to be the product of a
linear factor and two distinct irreducible quadratic factors. Finally if f = 1, then
R(X) mod p splits into five distinct linear factors. The value of f determines the
conjugacy class of Frobenius elements in A5 completely unless f = 5. For f = 5,
there are two conjugacy classes; see [Buhl, p. 53] for how to distinguish them.
In order to encode the full information about the conjugacy classes of the Frobe-
nius elements in one or more generating functions, we can proceed as follows. Let
σ be a finite-dimensional representation of G over C. Then the generating function
is
 1
L(s, σ) = ,.
det(1 − p−s σ(Frp ))
p=2,5,11
except that suitable factors for p = 2, 5, 11 need to be included. The goal is to
recognize this function in another form and thereby to find the pattern of the
coefficients. This is carried out in [Buhl].
264 A. W. KNAPP

Let K/F be a Galois extension of number fields, with Galois group G, and let
σ : G → AutC (V ) be a finite-dimensional complex representation of G. (As always,
we build continuity into the definition of “representation.”) The Artin L function
is defined to be 
L(s, σ) = L(s, σ, K/F ) = L℘ (s, σ),
℘

the product being taken over the nontrivial prime ideals ℘ in OF . Here s is a
complex variable.
Fix ℘, put q = |OF /℘|, and let P be a prime ideal in OK lying over ℘. The
definition of L℘ (s, σ) is a little simpler if ℘ is unramified in K, and we consider
that case first. Then there is a well defined Frobenius element FrP in GP , and we
put1
L℘ (s, σ) = det(1 − σ(FrP )q −s )−1 for Re s > 0. (5.1a)
Let the eigenvalues of σ(FrP ) be ε1 , . . . , εdim V ; these are roots of unity since FrP
has finite order. Then
V
dim
L℘ (s, σ) = (1 − εi q −s )−1 .
i=1

As P varies, FrP moves in a conjugacy class of G. The eigenvalues of σ(FrP ) do

not change, and the second formula for L℘ (s, σ) shows that the function depends
only on ℘.
Now suppose ℘ is allowed to be ramified in K, so that the inertia group IP is
nontrivial. Let V IP be the subspace of V on which σ(IP ) acts as the identity; this
would be all of V in the unramified case. Then σ(GP ) preserves this space. If
FrP is one lift to GP of the canonical generator of the Galois group of the residue
field extension, then the most general lift is FrP iP with iP ∈ IP . Thus σ(FrP ) is
unambiguous as a linear transformation on V IP , and we define2

L℘ (s, σ) = det(1 − σ(FrP )|V IP q −s )−1 for Re s > 0. (5.1b)

Again we can rewrite this using eigenvalues, and we see that the result is indepen-
dent of P .
Each Artin L function converges for Re s > 1. Artin L functions have the
following additional properties (see [Hei, pp. 222–223], [La, pp. 236–239], and [Mar,
p. 9]):
1) L(s, σ1 ⊕ σ2 , K/F ) = L(s, σ1 , K/F )L(s, σ2 , K/F ).
2) Suppose that F ⊂ E ⊂ K and that E is Galois over F . Let H = Gal(K/E),
a normal subgroup of G = Gal(K/F ). If σ̃ is a representation of G lifted
from a representation σ of G/H, then

L(s, σ̃, K/F ) = L(s, σ, E/F ).

Consequently an Artin L function depends only on s and a continuous finite-

dimensional representation of Gal(F /F ).

1 Concerning the choice of Fr

P or its inverse in this formula (i.e., σ or its contragredient), see
Remark 1 after Corollary 2.5.
2 See the footnote with (5.1a).
 INTRODUCTION TO THE LANGLANDS PROGRAM 265

3) Suppose that F ⊂ E ⊂ K with E/F possibly not Galois. Let σ0 be a

representation of G0 = Gal(K/E). Then
L(s, indG
G0 (σ0 ), K/F ) = L(s, σ0 , K/E).

A quasicharacter is a continuous homomorphism into C× . Let F be a number

field. By a Grossencharacter of F is meant a quasicharacter of CF .
Example 1. Fix a positive integer m. A Dirichlet character modulo m is the
lift χ to Z of a character of the multiplicative group (Z/mZ)× , with χ(a) set equal
to 0 if a and m are not relatively prime. Fix such a χ. With F = Q, let S consist
of ∞ and the primes dividing m. In the notation of §3, define a homomorphism
ϕ : I S → S 1 by ϕ((a/b)S ) = χ(a)/χ(b) whenever a and b are integers relatively
prime to m. If p is a prime dividing m and pc is the exact power of p dividing
m, then | ab − 1|p < p−c implies that a and b are congruent modulo pc . Hence if
| ab − 1|p < m−1 for all p dividing m, then a and b are congruent modulo m, and
it follows that χ(a) = χ(b). In other words ϕ is admissible in the sense of §3. By
Theorem 3.8 there exists a unique unitary Grossencharacter ϕ̃ of Q such that
ϕ̃(1, . . . , 1, p, 1, . . . ) = χ(p)
for all primes p not dividing m; here (1, . . . , 1, p, 1, . . . ) denotes the idele that is p
in the pth place and is 1 elsewhere.
Example 2. If ω is any (continuous) character of Gab F , then the composition
ω ◦θF with the Artin map of F is a Grossencharacter. Since ω has to factor through
F , it is the same to consider compositions ω0 ◦ θK/F , where
a finite quotient of Gab
K is a finite abelian Galois extension of F and ω0 is a character of Gal(K/F ).
Let ϕ be a Grossencharacter of F . Following Hecke in spirit, we shall associate
an L function L(s, ϕ) to ϕ. For each finite place v with corresponding prime
ideal ℘ of OF , we can restrict ϕ to the coordinate F℘× , obtaining a quasicharacter
ϕ℘ : F℘× → C× . Let O℘ be the ring of integers in F℘ . We say that ϕ℘ is ramified
if ϕ℘ |O℘× is nontrivial. For each ℘, let ℘ be a prime element in O℘ . Then the
definition is3
   −1
ϕ℘ (℘ )
L(s, ϕ) = L℘ (s, ϕ) = 1− . (5.2)
|OF /℘|s
℘ unramified ℘ unramified
for ϕ for ϕ

The functions L(s, ϕ) have nice analytic properties. They have meromorphic
continuations to C and satisfy a functional equation relating the values at s and
1 − s. The only possible pole is at s = 1 and is at most simple; there is no pole
if ϕ is nontrivial on (A× 1
F ) . These results are essentially due to Hecke. Later Tate
[Ta1] found an important different proof that uses local-global methods. For an
exposition of Tate’s work and a higher-dimensional generalization, see Jacquet’s
lecture [Ja2].
Sometimes authors include in L(s, ϕ) extra factors for the infinite places that
involve a gamma function and powers of certain numbers. See [Kna2] for a descrip-
tion of these. Shortly we shall use this kind of completed L function, writing Λ
for it. Inclusion of factors for the infinite places affects the poles of L(s, ϕ) only
slightly and makes the functional equation much simpler.
3 See the footnote with (5.1a).
266 A. W. KNAPP

Theorem 5.1. If K/F is a finite abelian Galois extension of number fields and
ω is a character of Gal(K/F ), then the Artin L function L(s, ω, K/F ) equals the
Hecke L function L(s, ϕ) of the Grossencharacter ϕ = ω ◦ θK/F .
Proof. Let H be the kernel of ω in G = Gal(K/F ), and let E be the fixed field
of H in K. Then E/F is a finite abelian Galois extension of F , and ω descends to
a one-one character ω0 of G/H = Gal(E/F ). By property (2) of Artin L functions,
we have L(s, ω, K/F ) = L(s, ω0 , E/F ). Also ϕ = ω0 ◦ ϕE/F . Thus it is enough to
prove that L(s, ω0 , E/F ) = L(s, ω0 ◦ θE/F ). We do so factor by factor.
Let ℘ be a nontrivial prime ideal in OF . We show that ℘ is unramified in E if
and only if ℘ is unramified for ϕ. With S as the set of infinite places and places
that ramify in E, first suppose that ℘ is not in S. If x℘ is in O℘× , then the proof of
Theorem 3.8 (with an = 1 for all n) shows that
 
E/F
ϕ℘ (x℘ ) = ϕ(. . . , 1, x℘ , 1, . . . ) = ϕ(({x℘ })S ) = ω0 = 1.
(1)
Hence ℘ is unramified for ϕ.
Conversely suppose ℘ is unramified for ϕ = ω0 ◦ θE/F . This means that ϕ℘ =
ω0 ◦ θE/F ◦ i℘ is 1 on O℘× . By Theorem 4.5, ω0 ◦ θEP /F℘ is 1 on O℘× . Since ω0 is
one-one, θEP /F℘ is 1 on O℘× . Referring to (2.3) and Corollary 2.5, we see that the
inertia group IP is trivial. Thus ℘ is not in S.
Now suppose that ℘ is ramified (in both senses). Then the ℘ factor of L(s, ϕ) is
1 by definition. Meanwhile the ℘ factor of L(s, ω0 , E/F ) is (1 − ω0 (FrP )|CIP q −s )−1 .
Since ω0 is one-one and IP is nontrivial, CIP = 0. Thus the ℘ factor for each L
function is 1 in the ramified case.
Finally suppose that ℘ is unramified (in both senses). Theorem 4.5 gives
ϕ℘ = ϕ ◦ i℘ = ω ◦ θK/F ◦ i℘ = ω0 ◦ θE/F ◦ i℘ = ω0 ◦ θEP /F℘ . (5.3)

Then L℘ (s, ω0 , E/F ) = (1 − ω0 (FrP )q −s )−1

by definition, while

L℘ (s, ϕ) = (1 − ϕ℘ (℘ )q −s )−1 by definition

−s −1
= (1 − ω0 θEP /F℘ (℘ )q ) by (5.3)
−s −1
= (1 − ω0 (FrP )q ) by Theorem 2.2.
Hence the ℘ factors for the two L functions match in the unramified case.
Corollary 5.2. If K/F is a finite abelian Galois extension of number fields and
ω is a character of Gal(K/F ), then L(s, ω, K/F ) extends to be entire in C. For the
trivial character ω = 1, L(s, 1, K/F ) extends to be meromorphic in C with a simple
pole at s = 1.
Of course, it is immediate from the properties of Artin L functions that many
more such functions are entire. For example, let K/F be any finite Galois extension.
Then the Artin L function for a representation of Gal(K/F ) induced from a non-
trivial one-dimensional representation is entire. Thus for any monomial group,
i.e., any finite group whose irreducible representations are all induced from one-
dimensional representations of subgroups, the Artin L function of any nontrivial
irreducible representation is entire. Dihedral groups are examples of monomial
groups.
 INTRODUCTION TO THE LANGLANDS PROGRAM 267

Artin Conjecture. Let K/F be any finite Galois extension of number fields,
and let σ be a nontrivial irreducible representation of Gal(K/F ). Then L(s, σ, K/F )
extends to be entire in C.
The Artin L function, which we defined to include only factors (5.1) from the
finite places, has a natural completion by adjoining some gamma factors for the
infinite places (see [Kna2, (3.6) and (4.6)], and we shall denote the completed L
function by Λ(s, σ, K/F ). It is actually conjectured that Λ(s, σ, K/F ) is entire if σ
is irreducible and nontrivial.
Despite the fact that the Artin Conjecture is not known, Brauer’s Induction
Theorem says that an Artin L function has a continuation to all of C that is
at least meromorphic and that the continued function satisfies the same kind of
functional equation as a Hecke L function. A more detailed statement of Brauer’s
result is as follows.
Theorem 5.3. The group character of any complex finite-dimensional represen-
tation of a finite group is an integer combination of group characters of represen-
tations induced from one-dimensional representations of subgroups. Consequently
any Artin L function has a continuation to all of C that is at least meromorphic.
Moreover each Artin L function satisfies a functional equation of the form

Λ(s, σ, K/F ) = ε(s, σ, K/F )Λ(1 − s, σ ∨ , K/F ), (5.4)

where σ ∨ is the contragredient and ε(s, σ, K/F ) is entire and nonvanishing.

The Langlands theory proposes addressing the Artin Conjecture by introducing
L functions that generalize those of Hecke and by showing that Artin L functions
are always of this kind. Aspects of this theory occupy much of the remainder of
this article. For another exposition of this kind, see [Gelb2].

6. Linear Reductive Algebraic Groups

Let F be a number field. Informally a linear algebraic group of n-by-n
matrices over F is a group G of n-by-n matrices defined by polynomial equations
in n2 matrix variables with coefficients in F . For a precise definition, see [Bo2],
[Bo3], or [We1].
Clearing fractions in the defining equations of G, we may assume that the coef-
ficients are all in OF . If R is a torsion-free commutative ring containing OF as a
subring, then the group of R points of G is well defined, independently of how we
cleared fractions, and we denote this group by G(R). The group G will be said to
be unipotent if G(C) consists entirely of unipotent matrices, while G is reductive
if {1} is the only connected unipotent normal subgroup of G(C). The reader may
wish to think of a reductive G as being GLn or SLn , and little will be lost for
current purposes by doing so.
A simple example of a linear algebraic group is the affine line, whose R points
are the elements of R. To work with this algebraic group as a linear group, we may
view it as the group of all matrices 10 ∗1 . Another example is the multiplicative
group, whose R points are the elements of R× ; this group is also called GL1 . The
affine line is unipotent, and the multiplicative group is reductive.
If v is a place of F , we can topologize GLn (Fv ) as an open subset of n2 -
dimensional space, and the result is a locally compact group. Then G(Fv ) is given
268 A. W. KNAPP

the relative topology from GLn (Fv ) and is a locally compact group. If v is finite,
then G(Ov ) is a compact subgroup of G(Fv ).
For each finite set S of places containing the set S∞ of infinite places, the group
G(AF (S)) is well defined since the ring AF (S) contains diagonally embedded OF .
The group G(AF (S)) is nothing more than the direct product of all G(Fv ) for v ∈ S
and all G(Ov ) for v ∈ / S. It is locally compact. Its topology may be described
alternatively as the relative topology from GLn (AF (S)).
Similarly, the group G(AF ) of AF points of G is well defined since AF contains
diagonally embedded F . When G is the affine line, G(AF ) is just the group of
adeles of F ; when G is the multiplicative group, G(AF ) is the group of ideles of F .
In every case, a member of G(AF ) may be regarded as a tuple of matrices indexed
by the places of F , the matrix in the v th place being in G(Fv ), with almost all such
matrices lying in G(Ov ).
As is the case with the ideles, the topology on G(AF ) is not necessarily the rela-
tive topology from n-by-n matrices over AF . Instead the topology is the restricted
direct product topology of the G(Fv ) relative to the G(Ov ). In other words, it
is the direct limit topology from the subgroups G(AF (S)), which areto be open. 
x 0
Alternatively we can topologize G(AF ) by embedding it in matrices 0 (det x)−1
of size n + 1 and giving it the relative topology from (n + 1)2 -dimensional space
over AF . With these definitions, G(AF ) gets the relative topology from GLn (AF ).
Because of this second way of realizing the topology, it follows that the group
G(diagonally embedded F ), which equals diagonally embedded G(F ), is a discrete
subgroup of G(AF ). We write G(F ) for this subgroup. The first theorem generalizes
Corollary 3.6.
Theorem 6.1. If G is a reductive linear algebraic group, then the number of
double cosets in G(F )\G(AF )/G(AF (S∞ )) is finite. For G = GLn , the number of
double cosets is the class number of F . For G = SLn , the number of double cosets
is 1.
References.
1) In a classical setting with G(Z) ⊂ G(Q), this theorem is due to Borel and
Harish-Chandra [Bo-HC]. The result in the classical setting says in part that if G is
a semisimple linear algebraic group over Q, then G(Z)\G(R) has finite volume. The
result in an adelic setting appears in Borel [Bo1, p. 19]. The relationship between
the two settings will be described below.
2) If the group G(AF (S∞ )) in the statement of the theorem is replaced by an
open subgroup of finite index, then it is clear that the number of double cosets
remains finite.

More is true than is asserted in the theorem. The hypothesis that G is reductive
is unnecessary. In any event, let G∞ = G(F∞ ) be the archimedean component of
G(AF ) (equal to 1 at every finite place). We say that G has the strong approxi-
mation property if G(F )G∞ is dense in G(AF ). In this case the number of double
cosets in G(F )\G(AF )/G(AF (S∞ )) is automatically 1. The (unipotent) affine line
has the strong approximation property [Cas, p. 67], and so does the reductive SLn .
For more discussion of this property, see [Kne].
In the classical setting for automorphic forms, one works with the quotient
G(Z)\G(R). Following [Bo-Ja, p. 195], let us see how this quotient space is related
to the double coset decomposition in Theorem 6.1. We decompose G(AF (S∞ )) =
 INTRODUCTION TO THE LANGLANDS PROGRAM 269

G∞ × K1 , where

 G∞ = G(F∞ ) is the archimedean component and where K1 =
G(U1 ) = G v finite Ov is the nonarchimedean component (equal to 1 at every
infinite place). Let K be any open subgroup of finite index in K1 , and use Theorem
6.1 to write 
G(AF ) = G(F )cG∞ K (6.1)
c∈C
as a disjoint union, for some finite subset C of G(AF ). Without loss of generality,
we may assume that the members of C all have component 1 at the infinite place.
For c ∈ C, define Gc = G∞ cKc−1 and Γc = Gc ∩ G(F ). The group Gc is open
in G(AF ), and the discreteness of G(F ) in G(AF ) implies that Γc is a discrete
subgroup of Gc . Since cKc−1 is compact, we may use projection on the infinite
places to identify Γc with a discrete subgroup of G∞ . If f is a right K invariant
function on G(AF ) and if c is in C, let fc be the function x → f (cx) on G∞ . Then
we readily check that the map
f → {fc }c∈C (6.2)

 of the space of functions on G(F )\G(AF )/K with the space of func-
is a bijection
tions on c∈C (Γc \G∞ ). Thus we obtain an identification

G(F )\G(AF )/K = (Γc \G∞ ). (6.3)
c∈C

Formula (6.3) is especially simple in cases where C = {1}. Examples, all with
K = K1 , are when G is the affine line and F is arbitrary, when G = SLn and F is
arbitrary, and when G = GLn and F = Q. For these examples, (6.3) becomes

F \AF / OF℘ = OF \F∞ , (6.4a)
℘

SLn (F )\SLn (AF )/ SLn (OF℘ ) = SLn (OF )\SLn (F∞ ) (6.4b)
℘

and GLn (Q)\GLn (AQ )/ GLn (Zp ) = GLn (Z)\GLn (R). (6.4c)
p

The right side of (6.3) is more concrete than the left side, but part of the action
is lost in working with the right side rather than with the adeles. For instance, in
the adelic picture of (6.4c), each of the groups GLn (Qp ) acts on GLn (Q)\GLn (AQ ),
and the corresponding action on this space by functions on GLn (Qp ) biinvariant
under GLn (Zp ) descends to an action on the left side of (6.4c). This action is
hidden, however, in the realization as GLn (Z)\GLn (R).
Let us describe the functor “restriction of the ground field,” which has the
property of reducing aspects of the theory over the number field F to the theory
over Q. We follow [We1]. Let d = [F : Q], and let σ1 , . . . , σd be the distinct field
maps of F into Q fixing Q. Let V be an affine variety over F . A pair (W, p), in
which W is an affine variety over Q and p : W → V is an algebraic map defined
over F , is said to be a variety obtained from V by restriction of the ground
field from F to Q if the map

d
(pσ1 , . . . , pσd ) : W → V σj ,
j=1
270 A. W. KNAPP

which is defined over Q, is an isomorphism of varieties. Such a pair (W, p) exists

and is unique up to a canonical isomorphism over Q. It is customary to drop the p
from the notation and write W = RF/Q V , regarding RF/Q as a functor from affine
varieties over F to affine varieties over Q. Restriction of the ground field has the
key property that (RF/Q V )(A) = V (F ⊗Q A) for any Q algebra A.
Examples.
1) If V is the affine line over F , then W is the vector space V regarded as a Q
vector space of dimension d. To put this example in the above context, let {αj }dj=1
be a basis of F over Q, and define p(x1 , . . . , xd ) = j αj xj . Then pσ (x1 , . . . , xd ) =
 σ
j αj xj , and

 σ1  σ 
(pσ1 , . . . , pσd )(x1 , . . . , xd ) = αj x1 , . . . , αj d xj .
j j

The fact that (pσ1 , . . . , pσd ) is an isomorphism follows from the fact that det{αjσi } =
0.
2) The previous example may be extended to an n-dimensional affine space V
over F in obvious fashion, with


d 
d

p(x11 , . . . , x1d , x21 , . . . , xnd ) = αj x1j , . . . , αj xnj .
j=1 j=1

More generally if V is the variety in n-dimensional affine space defined by some

polynomials P (X1 , . . . , Xn ), we can define a variety W in the nd-dimensional affine
space W by rewriting


d 
d

P αj x1j , . . . , αj xnj = α1 p1 (x11 , . . . , xnd ) + · · · + αd pd (x11 , . . . , xnd )
j=1 j=1

and replacing P by p1 , . . . , pd .
Restriction of the ground field is a functor of linear algebraic groups. Specifically
if V = G is a linear algebraic group over F , then the pair (RF/Q , p) can be taken
to consist of a linear algebraic group over Q and a homomorphism over F , and the
linear algebraic group structure is unique up to a canonical isomorphism of algebraic
d
groups. The map (pσ1 , . . . , pσd ) : RF/Q G → j=1 Gσj is then an isomorphism over
Q of algebraic groups. Because of the formula (RF/Q G)(A) = G(F ⊗Q A), the
isomorphism (pσ1 , . . . , pσd ) induces isomorphisms
(RF/Q G)(Q) ∼
= G(Q) × · · · × G(Q),
∼ G(F ),
(RF/Q G)(Q) =
(6.5)
(RF/Q G)(R) ∼= G∞ ,
(RF/Q G)(AQ ) ∼
= G(AF ).
Similarly we can define RE/F G as a reductive group over F whenever E/F is an
extension of number fields and G is a reductive group over E.
With G defined over F , let X ∗ (G)F be the set of all F rational homomorphisms
of G into GL1 . If χ is in X ∗ (G)F , then χ extends at each place to a continuous
homomorphism χv : G(Fv ) → Fv× . Let χAF : G(AF ) → A× F be the product of
 INTRODUCTION TO THE LANGLANDS PROGRAM 271

×
 χv . Then |χAF |AF is a homomorphism of G(AF ) into R+ . Define G(AF ) =
1
the
χ∈X ∗ (G)F ker |χAF |AF .
For example, if G = GLn , then  X ∗ (G)F consists of the integral powers of the
determinant, and | detAF (gv )|AF = v | det gv |v . Hence G(AF )1 consists of the n-
by-n matrices over AF for which the module of the determinant is 1. In the special
case that n = 1, this reduces to the group (A× 1
F ) that appears in Theorem 3.5. The
following theorem generalizes Theorem 3.5.
Theorem 6.2. Suppose that G(C) is connected. Then the group G(F ) lies
in G(AF )1 , and the quotient space G(F )\G(AF )1 has finite volume. Moreover,
G(F )\G(AF )1 is compact if and only if every unipotent element of G(F ) belongs
to the radical of G(F ).
References. [Bo-HC] and [Bo1, p. 22].

An example of a nonabelian G for which compactness of G(F )\G(AF )1 follows

from this theorem is the multiplicative group D× of a finite-dimensional division
algebra D of F with center F . For a direct proof of the compactness in this case,
see [Gf-Gr-P] or [We4, p. 74]. To prove the compactness from Theorem 6.2, we note
that a unipotent element of G(F ) is of the form 1 + x with x nilpotent. Since D is
closed under addition, x must be in D(F ). Since x is nilpotent, x is not invertible.
Thus x = 0, and the theorem applies.

7. Automorphic Forms and Automorphic Representations

Historically the theory of automorphic forms began with modular and cusp forms
for the group SL2 (Z).
 For a detailed discussion of such forms, see [Kna1] or [Shi].
a b
Briefly if g = c d is in GL(2, R), let g(z) = az+b
cz+d for all nonreal complex numbers
z, and define
j(g, z) = (cz + d)(det g)1/2 .
In our discussion, j(g, z) will occur only in the form j(g, z)2 , and we may conse-
quently use either determination of the square root of (det g)1/2 .
A modular form of weight k (an even integer) for SL2 (Z) is an analytic function
f (z) on the upper half plane such that

f (γ(z)) = j(γ, z)k f (z) for all γ ∈ SL2 (Z)

and such that f is analytic at ∞ in the following sense: The function

∞ f , being
analytic and periodic under z → z + 1, has an expansion f (z) = n=−∞ cn e2πinz ,
and the condition of analyticity at ∞ is that cn = 0 for n < 0. The condition at ∞
for a modular form can alternatively be formulated as the slow-growth condition

|f (x + iy)| ≤ Cy N for some C and N as y → +∞.

A cusp form is a modular form that vanishes at ∞ in the sense that c0 = 0.

The additional condition that a modular form is a cusp form can be reformulated
as the vanishing of an integral:

1
f (x + iy) dx = 0 for some or equivalently every y > 0.
0
272 A. W. KNAPP

A cusp form satisfies the rapid-decrease condition that for each N

|f (x + iy)| ≤ Cy −N for some C as y → +∞.

∞
A cusp form of weight k can be expanded asf (z) = n=1 cn e2πinz , and the
∞
associated L function of f , defined by L(s, f ) = n=1 cn /ns , satisfies a functional
equation relating the values at s and k − s. Hecke introduced what we now call
Hecke operators on the space of cusp forms of weight k. A computation using
contour integrals shows that this space is finite-dimensional. The Hecke operators
commute and are simultaneously diagonable. The eigenfunctions all have c1 = 0,
and if c1 is normalized to be 1 for an eigenfunction, then the corresponding L
function has an Euler product expansion, the product being taken over all primes.
Gelfand and Fomin were the first to notice that cusp forms could be realized
as smooth vectors in representations of a certain ambient Lie group. It is possible
to arrange for modular and cusp forms to lift both to SL2 (Z)\SL2 (R) and to
GL2 (Z)\GL2 (R), and we shall indicate both liftings.
The details for the lifting to SL2 (Z)\SL2 (R) are in [Gelb1] and also [Bump].
Given a modular form f as above, we define

φf,∞ (g) = f (g(i))j(g, i)−k (7.1)

for g ∈ SL2 (R). Then φ∞ = φf,∞ has the properties that

(i) φ∞ (γg) = φ∞ (g) for all γ ∈ SL2 (Z)  
−ikθ cos θ − sin θ
(ii) φ∞ (gr(θ)) = e φ∞ (g) for all r(θ) =
sin θ cos θ
(iii) φ∞ (g) satisfies the slow-growth condition that
     1/2  
 1x 
φ∞ 0 1
y 0
0 y −1/2
r(θ)  ≤ Cy N for some C and N as y → +∞.
 
k k
(iv) Ωφ∞ = − − 1 φ∞ for a suitable normalization of the Casimir oper-
2 2
ator Ω of SL2 (R)
(v) if f is a cusp form, then φ∞ is cuspidal in the sense that

1   
1x
φ∞ 0 1
g dx = 0 for all g.
0

For the lifting to GL2 (Z)\GL2 (R), we start by extending f to all nonreal complex
numbers by putting f (−z) = f (z). Then we define φf,∞ (g) by (7.1) for g ∈ GL2 (R).
The invariance property in (i) extends to be valid for γ ∈ GL2 (Z), properties (ii)
through (v) are unchanged, and there is one new property:
(vi) φ∞ (zg) = φ∞ (g) for all g in the center of GL2 (R).
It is natural to expect at first that SL2 (R) is the better group to use for the above
lifting, but it has become customary to lift to GL2 (R) or the positive-determinant
subgroup rather than SL2 (R) in order to be able to incorporate Hecke operators
more conveniently into the theory.
There are some other classical theories of automorphic forms that can be lifted to
Lie groups in the same way. The theory of Maass forms [Maa] concerns certain non-
holomorphic functions on the upper half plane, and these lift to GL2 (Z)\GL2 (R).
 INTRODUCTION TO THE LANGLANDS PROGRAM 273

A theory [We3] begun by Hecke for analytic functions on the upper half plane
transforming under the group
   
a b 
Γ0 (N ) = c d
∈ SL2 (Z)  N divides c

leads
  functions on GL2 (R) invariant under the group generated by Γ0 (N ) and
to
1 0
0 −1
. In addition, the theory of Hilbert modular forms [Ga] leads to quotients
of products of several copies of GL2 (R), and the theory of Siegel modular forms
[Si] leads to quotients of real symplectic groups.
In each case the theory can be reinterpreted in an adelic setting by means of
(6.3). For example, with classical modular and cusp forms with respect to SL2 (Z),
the isomorphism (6.4c) tells us that f should be lifted to

φf (g) = f (g∞ (i))j(g∞ , i)−k for g ∈ GL2 (AQ )

if g = γg∞ k1 is the decomposition of g ∈ GL2 (AQ ) according to (6.4c) reformulated
as 
GL2 (AQ ) = GL2 (Q)GL2 (R) GLn (Zp ).
p
The function φ = φf on GL2 (AQ ) has the properties that
(i) φ(γg) = φ(g) for all γ ∈ GL2 (Q)
(ii) φ(gk1 ) = φ(g) for all k1 ∈ p GL2 (Zp )
 
−ikθ cos θ − sin θ
(iii) φ(gr∞ (θ)) = e φ(g) for all r∞ (θ) = at the infinite
sin θ cos θ
place

of the variable in the infinite place, φ satisfies the equation
(iv) as a function
Ωφ = − k2 k2 − 1 φ for a suitable normalization of the Casimir operator Ω
of SL2 (R)
(v) φ(zg) = φ(g) for all scalar z in GL2 (AQ )
(vi) φ(g) satisfies the following slow-growth condition: for each c > 0 and
compact subset ω of GL2 (AQ ), there exist constants C and N such that
  
a0
φ 0 1
g ≤ C|a|N

for all g ∈ ω and a ∈ A×Q with |a|AQ > c

(vii) if f is a cusp form, then φ is cuspidal in the sense that

  
1x
φ 0 1
g dx = 0 for all g ∈ GL2 (AQ ).
Q\AQ

The group SL2 (Z), relative to which f satisfies an invariance property, is captured
by the compact group in (ii). The relevant identity is

 
GL2 (Q) ∩ GL2 (R) × GL2 (Zp ) = GL2 (Z).
p

For Hecke’s
 theory with Γ0 (N ), the corresponding compact group that appears in
(ii) is p Kp , where
   

Kp = a b
c d
∈ GL 2 (Zp  c/N ∈ Zp .
)
274 A. W. KNAPP

The group Kp coincides with GL2 (Zp ) for all p prime to N , and the relevant identity
is  

 
Kp = Γ0 (N ) ∪ 0 −1 Γ0 (N ).
1 0
GL2 (Q) ∩ GL2 (R) ×
p

This adelic setting is what we generalize to arbitrary reductive groups, following

[Bo-Ja]. Let F be a number field, let A = AF be the adeles of F , and let G be
a reductive group over F such that G(C) is connected. Let Z be a maximal F
split torus of the center of G. Let G∞ = G(F∞ ) be the archimedean component of
G(A), and let G(Af ) be the nonarchimedean component, so that

G(A) = G∞ × G(Af ). (7.2)

Let

K∞ = a maximal compact subgroup of the Lie group G∞

g = complexification of the (real) Lie algebra of G∞
U (g) = universal enveloping algebra of g
Z(g) = center of U (g).

 
Let K1 be the open compact subgroup G v finite Ov of G(Af ).
A complex-valued function f on G(A) is smooth if it is continuous and, when
viewed as a function of two arguments (x, y) as in (7.2), it is smooth in x for each
fixed y and is locally constant of compact support in y for each fixed x. Let ρ
be a finite-dimensional representation of K∞ , let J ⊂ Z(g) be an ideal of finite
codimension, and let K be an open subgroup of K1 . A smooth function f on G(A)
is automorphic relative to (ρ, J, K) if
(i) f (γg) = f (g) for all γ ∈ G(F )
(ii) f (gk) = f (g) for all k ∈ K
(iii) the span of the right translates of f by members of K∞ is finite-dimensional,
and every irreducible constituent of this representation of K∞ is a con-
stituent of ρ
(iv) the ideal J, acting in the G∞ variable of (7.2), has Jf = 0
(v) for each y ∈ G(Af ), the function x → f (xy) on G∞ satisfies a certain
slow-growth condition.
We shall not make (v) any more precise, but instead we refer to [Bo-Ja]; Theorem
7.3 below will clarify condition (v) in the principal case of interest. The set of
automorphic functions relative to (ρ, J, K) will be denoted A(ρ, J, K).
When G = GL1 , any Grossencharacter gives an example of an automorphic form
relative to a suitable triple.
Theorem 7.1. For every (ρ, J, K), A(ρ, J, K) is finite-dimensional.
This theorem is fundamental. It is due to Harish-Chandra; see [HC1] and [HC2,
p. 8], where it is proved in the setting of the right side of (6.3). The translation
into the current setting (which is the left side of (6.3)) is in [Bo-Ja, p. 195].
Theorem 7.2. Let a smooth function f on G(A) satisfy (i) through (iv) above
and also
f (zx) = χ(z)f (x) for all z ∈ Z(A) and x ∈ G(A) (7.3)
 INTRODUCTION TO THE LANGLANDS PROGRAM 275

for some (unitary) character of Z(F )\Z(A), so that |f | may be regarded as a func-
tion on (Z(A)G(F ))\G(A). If |f | is in Lp ((Z(A)G(F ))\G(A)) for some p ≥ 1,
then f satisfies condition (v) and hence is an automorphic form.
Reference. See [Bo-Ja, pp. 191 and 195]. The proof makes use of Theorem
6.2 and [HC2].

A cusp form is an automorphic form f such that (7.3) holds for some unitary
character χ of Z(F )\Z(A) and such that


f (ng) dn = 0 (7.4)
N (F )\N (A)

for the unipotent radical N of every proper parabolic subgroup of G and for all
g ∈ G(A). For G = GL1 , the condition (7.4) is empty, and therefore all unitary
Grossencharacters are cusp forms for GL1 . The classical analytic cusp forms relative
to SL2 (Z) yield cusp forms for G = GL2 in the sense of (7.4), and so do Hecke’s
cusp forms relative to the subgroup Γ0 (N ) of SL2 (Z). For general G, let 0A(ρ, J, K)
be the space of cusp forms relative to (ρ, J, K).
Theorem 7.3. Let a smooth function f on G(A) satisfy (i) through (iv) above,
as well as the cuspidal condition (7.4) and the condition (7.3) for some (unitary)
character of Z(F )\Z(A). Then the following conditions are equivalent:
(i) f satisfies (v) and hence is a cusp form
(ii) f is bounded
(iii) |f | is in L2 ((Z(A)G(F ))\G(A)).

Actually there is a sharper result: A smooth function f satisfying (i) through

(iv) of an automorphic form, as well as the slow-growth condition in (v) and the
condition (7.3) for some unitary χ, is automatically rapidly decreasing as soon as it
satisfies the cuspidal condition (7.4). We already noted this fact for classical cusp
forms relative to SL2 (Z). The result for general G requires precise definitions of
“slow growth” and “rapidly decreasing,” which we omit.
We want to define the notion of an automorphic representation of G(A). Put
A = A(ρ, J, K). The idea is that an automorphic representation is any irreducible
subquotient of A, but the trouble is that A need not be mapped to itself under
right translation by G(A). Specifically, K∞ finiteness need not be preserved under
right translation by G∞ .
The idea is to make A into a module for an algebra H (the Hecke algebra)
that reflects the action by G(Fv ) for each finite place v and reflects the action
by U (g) and K∞ at the infinite places. We summarize the construction, which
is given in more detail in [Fl]. For each finite place v, let Hv be the algebra
(under convolution) of all complex-valued locally constant functions of compact
support on G(Fv ). Haar measure on G(Fv ) is to be normalized so that G(Ov ) has
measure 1, and then the characteristic function Iv of G(Ov ) is an idempotent in
Hv . The algebra Hv contains a directed system of further idempotents, namely the
normalized characteristic function of each open subgroup of G(Ov ). An Hv module
is approximately unital if, for each member of the module, all idempotents
corresponding to sufficiently small open subgroups act as the identity. It is fairly
easy to see that smooth representations of G(Fv ), i.e., those in which each vector is
fixed by some compact open subgroup, correspond exactly to approximately unital
276 A. W. KNAPP

Hv modules. Such a representation is called admissible if the set of vectors fixed

by any compact open subgroup is finite-dimensional.
There is a natural way of forming a restricted tensor product of the algebras
Hv with respect to the idempotents Iv . The resulting algebra Hf is the part of H
corresponding to the finite places of F and is generated by product functions that
equal Iv at almost every place. A tuple of local idempotents, one for each Hv with
almost all of them being Iv , yields another idempotent in Hf , and the idempotents
obtained in this way are indexed by a directed set. A right Hf module is smooth
if each member of the module is fixed by all idempotents corresponding to members
of the directed set that are sufficiently large. The module is admissible if the set
of vectors fixed by any of these idempotents is finite-dimensional.
Next we let H∞ be the convolution algebra of all K∞ finite distributions on
G∞ that are supported on K∞ . This algebra is studied extensively in [Kna-Vo]. It
contains a directed family of idempotents as follows. Let dk denote normalized Haar
measure on K∞ . For each class of irreducible representations τ of K∞ , let χτ be the
character and let dτ be the degree. The directed family of idempotents is indexed
by all finite subsets of τ ’s, the idempotent corresponding to a given set being the
sum of dτ χτ for all τ in the set. A right H∞ module is approximately unital if,
for each member of the module, all sufficiently large idempotents act as the identity.
It is shown in [Kna-Vo, pp. 75 and 90] that (g, K∞ ) modules correspond exactly
to approximately unital H∞ modules. Such a module is admissible if the set of
vectors fixed by any of these idempotents is finite-dimensional, i.e., if each K∞ type
has finite multiplicity.
We define H = H∞ ⊗ Hf . Smoothness and admissibility of right H modules
are defined using idempotents that are pure tensors from H∞ and Hf . Then A
is a smooth right H module. An automorphic representation of H is any
irreducible subquotient of A. Similarly if we put 0A = 0A(ρ, J, K), then a cuspidal
automorphic representation of H is any irreducible subquotient of 0A.
If f is an automorphic form, then it is immediate from Theorem 7.1 that f ∗H is a
smooth admissible H module. It follows that every automorphic representation of H
is smooth and admissible. Such representations are commonly called automorphic
representations of G(A) even though not all of G(A) really acts.
More generally a topologically irreducible G(A) module is said to be automor-
phic if its underlying space of smooth vectors is an automorphic representation of
H. According to [Fl, Theorem 4], if χ is any (unitary) character of Z(F )\Z(A),
then any G(A) invariant irreducible closed subspace of

L2 (G(F )\G(A))χ
 !
= f  |f | ∈ L2 ((Z(A)G(F ))\G(A)) and f (zx) = χ(z)f (x) for z ∈ Z(A), x ∈ G(A)
is automorphic in this sense.
Theorem 7.4. The subspace of cuspidal functions in L2 (G(F )\G(A))χ decom-
poses discretely with finite multiplicities. Consequently whenever f is a cusp form,
f ∗ H is a finite direct sum of cuspidal automorphic representations.
Reference. This theorem is due to Gelfand and Piatetski-Shapiro [Gelf-Pi].
See [Gelb1, p. 33] for a discussion when G = GL2 , and see [HC2, p. 9] for the
general case.
 INTRODUCTION TO THE LANGLANDS PROGRAM 277

Remark. It follows from the theorem that cuspidal automorphic representa-

tions are unitarizable. That is, they are the underlying smooth representations for
irreducible unitary representations of G(A).

Hecke’s cusp forms discussed earlier in the section lead to cusp forms in the adelic
sense, by Theorem 7.3. Those whose L function has an Euler product expansion
lead to adelic cusp forms that generate single (irreducible) cuspidal automorphic
representations.

8. Langlands Theory for GLn

In this section we describe what the Langlands theory proposes for GLn . The
theory for G = GLn may be regarded as a special case of the general theory, which
will be discussed in the next section.
Fix a number field F . The theory wants to associate to each automorphic rep-
resentation of GLn (AF ) an L function given initially as a product of elementary L
factors, one for each place of F . The method is arranged to be a direct generalization
of the way in which an L function is attached to a Grossencharacter in (5.2), the
case of a unitary Grossencharacter being the special case n = 1. At the same time
it generalizes the way in which an L function is attached to a classical cusp form,
except that the starting point is the Euler product expansion and not the series
expansion; this situation is an instance of the special case n = 2 of the Langlands
theory. For general n and a given automorphic representation of GLn (AF ), the
Langlands theory gives a precise definition of the elementary L factors at almost
every place, and their product is convergent in a half plane. The definition at the
remaining finite set of places hinges on a conjecture known as the Local Langlands
Conjecture, which we shall discuss shortly.
By a theorem of Flath [Fl], any irreducible smooth admissible representation π of
GLn (A"F ) (or, more precisely, of the Hecke algebra) is a “restricted tensor product”
π = v πv of irreducible smooth admissible representations πv of the respective
factors GLn (Fv ) of GLn (AF ). To define the L function of π, it is therefore enough
to describe the elementary L factor of the representation πv of GLn (Fv ).
Thus let k be a local field of characteristic 0, and let π be an irreducible represen-
tation of GLn (k). Approximately, the theory proposes attaching to π a continuous
homomorphism ϕ : Wk → GLn (C) with certain properties. Here Wk is the Weil
group of k. Then the L factor for π is defined to be a certain elementary function
attached to ϕ that generalizes any of the factors on the right side of (5.2).
The above description is correct if k is archimedean, but the use of the Weil
group Wk is insufficient if k is nonarchimedean. For reasons that we illustrate by
example later in this section, the group Wk in the nonarchimedean case is replaced
by the Weil-Deligne group Wk , which we shall define now.
Let θk : k × → Wkab be the local reciprocity map of Corollary 2.5. Let w denote
the effect on w ∈ Wk of the composition of passing from Wk → Wkab , followed by
passing to k × by θk−1 , followed by passing to R×
+ by | · |k . The Weil-Deligne group
Wk is the semidirect product of Wk and C, where Wk acts on C by wxw−1 = wx.
Thus the multiplication rule on C × Wk is (a1 , w1 )(a2 , w2 ) = (a1 + w1 a2 , w1 w2 ).
In the nonarchimedean case, a continuous homomorphism ϕ : Wk → GLn (C)
is called an admissible homomorphism if ϕ is holomorphic in the C vari-
able, if ϕ(C) consists of unipotent matrices, and if ϕ(Wk ) consists of semisimple
matrices. Such a homomorphism amounts to specifying a pair (ρ, X), where
278 A. W. KNAPP

ρ : Wk → GL(n, C) is a continuous homomorphism such that ϕ(Wk ) consists

of semisimple matrices and X is a nilpotent endomorphism of Cn such that
ρ(w)Xρ(w)−1 = wX for all w ∈ Wk . Let Fr be an element of Wk such that
Fr = q −1 , where q is the order of the residue field; the element Fr is uniquely
determined modulo the inertia group Ik of (1.4). If VXI is the subspace of ker X
fixed by ρ(Ik ), then the elementary L factor associated to ϕ is4
L(s, ϕ) = det(1 − (ρ(Fr)|VXI )q −s )−1 . (8.1)

Parenthetically let us mention a substitute for the Weil-Deligne group that one
encounters in the literature. With the right definition the admissible homomor-
phisms into GLn (C) for the substitute group will correspond to the admissible
homomorphisms of Wk , and ultimately no semidirect products will be involved
in
 the1/2 definition of the substitute group. For w ∈ Wk ,  let hw be the matrix
w 0 1 z
in SL2 (C), and identify z ∈ C with . The above
0 w−1 0 1  
1 z
action of Wk on C by automorphisms translates into conjugation of by
0 1
hw . Since conjugation by hw extends to an automorphism of all of SL2 (C), we can
identify the action of Wk on C with a subaction of the action of Wk on SL2 (C)
by automorphisms. The semidirect product of SL2 (C) by Wk with respect to
this action then consists of pairs (x, w) with x ∈ SL2 (C) and w ∈ Wk , where
(x1 , w1 )(x2 , w2 ) = (x1 hw1 x2 h−1
w1 , w1 w2 ). This group is isomorphic to the direct
product SL2 (C) × Wk by (x, w) → (xhw , w), and SL2 (C) × Wk is then used as the
substitute for the Weil-Deligne group. We shall not use this substitute, however.
In the archimedean case, an admissible homomorphism ϕ : Wk → GLn (C) is
just a continuous homomorphism such that ϕ(Wk ) consists of semisimple matrices.
The elementary L factor associated to such a ϕ is a nowhere-zero meromorphic
function involving Γ functions and may be found in [Kna2, p. 404].
Two admissible homomorphisms ϕ1 and ϕ2 are said to be equivalent if they are
conjugate via GLn (C), i.e., if there exists g ∈ GLn (C) with gϕ1 (x)g −1 = ϕ2 (x) for
all x in the Weil-Deligne group or Weil group, as appropriate. The set of equivalence
classes of admissible homomorphisms is denoted Φ(GLn (k)).
For each local field k of characteristic 0, let Π(GLn (k)) be the set of equiv-
alence classes of (smooth) irreducible admissible representations of GLn (k). If
k is nonarchimedean, “equivalence” here means equivalence as representations of
the group (or of the Hecke algebra). If k is archimedean, “equivalence” means
infinitesimal equivalence (or equivalence as representations of the Hecke algebra).
Leaving aside some further definitions for the moment, we can state the Local
Langlands Conjecture as follows.
Local Langlands Conjecture. Π(GLn (k)) is indexed by Φ(GLn (k)) in a
natural way that is compatible with twisting by Grossencharacters and respects L
factors and ε factors.
For ϕ ∈ Φ(GLn (k)) and π ∈ Π(GLn (k)), the “twists” ϕ ⊗ α and π ⊗ α by a
Grossencharacter α are defined toward the end of this section. The requirement
about twists is that if ϕ corresponds to π, then ϕ ⊗ α corresponds to π ⊗ α for every
α.
4 See the footnote for (5.1a).
 INTRODUCTION TO THE LANGLANDS PROGRAM 279

Let us discuss the requirement on L factors and ε factors. We can define the
Langlands elementary L factor of an irreducible π to be the elementary L factor
of the corresponding ϕ:

L(s, π) = L(s, ϕ) if ϕ ↔ π, (8.2)

with L(s, ϕ) as in (8.1). In Jacquet’s lecture [Ja2] may be found a completely

different construction of elementary L factors attached to irreducible admissible
representations (due to Godement and Jacquet [Go-Ja]), and this construction
does not depend on any conjectures. We require that the two kinds of L factors
match. Properties of the Godement-Jacquet L functions will be discussed below
after Corollary 8.6.
We mentioned ε functions attached to Galois representations in connection with
Theorem 5.3. These are products of local ε factors that we have not defined, are 1
at almost every place, and generalize the Gauss sums that occur in the functional
equations of Dirichlet L functions [Kna1, p. 216]. Local ε factors depend upon
additional data, but let us suppress this point. In a fashion similar to that for
Galois representations, a local ε factor may be associated to each ϕ ∈ Φ(GLn (k)).
The correspondence of the Local Langlands Conjecture then allows one to define a
Langlands ε factor for π ∈ Π(GLn (k)) by

ε(s, π) = ε(s, ϕ) if ϕ ↔ π.

Meanwhile the Godement-Jacquet construction also defines an ε factor for each π

in Π(GLn (k)). We require that the two constructions match.
The paper [Kud, p. 380] gives a full list of the requirements that the conjectural
local Langlands correspondence is supposed to satisfy.
The Local Langlands Conjecture is a theorem when n = 1. In this case the result
amounts to a restatement of Corollary 2.5. For general n and for k archimedean,
the conjecture is a theorem of Langlands; see [Kna2] for an exposition. Much is
known in the nonarchimedean case. See [Kud] for an exposition. The conjecture is
known for n = 2 ([Kut] and [Tu]), for n = 3 ([Hen1]), and for all relatively prime
n and p [Moy]. For more recent work, see [Ha]. Henniart [Hen2] has shown that
there is at most one candidate for the local Langlands correspondence respecting ε
factors for “pairs” in the sense of [Ja-P-S]. See [Kud, §4.2] for more detail.
A part of the correspondence is easy to understand. For definiteness, let k be
nonarchimedean. Let ρ be a continuous homomorphism of Wk into the diagonal
subgroup of GLn (C). Composing with projection to each diagonal entry, we see
that ρ is completely determined by n quasicharacters χ1 , . . . , χn of Wkab ∼
= k × . Let
us write ρ = ρ(χ1 ,...,χn ) . If we put X = 0, then

ϕ = ϕ(χ1 ,...,χn ) = (ρ(χ1 ,...,χn ) , 0)

is an admissible homomorphism. Let T be the diagonal subgroup of GLn (k), and let
N be the upper-triangular subgroup with 1’s on the diagonal. The data (χ1 , . . . , χn )
give us a quasicharacter of T , and it is natural to associate to ϕ(χ1 ,...,χn ) the induced
representation
GL (k)
indT Nn ((χ1 , . . . , χn ) ⊗ 1) (8.3)
given by normalized induction (a member of the nonunitary principal series).
This representation may be taken to be the π that corresponds to ϕ when π is
280 A. W. KNAPP

irreducible, and it is known that irreducibility occurs if and only if there is no pair
of indices i and j such that χi χ−1
j equals | · |k .
When (8.3) is reducible, the π that is associated to ϕ(χ1 ,...,χn ) is a certain irre-
ducible subquotient of (8.3) known as the Langlands subquotient. To describe
the Langlands subquotient, we first remark that the set of irreducible subquotients
of (8.3) is unchanged (apart from equivalence) when χ1 , . . . , χn are permuted. In
order to have the indexing Φ ↔ Π depend only on equivalence classes, we require
that the Langlands subquotient not be affected by permutation. Next, we can
introduce complex numbers s1 , . . . , sn such that |χi | = | · |ski . The numbers si
are not unique, but their real parts are unique. If the real parts of s1 , . . . , sn are
nonincreasing, then the Langlands subquotient is the (unique) irreducible quotient
of (8.3). The result is that the Langlands subquotient is determined in every case.
See [Moe2] in this volume for further discussion.
A special case of the nonunitary principal series of particular interest is the
unramified principal series, those members of the nonunitary principal series
having a nonzero fixed vector under GLn (Ok ). By Frobenius reciprocity these are
just the representations (8.3) for which each χi (x) depends only on the module of
x. Thus each χi is of the form | · |si for some complex si . The Langlands sub-
quotient of an unramified principal series is the irreducible subquotient containing
a nonzero vector fixed by GLn (Ok ). All members of Π(GLn (k)) having a nonzero
vector fixed by GLn (Ok ) are of this form, up to equivalence. Up to equivalence,
they are parametrized by orbits of the symmetric group on tuples (s1 , . . . , sn ) of
complex numbers modulo 2πi(log q)−1 Z. In view of (8.2), the elementary L factor
of Langlands is well defined in the case of a member of Π(GLn (k)) having a nonzero
vector fixed by GLn (Ok ). Such a member of Π(GLn (k)) is said to be unramified.
Let us now discuss the need for the Weil-Deligne group Wk rather than just the
Weil group Wk . Suppose that n = 2. The irreducible admissible representations of
GL2 (k) are then of three kinds—the supercuspidal representations (those whose
matrix coefficients are compactly supported modulo the center), the Langlands
subquotients of the nonunitary principal series, and the special representations.
In (8.3), reducibility occurs for GL2 (k) exactly when χ1 χ−1 2 = | · |±1
k . In this
case, there are two irreducible subquotients, the Langlands subquotient and one
other. These “other” representations are the special representations. Their matrix
coefficients are square integrable modulo the center, but these representations are
not supercuspidal.
For n = 2, the Godement-Jacquet L factors of supercuspidal representations are
1, of special representations involve one factor with q −s in it, and of Langlands
subquotients of nonunitary principal series involve two factors with q −s in them.
For the Local Langlands Conjecture to be valid, it is necessary to arrange for some
other two-dimensional representations (of Wk or a substitute) than the irreducible
ones (which give 1 as L factor) and the direct sums of one-dimensional ones (which
give L factors that are the product of two expressions involving q −s ). Englargement
to the Weil-Deligne group allows the existence of indecomposable yet reducible two-
dimensional representations of Wk . For one of these representations, the space is
C2 with a basis e1 , e2 and with action
ρ(w)ei = wi−1 ei , Xe1 = e2 , Xe2 = 0.
The most general indecomposable yet reducible two-dimensional representation of
Wk , up to equivalence, is the tensor product of this one with a one-dimensional
 INTRODUCTION TO THE LANGLANDS PROGRAM 281

representation. The equivalence classes of indecomposable two-dimensional repre-

sentations are used to parametrize the special representations up to equivalence.
Use of Wk to settle GL2 (k) looks somewhat artificial, but Wk looks more rea-
sonable when one considers what is known about GLn . For GLn (k) with k nonar-
chimidean, the irreducible admissible representations have been classified. The
starting point is the supercuspidal representations, which have been classified by
Bushnell-Kutzko and by Corwin. Bernstein and Zelevinski showed how to classify
the irreducible admissible representations in terms of the supercuspidal represen-
tations. See [Kud] and [Moe2] for more detail and for references. The nature of
the Bernstein-Zelevinski part of the classification implies that if the supercuspidal
representations of GLn (k) correspond to the n-dimensional irreducible admissible
homomorphisms Wk → GLn (C) of the Weil group Wk for all n, then Π(GLn (k)) is
parametrized by the n-dimensional admissible homomorphisms Wk → GLn (C) of
the Weil-Deligne group Wk . (See [Ja1, (3.7)] and"[Kud, p. 381].)
Now we return to the number field F . Let π = v πv be an irreducible admissible
representation of GLn (AF ). Then we can define the Langlands L functions by

L(s, π) = L(s, πv )
finite v
 (8.4)
Λ(s, π) = L(s, πv ),
all v

where the factors are given by (8.2) and ultimately

"(8.1). The first question is one of
convergence. Built into the factorization of π as v πv is the following addendum,
which simplifies questions of convergence greatly.
"
Theorem 8.1. If π = v πv is an irreducible admissible representation of
GLn (AF ), then almost every πv is unramified, i.e., has a nonzero fixed vector under
GLn (Ov ) and is therefore the Langlands subquotient of an unramified principal
series.
Reference. [Fl, p. 181].

Corollary 8.2. If π is a unitarizable irreducible admissible representation of

GLn (AF ), then L(s, π) converges absolutely for Re s sufficiently large.
Reference. This is essentially due to Langlands. See [Bo4, p. 50].

In fact, Theorem 8.1 shows that it is enough to consider the factors of π that
are unramified principal series. Flath’s results show that we may take each factor
to be unitarizable, and then all that is needed is an estimate on the L factor for a
unitarizable unramified principal series.
Corollary 8.3. If π is a cuspidal automorphic representation of GLn (AF ), then
L(s, π) converges absolutely for Re s sufficiently large.
Reference. This kind of result is due to Langlands [Lgl2] and predates [Fl].
The proof is immediate from the above results: The cuspidal representation is
unitarizable by Theorem 7.4, and convergence follows from Corollary 8.2.

With considerably more effort, Langlands has addressed convergence of L(s, π)

for general automorphic π. We need another corollary of Theorem 8.1.
282 A. W. KNAPP

Corollary 8.4. Let P = M N be the usual Levi decomposition of a standard

parabolic subgroup of GLn . Let σ be a unitarizable irreducible admissible represen-
GL (AF )
tation of M (AF ), and let π be an irreducible subquotient of indM (An F )N (AF ) (σ ⊗ 1).
Then L(s, π) converges absolutely for Re s sufficiently large.
Reference. [Bo4, p. 52].

Theorem 8.5. An irreducible admissible representation π of GLn (AF ) is auto-

morphic if and only if π is equivalent with an irreducible subquotient of
GL (A )
indM (An F )N (AF ) (σ ⊗ 1)
F

for some cuspidal automorphic representation σ of M (AF ).

Reference. [Lgl4, p. 204].

Corollary 8.6. If π is an automorphic representation of GLn (AF ), then L(s, π)

converges absolutely for Re s sufficiently large.
Reference. This result of Langlands combines the above ideas. See [Bo4,
p. 52]. By Theorem 8.5, π is a constituent of a representation induced from a
cuspidal representation, and the cuspidal representation is unitarizable by Theorem
7.4. Then convergence follows from Corollary 8.4.

The next question concerns the analytic properties of automorphic representa-

tions. We shall be content with the results in the cuspidal case. For the general
automorphic case, see [Ja1, p. 83]. As we mentioned above, Godement and Jacquet
[Go-Ja] have constructed L and Λ functions in a way that is completely different
from using the Local Langlands Conjecture. Their theory is summarized in this
volume in [Ja2], and conjecturally their functions, which we denote LGJ (s, π) and
ΛGJ (s, π), coincide with those in (8.4). What is known is that the Godement-
Jacquet L functions agree with the Langlands L functions at every place where πv
is unramified. (This condition holds for almost every place by Theorem 8.1.) The
Godement-Jacquet L functions have good global analytic properties, as follows.
Theorem 8.7. If π is a cuspidal representation of GLn (AF ), the Godement-
Jacquet function ΛGJ (s, π) extends to be meromorphic in C with singularities given
at most by simple poles at s = 0 and s = 1. Moreover, ΛGJ (s, π) satisfies a
functional equation of the form
ΛGJ (s, π) = εGJ (s, π)ΛGJ (1 − s, π ∨ ),
where π ∨ is the contragredient and εGJ (s, π) is a multiple of N −s for some integer
N . The function ΛGJ (s, π) is actually entire unless π is a unitary character of
GL1 (AF ) trivial on the elements of module 1.
When n = 1, the L functions in Theorem 8.7 (with the elementary factors from
the infinite places dropped) are the ones attached to unitary Grossencharacters
by Hecke and Tate, and Theorem 8.7 generalizes the work in Tate’s thesis [Ta1].
When n = 2, these L functions (after a change of parameter) generalize the L
functions attached by Hecke to cusp forms relative to SL2 (Z); Hecke proved that
his L functions are entire and satisfy a functional equation.
 INTRODUCTION TO THE LANGLANDS PROGRAM 283

An important aspect of the Langlands theory for GLn is a reciprocity conjecture

that implies the Artin Conjecture (end of §5). We state this conjecture of Langlands
in two forms—the first version involving almost all places and either Langlands L
functions or Godement-Jacquet L functions, the second version involving all places
and Godement-Jacquet L functions. For the second version we need to complete
the Artin L function L(s, σ) to a function Λ(s, σ) by adjoining elementary L factors
for the infinite places; these factors are listed explicitly in [Kna2, (3.6) and (4.6)].
Langlands Reciprocity Conjecture, first version. For any irreducible rep-
resentation σ of Gal(F /F ) in GLn (C), there exists a cuspidal automorphic rep-
resentation π of GLn (AF ) such that the Artin L function of σ agrees with the
Langlands L function of π at almost every place where π is unramified.
Langlands Reciprocity Conjecture, second version. For any irreducible
representation σ of Gal(F /F ) in GLn (C), there exists a cuspidal automorphic rep-
resentation π of GLn (AF ) such that the completed Artin L function Λ(s, σ) of σ
is identical with the Godement-Jacquet L function ΛGJ (s, π) of π. In more detail,
Lv (s, σ) = LGJ GJ
v (s, π) for every finite place v, and L∞ (s, σ) = L∞ (s, π) if L∞ and
LGJ
∞ denote the products of the factors for the infinite places.
"
If π = v πv , then we have noted that the Langlands L factor for πv agrees
with the Godement-Jacquet L factor for πv when πv is unramified and that πv is
unramified at almost every place. Therefore the second version of the conjecture
implies the first. The converse is addressed by the following.
Theorem 8.8. The first version of the Langlands Reciprocity Conjecture for σ
implies the second for σ. Consequently either version of the Langlands Reciprocity
Conjecture for σ implies the Artin Conjecture (end of §5) for σ.
We postpone the discussion of Theorem 8.8 to the end of this section except to
note that the second sentence of the theorem follows by combining Theorem 8.7
and the second version of the conjecture.
Let us discuss the conjecture itself. The basic case for which this conjecture
is known is Theorem 5.1, which handles n = 1. That theorem says that abelian
Artin L functions are Hecke L functions of Grossencharacters. On the one hand,
this result is essentially equivalent with Artin reciprocity. On the other hand, it
establishes that the pattern of ϕ℘ (π℘ ), which governs the factorization of certain
polynomials modulo ℘, is given in terms of arithmetic progressions in the abelian
case. The general conjecture may therefore be regarded as a statement about
reciprocity on the one hand and a statement about the pattern of factorization of
polynomials modulo ℘ on the other hand.
It is natural to expect that the Langlands Reciprocity Conjecture holds for σ =
ind ρ whenever it holds for ρ and ind ρ is irreducible. But it is an open problem
to prove such a statement, even for ρ one-dimensional. In the special case that
ρ is one-dimensional and the induced representation is two-dimensional, Jacquet
and Langlands [Ja-Lgl] did manage to prove the statement, and it follows that the
Langlands Reciprocity Conjecture holds when n = 2 for σ of “dihedral type” in the
sense of [Ro2, §16].
In fundamental work [Lgl5], Langlands handled some additional cases when n =
2, and later Tunnell [Tu] was able to deduce an improved result using the methods
of Langlands. These results of Langlands and Tunnell are the subject of the lectures
[Ro2] by Rogawski. The statement is as follows.
284 A. W. KNAPP

Theorem 8.9. If σ is a two-dimensional complex representation of Gal(F /F )

with solvable image, then the Langlands Reciprocity Conjecture holds for σ.
For further work with n = 2, see [Buhl] and [Fre].
Any initial attempt at proving the Langlands Reciprocity Conjecture raises the
" have irreducible admissible representations πv for each
following question: If we
place v of F and if π = v πv is well defined, how do we tell whether π is automor-
phic? In other words, how is the global behavior relative to GLn (F ) reflected in
the system of representations πv ? Theorem 8.1 gives a necessary condition; almost
all πv must be Langlands subquotients of unramified principal series. What else
can be said?
We are especially
" interested in the cuspidal case. If we are given an irreducible
admissible π = v πv , the Multiplicity One Theorem below says that there is
at most one way that π can occur as a cuspidal representation. Recall from Theorem
7.4 that the cuspidal part of L2 (GLn (F )\GLn (AF ))χ decomposes discretely with
finite multiplicities.
Theorem 8.10. If π is an irreducible admissible representation of GLn (AF ) oc-
curring in the cuspidal part of L2 (GLn (F )\GLn (AF ))χ for some unitary character
χ, then π occurs with multiplicity one.
Reference. [Sha].
" "
For two Grossencharacters χ = v χv and χ = v χv , an equality χv = χv for
almost all v implies χ = χ . In fact, we may assume that χv = 1 for almost all v.
Applying the Weak Approximation Theorem (Theorem 3.7) and using that χ is 1
on F × and is continuous, we see that χ equals 1 everywhere. The following Strong
Multiplicity One Theorem generalizes this result from GL1 to GLn .
" "
Theorem 8.11. Let π = v πv and π  = v πv be irreducible admissible rep-
resentations of GLn (AF ) occurring in the cuspidal part of L2 (GLn (F )\GLn (AF ))χ
for some unitary character χ. If πv is equivalent with πv for almost all v, then
π = π .
Reference. [Ja-Sha, p. 553].

We do not conclude merely that π and π  are equivalent. Theorem 8.10 allows
us to deduce equality from equivalence. It follows from Theorem 8.11 that the
cuspidal automorphic representation π in the Langlands Reciprocity Conjecture is
unique if it exists.
Thus cuspidal representations are rather rigid. But how do we tell when we have
one? When F = Q, we can phrase the question in a related way. If L(s) is a Dirichlet
series obtained by expanding out a product of elementary factors as in (8.1), when
is L(s) the L function of a cuspidal automorphic representation? Weil [We2],
generalizing
∞ work of Hecke, answeredthe question about when a Dirichlet series
s ∞
n=1 a n /n comes from a cusp form n=1 an e2πinτ for some Γ0 (N ), and Jacquet-
Langlands [Ja-Lgl] answered the corresponding question about general cuspidal
representations of GL ∞ 2 . Weil’s result, known as the Weil Converse Theorem, says
that if every twist n=1 an χ(n)/ns by a primitive Dirichlet character χ modulo
r with r prime to N extends to an entire function ∞bounded in vertical strips and
satisfying a suitable functional equation, then n=1 an e2πinτ is a cusp form for
Γ0 (N ).
 INTRODUCTION TO THE LANGLANDS PROGRAM 285

The Jacquet-Langlands result [Ja-Lgl, p. 397] below uses the Weil group WF of
the number field F , a notion treated in [Ta3] whose precise definition we do not
need. We list the properties of WF that we shall use:
1) There is a natural continuous homomorphism WF → Gal(F /F ), and it of
course induces a homomorphism WFab → Gal(F /F )ab .
2) The Artin map F × \A×F → Gal(F /F )
ab
factors as
∼
F × \A×
F −
−−→ WFab −−−→ Gal(F /F )ab .

3) There is a canonical continuous homomorphism jv : WFv → WF such that

the diagram
jv
WFv −−−−→ WF
 
 
 
Gal(F v /Fv ) −−−−→ Gal(F /F )
commutes.
∼ × ∼
4) The isomorphisms F × \A× F → WF of (2) and Fv → WFv of Corollary 2.5
ab ab

are such that the diagram

Fv× −−−−→ F × \A×

F
 
 
 
WFabv −−−−→ WFab

commutes. (This strengthens Theorem 4.5.)

The Jacquet-Langlands Converse Theorem also uses Weil’s generalization of Artin
L functions in which a representation σ of the Galois group Gal(F /F ) is replaced
by an admissible homomorphism of the Weil group WF into some GLn (C). The
local definition is in (8.1), and (3) says that it is consistent with Artin’s definition.
Further properties of the Weil L functions L(s, σ) and Λ(s, σ) are given in [Ja-Lgl,
pp. 393–394]. Like Artin L functions, the Weil L functions are known to have
meromorphic continuations and to satisfy a functional equation (5.4). Thus a
hypothesis about a functional equation need not be included in the theorem.
To state the theorem we need to explain a notion of twisting in this context.
Let σ be an n-dimensional representation of Gal(F /F ) or, more generally as a
consequence of (1) above, of the Weil group WF of the number field F . Let α be a
Grossencharacter.
The twist σ ⊗ α will be an admissible n-dimensional representation of WF . To
define σ ⊗ α, all we have to do is interpret α as a one-dimensional representation of
WF . This we can do since α is a one-dimensional representation of F × \A× F and this
group is isomorphic with WFab by (2). Thus σ ⊗α is meaningful, and so are the Weil
L function
" L(s, σ ⊗ α) and its completed version Λ(s, σ ⊗ α). The decomposition
α= αv is unambiguous in the two interpretations as a consequence of (4).
Theorem 8.12 (Jacquet-Langlands Converse Theorem). Suppose that σ is a
two-dimensional representation of WF , and suppose that π(σv ) is the irreducible
admissible representation of GL2 (Fv ) corresponding to σ ◦ jv . If, for every unitary
Grossencharacter α, the Weil L functions Λ(s, σ ⊗ α) and Λ(s, σ ∨ ⊗ ω −1 ) are
286 A. W. KNAPP

"
entire functions bounded in vertical strips, then v π(σv ) is a cuspidal automorphic
representation of GL2 (AF ).
A corresponding theorem is known for GL3 . See [Pi] and [Co-Pi] for results for
GLn with n ≥ 4, where additional hypotheses are needed. Converse theorems play
a role in the work of Langlands on the Artin Conjecture; see Rogawski’s lectures
[Ro2]).
Let us return to Theorem 8.8. A proof for n = 2 appears in [Lgl5, pp. 23–24].
See also [De-S]. The result for general n appears to be a folk theorem, with no proof
appearing in the literature. We are grateful to Dinakar Ramakrishnan for supplying
us with the proof that follows and for giving permission that it be reproduced here.
Let σ be an n-dimensional representation of Gal(F /F ), and let π be a cuspidal
automorphic representation of GLn (AF ), with σ and π as in the first version of
the Langlands Reciprocity Conjecture. We write σv for the restriction of σ to
Gal(F v /Fv ). The same representation π of GLn (AF ) will occur in the two versions
"
of the conjecture, as it must by Theorem 8.11, and we decompose π as π = πv .
If τ is any finite-dimensional representation of WFv , we say that τ is ramified
if the restriction of τ to the inertia group in (1.4) is nontrivial. If τ is unramified,
τ descends to a representation of the abelian group Z in (1.4). A representa-
tion of Gal(F v /Fv ) yields a representation of WFv by restriction, with the inertia
group unchanged, and we may thus speak of ramification for a representation of
Gal(F v /Fv ).
We say that our given representation σ of Gal(F /F ) is ramified at v if σv is
ramified. The continuity of σ implies that σ descends to Gal(K/F ) for some finite
Galois extension K of F . Almost every finite place of F is unramified in K, and
each of these places is a place where σv is unramified. Therefore σ is unramified at
almost every place.
The statement of the conjecture gives us a finite set of places where π is unram-
ified but the L functions of σ and π are not known to agree. We enlarge this set to
a finite set S so that S contains all finite places where σ or π is ramified, as well as
all infinite places. Since the Langlands and Godement-Jacquet L factors of π agree
at all places v ∈/ S, we may write our given equality as
Lv (s, σ) = LGJ
v (s, π) for v ∈
/ S, (8.5)
with an Artin L factor on the left side and a Godement-Jacquet L factor on the
right side. The idea is to get information by twisting σ and π by suitable unitary
Grossencharacters α.
The twist σ ⊗ α was discussed above. A little explanation is in order for π ⊗ α.
The representation π is an irreducible constituent of some L2 (GLn (F )\GLn (A))χ
generated by cuspidal functions. We can regard α as a representation of the
center Z(A) trivial on Z(F ), and then π ⊗ α is an irreducible constituent of
L2 (GLn (F )\GLn (A))χα generated by cuspidal functions.
If v is a finite place, let mFv be the maximal ideal of OFv . The conductor of a
character β of Fv× will be defined to be the integer m ≥ 0 such that β is trivial on
m−1
1 + mmFv but not on 1 + mFv . (For m = 0, we understand the condition to be that
β is trivial on OF×v ; for m = 1, we understand the condition to be that β is trivial on
1 + mFv but not on OF×v .) By Corollary 2.5 we may regard β as a character of WFabv ,
hence as a one-dimensional representation of WFv . The condition that m = 0 for β
is the same as the condition that this one-dimensional representation be trivial on
 INTRODUCTION TO THE LANGLANDS PROGRAM 287

the inertia group in (1.4); thus we may consistently define β to be unramified if

m = 0.

v (s, π) for v ∈
Lemma 1. Under the assumption that Lv (s, σ) = LGL / S,
Lv (s, σ ⊗ α) = LGJ
v (s, π ⊗ α) for v ∈
/S (8.6)
for any Grossencharacter α.
Proof. First suppose that α is unramified at v. Since σ is unramified at v,
(8.1) gives the formulas
Lv (s, σ) = det(1 − σv (Fr)q −s )−1 , (8.7a)
−s −1
Lv (s, σ ⊗ α) = det(1 − σv (Fr)αv (v )q ) , (8.7b)
where v is a member of OFv of module q −1 and σv (Fr) is some unitary matrix. If
the eigenvalues of σv (Fr) are a1 , . . . , an , then

n
Lv (s, σ) = (1 − aj q −s )−1 , (8.8a)
j=1


n
Lv (s, σ ⊗ α) = (1 − aj αv (v )q −s )−1 . (8.8b)
j=1

Since π is unramified at v, the discussion with (8.3) shows that πv is the Lang-
lands subquotient of some unramified
GL (Fv )
indT Nn ((χ1 , . . . , χn ) ⊗ 1) (8.9)
and that the associated L factor is the factor for (ρ(χ1 ,...,χn ) , 0). Thus

n
LGJ
v (s, π) = (1 − bj q −s )−1 , (8.10)
j=1

where bj = χj (v ). The twist of (8.9) by αv is the unramified

GL (Fv )
indT Nn ((χ1 αv , . . . , χn αv )) ⊗ 1), (8.11)
and it follows from [Ja1, (3.4)] that

n

v (s, π ⊗ α) =
LGJ (1 − bj αv (v )q −s )−1 . (8.12)
j=1

The assumed equality of (8.8a) and (8.10) forces every symmetric polynomial in n
variables to agree at (a1 , . . . , an ) and (b1 , . . . , bn ), and hence we have an equality
of unordered sets {a1 , . . . , an } = {b1 , . . . , bn }. Therefore (8.8b) equals (8.12).
Now suppose that α is ramified at v. We shall show that Lv (s, σ ⊗ α) and
v (s, π ⊗ α) are both 1. In the case of σ, decompose σv into the direct sum of
LGJ
irreducibles τ . Since σ is unramified at v, each τ is unramified and thus τ ⊗ αv is
ramified. Let Vτ be the space on which τ acts. The invariant subspaces of Vτ under
τ are the same as those under τ ⊗ αv , and the inertial invariants form a proper
such subspace. By irreducibility the inertial invariants are 0, and Lv (s, τ ⊗ αv ) = 1.
Since Lv (s, σ ⊗ α) is the product of such factors, Lv (s, σ ⊗ α) = 1.
288 A. W. KNAPP

In the case of π, we stillhave that LGJv (s, π ⊗ α) is to be computed from (8.12)

n
and that LGJ
v (s, π ⊗ α) = L
j=1 v
GJ
(s, χ j αv ) by [Ja1, (3.4)]. But this time χj αv is
one-dimensional and ramified. The subspace of invariants is 0, LGJ v (s, χj αv ) is 1,
and LGJ
v (s, π ⊗ α) = 1.

Lemma 2. Let T and T  be finite sets of places, with T containing only finite
places. Fix integers mu ≥ 0 for u ∈ T . Then there exists a Grossencharacter α
such that
(i) αv = 1 for all v in T  and
(ii) for each u in T the conductor of αu is ≥ mu .

Reference. This is a special case of [Ar-Ta, Theorem 5, p. 103].

From the end of §7, we know that the cuspidal automorphic representation π
is unitary. The results of Flath [Fl] therefore imply that every local component
πv is unitary. Although πv is not known to be tempered (see [Moe1] and [Moe2]
for “tempered”), πv does satisfy another property—it is “generic” in a sense to be
defined below.
Let N be the algebraic group of matrices x = (xij ) with xij equal to 0 for i > j,
equal to 1 for i = j, and unrestricted for i < j. Fix a place v and a nontrivial
unitary character ψv of the additive group Fv , and consider a unitary character θv
of N (Fv ) of the form
θv (x) = ψ(c1 x12 + · · · + cn−1 xn−1,n ).
We say that θv is nondegenerate if c1 · · · cn−1 = 0. Let Π(v) be an irreducible
unitary admissible representative of GLn (Fv ), and let V (v) be the space of its
underlying smooth representation (the space of C ∞ vectors if v is infinite, the
space of vectors fixed by some idempotent of Hv if v is finite). The representation
Π(v) is said to be generic if there exists a nonzero continuous linear functional λ
on V (v) and some nondegenerate θv as above such that
λ(Π(v) (ng)x) = θv (n)λ(Π(v) (g)x) for all n ∈ N (Fv ), g ∈ GLn (Fv ), x ∈ V (v) .
"
Lemma 3. If Π = v Πv is a cuspidal automorphic representation of
GLn (AF ), then each Πv is a unitary generic representation of GLn (Fv ).
Reference. [Sha, Corollary, p. 190].

Proof of Theorem 8.8.

Step 1. If v is any finite place, then there exists an integer m = m(v, σ) ≥ 0 such
that Lv (s, σv ⊗ β) = 1 for every unitary character β of Fv× of conductor ≥ m.
In fact, since the L factor of a direct sum of representations is the product of the
L factors of the summands, it is enough to prove that the local factor Lv (s, τ ⊗ β)
is 1 for an irreducible τ when m is large enough. Fix τ .
If there exists β0 with τ ⊗ β0 unramified (at v), then τ ⊗ β0 factors through the
abelian group Z in (1.4), and τ must be one-dimensional. So we can regard τ as a
character of Fv× , say with conductor m0 . If β has conductor > m0 , then τ ⊗ β will
be ramified, and Lv (s, τ ⊗ β) will be 1.
Otherwise τ ⊗ β is ramified (at v) for every unitary character β of Fv× . Let τ
act on Vτ . The invariant subspaces of Vτ under τ are the same as under τ ⊗ β, and
 INTRODUCTION TO THE LANGLANDS PROGRAM 289

the inertial invariants form a proper such subspace. By irreducibility the inertial
invariants are 0, and therefore Lv (s, τ ⊗ β) = 1.
Step 2. If v is any finite place, then there exists an integer m = m (v, π) ≥ 0
×
such that LGJv (s, πv ⊗ β) = 1 for every unitary character β of Fv of conductor

≥m.
In fact, the Langlands classification ([Moe2]) shows that πv is the Langlands
quotient of a representation induced from a parabolic subgroup with the product of
a unitary discrete series and a quasicharacter on the Levi factor. The representation
on the Levi factor is just the tensor product of similar representations on the
component subgroups GLni (Fv ), and [Ja1, (3.4)] shows that the L factor of πv is
the product of the L factors of these representations on the component subgroups
GLni (Fv ). Thus it is enough to handle a representation of GLk (Fv ) that is the
product of a unitary discrete series and a quasicharacter. Work of Zelevinsky
discussed in [Moe2] shows that such a representation η is the unique irreducible
submodule of an induced representation of the form
GL(k,Fv )

((µ ⊗ γ) ⊗ (| · | 2 , . . . , | · |− 2 )) ⊗ 1 .
k−1 k−1
indP
Here P = M U is the standard parabolic subgroup associated to the partition
(d, d, . . . , d) for some divisor d of n, M is the product of copies of GL(d, Fv ), and
the representation µ ⊗ γ of M is the product of a (unitary) supercuspidal µ and a
quasicharacter γ. By Theorem 8.2 of [Ja-P-S],
d −1 ))
k
v (s, η ⊗ β) = Lv (s, µ ⊗ (γβ| · |
LGJ GJ

for any quasicharacter β of Fv× . If d > 1, then this L factor is 1 for any β by [Ja1,
(1.3.5)]. Thus we may assume that d = 1, in which case µ is just a unitary character
of Fv× and the argument is finished as in the case of the Galois representation σ.
Step 3.
∨ ∨
∞ (1 − s, π ) = ε(s)L∞ (s, π)L∞ (1 − s, σ )
L∞ (s, σ)LGJ GJ
(8.13)
for an entire nonvanishing function ε(s).
In fact, let T be the subset of finite places in S, and let T  be the subset of
infinite places. For each u in T , let mu denote the maximum of the numbers
m(u, σ), m(u, σ ∨ ), m (u, π), and m (u, π ∨ ) given by Steps 1 and 2. Let α be a
Grossencharacter chosen by Lemma 2 for the data T , T  , and {mu | u ∈ T }. Then
we have
Lu (s, σ ⊗ α) = LGJu (s, π ⊗ α) = 1 for u ∈ T
Lv (s, σ ⊗ α) = Lv (s, σ) for v ∈ T  (8.14)

LGJ
v (s, π ⊗ α) = LGJ
v (s, π) for v ∈ T .
∨ ∨
Similar formulas are valid for σ and π . The global functional equations of
Λ(s, σ ⊗α) and ΛGJ (s, π ⊗α) given in (5.4) (as generalized by Weil) and in Theorem
8.7 imply that
Λ(s, σ ⊗ α)ΛGJ (1 − s, π ∨ ⊗ α∨ ) = ε(s)ΛGJ (s, π ⊗ α)Λ(1 − s, σ ∨ ⊗ α∨ ), (8.15)
where ε(s) is entire and nonvanishing. Let LS or LGJS denote a product of factors
corresponding to the places not in S. Substituting from (8.14) into (8.15), we obtain
∨
L∞ (s, σ)LS (s, σ ⊗ α)LGJ
∞ (1 − s, π )L
GJS
(1 − s, π ∨ ⊗ α∨ )
= ε(s)LGJ
∞ (s, π)L
GJS
(s, π ⊗ α)L∞ (1 − s, σ ∨ )LS (1 − s, σ ∨ ⊗ α∨ ).
290 A. W. KNAPP

Use of Lemma 1 allows us to cancel all the factors LS and LGJS in this formula,
and we arrive at (8.13).
Step 4.
L∞ (s, σ) = LGJ
∞ (s, π) and L∞ (1 − s, σ ∨ ) = LGJ ∨
∞ (1 − s, π ).

In fact, inspection of the formulas in [Kna2, (3.6) and (4.6)] shows for each
infinite place w that Lw (s, σw ) is nowhere vanishing and has no poles for Re s > 0.
∨
Similarly Lw (1 − s, σw ) is nowhere vanishing and has no poles for Re s < 0.
We expect corresponding properties for π, but we get less. By [Ja1, (5.1)], each
LGJ
w (s, πw ) is an Artin L factor and hence is nowhere vanishing. The conclusions
about poles are more difficult to prove: Lw (s, πw ) has no poles for Re s ≥ 12 , and
∨
Lw (1 − s, πw ) has no poles for Re s ≤ 12 . This result is stated as [Ba-R, Proposition
2.1]. Its proof combines the fact that πw is unitary and generic (as follows from
Lemma 3) with a classication of (irreducible) unitary generic representations (which
follows readily from Vogan’s classification of irreducible unitary representations.
(See [Moe2] for a qualitative discussion of Vogan’s classification in this volume.)
If we rewrite (8.13) as
∨ ∨
∞ (s, π) = ε(s)L∞ (1 − s, σ )/L∞ (1 − s, π ),
L∞ (s, σ)/LGJ GJ GJ

then we see that the left side has no poles or zeros for Re s ≥ 12 and the right side has
no poles or zeros for Re s ≤ 12 . Therefore both sides are entire and nonvanishing.
To complete Step 4, it is enough to show that a product of elementary L factors
for archimedean places is determined by its poles. Referring to [Kna2,
(3.6) and
π − 2 −t Γ 2s + t or
s
(4.6)], we see that each such elementary L factor is of the form

2(2π)−s−u Γ(s + u) for some t or u. If we put ΓR (s) = π − 2 Γ 2s , then we conclude
s



from the well known formula Γ z2 Γ z2 + 12 = π 1/2 2−z+1 Γ(z) that

2(2π)−s−u Γ(s + u) = ΓR (s + u)ΓR (s + u + 1).

Hence
r every product of elementary L factors for archimedean places is of the form
j=1 R (s + tj ) for suitable tj ’s. No factor has a zero, and the pole of ΓR (s + tj )
Γ
with Re s largest occurs at s = −tj and has residue 2. Thus we can decompose
the product functions one factor at a time by finding the pole with Re s as large as
possible and dividing off a corresponding ΓR (s + tj ).
Step 5. If v is any finite place in S, then Lv (s, σ) = LGJ v (s, π).
In fact, let T be the set of finite places in S other than v, and let T  be the union
of {v} and the set of infinite places. For each u in T , let mu denote the maximum of
the numbers m(u, σ), m(u, σ ∨ ), m (u, π), and m (u, π ∨ ) given by Steps 1 and 2. Let
α be a Grossencharacter chosen by Lemma 2 for the data T , T  , and {mu | u ∈ T }.
Arguing as in Step 3 and taking into account the result of Step 4, we are led to an
equation
∨ ∨
v (s, π) = ε1 (s)Lv (1 − s, π )/Lv (1 − s, σ ).
Lv (s, σ)/LGJ GJ GJ
(8.16)
where ε1 (s) is an entire nonvanishing function.
Using the definition of Lv (s, σv ) in (5.1b) and taking into account that σ is
unitary, we see that Lv (s, σv ) is a product of factors (1 − cqv−s )−1 with |c| = 1.
Hence Lv (s, σv ) is nowhere vanishing and has no poles for Re s > 0. Similarly
Lv (1 − s, σ ∨ ) is nowhere vanishing and has no poles for Re s < 1.
 INTRODUCTION TO THE LANGLANDS PROGRAM 291

Again we expect corresponding formulas for π, but matters are not so simple.
∨
v (s, π) has no poles for Re s ≥ 2 and Lv (1 − s, π )
1
The conclusions are that LGJ GJ

has no poles for Re s ≤ 2 , and again the result is stated as [Ba-R, Proposition
1

2.1]. Its proof uses Lemma 3 and a classification of the irreducible unitary generic
representations. (A somewhat different proof may be found in [Ja-P-S].)
Arguing as in Step 4, we write (8.16) as
∨ ∨
v (s, π) = ε1 (s)Lv (1 − s, σ )/Lv (1 − s, π ).
Lv (s, σ)/LGJ GJ GJ

The restrictions on zeros and poles imply that each side is entire and nonvanishing.
The left side is of the form (1 + P (qv−s ))/(1 + Q(qv−s )), where P and Q are polyno-
mials without constant term. For this expression to have neither poles nor zeros,
we must have P = Q. Therefore Lv (s, σ) = LGJ v (s, π). This completes Step 5 and
the proof of Theorem 8.8.

9. L Groups and General Langlands L Functions

With GLn in place as a model, we can now describe what the Langlands theory
proposes for an arbitrary linear reductive group G over a number field F . It will
be assumed throughout that G(C) is connected.
The material in this section is largely due to Langlands [Lgl2] and [Lgl3], and
the presentation amounts to a summary of the exposition [Bo4]. The reader may
wish to consult [Bo4] for a more precise and detailed account and [Bl-Ro] for a
discussion that includes a number of examples.
The theory introduces a group L G that is the semidirect product of a certain
complex reductive group G  and the Galois group Gal(F /F ), with G normal.5
When G = GLn , G  is GLn (C), and the semidirect product is a direct product;
the definitions will show that the Galois group can often be ignored in this case,
and we are reduced to the situation in §8.
The same construction as for the L group of G(F ) yields, for each place v of F ,
an L group for G(Fv ). The complex group G  is unchanged, and the Galois group
Gal(F /F ) is cut down to the decomposition subgroup Gal(F v /Fv ).
Thinking in terms of Fv , let us consider the case of an arbitrary local field k
of characteristic 0 containing F . Then we can form an L group for G(k). An
“admissible homomorphism” is a certain kind of homomorphism ϕ (to be described
below) of the Weil group Wk or the Weil-Deligne group Wk , according as k is
archimedean or nonarchimedean, into the L group of G(k) that “covers the identity
mapping on Gal(k/k).” Here “ϕ covers the identity mapping on Gal(k/k)” means
that when ϕ is followed by the map of the L group to Gal(k/k), the result is
the usual map of Wk or Wk into Gal(k/k). Two admissible homomorphisms are
 and the set of equivalence classes is denoted
equivalent if they are conjugate via G,
Φ(G(k)).
We associate an elementary L factor L(s, ϕ, r) to this situation whenever r is a
representation of L G into some GLn (C) that is holomorphic in the G variable. The

5 To handle more advanced topics such as endoscopy, some adjustment in the definition of L G

is needed. The traditional adjustment is to replace the Galois group by a Weil group. Another
 by the Galois group that is not
possibility, discussed in [Ad-Ba-V], is to use an extension of G
necessarily a semidirect product.
292 A. W. KNAPP

definition is simply
L(s, ϕ, r) = L(s, r ◦ ϕ), (9.1)
where the right side is given by (8.1). When G is GLn and r is the standard
representation, this definition reduces to the definition (8.1) for the group GLn .
As with GLn , we let Π(G(k)) be the set of equivalence classes of irreducible
admissible representations of G(k).
Local Langlands Conjecture. Π(G(k)) is partitioned in a natural way into
finite nonempty subsets Πϕ indexed by all ϕ ∈ Φ(G(k)).
As in the case of GLn , the correspondence is to be consistent with a number
of conditions. See [Bo4, p. 43] for details. Ideally the correspondence should
be consistent with functoriality, which we shall define in §10. Consistency with
functoriality implies that the sets Πϕ are not necessarily singleton sets, unlike the
case of GLn . (See §10 below.) A set Πϕ is called an L packet, and members of
the same Πϕ are said to be L indistinguishable.
The conjecture is known to be true if G is a torus ([Lgl1], summarized in [Bo4,
p. 41]) and if k is archimedean ([Lgl3], summarized in [Bo4, p. 46]). Cases with
G = GLn for which it is true were discussed in §8. Also we shall insist that
representations with a nonzero G(Ok ) fixed vector (i.e., the Langlands subquotients
of the unramified principal series) be parametrized by ϕ’s in a particular way; this
parametrization we shall discuss below.
The Local Langlands Conjecture allows us to define an elementary L factor
whenever π is an irreducible admissible representation of G(k) and r is a finite-
dimensional holomorphic representation of L G. The definition is simply
L(s, π, r) = L(s, ϕ, r) if ϕ ↔ π, (9.2)
with L(s, ϕ, r) as in (9.1). When G is GLn and r is the standard representation,
this definition reduces to the definition (8.2) for GLn .
Before continuing, let us fill in some details in the above discussion.
We begin with a rough description of the L group L G. The group G  is a connected
complex reductive group, having the same dimension as G and having root system
equal to the system of coroots 2α/α, α of G. The question of the exact size of the
center is somewhat involved and will be described in a moment. For our purposes,
it will be enough to know that
(i) if G = GLn , then G = GLn (C)
(ii) if G(C) is simply connected, then G is an adjoint group

(iii) if G(C) is an adjoint group, then G is simply connected.
Here are some examples:

G 
G
SLn P GLn (C)
P GLn SLn (C)
Sp2n SO2n+1 (C)
SO2n+1 Sp2n (C)

 by working with
Langlands [Lgl3] captured the exact size of the center of G
weight lattices, but the treatment in [B4] makes matters axiomatic with the
 INTRODUCTION TO THE LANGLANDS PROGRAM 293

notion of root datum, which is described in [Sp]. A root datum is a 4-tuple

(X, ∆, X ∨ , ∆∨ ), where
(i) X and X ∨ are free abelian groups of finite rank in duality by a pairing
X × X ∨ → Z denoted  · , · ,
(ii) ∆ and ∆∨ are reduced root systems lying in subspaces of X ⊂ X ⊗ R and
X ∨ ⊂ X ∨ ⊗ R, respectively,
(iii) ∆ and ∆∨ are in bijection by a map α → α∨ such that α, α∨ = 2 for all
α ∈ ∆,
(iv) ∆ is preserved by the maps sα : X → X given by sα (x) = x − x, α∨ α,
and
(v) ∆∨ is preserved by the maps sα∨ : X ∨ → X ∨ given by sα∨ (x) =
x − x, α α∨ .
To our reductive group G and a maximal torus T is associated a root datum
ψ(G, T ) = (X, ∆, X ∨ , ∆∨ ) as follows: X is the group of rational characters X ∗ (T )
(i.e., algebraic homomorphisms T → GL1 ), X ∨ is the group X∗ (T ) of one-
parameter subgroups (i.e., algebraic homomorphisms GL1 → T ), the form x, u
×
for x ∈ X ∗ (T ) and u ∈ X∗ (T ) is given by x(u(t)) = tx,u for all t ∈ F , ∆ is
the root system of (G, T ), and ∆∨ is the system of coroots described as a subset
of X∗ (T ) in [Sp, pp. 6–7]. The theorem below says that all root data arise in this
way.
Theorem 9.1. For any root datum Ψ, there exists a connected reductive group
G and a maximal torus T in G such that Ψ = ψ(G, T ). The pair (G, T ) is unique
up to isomorphism over F .
Reference. [Sp, p. 9].

If Ψ = (X, ∆, X ∨ , ∆∨ ) is a root datum, then Ψ∨ = (X ∨ , ∆∨ , X, ∆) is easily seen

to be a root datum. Starting from (G, T ), we assemble the root datum ψ(G, T )
and apply Theorem 9.1 to ψ(G, T )∨ . The group G  is defined to be the C points of
the reductive group produced by the theorem. Also we define T to be the C points
of the maximal torus produced by the theorem.
All this discussion really involved only groups defined over F . We bring in F
through an action of Gal(F /F ). Fix a positive system ∆+ . If γ is in Gal(F /F ),
then γ carries T (F ) to T (F ) and ∆ to ∆, and there exists a member gγ of G(F )
that normalizes T (F ) such that gγ γ carries ∆+ to itself. Any two such elements
gγ are in the same coset relative to T (F ), and thus we can associate to γ a unique
permutation of ∆+ . This composite element also acts as an automorphism of X ∗ (T )
and X∗ (T ). As γ varies, we obtain compatible group actions of Gal(F /F ) on ∆+ ,
X ∗ (T ), and X∗ (T ). Thus we obtain a group action of Gal(F /F ) on the dual root
datum (X∗ (T ), ∆∨ , X ∗ (T ), ∆) preserving positive roots. This action lifts uniquely
to an action of Gal(F /F ) on G  once we choose root vectors for the simple roots.
Taking all the choices into account, we find that this action is canonical up to inner
automorphism by a member of T. In any event, L G is defined to be the semidirect
product of Gal(F /F ) and G.  Henceforth we shall usually discuss only groups G
that are split over F ; then the action is trivial, and L G is the direct product and
G and the Galois group.
Let us now return to G(k) with k a local field of characteristic 0 that contains F .
We shall define admissible homomorphisms more precisely, but only in the case that
294 A. W. KNAPP

G(k) is quasisplit over k. An element of L G is said to be semisimple or unipotent

 component is semisimple or unipotent. If k is nonarchimedean,
according as its G
an admissible homomorphism is a continuous homomorphism ϕ of the Weil-
Deligne group Wk into L G with the following properties:
(i) ϕ covers the identity mapping of Gal(k/k),
(ii) ϕ is holomorphic in the C variable of Wk , and every member of ϕ(C) is
unipotent, and
(iii) every member of ϕ(Wk ) is semisimple.
(Another condition, involving “relevance” of parabolic subgroups, is needed if G(k)
is not quasisplit. See [Bo4, p. 40].) If k is archimedean, admissible homomorphisms
are defined in the same way except that Wk replaces Wk and condition (ii) is
dropped.
The definitions are now complete in the case of G(k), and the Local Langlands
Conjecture allows us to associate to each irreducible admissible representation π of
G(k) and holomorphic representation r of L G an elementary L factor L(s, π, r).
It is known what these L factors should be in the case of unramified principal se-
ries. To keep matters simple, let us suppose that G is split over the nonarchimedean
field k. Let Q be the compact subgroup Q = G(Ok ). We shall assume that Q
is a “hyperspecial maximal compact subgroup” in the sense of the exposition [Ti,
pp. 35, 55]. Examples where this condition is satisfied are where G(Ok ) is GLn (Ok )
in GLn (k), Spn (Ok ) in Spn (k), or SOn (Ok ) in SO(k); here SOn is the group of
matrices of determinant one preserving a quadratic form
x1 xp+1 + · · · xp x2p + x22p+1
# $
in which p = n2 and the last term is absent if n is even. See [Car, p. 140] and [Bo4,
pp. 38–39, 45] for further discussion of the terms “special” and “hyperspecial.”
Suppose that T is a k split maximal torus chosen so that T (k) ∩ Q is maximal
compact in T (k). Let H(G(k), Q) be the subalgebra of bi-Q-invariant functions in
the Hecke algebra of G(k). This subalgebra is known to be commutative, and the
characteristic function of Q is the identity. Choose a Borel subalgebra B = T N
containing T , normalize Haar measure dm on N (k) so that N ∩Q gets total measure
1, and let ∆(t) be the positive function on T (k) defined by ∆(t) = d(tnt−1 )/dn.
For f ∈ H(G(k), Q), define


Sf (t) = ∆(t) 1/2
f (tn) dn for t ∈ T (k).
N (k)

Theorem 9.2. With the above assumptions and notation, the mapping S is
an algebra isomorphism of H(G(k), Q) onto the subalgebra H(T (k), T (k) ∩ Q)W of
Weyl-group invariants in H(T (k), T (k) ∩ Q).
Reference. This theorem is due to Satake, and the mapping S is called the
Satake isomorphism. See [Car, p. 147].

Now let π be an irreducible admissible representation with a nonzero vector

φ fixed by Q. The space of all vectors fixed by Q is one-dimensional, and it
follows that π(H(G(k), Q)) operates on φ by scalars. The result is an algebra
homomorphism λ = λ(π) of H(G(k), Q) into C. Sorting out the notation with the
aid of [Car, p. 134], we see that the algebra homomorphisms of H(T (k), T (k) ∩ Q)
into C are simply the members of T, and it follows that we may identify an algebra
 INTRODUCTION TO THE LANGLANDS PROGRAM 295

homomorphism of H(T (k), T (k) ∩ Q)W into C with a W orbit in T. Thus Theorem
9.2 implies that we may regard λ(π) as a W orbit in T.
We may understand this construction in other terms as follows. The Hecke
algebra H(T (k), T (k) ∩ Q) is just the C group ring of T, which we write as C[T].
The subspace C[T]W of Weyl-group invariants is the coordinate ring of the affine
variety T/W , and then it becomes clear that a homomorphism of this algebra into C
is simply a W orbit in T. If π, as above, is an irreducible admissible representation
of G(k) with a nonzero vector fixed by Q, then the point it defines in T/W is called
the Langlands class of π. For G = GLn , this point can be represented by an
n-tuple (ε1 , . . . , εn ) upto permutation, and the local L factor that we used in §8
amounts to L(s, π) = i=1 (1 − εi q −s )−1 in these terms.
n

In the general case that we are studying, the elementary L factor associated to
π and a holomorphic representation r of L G is taken to be

L(s, π, r) = det(1 − r(λ(π))q −s )−1 . (9.3)

(See [Bo4, p. 39].) In (9.3) any representative of the W orbit λ(π) in T is to be

used, and the determinant is independent of the choice. To arrange for this L factor
to come from an admissible homomorphism ϕ, choose Fr in Wk with Fr = q −1 ,
and fix a representative of λ(π) in T. The idea is to make ϕ(Fr) = (λ(π), Fr) and
to make ϕ behave in an “unramified” fashion. To do so, define ε : Wk → Z by

Frε(z,w) = w for z ∈ C and w ∈ Wk ,

and let
ϕ(z, w) = (λ(π), Fr)ε(z,w) .
With this ϕ associated to π, the definition (9.2) attaches (9.3) to π and r as
elementary L factor.
When G(k) is not k split, the treatment of unramified principal series involves
considerably more structure theory, and the Satake isomorphism is more compli-
cated to state. For an exposition of the structure theory, see [Ti] and [Car]. For
the definition of the elementary L factor associated to an irreducible admissible
representation with a nonzero fixed vector under G(Ok ), see [Bo4, pp. 39 and
44–45].
Let us return to the number field F . As was true for GLn , the results of [Fl]
irreducible admissible representation π of G(AF ) is a restricted tensor
show that an "
product π = v πv , and moreover almost every πv has a nonzero vector fixed under
G(Ov ). Let r be a holomorphic representation of the L group of G(F ). Since the
L group of G(Fv ) may be taken to be a subgroup of G(F ), we obtain by restriction
a holomorphic representation r"of the L group of each G(Fv ). For the irreducible
admissible representation π = v πv of G(AF ), we can then define

L(s, π, r) = L(s, πv , r)
finite v
 (9.4)
Λ(s, π, r) = L(s, πv , r),
all v

with L(s, πv , r) as in (9.2) (and almost always as in (9.3)).

296 A. W. KNAPP

Theorem 9.3. If π is an automorphic representation of G(AF ) and r is a

holomorphic representation of L G, then L(s, π, r) converges absolutely for Re s
sufficiently large.

Reference. This theorem is due to Langlands. The line of argument was given
in Corollary 8.2 through Corollary 8.6.

The global analytic properties of Λ(s, π, r), if any, are known only in special
cases. It is expected that these functions are meromorphic and satisfy a functional
equation. For more information, see [Bo4].

10. Functoriality
Functoriality refers to translating knowledge of a holomorphic homomorphism
of L groups into results about automorphic representations. Much of [Lgl2] raises
specific questions about this problem, and [Bo4, §§ 15–17] discusses progress as of
the late 1970s. More recent progress is the subject of [Lgl7].
A homomorphism ω : G → H between reductive Lie groups over a local or
global field induces a holomorphic homomorphism ψ : L H → L G if the image of ω
is normal [Bo4, p. 29], and moreover ψ covers the identity mapping of the Galois
group. For example, the inclusion ω of SL2 (R) into GL2 (R) induces the natural
quotient map
ψ : GL2 (C) × Γ → P GL2 (C) × Γ
of L groups, where Γ = Gal(C/R). An admissible homomorphism

ϕ : WR → GL2 (C) × Γ

induces by composition the admissible homomorphism

ψ ◦ ϕ : WR → P GL2 (C).

The resulting map Φ(GL2 (R)) → Φ(SL2 (R)) gives us a correspondence (not a
function!) Π(GL2 (R)) → Π(SL2 (R)) since the Local Langlands Conjecture is a
theorem in the archimedean case. Examining matters, we see that we associate
to each irreducible admissible representation of GL2 (R) all of its constituents on
restriction to SL2 (R). Nothing very deep is happening here, but the fact that
each discrete series of GL2 (R) and certain principal series decompose into two
inequivalent pieces on restriction to SL2 (R) forces some members of Π(SL2 (R)) to
have more than one element.
Let us call a holomorphic homomorphism ψ : L H → L G covering the identity of
the Galois group an L homomorphism. Not every L homomorphism arises from
a homomorphism G → H; in fact, most do not. We can still ask
(a) in the local case whether such a ψ induces a correspondence Π(H) → Π(G)
and
(b) in the global case whether such a ψ induces a correspondence of automorphic
representations to automorphic representations (or cuspidal representations
to cuspidal representations).
In each case we ask that the correspondence respect L functions, or as much of
 INTRODUCTION TO THE LANGLANDS PROGRAM 297

L functions as is known. (Also we ask that the correspondence respect ε factors,

which we have largely ignored in this article.)
To fix the ideas, let us continue with the notation of §9. Thus let F be a number
field, and let G be a reductive group over F . We denote by k any local field
containing F , so that G(k) is defined.
First let us consider the local case. If we accept the Local Langlands Conjec-
ture, then (a) is solved, by the same argument as in the example above: The L
homomorphism ψ : L H → L G induces a map ψ∗ : Φ(H) → Φ(G) (at least if G is
quasisplit, so that the all the conditions on a member of Φ(G) were given in §9). In
turn, ψ∗ induces a correspondence Π(H) → Π(G), and the correspondence respects
L factors.
From an organizational point of view, we could insist that the Local Langlands
Conjecture is to be proved first (or else taken as a working hypothesis), and then
functoriality is to be addressed. But [Bo4] points out some early cases in which
a partial result about local functoriality was established and then used to obtain
a partial result about the Local Langlands Conjecture. If local functoriality and
the Local Langlands Conjecture ultimately turn out to be true, the proofs may
therefore have to start from some basic information (the ϕ’s for the unramified
principal series, the L and ε factors for GLn and the standard representation, and
some other conditions) and establish local functoriality and the Local Langlands
Conjecture together.
Nevertheless, to keep matters brief, we shall take local functoriality as a working
hypothesis and consider the global case (b). Here is one possible statement of the
problem.

Question (Global functoriality). Let G and H be reductive groups over F with

G quasisplit, and let ψ : L H → L G be an L homomorphism. For each place v of F ,
let ψv be the restriction of ψ to a map between the L groups of H(Fv ) and G(Fv ), let
(ψv )∗ : Φ(H(Fv )) → Φ(G(Fv )) be the induced map on admissible homomorphisms,
and let (ψv )∗ also denote the correspondence" Π(H(Fv )) → Π(G(Fv )) obtained from
the Local Langlands Conjecture. Let π = v πv be an automorphic representation
of H(AF ).
"
(i) Does there exists a choice Πv ∈ (ψv )∗ (πv ) for every v such that Π = v Πv
is an automorphic representation of G(AF )?
(ii) If so, and if π is cuspidal, under what conditions is Π cuspidal?

This is an extremely deep question, even if the condition Πv ∈ (ψv )∗ (πv ) is re-
quired only at almost every place. (Relaxing the requirement in this way allows one
to address the question without first establishing the Local Langlands Conjecture.)
Here are two illustrations, taken from [Lgl2], of just how deep it is. Let us write
Γ = Gal(F /F ) and Γv = Gal(F v /Fv ) with Γv ⊂ Γ.

Example 1. Let G = GLn and H = {1}, so that L H = {1} × Γ and L G =

GLn (C) × Γ. Fix an n-dimensional representation σ of Γ. The map ψ : L H → L G
given by ψ(1, γ) = (σ(γ), γ) is an L homomorphism. If v is a finite place of F ,
(v)
then the only admissible homomorphism for H(Fv ) is ϕ0 (z, w) = (1, ι(w)), where
ι : WFv → Γv is the inclusion. Let r be the representation 1 × σ of L H, and let
p : Wk → Wk be the natural quotient map. Then the elementary L factor for the
298 A. W. KNAPP

representation r at the finite place v is

(v)
L(s, 1, r) = L(s, r ◦ ϕ0 ) by (9.1)
= L(s, σ ◦ ι ◦ p) (10.1)
−s −1
= det(1 − (σ ◦ ι(Fr)|(Cn )I )q ) by (8.1),

and the right side coincides with the Artin L factor (5.1) for r. Hence the global L
function for the cuspidal representation 1 of H(AF ) = {1} and for the representation
r coincides with the Artin L function of r. If the answer to (i) is affirmative, let
Π ∈ ψ∗ (1) be an automorphic representation of GLn (AF ) for which Πv ∈ (ψv )∗ (1v )
(v)
for each v. The admissible homomorphism Φv for Πv has Φv = (ψv )∗ (ϕ0 ) =
(v)
ψv ◦ ϕ0 . Let R be the standard representation of L G (trivial on Γ). Then we have

(v)
L(s, Πv ) = L(s, Πv , R) = L(s, ψv ◦ ϕ0 , R) by (9.2)
(v)
= L(s, R ◦ ψv ◦ ϕ0 ) by (9.1)
= L(s, σ ◦ ι ◦ p)
= L(s, 1, r) by (10.1).

Thus an affirmative answer to (i) for this situation implies that the Artin L function
for any n-dimensional representation of Γ is the L function of an automorphic
representation of GLn . An affirmative answer to (ii) implies that the latter repre-
sentation is cuspidal, i.e., gives an affirmative answer to the Langlands Reciprocity
Conjecture of §8.
Example 2. Let H be general, and let r be a holomorphic representation of
L
H into GLn (C). Put G = GLn , and define ψ(x, γ) = (r(x, γ), γ). Let π be an
automorphic representation of H. If the answer to (i) is affirmative, let Π ∈ ψ∗ (π)
be a corresponding automorphic representation of GLn (AF ). Tracking down the
definitions as in Example 1, we find that

L(s, π, r) = L(s, Π),

i.e., the L function of π and the representation r is a standard L function for GLn .
If we assume about GLn that Langlands L functions and Godement-Jacquet L
functions coincide, then it follows from a generalization of Theorem 8.7 proved in
[Ja1] that Λ(s, π, r) has a meromorphic continuation to C and satisfies a functional
equation. Moreover if π is cuspidal and the answer to (ii) is affirmative, then (with
some exceptions that can be sorted out) L(s, π, r) and Λ(s, π, r) are entire.
We mention two situations in which substantial progress has been made in es-
tablishing global functoriality. Both these situations are discussed in more detail
in Rogawski’s lectures [Ro2].
1) Adjoint representation for GL2 . This is an instance where Example 2 can be
carried out. For automorphic representations π of GL2 , we consider L functions
L(s, π, Ad), where Ad : GL2 (C) → GL3 (C) is the adjoint representation. As in
Example 2, the goal is to exhibit these L functions as standard L functions L(s, Π)
with Π automorphic for GL3 . This amounts to establishing global functoriality
when G = GL3 , H = GL2 , and ψ : L H → L G is given by ψ(x, γ) = (Ad(x), γ).
The positive result here is due to Gelbart and Jacquet [Gelb-Ja].
 INTRODUCTION TO THE LANGLANDS PROGRAM 299

2) Base change for GL2 . Let E be any finite extension of the number field
F . Let H = GL2 over F , and let G = RE/F (GL2 ) be the group over F given
by restriction of ground field as in (6.5). As with (6.5), for any F algebra A,
(RE/F )(GL2 (A)) = GL2 (E ⊗F A). Thus

G(F ) = GL2 (E ⊗F F ) = GL2 (F ) × · · · × GL2 (F )

and G(F ) = GL2 (E ⊗F F ) = GL2 (E).

The L groups are

L
H = GL2 (C) × Gal(F /F )

and L
G = (GL2 (C) × · · · × GL2 (C))  Gal(F /F ),

the second one being a semidirect product. The action of Gal(F /F ) on the product
GL2 (C) × · · · × GL2 (C) permutes the coordinates. This action factors through
Gal(E/F ) if E is Galois over F . The map ψ : L H → L G is given by the diagonal
map on the identity component and by the identity map on the Galois group.
Langlands [Lgl5] proved global functoriality in this setting when E/F is a cyclic
Galois extension of prime degree; this is the long step in the proof of new cases of
Artin’s Conjecture established by Langlands. Arthur and Clozel [Ar-Cl] proved the
corresponding instance of global functoriality for GLn when E/F is cyclic Galois
of prime degree.
There is also a considerable amount of more recent progress. Various newer
results on the analytic properties of L(s, π, r) are summarized in [Ra]. Rogawski
[Ro1] has made an extensive study of automorphic representations of the group U3 .
Here E/F is a quadratic extension of number fields, and G = U3 is the associated
unitary group. For the group H = U2 × U1 , there is an embedding L H → L G,
and Rogawski’s work addresses functoriality for this map. Rogawski also studies
% where G
functoriality for the map L G → L G, % = RE/F (G); this is the base change
lifting from U3 to GL3 over E. This work is applied to arithmetic geometry in [Lgl-
Ra]; the forward of [Lgl-Ra] puts a number of aspects of the Langlands program in
perspective.

References
[Ad-Ba-V] Adams, J., D. Barbasch, and D. A. Vogan, The Langlands Classification and
Irreducible Characters for Real Reductive Groups, Birkhäuser, Boston, 1992.
[Ar-Cl] Arthur, J., and L. Clozel, Simple Algebras, Base Change, and the Advanced Theory
of the Trace Formula, Princeton University Press, Princeton, 1989.
[Ar-Ta] Artin, E., and J. Tate, Class Field Theory, W. A. Benjamin Inc., New York, 1967.
[Ba-R] Barthel, L., and D. Ramakrishnan, A non-vanishing result for twists of L-functions of
GL(n), Duke Math. J. 74 (1994), 681–700.
[Bl-Ro] Blasius, D., and J. D. Rogawski, Zeta functions of Shimura varieties, Motives, Proc.
Symp. Pure Math., vol. 55, Part II, American Mathematical Society, Providence, 1994,
pp. 525–571.
[Bo1] Borel, A., Some finiteness properties of adele groups over number fields, Publ. Math.
I.H.E.S. 16 (1963), 5–30.
[Bo2] Borel, A., Linear algebraic groups, Algebraic Groups and Discontinuous Subgroups,
Proc. Symp. Pure Math., vol. 9, American Mathematical Society, Providence, 1966,
pp. 3–19.
[Bo3] Borel, A., Linear Algebraic Groups, W. A. Benjamin, New York, 1969.
300 A. W. KNAPP

[Bo4] Borel, A., Automorphic L-functions, Automorphic Forms, Representations, and L-

Functions, Proc. Symp. Pure Math., vol. 33, Part II, American Mathematical Society,
Providence, 1979, pp. 27–61.
[Bo-HC] Borel, A., and Harish-Chandra, Arithmetic subgroups of algebraic groups, Annals of
Math. 75 (1962), 485–535.
[Bo-Ja] Borel, A., and H. Jacquet, Automorphic forms and automorphic representations,
Automorphic Forms, Representations, and L-Functions, Proc. Symp. Pure Math.,
vol. 33, Part I, American Mathematical Society, Providence, 1979, pp. 189–202.
[Buhl] Buhler, J. P., Icosahedral Galois Representations, Lecture Notes in Mathematics,
vol. 654, Springer-Verlag, Berlin, 1978.
[Bump] Bump, D., Automorphic Forms and Representations, Cambridge University Press,
Cambridge, 1997.
[Car] Cartier, P., Representations of p-adic groups: A survey, Automorphic Forms, Rep-
resentations, and L-Functions, Proc. Symp. Pure Math., vol. 33, Part I, American
Mathematical Society, Providence, 1979, pp. 111–155.
[Cas] Cassels, J. W. S., Global fields, Algebraic Number Theory (J. W. S. Cassels and A.
Fröhlich, eds.), Academic Press, London, 1967, pp. 42–84.
[Cas-Fr] Cassels, J. W. S., and A. Fröhlich, Algebraic Number Theory, Academic Press, London,
1967.
[Ch] Chevalley, C., La théorie du corps de classes, Annals of Math. 41 (1940), 394–418.
[Co-Pi] Cogdell, J. W., and I. I. Piatetski-Shapiro, Converse theorems for GLn , Publ. Math.
I.H.E.S. 79 (1994), 157–214.
[De-S] Deligne, P., and J.-P. Serre, Formes modulaires de poids 1, Annales Scient. Ecole
Norm. Sup. 7 (1974), 507–530.
[Fl] Flath, D., Decomposition of representations into tensor products, Automorphic Forms,
Representations, and L-Functions, Proc. Symp. Pure Math., vol. 33, Part I, American
Mathematical Society, Providence, 1979, pp. 179–183.
[Fre] Frey, G. (ed.), On Artin’s Conjecture for Odd 2-dimensional Representations, Lecture
Notes in Mathematics, vol. 1585, Springer-Verlag, Berlin, 1994.
[Frö] Fröhlich, A., Local fields, Algebraic Number Theory (J. W. S. Cassels and A. Fröhlich,
eds.), Academic Press, London, 1967, pp. 1–41.
[Ga] Garrett, P. B., Holomorphic Hilbert Modular Forms, Wadsworth & Brooks/Cole,
Pacific Grove, CA, 1990.
[Gelb1] Gelbart, S. S., Automorphic Forms on Adele Groups, Princeton University Press,
Princeton, NJ, 1975.
[Gelb2] Gelbart, S. S., An elementary introduction to the Langlands program, Bull. Amer.
Math. Soc. 10 (1984), 177–219.
[Gelb-Ja] Gelbart, S. S., and H. Jacquet, A relation between automorphic representations of
GL(2) and GL(3), Annales Scient. École Norm. Sup. 11 (1978), 471–542.
[Gelf-Pi] Gelfand, I. M., and I. Piatetski-Shapiro, Automorphic forms and representation theory,
Trudy Moscov. Mat. Obšč. 12 (1963), 389–412.
[Gf-Gr-P] Gelfand, I. M., M. I. Graev, and I. I. Piatetskii-Shapiro, Representation Theory and
Automorphic Functions, W. B. Saunders, Philadelphia, 1969.
[Go-Ja] Godement, R., and H. Jacquet, Zeta-functions of simple algebras, Lecture Notes in
Mathematics, vol. 260, Springer-Verlag, New York, 1972.
[Ha] Harris, M., Supercuspidal representations in the cohomology of Drinfel d upper half
spaces; elaboration of Carayol’s program, preprint, 1994; Invent. Math. (to appear).
[HC1] Harish-Chandra, Automorphic forms on a semisimple Lie group, Proc. Nat. Acad. Sci.
USA 45 (1959), 570–573.
[HC2] Harish-Chandra, Automorphic Forms on Semisimple Lie Groups, Lecture Notes in
Mathematics, vol. 62, Springer-Verlag, Berlin, 1968.
[Hei] Heilbronn, H., Zeta-functions and L-functions, Algebraic Number Theory (J. W. S.
Cassels and A. Fröhlich, eds.), Academic Press, London, 1967, pp. 204–230.
[Hen1] Henniart, G., La conjecture de Langlands locale pour GL(3), Mem. Soc. Math. France
11–12 (1984), 1–186.
[Hen2] Henniart, G., Caractérisation de la correspondence de Langlands locale par les facteurs
ε de paires, Invent. Math. 113 (1993), 339–350.
 INTRODUCTION TO THE LANGLANDS PROGRAM 301

[Ja1] Jacquet, H., Principal L functions of the linear group, Automorphic Forms, Repre-
sentations, and L-Functions, Proc. Symp. Pure Math., vol. 33, Part II, American
Mathematical Society, Providence, 1979, pp. 63–86.
[Ja2] Jacquet, H., Principal L-functions for GL(n), these Proceedings, pp. 321–329.
[Ja-Lgl] Jacquet, H., and R. P. Langlands, Automorphic Forms on GL(2), Lecture Notes in
Mathematics, vol. 114, Springer-Verlag, Berlin, 1970.
[Ja-P-S] Jacquet, H., I. I. Piatetski-Shapiro, and J. A. Shalika, Rankin-Selberg convolutions,
Amer. J. Math. 105 (1983), 367–464.
[Ja-Sha1] Jacquet, H., and J. A. Shalika, On Euler products and the classification of automorphic
representations I, Amer. J. Math. 103 (1981), 499–558.
[Ja-Sha2] Jacquet, H., and J. Shalika, Rankin-Selberg convolutions: Archimedean theory,
Festschrift in Honor of I. I. Piatetski-Shapiro on the Occasion of His Sixtieth
Birthday, Part I (Ramat-Aviv, 1989), Israel Math. Conf. Proc., vol. 2, Weizmann,
Jerusalem, 1990, pp. 125–207.
[Kna1] Knapp, A. W., Elliptic Curves, Princeton University Press, Princeton, 1992.
[Kna2] Knapp, A. W., Local Langlands correspondence: the archimedean case, Motives, Proc.
Symp. Pure Math., vol. 55, Part II, American Mathematical Society, Providence, 1994,
pp. 393–410.
[Kna-Vo] Knapp, A. W., and D. A. Vogan, Cohomological Induction and Unitary Representa-
tions, Princeton University Press, Princeton, NJ, 1995.
[Kne] Kneser, M., Strong approximation, Algebraic Groups and Discontinuous Subgroups,
Proc. Symp. Pure Math., vol. 9, American Mathematical Society, Providence, 1966,
pp. 187–196.
[Kud] Kudla, S. S., The local Langlands correspondence: the non-archimedean case, Motives,
Proc. Symp. Pure Math., vol. 55, Part II, American Mathematical Society, Providence,
1994, pp. 365–391.
[Kut] Kutzko, P., The local Langlands conjecture for GL(2) of a local field, Annals of Math.
112 (1980), 381–412.
[La] Lang, S., Algebraic Number Theory, Springer-Verlag, New York, 1986.
[Lgl1] Langlands, R. P., Representations of abelian algebraic groups, duplicated notes, Yale
University, New Haven, CT, 1968.
[Lgl2] Langlands, R. P., Problems in the theory of automorphic forms, Lectures in Modern
Analysis and Applications III, Lecture Notes in Mathematics, vol. 170, Springer-
Verlag, Berlin, 1970, pp. 18–61.
[Lgl3] Langlands, R. P., On the classification of irreducible representations of real algebraic
groups, mimeographed notes, Institute for Advanced Study, Princeton, NJ, 1973,
Representation Theory and Harmonic Analysis on Semisimple Lie Groups (P. J.
Sally and D. A. Vogan, eds.), Math. Surveys and Monographs, vol. 31, American
Mathematical Society, Providence, 1989, pp. 101–170.
[Lgl4] Langlands, R. P., On the notion of an automorphic representation, Automorphic
Forms, Representations, and L-Functions, Proc. Symp. Pure Math., vol. 33, Part
I, American Mathematical Society, Providence, 1979, pp. 203–207.
[Lgl5] Langlands, R. P., Base Change for GL(2), Princeton University Press, Princeton, 1980.
[Lgl6] Langlands, R. P., Representation theory: its rise and its role in number theory,
Proceedings of the Gibbs Symposium (New Haven, CT, 1989), American Mathematical
Society, Providence, RI, 1990, pp. 181-210.
[Lgl7] Langlands, R. P., Where stands functoriality today?, these Proceedings, pp. 457–471.
[Lgl-Ra] Langlands, R. P., and D. Ramakrishnan (eds.), The Zeta Functions of Picard Modular
Surfaces, Les Publications CRM, Montréal, 1992.
[Maa] Maass, H., Über eine neue Art von nicht analytischen automorphen Funktionen und
die Bestimmung Dirichletscher Reihen durch Funktionalgleichungen, Math. Annalen
121 (1949), 141–183.
[Mar] Martinet, J., Character theory and Artin L-functions, Algebraic Number Fields:
L-Functions and Galois Properties (A. Fröhlich, ed.), Academic Press, London, 1977,
pp. 1–87.
[Moe1] Mœglin, C., Representations of GL(n) over the real field, these Proceedings,
pp. 157–166.
302 A. W. KNAPP

[Moe2] Mœglin, C., Representations of GL(n,F) in the nonarchimedean case, these Proceed-
ings, pp. 303–319.
[Moy] Moy, A., Local constants and the tame Langlands correspondence, Amer. J. Math.
108 (1986), 863–929.
[Ne] Neukirch, J., Class Field Theory, Springer-Verlag, Berlin, 1986.
[Pi] Piatetski-Shapiro, I., Festschrift in Honor of I. I. Piatetski-Shapiro on the Occasion
of His Sixtieth Birthday, Part II, Israel Math. Conf. Proc., vol. 3, 1990, pp. 185–195.
[Ra] Ramakrishnan, D., Pure motives and automorphic forms, Motives, Proc. Symp. Pure
Math., vol. 55, Part II, American Mathematical Society, Providence, 1994, pp. 411–446.
[Ro1] Rogawski, J. D., Automorphic Representations of Unitary Groups in Three Variables,
Princeton University Press, Princeton, 1990.
[Ro2] Rogawski, J. D., Functoriality and the Artin conjecture, these Proceedings,
pp. 331–353.
[Se1] Serre, J.-P., Local class field theory, Algebraic Number Theory (J. W. S. Cassels and
A. Fröhlich, eds.), Academic Press, London, 1967, pp. 129–161.
[Se2] Serre, J.-P., Local Fields, Springer-Verlag, New York, 1979.
[Sha] Shalika, J. A., The multiplicity one theorem for GLn , Annals of Math. 100 (1974),
171–193.
[Shi] Shimura, G., Introduction to the Arithmetic Theory of Automorphic Functions, Prince-
ton University Press, Princeton, NJ, 1971.
[Si] Siegel, C. L., Topics in Complex Function Theory, vol. III, Wiley-Interscience, New
York, 1973.
[Sp] Springer, T. A., Reductive groups, Automorphic Forms, Representations, and L-
Functions, Proc. Symp. Pure Math., vol. 33, Part I, American Mathematical Society,
Providence, 1979, pp. 3–27.
[Ta1] Tate, J. T., Fourier analysis in number fields and Hecke’s zeta-functions, Ph.D. The-
sis, Princeton University, 1950, Algebraic Number Theory (J. W. S. Cassels and A.
Fröhlich, eds.), Academic Press, London, 1967, pp. 305–347.
[Ta2] Tate, J. T., Global class field theory, Algebraic Number Theory (J. W. S. Cassels and
A. Fröhlich, eds.), Academic Press, London, 1967, pp. 162–203.
[Ta3] Tate, J., Number theoretic background, Automorphic Forms, Representations, and L-
Functions, Proc. Symp. Pure Math., vol. 33, Part II, American Mathematical Society,
Providence, 1979, pp. 3–26.
[Ti] Tits, J., Reductive groups over local fields, Automorphic Forms, Representations, and
L-Functions, Proc. Symp. Pure Math., vol. 33, Part I, American Mathematical Society,
Providence, 1979, pp. 29–69.
[Tu] Tunnell, J., On the local Langlands conjecture for GL(2), Invent. Math. 46 (1978),
179–200.
[We1] Weil, A., Adeles and Algebraic Groups, bound notes, Institute for Advanced Study,
Princeton, NJ, 1961.
[We2] Weil, A., Über die Bestimmung Dirichletscher Reihen durch Funktionalgleichungen,
Math. Annalen 168 (1967), 149–156.
[We3] Weil, A., Dirichlet Series and Automorphic Forms, Lecture Notes in Mathematics,
vol. 189, Springer-Verlag, Berlin, 1971.
[We4] Weil, A., Basic Number Theory, Springer-Verlag, New York, 1973.

Department of Mathematics, State University of New York, Stony Brook, New York
11794, U.S.A.
E-mail address: [email protected]