Automorphic Forms on GL(2)

Herve Jacquet and Robert P. Langlands

Formerly appeared as volume #114 in the Springer Lecture Notes in Mathematics, 1970, pp. 1-548

Chapter 1

Table of Contents
Introduction

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii

Chapter I: Local Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.
2.
3.
4.
5.
6.
7.
8.

Weil representations . . . . . . . . . . . . . . . . .
Representations of GL(2, F ) in the non-archimedean case
The principal series for non-archimedean fields . . . . .
Examples of absolutely cuspidal representations . . . . .
Representations of GL(2, R) . . . . . . . . . . . . .
Representation of GL(2, C) . . . . . . . . . . . . . .
Characters . . . . . . . . . . . . . . . . . . . . .
Odds and ends . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

. 1
15
58
77
96
138
151
173

Chapter II: Global Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189

9.
10.
11.
12.

The global Hecke algebra . . . . .

Automorphic forms . . . . . . .
Hecke theory . . . . . . . . . .
Some extraordinary representations

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

189
204
221
251

Chapter III: Quaternion Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . 267

13.
14.
15.
16.

Zeta-functions for M (2, F ) . . . . . . . .

Automorphic forms and quaternion algebras
Some orthogonality relations . . . . . . .
An application of the Selberg trace formula .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

267
294
304
320

Chapter 1

ii
Introduction

Two of the best known of Heckes achievements are his theory of L-functions with grossencharakter, which are Dirichlet series which can be represented by Euler products, and his theory of the
Euler products, associated to automorphic forms on GL(2). Since a grossencharakter is an automorphic
form on GL(1) one is tempted to ask if the Euler products associated to automorphic forms on GL(2)
play a role in the theory of numbers similar to that played by the L-functions with grossencharakter.
In particular do they bear the same relation to the Artin L-functions associated to two-dimensional
representations of a Galois group as the Hecke L-functions bear to the Artin L-functions associated
to one-dimensional representations? Although we cannot answer the question definitively one of the
principal purposes of these notes is to provide some evidence that the answer is affirmative.
The evidence is presented in 12. It come from reexamining, along lines suggested by a recent
paper of Weil, the original work of Hecke. Anything novel in our reexamination comes from our point
of view which is the theory of group representations. Unfortunately the facts which we need from the
representation theory of GL(2) do not seem to be in the literature so we have to review, in Chapter I,
the representation theory of GL(2, F ) when F is a local field. 7 is an exceptional paragraph. It is not
used in the Hecke theory but in the chapter on automorphic forms and quaternion algebras.
Chapter I is long and tedious but there is nothing hard in it. Nonetheless it is necessary and
anyone who really wants to understand L-functions should take at least the results seriously for they
are very suggestive.
9 and 10 are preparatory to the Hecke theory which is finally taken up in 11. We would like to
stress, since it may not be apparent, that our method is that of Hecke. In particular the principal tool is
the Mellin transform. The success of this method for GL(2) is related to the equality of the dimensions
of a Cartan subgroup and the unipotent radical of a Borel subgroup of P GL(2). The implication is that
our methods do not generalize. The results, with the exception of the converse theorem in the Hecke
theory, may.
The right way to establish the functional equation for the Dirichlet series associated to the
automorphic forms is probably that of Tate. In 13 we verify, essentially, that this method leads to the
same local factors as that of Hecke and in 14 we use the method of Tate to prove the functional equation
for the L-functions associated to automorphic forms on the multiplicative group of a quaternion
algebra. The results of 13 suggest a relation between the characters of representations of GL(2) and
the characters of representations of the multiplicative group of a quaternion algebra which is verified,
using the results of 13, in 15. This relation was well-known for archimedean fields but its significance
had not been stressed. Although our proof leaves something to be desired the result itself seems to us
to be one of the more striking facts brought out in these notes.
Both 15 and 16 are after thoughts; we did not discover the results in them until the rest of the
notes were almost complete. The arguments of 16 are only sketched and we ourselves have not verified
all the details. However the theorem of 16 is important and its proof is such a beautiful illustration
of the power and ultimate simplicity of the Selberg trace formula and the theory of harmonic analysis
on semi-simple groups that we could not resist adding it. Although we are very dissatisfied with the
methods of the first fifteen paragraphs we see no way to improve on those of 16. They are perhaps
the methods with which to attack the question left unsettled in 12.
We hope to publish a sequel to these notes which will include, among other things, a detailed
proof of the theorem of 16 as well as a discussion of its implications for number theory. The theorem
has, as these things go, a fairly long history. As far as we know the first forms of it were assertions about
the representability of automorphic forms by theta series associated to quaternary quadratic forms.

Chapter 1

iii

As we said before nothing in these notes is really new. We have, in the list of references at
the end of each chapter, tried to indicate our indebtedness to other authors. We could not however
acknowledge completely our indebtedness to R. Godement since many of his ideas were communicated
orally to one of us as a student. We hope that he does not object to the company they are forced to keep.
The notes were typed by the secretaries of Leet Oliver Hall. The bulk of the work was done by
Miss Mary Ellen Peters and to her we would like to extend our special thanks. Only time can tell if the
mathematics justifies her great efforts.
New York, N.Y.
New Haven, Conn.

that appeared in the SLM volume

August, 1969

Chapter I: Local Theory

1 Weil representations. Before beginning the study of automorphic forms we must review the representation theory of the general linear group in two variables over a local field. In particular we have to
prove the existence of various series of representations. One of the quickest methods of doing this is
to make use of the representations constructed by Weil in [1]. We begin by reviewing his construction
adding, at appropriate places, some remarks which will be needed later.
In this paragraph F will be a local field and K will be an algebra over F of one of the following
types:
(i) The direct sum F F .
(ii) A separable quadratic extension of F .
(iii) The unique quaternion algebra over F . K is then a division algebra with centre F .
(iv) The algebra M (2, F ) of 2 2 matrices over F .
In all cases we identify F with the subfield of K consisting of scalar multiples of the identity. In
particular if K = F F we identify F with the set of elements of the form (x, x). We can introduce an
involution of K , which will send x to x , with the following properties:
(i) It satisfies the identities (x + y) = x + y and (xy) = y x .
(ii) If x belongs to F then x = x .
(iii) For any x in K both (x) = x + x and (x) = xx = x x belong to F .
If K = F F and x = (a, b) we set x = (b, a). If K is a separable quadratic extension of F the
involution is the unique non-trivial automorphism of K over F . In this case (x) is the trace of x and
(x) is the norm of x. If K is a quaternion algebra a unique with the required properties is known to
exist. and are the reduced trace and reduced norm respectively. If K is M (2, F ) we take to be the
involution sending




x=

to

x=

a b
c d

d b
c a

Then (x) and (x) are the trace and determinant of x.

If = F is a given non-trivial additive character of F then K = F is a non-trivial additive
character of K . By means of the pairing

x, y = K (xy)
we can identify K with its Pontrjagin dual. The function is of course a quadratic form on K which is
a vector space over F and f = F is a character of second order in the sense of [1]. Since

(x + y) (x) (y) = (xy )

and

f (x + y)f 1 (x)f 1 (y) = x, y 

the isomorphism of K with itself associated to f is just . In particular and f are nondegenerate.

Chapter 1

Let S(K) be the space of Schwartz-Bruhat functions on K . There is a unique Haar measure dx
on K such that if belongs to S(K) and

(x) =

(y) K (xy) dy
K

then

 (x) dx.

(0) =
K

The measure dx, which is the measure on K that we shall use, is said to be self-dual with respect to K .
Since the involution is measure preserving the corollary to Weils Theorem 2 can in the present
case be formulated as follows.
Lemma 1.1. There is a constant which depends on the F and K, such that for every function

in S(K)

( f )(y) K (yx) dy = f 1 (x )  (x)

f is the convolution of and f . The values of are listed in the next lemma.
Lemma 1.2 (i) If K = F F or M (2, F ) then = 1.

(ii) If K is the quaternion algebra over F then = 1.

(iii) If F = R, K = C, and

F (x) = e2iax ,

then
=

a
i
|a|

(iv) If F is non-archimedean and K is a separable quadratic extension of F let be the quadratic

character of F associated to K by local class-eld theory. If UF is the group of units of F

let m = m() be the smallest non-negative integer such that is trivial on
UFm = {a UF | 1 (mod pm
F )}

and let n = n(F ) be the largest integer such that F is trivial on the ideal pn
F . If a is any
generator on the ideal pm+n
then
F


1 () F (a1 ) d
.
= (a) UF

 UF 1 () F (a1 ) d
The first two assertions are proved by Weil. To obtain the third apply the previous lemma to the
function

(z) = e2zz .

We prove the last. It is shown by Weil that || = 1 and that if  is sufficiently large differs from


p
K

F (xx ) dx

Chapter 1

by a positive factor. This equals


p
K

F (xx ) |x|K d x =

p
K

F (xx )|xx |F d x

if d x is a suitable multiplicative Haar measure. Since the kernel of the homomorphism is compact
the integral on the right is a positive multiple of


(p
)
K

F (x) |x|F d x.

k
Set k = 2 if K/F is unramified and set k =  if K/F is ramified. Then (p
K ) = pF (K).

Since 1 + is twice the characteristic function of (K ) the factor is the positive multiple of

pk
F

F (x) dx +

pk
F

F (x) (x) dx.

For  and therefore k sufficiently large the first integral is 0. If K/F is ramified well-known properties
of Gaussian sums allow us to infer that the second integral is equal to


F



UF


a

d.

Since = 1 we obtain the desired expression for by dividing this integral by its absolute value. If
K/F is unramified we write the second integral as


j=0

In this case m = 0 and


(1)jk


pk+j
F

F (x) dx

pk+j+1
F

F (x) dx


pk+j
F

F (x) dx

is 0 if k j > n but equals q kj if k j n, where q is the number of elements in the residue class
field. Since (a) = (1)n the sum equals




1
(a) q m +
(1)j q mj 1

q
j=0

2(a)qm+1

A little algebra shows that this equals

so that = (a), which upon careful inspection is
q+1
seen to equal the expression given in the lemma.
In the notation of [19] the third and fourth assertions could be formulated as an equality

= (K/F, F ).
It is probably best at the moment to take this as the definition of (K/F, F ).
If K is not a separable quadratic extension of F we take to be the trivial character.

Chapter 1

Proposition
is a unique representation r of SL(2, F ) on S(K) such that


1.3 There


0
1/2
(x) = () ||K (x)
1




1 z
(ii) r
(x) = F (z(x))(x)
0 1




0 1
(iii) r
(x) =  (x ).
1 0
If S(K) is given its usual topology, r is continuous. It can be extended to a unitary representation
of SL(2, F ) on L2 (K), the space of square integrable functions on K. If F is archimedean and
belongs to S(K) then the function r(g) is an indenitely dierentiable function on SL(2, F ) with
values in S( K).
(i) r

This may be
from
of Weil.
We sketch

deduced

the results

 a proof. SL(2, F ) is the group generated
by the elements

0
,
0 1

1 z
, and w =
0 1

relations

(a)

0
0 1

(b)

w =

(c)

1 a
0 1

w=

a1
0

=

2

0 1
1 0
1
0

1 0
0 1

0
a



with in F and z in F subject to the


w

1 a
0 1

w

1 a1
0
1






0
1 z
together with the obvious relations among the elements of the form
and
. Thus
0 1
0 1
the uniqueness of r is clear. To prove the existence one has to verify that the mapping specified by
(i), (ii), (iii) preserves all relations between the generators. For all relations except (a), (b), and (c) this
can be seen by inspection. (a) translates into an easily verifiable property of the Fourier transform. (b)
translates into the equality 2 = (1) which follows readily from Lemma 1.2.
If a = 1 the relation (c) becomes


 (y ) F ((y))y, x  dy = F ((x))
(y)F ((y))y, x  dy
(1.3.1)
K

which can be obtained from the formula of Lemma 1.1 by replacing (y) by  (y ) and taking the
inverse Fourier transform of the right side. If a is not 1 the relation (c) can again be reduced to (1.3.1)
provided F is replaced by the character x F (ax) and and dx are modifed accordingly. We refer
to Weils paper for the proof that r is continuous and may be extended to a unitary representation of
SL(2, F ) in L2 (K).
Now take F archimedean. It is enough to show that all of the functions r(g) are indefinitely
differentiable in some neighborhood of the identity. Let



NF =

1 x
0 1




x F


Chapter 1



and let

AF =

0
0 1





 F


Then NF wAF NF is a neighborhood of the identity which is diffeomorphic to NF AF NF . It is

enough to show that

(n, a, n1 ) = r(nwan)
is infinitely differentiable as a function of n, as a function of a, and as a function of n1 and that
the derivations are continuous on the product space. For this it is enough to show that for all all
derivatives of r(n) and r(a) are continuous as functions of n and or a and . This is easily done.
The representation r depends on the choice of F . If a belongs to F and F (x) = F (ax) let

r be the corresponding representation. The constant  = (a) .
Lemma 1.4 (i) The representation r  is given by


r (g) = r

a
0

0
1


1
a
g
0

0
1



(ii) If b belongs to K let (b)(x) = (b1 x) and let (b)(x) = (xb). If a = (b) then

r  (g)(b1 ) = (b1 )r(g)

and

r  (g)(b) = (b)r(g).

In particular if (b) = 1 both (b) and (b) commute with r.

We leave the verification of this lemma to the reader. Take K to be a separable quadratic extension
of F or a quaternion algebra of centre F . In the first case (K ) is of index 2 in F . In the second case
(K ) is F if F is non-archimedean and (K ) has index 2 in F if F is R.
Let K  be the compact subgroup of K consisting of all x with (x) = xx = 1 and let G+ be the
subgroup of GL(2, F ) consisting of all g with determinant in (K ). G+ has index 2 or 1 in GL(2, F ).
Using the lemma we shall decompose r with respect to K  and extend r to a representation of G+ .
Let be a finite-dimensional irreducible representation of K in a vector space U over C. Taking
the tensor product of r with the trivial representation of SL(2, F ) on U we obtain a representation on

S(K) C U = S(K, U )
which we still call r and which will now be the centre of attention.
Proposition 1.5 (i) If S(K, ) is the space of functions in S(K, U ) satisfying

(xh) = 1 (h)(x)
for all h in K  then S(K, ) is invariant under r(g) for all g in SL(2, F ).
(ii) The representation r of SL(2, F ) on S(K, ) can be extended to a representation r of G+
satisfying




a 0
1/2
r
(x) = |h|K (h)(xh)
0 1
if a = (h) belongs to (K ).

Chapter 1

(iii) If is the quasi-character of F such that

(a) = (a)I
for a in F then

r

a
0

0
a


= (a) (a)I

(iv) The representation r is continuous and if F is archimedean all factors in S(K, ) are

innitely dierentiable.
(v) If U is a Hilbert space and is unitary let L2 (K, U ) be the space of square integrable functions
from K to U with the norm

2 =

(x)2 dx

If L2 (K, ) is the closure of S(K, ) in L2 (K, U ) then r can be extended to a unitary

representation of G+ in L2 (K, ).
The first part of the proposition is a consequence of the previous lemma. Let H be the group of
matrices of the form




a 0
0 1

with a in (K ). It is clear that the formula of part (ii) defines a continuous representation of H on
S(K, ). Moreover G+ is the semi-direct of H and SL(2, F ) so that to prove (ii) we have only to show
that















a 0
0 1

a1
0

0
1

= r

a 0
0 1

r (g) r

a1
0

0
1

Let a = (h) and let r be the representation associated F (x) = F (ax). By the first part of the
previous lemma this relation reduces to

r
(g) = (h) r (g) 1 (h),

which is a consequence of the last part of the previous lemma.

To prove (iii) observe that

a 0
0 a

=

a2
0

0
1



a1
0

0
1

and that a2 = (a) belongs to (K ). The last two assertions are easily proved.
We now insert some remarks whose significance will not be clear until we begin to discuss the
local functional equations. We associate to every in S(K, ) a function

W (g) = r (g) (1)

on G+ and a function

(a) = W

on (K ). The both take values in U .

a 0
0 1

(1.5.1)


(1.5.2)

Chapter 1

It is easily verified that

W

1 x
0 1

 
g = F (x)W (g)

If g G+ and F is a function on G+ let (g)F be the function h F (hg). Then

(g) W = Wr (g)
Let B+ be the group of matrices of the form

a x
0 1

with a in (K ). Let be the representation of B+ on the space of functions on (K ) with values in

U defined by





a 0
0 1

and

1 x
0 1

(b) = (ba)


(b) = F (bx) (b).

Then for all b in B+

(b) = r (b).

(1.5.3)

The application , and therefore the application W , is injective because

1/2

((h)) = |h|K (h) (h).

(1.5.4)

Thus we may regard r as acting on the space

of functions , S(K, ). The effect of a

V 
a 0
corresponds to the operator (a) (a)I . Since
matrix in B+ is given by (1.5.3). The matrix

0 a


0 1
G+ is generated by B+ , the set of scalar matrices, and w =
the representation r on V is
1 0
determined by the action of w. To specify this we introduce, formally at first, the Mellin transform of
= .
If is a quasi-character of F let

()

=
() () d .
(1.5.5)
(K )

Appealing to (1.5.4) we may write this as

 () = ()

=

|h|K ((h)) (h) (h) dh.

1/2

(1.5.6)

If is a quasi-character of F we sometimes write for the associated quasi-character of K .

The tensor product of and is defined by

( )(h) = (h) (h).

Chapter 1

If K : h |h|K is the module of K then

1/2

1/2

K (h) = |h|K ((h)) (h).

We also introduce, again in a purely formal manner, the integrals

(h) (h) dh

Z(, ) =
K

and

Z(

1 (h) (h) dh

, ) =
K

so that

1/2

()

= Z(K , ).

(1.5.7)

Now let  = r (w) and let  be the Fourier transform of so that r (w) (x) =  (x ). If

0 =





1 1 1/2

 1 1
=
Z

,
r
(w)

0
0
K

which equals

1 1
0 ((h))(h) (h ) d h.

Since 0 ((h)) = ((h)) = (h h) = (h )(h) this expression equals

((h))

1 ((h))1(h)  (h) d h

(h ) (h ) d h =

so that


K



1 1/2
1

.

 (1 1
)
=
Z

0
K

(1.5.8)

Take = 1 sF where 1 is a fixed quasi-character and s is complex number. If K is a separable

quadratic extension of F the representation is one-dimensional and therefore a quasi-character. The
integral defining the function
1/2

Z(K , )
is known to converge for Re s sufficiently large and the function itself is essentially a local zeta-function
in the sense of Tate. The integral defining

Z(1 K 1 ,  )
1/2

converges for Re s sufficiently small, that is, large and negative. Both functions can be analytically
continued to the whole s-plane as meromorphic functions. There is a scalar C() which depends
analytically on s such that

Z(K , ) = C()Z(1 K 1 ,  ).
1/2

1/2

All these assertions are also known to be valid for quaternion algebras. We shall return to the verification
later. The relation

()

= 1 C()
 (1 1
0 )

Chapter 1

determines  in terms of .
If is a quasi-character of F and 1 = then S(K, 1 ) = S(K, ) and

r1 (g) = (detg)r (g)

so that we may write

r1 = r

However the space V1 of functions on (K ) associated to r1 is not necessarily V . In fact

V1 = { | V }
and r1 (g) applied to is the product of (detg) with the function r (g) . Given one can always
find a such that is equivalent to a unitary representation.
If is unitary the map is an isometry because


(K )

 (a) d a =
2

(h) (h) |h|K d h =

2

(h)2 dh
K

if the measures are suitably normalized.

We want to extend some of these results to the case K = F F . We regard the element of K
as defining a row vector so that K becomes a right module for M (2, F ). If belongs to S(K) and g
belongs to GL(2, F ), we set

(g) (x) = (xg).

Proposition 1.6 (i) If K = F F then r can be extended to a representation r of GL(2, F ) such

that
r

a
0

0
1



=

a
0

0
1



for a in F .
 is the partial Fourier transform
(ii) If
 b) =
(a,


(a, y) F (by) dy
F

and the Haar measure dy is self-dual with repsect to F then

 = (g)

[r(g)]
for all in S(K) and all g in GF .
It is easy to prove part (ii) for g in SL(2, F ). In fact one has just to check it for the standard
generators and for these it is a consequence of the definitions of Proposition 1.3. The formula of part (ii)
therefore defines an extension of r to GL(2, F ) which is easily seen to satisfy the condition of part (i).
Let be a quasi-character of K . Since K = F F we may identify with a pair (1 , 2 )
of quasi-characters of F . Then r will be the representation defined by
1/2

r (g) = |detg|F 1 (detg)r(g).

Chapter 1

10

If x belongs to K and (x) = 1 then x is of the form (t, t1 ) with t in F . If belongs to S(K)

set

((t, t1 )) ((t, t1 )) d t.

(, ) =
F

Since the integrand has compact support on F the integral converges. We now associate to the
function

W (g) = (, r (g))
on GL(2, F ) and the function

(a) = W

a 0
0 1

(1.6.1)


(1.6.2)

on F . We still have

(g)W = Wr (g).


If

BF =

a x
0 1

| a F , x F

and if the representations of BF on the space of functions on F is defined in the same manner as
the representation of B+ then

(b) = r (b)
for b in BF . The applications W and are no longer injective.
If 0 is the quasi-character defined by

0 (a) = ((a, a)) = 1 (a) 2 (a)

then

a 0
0 a

 
g = 0 (a) W (g).

It is enough to verify this for g = e.

W

and

so that

r
Consequently

W

a 0
0 a

a 0
0 a

a 0
0 a
a 0
0 a



 
a 0
= , r

0 a

=

a2
0

0
1







1/2

0
a

1/2

(x, y) = |a2 |F 1 (a2 )|a|K

(ax, a1 y).

1 (a2 )1 (x)21 (x)(ax, a1 x1 ) d x


= 1 (a)2 (a)
1 (x)21 (x)(x, x1 ) d x

which is the required result.

a1
0

Chapter 1

11

Again we introduce in a purely formal manner the distribution

Z(, ) = Z(1 , 2 ) =

(x1 , x2 ) 1 (x2 ) 2 (x2 ) d x2 d x2 .

If is a quasi-character of F and = we set

()

=

() () d .
F

The integral is

 
0
() , r
d
0
1

F







0
1
1

(x, x )1 (x)2 (x) d x d

()
r
=
0
1

F
F

which in turn equals




1/2

()1 ()||F


(x, x1 )1 (x)21 (x) d x d .

Writing this as a double integral and then changing variables we obtain

(, x)1 ()2 (x)|x|F d d x

1/2



1/2
1/2
()

= Z 1 F , 2 F , .

so that

(1.6.3)

Let  = r (w) . Then



1 1 1/2
1 1 1/2

 (1 1
)
=
Z

,
r
(w)

0
2
1
F
F
which equals



 (y, x)1 21 (x)1 11 (y)|xy|F d x d y

1/2

so that

1 1

 (1 1
1 F , 1 21 F ,  ).
0 ) = Z(
1/2

1/2

(1.6.4)

Suppose = 1 sF where 1 is a fixed quasi-character and s is a complex number. We shall see that
the integral defining the right side of (1.6.3) converges for Re s sufficiently large and that the integral
defining the right side of (1.6.4) converges for Re s sufficiently small. Both can be analytically continued
to the whole complex plane as meromorphic functions and there is a meromorphic function C() which
is independent of such that

Z(1 F , 2 F ) = C()Z(111 F , 1 21 F ,  ).
1/2

Thus

1/2

1/2

()

= C()
 (1 1
0 )

The analogy with the earlier results is quite clear.

1/2

Chapter 1

12

2 Representations of GL(2, F ) in the non-archimedean case. In this and the next two paragraphs
the ground field F is a non- archimedean local field. We shall be interested in representations of
GF = GL(2, F ) on a vector space V over C which satisfy the following condition.
(2.1) For every vector v in V the stabilizer of v in GF is an open subgroup of GF .

Those who are familiar with such things can verify that this is tantamount to demanding that the
map (g, v) (g)v of GF V into V is continuous if V is given the trivial locally convex topology in
which every semi-norm is continuous. A representation of GF satisfying (2.1) will be called admissible
if it also satisfies the following condition
(2.2) For every open subgroup G of GL(2, OF ) the space of vectors v in V stablizied by G is

nite-dimensional.
OF is the ring of integers of F .
Let HF be the space of functions on GF which are locally constant and compactly supported.
Let dg be that Haar measure on GF which assigns the measure 1 to GL(2, OF ). Every f in HF may be
identified with the measure f (g) dg . The convolution product


f1 (g) f2(g 1 h) dg

f1 f2 (h) =
GF

turns HF into an algebra which we refer to as the Hecke algebra. Any locally constant function
on GL(2, OF ) may be extended to GF by being set equal to 0 outside of GL(2, OF ) and therefore
may be regarded as an element of HF . In particular if i , 1 i r , is a family of inequivalent
finite-dimensional irreducible representations of GL(2, OF ) and

i (g) = dim(i ) tr i (g 1 )
for g in GL(2, OF ) we regard i as an element of HF . The function

r


i=1

is an idempotent of HF . Such an idempotent will be called elementary.

Let be a representation satisfying (2.1). If f belongs to HF and v belongs to V then f (g) (g)v
takes on only finitely many values and the integral


f (g) (g)v dg = (f )v
GF

may be defined as a finite sum. Alternatively we may give V the trivial locally convex topology and use
some abstract definition of the integral. The result will be the same and f (f ) is the representation
of HF on V . If g belongs to GF then (g)f is the function whose value at h is f (g1 h). It is clear that

((g)f ) = (g) (f ).
Moreover

Chapter 1

13

(2.3) For every v in V there is an f in HF such that f (v) = v.

In fact f can be taken to be a multiple of the characteristic function of some open and closed
neighborhood of the identity. If is admissible the associated representation of HF satisfies
(2.4) For every elementary idempotent of HF the operator () has a nite-dimensional range.

We now verify that from a representation of HF satisfying (2.3) we can construct a representation of GF satisfying (2.1) such that


(f ) =

f (g) (g) dg.

GF

By (2.3) every vector v in V is of the form

r


v=

(fi ) vi

i=1

with vi in V and fi in HF . If we can show that

r


(fi ) vi = 0

(2.3.1)

i=1

implies that

w=

r



(g)fi vi
i=1

is 0 we can define (g)v to be

r



(g)fi vi
i=1

will clearly be a representation of GF satisfying (2.1).

Suppose that (2.3.1) is satisifed and choose f in HF so that (f )w = w. Then
r



w=
f (g)fi vi .
i=1

If (g)f (h) = f (hg) then

so that

f (g)fi = (g 1 )f fi

r
r


 1
 1  
w=
(g )f fi vi = (g )f
(fi )vi = 0.
i=1

i=1

It is easy to see that the representation of GF satisfies (2.2) if the representation of HF satisfies
(2.4). A representation of HF satisfying (2.3) and (2.4) will be called admissible. There is a complete
correspondence between admissible representations of GF and of HF . For example a subspace is
invariant under GF if and only if it is invariant under HF and an operator commutes with the action
of GF if and only if it commutes with the action of HF .

Chapter 1

14

>From now on, unless the contrary is explicitly stated, an irreducible representation of GF or HF
is to be assumed admissible. If is irreducible and acts on the space V then any linear transformation
of V commuting with HF is a scalar. In fact if V is assumed, as it always will be, to be different
from 0 there is an elementary idempotent such that () = 0. Its range is a finite-dimensional space
invariant under A. Thus A has at least one eigenvector and is consequently a scalar. In particular there
is a homomorphism of F into C such that

a 0
0 a



= (a)I

for all a in F . By (2.1) the function is 1 near the identity and is therefore continuous. We shall
refer to a continuous homomorphism of a topological group into the multiplicative group of complex
numbers as a quasi-character.
If is a quasi-character of F then g (detg) is a quasi-character of GF . It determines a
one-dimensional representation of GF which is admissible. It will be convenient to use the letter to
denote this associated representation. If is an admissible reprentation of GF on V then will be
the reprenentation of GF on V defined by

( )(g) = (detg)(g).
It is admissible and irreducible if is.
Let be an admissible representation of GF on V and let V be the space of all linear forms on
V . We define a representation of HF on V by the relation

v, (f )v  = (f)v, v 
where f(g) = f (g 1 ). Since will not usually be admissible, we replace V by V = (HF )V .
The space V is invariant under HF . For each f in HF there is an elementary idempotent such that
f = f and therefore the restriction
 of to V satisfies (2.3). It is easily seen that if is an elementary

 is admissible we have to verify that

idempotent so is . To show that

V () =
()V = ()V


= ()V
and let Vc = 1 ()
V . V is clearly the direct sum of V ()
,
is finite-dimensional. Let V ()
which is finite-dimensional, and Vc . Moreover V () is orthogonal to Vc because

()v,
v = 0.
v ()v,
()
v = ()v
and is therefore
It follows immediately that V () is isomorphic to a subspace of the dual of V ()

finite-dimensional. It is in fact isomorphic to the dual of V () because if v annihilates Vc then, for all
v in V ,
v  = 0
v, ()v  v, v  = v ()v,

so that ()v = v .

 will be called the representation contragradient to . It is easily seen that the natural map of
V into V is an isomorphism and that the image of this map is
 (HF )V so that may be identified
.
with the contragredient of
If V1 is an invariant subspace of V and V2 = V1 \ V we may associate to representations 1 and
2 on V1 and V2 . They are easily seen to be admissible. It is also clear that there is a natural embedding


of V2 in V . Moreover any element
v1 of

 V1 lies in V1 () for some and therefore is determined by its

effect on V1 (). It annihilates I () V1 . There is certainly a linear function v on V which annihilates



V and agrees with V1 on V1 ()
. v is necessarily in V so that V1 may be identified with V2 \ V .
I ()
Since every representation is the contragredient of its contragredient we easily deduce the following
lemma.

Chapter 1

15

2 is the annihilator of V1 in V then

Lemma 2.5 (a). Suppose V1 is an invariant subspace of V . If V
V1 is the annihilator of V2 in V .
(b) is irreducible if and only if
 is.
Observe that for all g in GF

(g)v, 
v = v,
(g 1 )
v.
If is the one-dimensional representation associated to the quasi-character then
 = 1 . Moreover
.
if is a quasi-character and any admissible representation then the contragredient of is 1
Let V be a separable complete locally convex space and a continuous representation of GF
on V . The space V0 = (HF )V is invariant under GF and the restriction 0 of to V0 satisfies (2.1).
Suppose that it also satisfies (2.2). Then if is irreducible in the topological sense 0 is algebraically
irreducible. To see this take any two vectors v and w in V0 and choose an elementary idempotent so
that ()v = v . v is in the closure of (HF )w and therefore in the closure of (HF )w ()V . Since,
by assumption, ()V is finite dimensional, v must actually lie in (HF )w.
The equivalence class of is not in general determined by that of 0 . It is, however, when
is unitary. To see this one has only to show that, up to a scalar factor, an irreducible admissible
representation admits at most one invariant hermitian form.
Lemma 2.6 Suppose 1 and 2 are irreducible admissible representations of GF on V1 and V2 respectively. Suppose A(v1 , v2 ) and B(v1 , v2 ) are non-degenerate forms on V1 V2 which are linear
in the rst variable and either both linear or both conjugate linear in the second variable. Suppose
moreover that, for all g in GF



A 1 (g)v1 , 2 (g)v2 = A(v1 , v2 )
and



B 1 (g)v1 , 2 (g)v2 = B(v1 , v2 )

Then there is a complex scalar such that

B(v1 , v2 ) = A(v1 , v2 )
Define two mappings S and T of V2 into V1 by the relations

A(v1 , v2 ) = v1 , Sv2 

and

B(v1 , v2 ) = v1 , T v2 ,

Since S and T are both linear or conjugate linear with kernel 0 they are both embeddings. Both take
V2 onto an invariant subspace of V1 . Since V1 has no non-trivial invariant subspaces they are both
isomorphisms. Thus S 1 T is a linear map of V2 which commutes with GF and is therefore a scalar I .
The lemma follows.
An admissible representation will be called unitary if it admits an invariant positive definite
hermitian form.
We now begin in earnest the study of irreducible admissible representations of GF . The basic
ideas are due to Kirillov.

Chapter 1

16

Proposition 2.7. Let be an irreducible admissible representation of GF on the vector space V .

(a) If V is nite-dimensional then V is one-dimensional and there is a quasi-character of F

such that
(g) = (detg)
 
(b) If V is innite dimensional there is no nonzero vector invariant by all the matrices 10 x1 ,
x F.

If is finite-dimensional its kernel H is an open subgroup. In particular there is a positive

number @ such that




1 x
0 1

belongs to H if |x| < @. If x is any element of F there is an a in F such that |ax| < @. Since

a1
0

0
a



1 ax
0 1

the matrix



a 0
0 1

=

1 x
0 1

1 x
0 1

belongs to H for all x in F . For similar reasons the matrices

1 0
y 1

do also. Since the matrices generate SL(2, F ) the group H contains SL(2, F ). Thus (g1 )(g2 ) =
(g2 )(g1 ) for all g1 and g2 in GF . Consequently each (g) is a scalar matrix and (g) is onedimensional. In fact

(g) = (detg)I
where is a homorphism of F into C . To see that is continuous we need only observe that

a 0
0 1


= (a)I.

Suppose V contains a nonzero vector v fixed by all the operators

1 x
0 1


.

Let H be the stabilizer of the space Cv . To prove the second part of the proposition we need only verify
that H is of finite index in GF . Since it contains the scalar matrices and an open subgroup of GF it will
be enough to show that it contains SL(2, F ). In fact we shall show that H0 , the stabilizer of v , contains
SL(2, F ). H0 is open and therefore contains a matrix

a b
c d

Chapter 1

17

with c = 0. It also contains

If x =

b0
c

1 ac1
0
1

y then



a b
c d
1 0
y 1



1 dc1
0
1

= w0

=

1 x
0 1

0 b0
c 0


= w0 .

w01

also belongs to H0 . As before we see that H0 contains SL(2, F ).

Because of this lemma we can confine our attention to infinite-dimensional representations. Let
= F be a nontrivial additive character of F . Let BF be the group of matrices of the form

b=

a x
0 1

with a in F and x in F . If X is a complex vector space we define a representation of BF on the

space of all functions of F with values in X by setting


(b) () = (x)(a).

leaves the invariant space S(F , X) of locally constant compactly supported functions. is
continuous with respect to the trivial topology on S(F , X).
Proposition 2.8. Let be an innite dimensional irreducible representation of GF on the space V .

Let p = pF be the maximal ideal in the ring of integers of F , and let V  be the set of all vectors v
in V such that





1 x
F (x)
v dx = 0
0 1
pn

for some integer n. Then

(i) The set V  is a subspace of V .
(ii) Let X = V  \ V and let A be the natural map of V onto X. If v belongs to V let v be the
function dened by




 
a 0
v (a) = A
v .
0 1
The map v v is an injection of V into the space of locally constant functions on F with
value in X.
(iii) If b belongs to BF and v belongs to V then
(b)v = (b)v .
If m n so that pm contains pn then

(x)
pm

is equal to


ypm /pn

(y)

1 y
0 1

1 x
0 1


v dx

 

(x)
pn

1 x
0 1


v dx.

Thus if the integral of the lemma vanishes for some integer n it vanishes for all larger integers. The
first assertion of the proposition follows immediately.
To prove the second we shall use the following lemma.

Chapter 1

18

Lemma 2.8.1 Let p m be the largest ideal on which is trivial and let f be a locally constant function

on p with values in some nite dimensional complex vector space. For any integer n  the
following two conditions are equivalent
(i) f is constant on the cosets of pn in p
(ii) The integral

(ax) f (x) dx
p

is zero for all a outside of pm+n .

Assume (i) and let a be an element of F which is not in pm+n . Then x (ax) is a
non-trivial character of pn and


(ax) f (x) dx =
p




(ax) dx f (y) = 0.

(ay)
pn

yp /pn

f may be regarded as a locally constant function on F with support in p . Assuming (ii) is

tantamount to assuming that the Fourier transform F  of f has its support in pm+n . By the Fourier
inversion formula

f (x) =
(xy) f  (y) dy.
pm+n

If y belongs to pm+n the function x (xy) is constant on cosets of pn . It follows immediately

that the second condition of the lemma implies the first.
To prove the second
of the proposition we show that if v vanishes identically then v

assertion
is fixed by the operator 10 x1 for all x in F and then appeal to Proposition 2.7.
Take





f (x) =

1 x
0 1

v.

The restriction of f to an ideal in F takes values in a finite-dimensional subspace of V . To show that

f is constant on the cosets of some ideal pn it is enough to show that its restriction to some ideal p
containing pn has this property.
By assumption there exists an n0 such that f is constant on the cosets of pn0 . We shall now
show that if f is constant on the cosets of pn+1 it is also constant on the cosets of pn . Take any ideal
p containing pn . By the previous lemma


(ax) f (x) dx = 0
p

if a is not in pm+n1 . We have to show that the integral on the left vanishes if a is a generator of
pm+n1 .
If UF is the group of units of OF the ring of integers of F there is an open subgroup U1 of UF
such that





b 0
0 1

v=v

for b in U1 . For such b

b 0
0 1




(ax) f (x) dx =
p

(ax)
p

b 0
0 1





1 x

v dx
0 1

Chapter 1

19

is equal to


(ax)
p

1 bx
0 1






 a 
b 0

v dx =
x f (x) dx.
0 1
b
p

Thus it will be enough to show that for some sufficiently large  the integral vanishes when a is taken
to be one of a fixed set of representatives of the cosets of U1 in the set of generators of pm+n1 . Since
there are only finitely many such cosets it is enough to show that for each a there is at least one  for
which the integral vanishes.
By assumption there is an ideal a(a) such that

(x)
a(a)

But this integral equals

|a|

a 0
0 1

1 x
0 1



a 0
0 1

 



(ax)

a1 a(a)

v dx = 0

1 x
0 1


v dx

so that  = (a) could be chosen to make

p = a1 a(a).
To prove the third assertion we verify that

 
1 y
A
v = (y) A(v)
0 1

(2.8.2)

for all v in V and all y in F . The third assertion follows from this by inspection. We have to show that

is in V  or that, for some n,

(x)
pn

1 x
0 1

(x)
pn



v (y)v










1 y
1 x

v dx
(x) (y)
v dx
0 1
0 1
pn

is zero. The expression equals

1 y
0 1

1 x+y
0
1




v dx

(x + y)
pn

1 x
0 1


v dx.

If pn contains y we may change the variables in the first integral to see that it equals the second.
It will be convenient now to identify v with v so that V becomes a space of functions on F
with values in X . The map A is replaced by the map (1). The representation now satisfies

(b) = (b)
if b is in BF . There is a quasi-character 0 of F such that

a 0
0 a



If

w=
the representation is determined by 0 and (w).

= 0 (a) I.
0 1
1 0



Chapter 1

20

Proposition 2.9 (i) The space V contains

V0 = S(F , X)
(ii) The space V is spanned by V0 and (w)V0 .
For every in V there is a positive integer n such that

a x
0 1


=

if x and a 1 belong to pn . In particular (a) = (a) if belongs to F and a 1 belongs to pn .

The relation

(x)() = ()
for all x in pn implies that () = 0 if the restriction of to pn is not trivial. Let pm be the largest
ideal on which is trivial. Then () = 0 unless || |C|mn if C is a generator of p.
Let V0 be the space of all C in V such that, for some integer  depending on , () = 0 unless
|| > |C| . To prove (i) we have to show that V0 = S(F , X). It is at least clear that S(F , X) contains
V0 . Moreover for every in V and every x in F the difference

=
is in V0 . To see this observe that

1 x
0 1





 () = 1 (x) ()

is identically zero for x = 0 and otherwise vanishes at least on x1 pm . Since there is no function in
V invariant under all the operators





1 x
0 1

the space V0 is not 0.

Before continuing with the proof of the proposition we verify a lemma we shall need.
Lemma 2.9.1 The representation of BF in the space S(F ) of locally constant, compactly sup-

ported, complex-valued functions on F is irreducible.

For every character of UF let be the function on F which equals on UF and vanishes off
UF . Since these functions and their translates span S(F ) it will be enough to show that any non-trivial
invariant subspace contains all of them. Such a space must certainly contain some non-zero function
which satisfies, for some character of UF , the relation

(a@) = (@) (a)

for all a in F and all @ in UF . Replacing by a translate if necessary we may assume that (1) = 0.
We are going to show that the space contains if is different from . Since UF has at least two
characters we can then replace by some with different from , and replace by and by to
see it also contains .

Chapter 1

21

Set


(@)

UF



@ 0
0 1

1 x
0 1


d@

where x is still to be determined. is to be different form .  belongs to the invariant subspace and

 (a@) = (@) (a)

for all a in F and all @ in UF . We have

1 (@)(@)(ax@) d@

(a) = (a)
UF

The character 1 has a conductor pn with n positive. Take x to be of order n m. The integral,
which can be rewritten as a Gaussian sum, is then, as is well-known, zero if a is not in UF but different
from zero if a is in UF . Since (1) is not zero  must be a nonzero multiple of .
To prove the first assertion of the proposition we need only verify that if u belongs to X then V0
contains all functions of the form ()u with in S(F ). There is a in V such that (1) = u.
Take x such that (x) = 1. Then





 =

1 x
0 1

is in V0 and  (1) = 1 (x) u. Consequently every u is of the form (1) for some in V0 .
If is a character of UF let V0 () be the space of functions in V0 satisfying

(a@) = (@)(a)
for all a in F and all @ in UF . V0 is clearly the direct sum of the space V0 (). In particular every vector
u in X can be written as a finite sum


u=

i (1)

where i belongs to some V0 (i ).

If we make use of the lemma we need only show that if u can be written as u = (1) where is
in V0 () for some then there is at least one function in V0 of the form ()u where is a nonzero
function in S(F ). Choose different from and let pn be the conductor of 1 . We again consider

 =

1 (@)

UF

where x is of order n m. Then



(a) = (a)

@ 0
0 1



1 x
0 1


d@

1 (@)(@)F (ax@) d@

UF

The properties of Gaussian sums used before show that  is a function of the required kind.
The second part of the proposition is easier to verify. Let PF be the group of upper-triangular
matrices in GF . Since V0 is invariant under PF and V is irreducible under GF the space V is spanned
by V0 and the vectors





 =

1 x
0 1

(w)

Chapter 1

22

with in V0 . But

 = { (w)} + (w)

and as we saw,  (w) is in V0 . The proposition is proved.

To study the effect of w we introduce a formal Mellin transform. Let C be a generator of p. If
is a locally constant function on F with values in X then for every integer n the function @ (@Cn )
on UF takes its values in a finite-dimensional subspace of X so that the integral


(@C n )(@) =
n ()
UF

is defined. In this integral we take the total measure of UF to be 1. It is a vector in X . (,

 t) will be
the Formal Laurent series


tn
n ()

If is in V the series has only a finite number of terms with negative exponent. Moreover the series
(,
 t) is different from zero for only finitely many . If belongs to V0 these series have only finitely
 t) vanish then = 0.
many terms. It is clear that if is locally constant and all the formal series (,
Suppose takes values in a finite-dimensional subspace of X , is a quasi-character of F , and
the integral


(a)(a) d a

(2.10.1)

is absolutely convergent. If  is the restriction of to UF this integral equals

(C n @)  (@) d@ =

zn

UF

zn
n (  )

if z = (C). Consequently the formal series (

  , t) converges absolutely for t = z and the sum is
equal to (2.10.1). We shall see that X is one dimensional and that there is a constant c0 = c0 () such
that if |(C)| = |C|c with c > c0 then the integral (2.10.1) is absolutely convergent. Consequently all
 t) have positive radii of convergence.
the series (,
If = F is a given non-trivial additive character of F , any character of UF , and x any element
of F we set


(, x) =

(@) (@x) d@
UF

The integral is taken with respect to the normalized Haar measure on UF . If g belongs to GF , belongs
to V , and  = (g) we shall set

(g) (,
 t) =
 (, t).

Proposition 2.10 (i) If belongs to UF and  belongs to Z then

C
0

0
1



 t)
(,
 t) = t 1 () (,

(ii) If x belongs to F then

1
0

x
1


(,
 t) =


n

, C x)
n ()
n

Chapter 1

23

where the inner sum is taken over all characters of UF

(iii) Let 0 be the quasi-character dened by

a
0

0
a


= 0 (a) I

for a in F . Let 0 be its restriction to UF and let z0 = 0 (C). For each character of UF
there is a formal series C(, t) with coecients in the space of linear operators on X such
that for every in V0

0
1

1
0



(,
 t) = C(, t) (
 1 01 , t1 z01 ).

Set


=
Then

 (, t) =

C 0
0
1


.


tn

(@) (C n+ @) d@.
UF

Changing variables in the integration and in the summation we obtain the first formula of the proposition.
Now set





1 x
0 1

 =

Then

 (, t) =


tn

(C n @x) (@) (C n@) d@.

UF

By Fourier inversion

(C n @) =

n () 1 (@).

The sum on the right is in reality finite. Substituting we obtain

 (, t) =

tn



1 (@) (@C n x) d@
n ()
UF

as asserted.
 t) = 0 unless = 1 01 . This
Suppose is a character of UF and in V0 is such that (,
means that

(a@) 0 (@) (a)

or that

@ 0
0 1


= 0 (@)

Chapter 1

24

for all @ in UF . If  = (w) then

@ 0
0 1



=

Since

1 0
0 @
the expression on the right is equal to

@ 0
0 1



=



(w) = (w)

@ 0
0 @



@1
0

0
1

1 0
0 @


.



1 (@)(w) = 1 (@) ,
so that
 (, t) = 0 unless = .
Now take a vector u in X and a character of UF and let be the function in V0 which is zero
outside of UF and on UF is given by

(@) = (@) 0 (@)u.



If = (w) then

n

(2.10.2)

is a function of n, , and u which depends linearly on u and we may write

n () = Cn ()u
where Cn () is a linear operator on X .
We introduce the formal series

C(, t) =

tn Cn ().

We have now to verify the third formula of the proposition. Since is in V0 the product on the right
is defined. Since both sides are linear in we need only verify it for a set of generators of V0 . This
set can be taken to be the functions defined by (2.10.2) together with their translates of power C . For
functions of the form (2.10.2) the formula is valid because of the way the various series C(, t) were
defined. Thus all we have to do is show that if the formula is valid for a given function it remains
valid when is replaced by





C 0
0 1

By part (i) the right side is replaced by

z0 t C(, t) (
 1 01 , t1 z01 ).

Since

(w)

C 0
0 1



=

1 0
0 C


(w)

and (w)(,
 t) is known we can use part (i) and the relation

1 0
0 C

C 0
0 C



C 0
0
1

to see that the left side is replaced by

z0 t (w)(,
 t) = z0 t C(, t)(
 1 01 , t1 z01 ).
For a given u in X and a given character of UF there must exist a in V such that

(,
 t) =

tn Cn ()u

Consequently there is an n0 such that Cn ()u = 0 for n < n0 . Of course n0 may depend on u and
. This observation together with standard properties of Gaussian sums shows that the infinite sums
occurring in the following proposition are meaningful, for when each term is multiplied on the right
by a fixed vector in X all but finitely many disappear.

Chapter 1

25

Proposition 2.11 Let p be the largest ideal on which is trivial.

(i) Let and be two characters of UF such that 0 is not 1. Let pm be its conductor. Then

( 1 , C n )( 1 , C p )Cp+n ()

is equal to

( 1 1 01 , C m )z0m+ 0 (1)Cnm ()Cpm ()

for all integers n and p.

(ii) Let be any character of UF and let  = 1 01 . Then

( 1 , C n )( 1 , Cp)Cp+n ()

is equal to
z0p 0 (1)n,p

+ (|C|

1)1 z0 +1 Cn1 ()Cp1 (

)

z r Cn+r ()Cp+r (
)

for all integers n and p.

The left hand sums are taken over all characters of UF and n,p is Kroneckers delta. The
relation















0 1
1 0

1 1
0 1

0 1
1 0

1 1
0 1

0 1
1 0

1 1
0 1

implies that

(w)

1 1
0 1



(w) = 0 (1)

1 1
0 1



(w)

1 1
0 1



for all in V0 . Since (w) is not necessarily in V0 we write this relation as














1 1
1 1
1 1
(w)
(w) (w) + 2 (w) = 0 (1)
(w)
.
0 1
0 1
0 1
The term 2 (w) is equal to 0 (1).
We compute the Mellin transforms of both sides

and

(w)

1 1
0 1

1 1
0 1


(,
 t) =


(,
 t) =

tn

tn

(1 , C n )
n ()

(1 1 01 , C p )z0P Cp+n ()
p ()

p,

so that the Mellin transform of the right side is

0 (1)


n

tn


p,,

( 1 , C n )(1 1 01 , C p )z0p Cp+n ()

p ().

(2.11.1)

Chapter 1

26

On the other hand

(w)(,
 t) =

tn

and

1 1
0 1


(w)(,
 t) =

so that

is equal to


n

tn

1 1
0 1


n

z0p Cp+n ()
p ( 1 01 )

tn

z0p (1 , C n )Cp+n ()
p (1 01 )

p,


(w)(,
 t) (w)(,
 t)

z0p [(0 , C n ) (0 )]Cp+n (1 01 )

p ().

p,

Here (0 ) is 1 if 0 is the trivial character and 0 otherwise. The Mellin transform of the left hand
side is therefore

tn

z0pr [( 1 , C r )( 1 )]Cn+r ()Cp+r (1 01 )

p ()+0 (1)

tn
n (). (2.11.2)

p,r,

The coefficient of tn
p () in (2.11.1) is

0 (1)

( 1 , C n ) (1 1 01 , C p )z01 Cp+n ()

(2.11.3)

and in (2.11.2) it is

[( 1 , C r ) ( 1 )]z0pr Cn+r ()Cp+r (1 01 ) + 0 (1)n, ( 1 )I

(2.11.4)

These two expressions are equal for all choice of n, p, , and .

If = and the conductor of 1 is pm the gaussian sum ( 1 , C r ) is zero unless r = m.
Thus (2.11.4) reduces to

( 1 , C m )z0pm Cnm ()Cpm (1 01 ).

Since

(, x) = (1) (, x)

the expression (2.11.3) is equal to

1 (1)

( 1 , C n ) (1 1 01 C p )z0p Cp+n ().

1 01

Replacing by
we obtain the first part of the proposition.
If = then ( 1 ) = 1. Moreover, as is well-known and easily verified, (1 , C r ) = 1 if
r ,

( 1 , C 1 ) = |C|(|C| 1)1

and ( 1 , C r ) = 0 if r  2. Thus (2.11.4) is equal to

0 (1)n,pI+(|C|1)1 z0p+ +1 Cn 1 ()Cn 1 ( 1 01 )


r= 2

The second part of the proposition follows.

z0pr Cn+r ()Cn+r ( 1 01 ).

Chapter 1

27

Proposition(2.12) (i) For every n, p, and

Cn ()Cp () = Cp ()Cn ()
(ii) There is no non-trivial subspace of X invariant under all the operators Cn ().
(iii) The space X is one-dimensional.
Suppose 0 = 1. The left side of the first identity in the previous proposition is symmetric in
the two pairs (n, ) and (p, ). Since (1 1 01 , C m ) is not zero we conclude that

Cnm () Cpm () = Cpm () Cnm ()

for all choices of n and p. The first part of the proposition is therefore valid in = .
Now suppose = . We are going to that if (p, n) is a given pair of integers and u belongs to X
then

Cn+r ()Cp+r (
)u = Cp+r (
)Cn+r ()u
for all r in Z. If r  0 both sides are 0 and the relation is valid so the proof can proceed by induction
on r . For the induction one uses the second relation of Proposition 2.11 in the same way as the first was
used above.
Suppose X1 is a non-trivial subspace of X invariant under all the operators Cn (). Let V1 be
the space of all functions in V0 which take values in X1 and let V1 be the invariant subspace generated
by V1 . We shall show that all functions in V1 take values in X1 so that V1 is a non-trivial invariant
subspace of V . This will be a contradiction. If in V takes value in X1 and g belongs to PF then (g)
also takes values in X1 . Therefore all we need to do is show that if is in V1 then (w) takes values
in X1 . This follows immediately from the assumption and Proposition 2.10.
To prove (iii) we show that the operators Cn () are all scalar multiples of the identity. Because
of (i) we need only show that every linear transformation of X which commutes with all the operators
Cn () is a scalar. Suppose T is such an operator. If belongs to V let T be the function from F to
X defined by



T (a) = T (a) .

Observe that T is still in V . This is clear if belongs to V0 and if = (w)0 we see on examining
the Mellin transforms of both sides that

T = (w)T 0 .
Since V = V0 + (w)V0 the observation follows. T therefore defines a linear transformation of V which
clearly commutes with the action of any g in PF . If we can show that it commutes with the action of w
it will follow that it and, therefore, the original operator on X are scalars. We have to verify that

(w)T = T (w)
at least for on V0 and for = (w)0 with 0 in V0 . We have already seen that the identity holds for
in V0 . Thus if = (w)0 the left side is

(w)T (w)0 = 2 (w)T 0 = 0 (1)T 0

and the right side is

T 2 (w)0 = 0 (1)T 0.
Because of this proposition we can identify X with C and regard the operators Cn () as complex
numbers. For each r the formal Laurent series C(, t) has only finitely many negative terms. We now
want to show that the realization of on a space of functions on F is, when certain simple conditions
are imposed, unique so that the series C(, t) are determined by the class of and that conversely the
series C(, t) determine the class of .

Chapter 1

28

Theorem 2.13 Suppose an equivalence class of innite-dimensional irreducible admissible represen-

tations of GF is given. Then there exists exactly one space V of complex-valued functions on F
and exactly one representation of GF on V which is in this class and which is such that
(b) = (b)
if b is in BF and is in V .

We have proved the existence of one such V and . Suppose V  is another such space of functions
and a representation of GF on V  which is equivalent to . We suppose of course that


 (b) = (b)
if b is in BF and is in V  . Let A be an isomorphism of V with V  such that A(g) =  (g)A for all g .
Let L be the linear functional

L() = A(1)
on V . Then

 
a 0
L
= A(a)
0 1

so that A is determined by L. If we could prove the existence of a scalar such that L() = (1) it
would follow that

A(a) = (a)
for all a such that A = . This equality of course implies the theorem.
Observe that

 




1 x
1 x

L
=
A(1) = (x)L().
0 1
0 1

(2.13.1)

Thus we need the following lemma.

Lemma 2.13.2 If L is a linear functional on V satisfying (2.13.1) there is a scalar such that

L() = (1).
This is a consequence of a slightly different lemma.
Lemma 2.13.3 Suppose L is a linear functional on the space S(F ) of locally constant compactly

supported functions on F such that

1
L
0

x
1

 
= (x) L()

for all in S(F ) and all x in F . Then there is a scalar such that L() = (1).
Suppose for a moment that the second lemma is true. Then given a linear functional L on V
satisfying (2.13.1) there is a such that L() = (1) for all in V0 = S(F ). Take x in F such that
(x) = 1 and in V . Then

 




 
1 x
1 x
L() = L
+L
.
0 1
0 1

Chapter 1

29

Since

1 x
0 1



is in V0 the right side is equal to

(1) (x)(1) + (x)L()

so that





1 (x) L() = 1 (x) (1)

which implies that L() = (1).

To prove the second lemma we have only to show that (1) = 0 implies L() = 0. If we set
(0) = 0 then becomes a locally constant function with compact support in F . Let  be its Fourier
transform so that


(ba)  (b) db.

(a) =
F

Let be an open compact subset of F containing 1 and the support of . There is an ideal a in F
so that for all a in the function  (b)(ba) is constant on the cosets of a in F . Choose an ideal b
containing a and the support of  . If S is a set of representatives of b/a and if c is the measure of a
then


(ba) (b).

(a) = c

bS

If 0 is the characteristic function of this relation may be written

b

bS

with b = c (b). If (1) = 0 then

1 b
0 1


0

b (b) = 0

bS

so that


b

1 b
0 1


0 (b)0

It is clear that L() = 0.

The representation of the theorem will be called the Kirillov model. There is another model
which will be used extensively. It is called the Whittaker model. Its properties are described in the next
theorem.

Chapter 1

30

Theorem 2.14 (i)

For any in V set



W (g) = (g) (1)

so that W is a function in GF . Let W (, ) be the space of such functions. The map W is

an isomorphism of V with W (, ). Moreover
W(g) = (g)W
(ii) Let W () be the space of all functions W on GF such that

W

1
0

x
1


g = (x)W (g)

for all x in F and g in G. Then W (, ) is contained in W () and is the only invariant

subspace which transforms according to under right translations.

Since

a 0
0 1



 
a 0
=
(1) = (a)
0 1

the function W is 0 only if is. Since

(g)W (h) = W (hg)

the relation

W(g) = (g)W
is clear. Then W (, ) is invariant under right translation and transforms according to .
Since



 






1 x
0 1

1 x
0 1

(g) (1) = (x){(g)(1)}

the space W (, ) is contained in W (). Suppose W is an invariant subspace of W () which transforms according to . There is an isomorphism A of V with W such that



A (g) = (g)(A).
Let

L() = A(1).
Since



L (g) = A(g)(1) = (g)A(1) = A(g)

the map A is determined by L. Also

 




1 x
1 x
L
= A
= (x)A(1) = (x)L()
0 1
0 1
so that by Lemma 2.13.2 there is a scalar such that

L() = (1).

Chapter 1

31

Consequently A = W and W = W (, ).
The realization of on W (, ) will be called the Whittaker model. Observe that the representation of GF on W () contains no irreducible finite-dimensional representations. In fact any such
representation is of the form

(g) = (detg).
If were contained in the representation on W () there would be a nonzero function W on GF such
that



 

1 x
0 1

= (x)(detg)W (e)

In particular taking g = e we find that

W

1 x
0 1


= (x)W (e)

However it is also clear that

W

1 x
0 1




1 x
= det
W (e) = W (e)
0 1

so that (x) = 1 for all x. This is a contradiction. We shall see however that is a constituent of the
representation on W (). That is, there are two invariant subspaces W1 and W2 of W () such that W1
contains W2 and the representation of the quotient space W1 /W2 is equivalent to .
Proposition 2.15 Let and  be two innite-dimensional irreducible representations of GF realized

in the Kirillov form on spaces V and V  . Assume that the two quasi-characters dened by

a
0

0
a


= (a)I

a
0

0
a



=  (a)I

are the same. Let {C(, t)} and {C  (, t)} be the families of formal series associated to the two
representations. If
C(, t) = C  (, t)
for all then =  .
If belongs to S(F ) then, by hypothesis,

(w)(,
 t) =  (w)(,
 t)
so that (w) =  (w). Since V is spanned by S(F ) and (w)S(F ) and V  is spanned by S(F )
and  (w)S(F ) the spaces V and V  are the same. We have to show that

(g) =  (g)
for all in V and all g in GF . This is clear if g is in PF so it is enough to verify it for g = w.
We have already observed that (w)0 =  (w)0 if 0 is in S(F ) so we need only show that
(w) =  (w) if is of the form (w)0 with 0 in S(F ). But (w) = 2 (w)0 = (1)0 and,
since (w)0 =  (w)0 ,  (w) =  (1)0 .

Chapter 1

32

Let NF be the group of matrices of the form

1 x
0 1

with x in F and let B be the space of functions on GF invariant under left translations by elements
of NF . B is invariant under right translations and the question of whether or not a given irreducible
representation is contained in B arises. The answer is obviously positive when = is onedimensional for then the function g (detg) is itself contained in B.
Assume that the representation which is given in the Kirillov form acts on B. Then there is a
map A of V into B such that

A(g) = (g)A
If L() = A(1) then

L

1 x
0 1

 
= L()

(2.15.1)

for all in V and all x in F . Conversely given such a linear form the map A defined by



A(g) = L (g)
satisfies the relation A(g) = (g)A and takes V into B. Thus is contained in B if an only if there is
a non-trivial linear form L on V which satisfies (2.15.1).
Lemma 2.15.2 If L is a linear form on S(F ) which satises (2.15.1) for all x in F and for all

in S(F ) then L is zero.

We are assuming that L annihilates all functions of the form

1 x
0 1



so it will be enough to show that they span S(F ). If belongs to S(F ) we may set (0) = 0 and
regard as an element of S(F ). Let  be its Fourier transform so that

 (b)(bx) db.

(x) =
F

Let be an open compact subset of F containing the support of and let pn be an ideal containing
. There is an ideal a of F such that  (b)(bx) is, as a function of b, constant on cosets of a for all x
in pn . Let b be an ideal containing both a and the support of  . If S is a set of representatives for the
cosets of a in b, if c is the measure of a, and if 0 is the characteristic function of then

(x) =

b (bx)0 (x)

bS

if b = c (b). Thus


b

b

1 b
0 1


0 .

Chapter 1

33

Since (0) = 0 we have

b = 0

so that


b






1 b
0 0
b
0 1

as required.
Thus any linear form on V verifying (2.15.1) annihilates S(F ). Conversely any linear form on
V annihilating S(F ) satisfies (2.15.1) because

1 x
0 1



is in S(F ) if is in V . We have therefore proved

Proposition 2.16 For any innite-dimensional irreducible representation the following two prop-

erties are equivalent:

(i) is not contained in B.
(ii) The Kirillov model of is realized in the space S(F ).
A representation satisfying these two conditions will be called absolutely cuspidal.
Lemma 2.16.1 Let be an innite-dimensional irreducible representation realized in the Kirillov
form on the space V . Then V0 = S(F ) is of nite codimension in V .

We recall that V = V0 + (w)V0 . Let V1 be the space of all in V0 with support in UF . An

element of (w)V0 may always be written as a linear combination of functions of the form

Cp 0
0 1


(w)

with in V1 and p in Z. For each character of UF let be the function in V1 such that (@) =
(@)0 (@) for @ in UF . Then

 (, t) = (0 )

and

(w)
 (, t) = (1 )C(, t).

Let = (w) . The space V is spanned by V0 and the functions

Cp 0
0 1



For the moment we take the following two lemmas for granted.

Chapter 1

34

F there is an integer n0 and a family of constants i ,

Lemma 2.16.2 For any character of U
1 i p, such that

Cn () =

p


i Cni ()

i=1

for n n0 .
Lemma 2.16.3 There is a nite set S of characters of UF such that for not in S the numbers
Cn () are 0 for all but nitely many n.

If is not in S the function is in V0 . Choose in S and let V be the space spanned by the
functions


Cp 0
0 1

and the functions in V0 satisfying (a@) = (a)1 (@) for all a in F and all @ in UF . It will be
enough to show that V /V V0 is finite-dimensional.
 t) = 0 unless = and we may identify with the sequence {
n ()}.
If is in V then (,
The elements of V V0 are the elements corresponding to sequences with only finitely many nonzero
terms. Referring to Proposition 2.10 we see that all of the sequences satisfying the recursion relation

n () =

p


a
ni ()

i=1

for n n1 . The integer n1 depends on .

Lemma 2.16.1 is therefore a consequence of the following elementary lemma whose proof we
postpone to Paragraph 8.
Lemma 2.16.4 Let i , 1 i p, be p complex numbers. Let A be the space of all sequences {an },

n Z for which there exist two integers n1 and n2 such that

an =

i ani

1ip

for n n1 and such that an = 0 for n n2 . Let A0 be the space of all sequences with only a nite
number of nonzero terms. Then A/A0 is nite-dimensional.
We now prove Lemma 2.16.2. According to Proposition 2.11

( 1 , C n )( 1 , C p )Cp+n ()

is equal to

z0p 0 (1)n,p

+ (|C|

1)1 z0 +1 Cn1 ()Cp1 (

)

z0r Cn+r ()Cp+r (

).

Remember that p is the largest ideal on which is trivial. Suppose first that  = .

Chapter 1

35

Take p =  and n > . Then (n p) = 0 and

( 1 , C n )( 1 , C p ) = 0
unless = . Hence

Cn () = (|C|

1)1 z0 +1 Cn1 ()C2 1 ()

z0r Cn+r ()C +r ()

which, since almost all of the coefficients C +r () in the sum are zero, is the relation required.
If =  take p  and n > p. Then (1 , C n ) = 0 unless = and (1 , C p ) = 0
unless = . Thus

(|C| 1)1 z0 +1 Cn1 ()Cp1 (

)

z0r Cn+r ()Cp+r (

) = 0.

(2.16.5)

There is certainly at least one i for which Ci (

) = 0. Take p 1  i. Then from (2.16.5) we deduce
a relation of the form

Cn+r () =

q


i Cn+ri ()

i=1

where r is a fixed integer and n is any integer greater than p.

Lemma 2.16.3 is a consequence of the following more precise lemma. If pm is the conductor of a
character we refer to m as the order of .
Lemma 2.16.6 Let m 0 be of the order 0 and let m1 be an integer greater than m0 . Write 0 in any

manner in the form 0 = 11 21 where the orders of 1 and 2 are strictly less than m1 . If the
order m of is large enough
C2m2 () = 21 (1)z0m

(11 , C m )
(2 1 , C m )

and Cp () = 0 if p = 2m 2.
Suppose the order of is at least m1 . Then 1 0 = 21 is still of order m. Applying
Proposition 2.11 we see that

( 1 1 , C n+m+ )( 1 , C p+m+ )Cp+n+2m+2 ()

is equal to

(11 1 01 , C m )z0m+ 1 0 (1)Cnm ()Cpm ()

for all integers n and p. Choose n such that Cn (1 ) = 0. Assume also that m + n +   or that
m 2 n. Then (1 1 , C n+m+ ) = 0 unless = 1 so that

(11 , C p+m+ )Cp+n+2m+2 (1 ) = (2 1 , C m z0m+ 1 0 (1)Cn (1 )Cp ().

Chapter 1

36

Since 11 is still of order m the left side is zero unless p = 2m 2. The only term on the right side
that can vanish is Cp (). On the other hand if p = 2m 2 we can cancel the terms Cn (1 ) from both
side to obtain the relation of the lemma.
Apart from Lemma 2.16.4 the proof of Lemma 2.16.1 is complete. We have now to discuss its
consequences. If 1 and 2 are two quasi-characters of F let B(1 , 2 ) be the space of all functions
on GF which satisfy
(i) For all g in GF , a1 , a2 in F , and x in F

a1
0

x
a

 1/2
 
 a1 
g = 1 (a1 )2 (a2 )   (g).
a2

(ii) There is an open subgroup U of GL(2, OF ) such that (gu) (g) for all u in U .
Since

GF = NF AF GL(2, OF )
where AF is the group of diagonal matrices the elements of B(1 , 2 ) are determined by their restrictions to GL(2, OF ) and the second condition is tantamount to the condition that be locally constant.
B(1 , 2 ) is invariant under right translations by elements of GF so that we have a representation
(1 , 2 ) of GF on B(1 , 2 ). It is admissible.
Proposition 2.17 If is an innite-dimensional irreducible representation of GF which is not absolutely cuspidal then for some choice of 1 and 2 it is contained in (1 , 2 ).

We take in the Kirillov form. Since V0 is invariant under the group PF the representation
defines a representation of PF on the finite-dimensional space V /V0 . It is clear that is trivial on
NF and that the kernel of is open. The contragredient representation has the same properties. Since
PF /NF is abelian there is a nonzero linear form L on V /V0 such that

a1
0

x
a2



1
L = 1
1 (a1 )1 (a2 )L

for all a1 , a2 , and x. 1 and 2 are homomorphisms of F into C which are necessarily continuous.
L may be regarded as a linear form on V . Then

a1
L
0
If is in V let A be the function

x
a2

 
= 1 (a1 )2 (a2 )L().



A(g) = L (g)

on GF . A is clearly an injection of V into B(1 , 2 ) which commutes with the action of GF .

Before passing to the next theorem we make a few simple remarks. Suppose is an infinitedimensional irreducible representation of GF and is a quasi-character of F . It is clear that W (
, ) consists of the functions

g W (g)(detg)

with W on W (, ). If V is the space of the Kirillov model of the space of the Kirillov model of
consists of the functions a (a)(a) with in V . To see this take in the Kirillov form and observe

Chapter 1

37

first of all that the map A : is an isomorphism of V with another space V  on which GF acts
by means of the representation  = A( )A1 . If

b

x
0 1

belongs to BF and  = then

 (b) (a) = (a){()(ax)(a)} = (ax) (a)

so that  (b) = (b) . By definition then  is the Kirillov model of . Let 1 be the restriction
of to UF and let z1 = (C). If  = then

 (, t) = (

1 , z1 t).
Thus

 (w) (, t) = (w)(

 1 11 01 , z01 z11 t1 ).
1 , z1 t) = C(1 , z1 t)(v

The right side is equal to

C(1 , z1 t)
 ( 1 01 11 , z01 z12 t1 )

so that when we replace by we replace C(, t) by C(1 , z1 t).

Suppose  (x) = (bx) with b in F is another non-trivial additive character. Then W (,  )
consists of the functions



 

b 0
0 1

W  (g) = W

with W in W (, ).
The last identity of the following theorem is referred to as the local functional equation. It is the
starting point of our approach to the Hecke theory.
Theorem 2.18 Let be an irreducible innite-dimensional admissible representation of G F .

(i) If is the quasi-character of GF dened by

a
0

0
a


= (a)I

then the contragredient representation

 is equivalent to 1 .
(ii) There is a real number s0 such that for all g in GF and all W in W (, ) the integrals


W
F


W
F

a
0

0
1

a
0

0
1

 
g |a|s1/2 d a = (g, s, W )

 
 s, W )
g |a|s1/2 1 (a) d a = (g,

converge absolutely for Re s > s0 .

(iii) There is a unique Euler factor L(s, ) with the following property: if

(g, s, W ) = L(s, )(g, s, W )

Chapter 1

38

then (g, s, W ) is a holomorphic function of s for all g and all W and there is at least one
W in W (, ) such that
(e, s, W ) = as
where a is a positive constant.
(iv) If

 s, W ) = L(s,
 s, W )
(g,
)(g,

there is a unique factor @(s, , ) which, as a function of s, is an exponential such that



0
1

1
0


g, 1 s, W

= @(s, , )(g, s, W )

for all g in GF and all W in W (, ).

To say that L(s, ) is an Euler product is to say that L(s, ) = P 1 (q s ) where P is a polynomial
with constant term 1 and q = |C|1 is the number of elements in the residue field. If L(s, ) and
L (s, ) were two Euler factors satisfying the conditions of the lemma their quotient would be an entire
function with no zero. This clearly implies uniqueness.
If is replaced by  where  (x) = (bx) the functions W are replaced by the functions W 
with



 

W  (g) = W

and

b 0
0 1

(g, s, W  ) = |b|1/2s (g, s, W )

while

 s, W ).
(g, s, W  ) = |b|1/2s (b)(g,

Thus L(s, ) will not depend on . However

@(s, ,  ) = (b) |b|2s1 @(s, , ).

According to the first part of the theorem if W belongs to W (, ) the function

 (g) = W (g) 1 (detg)

W
is in W (
, ). It is clear that

 s, W ) = (detg)(g, s, W
)
(g,

 s, W ) is a
so that if the third part of the theorem is valid when is replaced by
 the function (g,
 one sees that
holomorphic function of s. Combining the functional equation for and for
@(s, , )@(1 s,
, ) = (1).
Let V be the space on which the Kirillov model of acts. For every W in W (, ) there is a
unique in V such that





a 0
0 1

= (a).

Chapter 1

39

If is itself the canonical model

(w)(a) = W

where

w=

a 0
0 1

0 1
1 0

 
w


.

If is any quasi-character of F we set


()

=

(a)(a) d a
F

if the integral converges. If 0 is the restriction of to UF then



()

=
 0 , (C) .
Thus if F is the quasi-character F (x) = |x| and the appropriate integrals converge
s1/2

) = (1,
 q 1/2s )

s1/2

1 ) = (
 01 , z01 q 1/2s )

(e, s, W ) = (
 F
(e, s, W ) = (
 F

if 0 is the restriction of to UF and z0 = (). Thus if the theorem is valid the series (1,
 t) and
1
(
 0 , t) have positive radii of convergence and define functions which are meromorphic in the whole
t-plane.
It is also clear that

(w, 1 s, W ) = (w)(
 01 , z01 q s1/2 ).

If belongs to V0 then
1/2 1/2

(w)(
 01 , z01 q 1/2 t) = C(01 , z0

t)(1,
 q 1/2 t1 ).

Choosing in V0 such that (1,

 t) 1 we see that C(01 , t) is convergent in some disc and has an
analytic continuation to a function meromorphic in the whole plane.
Comparing the relation
1/2 1/2 s

(w)(
 01 z01 q 1/2 q s ) = C(01 , z0

q )(1,
 q 1/2 q s )

with the functional equation we see that

C(01 , z01 q 1/2 q s ) =

L(1 s,
)@(s, , )
.
L(s, )

Replacing by we obtain the formula

1 1 1/2 s
C(01 1
q )=
0 , z0 z1 q

)@(s, , )
L(1 s, 1
.
L(s, )

Appealing to Proposition 2.15 we obtain the following corollary.

(2.18.1)

Chapter 1

40

Corollary 2.19 Let and  be two irreducible innite-dimensional representations of GF . Assume

that the quasi-characters and  dened by

a
0

0
a


= (a)I

a
0

0
a



=  (a)I

are equal. Then and  are equivalent if and only if

)@(s, , )
 )@(s,  , )
L(1 s, 1
L(1 s, 1
=
L(s, )
L(s,  )
for all quasi-characters.
We begin the proof of the first part of the theorem. If 1 and 2 are numerical functions on F

we set

1 , 2  =

1 (a)2 (a) d a.

The Haar measure is the one which assigns the measure 1 to UF . If one of the functions is in S(F )
and the other is locally constant the integral is certainly defined. By the Plancherel theorem for UF

,   =


n

(1)
n ()
n ( 1 ).

The sum is in reality finite. It is easy to se that if b belongs to B

 (b), (b) ) = ,  .
Suppose is given in the Kirillov form and acts on V . Let  , the Kirillov model of 1 ,
act on V  . To prove part (i) we have only to construct an invariant non-degenerate bilinear form on
V V  . If belongs to V0 and  belongs to V  or if belongs to V and  belongs to V0 we set

(,  ) = ,  .
If and  are arbitrary vectors in V and V  we may write = 1 + (w)2 and  = 1 +  (w)2
with , 2 in V0 and 1 , 2 in V0 . We want to set

(,  ) = 1 , 1  + 1 ,  (w)2  + (w)2 , 1  + 2 , 2 .
The second part of the next lemma shows that is well defined.
Lemma 2.19.1 Let and  belong to V0 and V0 respectively. Then

(i)

(w),   = 0 (1),  (w) 

(ii) If either (w) belongs to V0 or  (w) belongs to V0 then

(w),  (w)  = ,  .

Chapter 1
The relation

41

(w)(,
 t) =

tn

implies that

(w),   =

Cn+p ()
p ( 1 01 )z0p

(1)Cn+p ()
p ( 1 01 )z0p
n ( 1 ).

(2.19.2)

n,p,

Replacing by  replaces by 1 , 0 by 01 , z0 by z01 , and C(, t) by C(01 , z01 t). Thus

, (w)  =

(1)Cn+p (01 )z0n

p ( 1 0 )
n ( 1 ).

(2.19.3)

n,p,

Replacing by 0 in (2.19.3) and comparing with (2.19.2) we obtain the first part of the lemma.
Because of the symmetry it will be enough to consider the second part when (w) belongs to
V0 . By the first part

(w),  (w)  = 0 (1) 2(w),   = ,  .

It follows immediately from the lemma that



(w),  (w) = (,  )
so that to establish the invariance of we need only show that



(p),  (p) = (,  )
for all triangular matrices p. If is in V0 or  is in V0 this is clear. We need only verify it for in
(w)V0 and  in  (w)V0 .
If is in V0 ,  is in V0 and p is diagonal then





(p)(w),  (p)  (w) = (w)(p1 ),  (w)  (p1 )
where p1 = w1 pw is also diagonal. The right side is equal to





(p1 ),  (p1 ) = (,  ) = (w),  (w)p .
Finally we have to show that*

 





1 x
1 x

 = (,  )

,
0 1
0 1

(2.19.2)

for all x in F and all and  . Let i , 1 < i < r , generate V modulo V0 and let j , 1 j r  , generate
V  modulo V0 . There certainly is an ideal a of F such that

1 x
0 1


i = i

* The tags on Equations 2.19.2 and 2.19.3 have inadvertently been repeated.

Chapter 1

42

and

1 x
0 1



j = j

for all i and j if x belongs to a. Then

 





1 x
1 x

j = (i , j ).

i ,
0 1
0 1
Since 2.19.2 is valid if x is in a and is in V0 or  is in V0 it is valid for all and  provided that x is
in a. Any y in F may be written as ax with a in F and x in a. Then

1 y
0 1

=

a 0
0 1



1 x
0 1



a1
0

0
1

and it follows readily that

 

1 y
1 y

 = (,  ).
C, 
0 1
0 1
Since is invariant and not identically zero it is non-degenerate. The rest of the theorem will now
be proved for absolutely cuspidal representations. The remaining representations will be considered
in the next chapter. We observe that since W (, ) is invariant under right translations the assertions
need only be established when g is the identity matrix e.
 
If is absolutely cuspidal then V = V0 = S(F ) and W a0 01 = (a) is locally constant with
 s, W ) are absolutely convergent
compact support. Therefore the integrals defining (e, s, W ) and (e,
for all values of s and the two functions are entire. We may take L(s, ) = 1. If is taken to be the
characteristic function of UF then (e, s, W ) = 1.
Referring to the discussion preceding Corollary 2.19 we see that if we take

@(s, , ) = C(01 , z01 q 1/2 q s )

the local functional equation of part (iv) will be satisfied. It remains to show that @(s, , ) is an
exponential function or, what is at least ast strong, to show that, for all , C(, t) is a multiple of a
power of t. It is a finite linear combination of powers of t and if it is not of the form indicated it has a
zero at some point different form 0. C(01 , z01 t1 ) is also a linear combination of powers of t and
so cannot have a pole except at zero. To show that C(, t) has the required form we have only to show
that

C(, t)C( 101 , z01 t1 ) = 0 (1).

(2.19.3)

Choose in V0 and set  = (w). We may suppose that  (, t) = 0. The identity is obtained by
combining the two relations

 (, t) = C(, t)(
 1 01 , z01 t1 )
and

0 (1)(
 1 01 , t) = C( 1 01 , t)
 (, z01 t1 ).

We close this paragraph with a number of facts about absolutely cuspidal representations which
will be useful later.

Chapter 1

43

Proposition 2.20 Let be an absolutely cuspidal representation of GF . If the quasi-character

dened by

a
0

0
a


= (a)I

is actually a character then is unitary.

As usual we take and
 in the Kirillov form. We have to establish the existence of a positivedefinite invariant hermitian form on V . We show first that if belongs to V and
 belongs to V then
there is a compact set in GF such that if



ZF =

a 0
0 a




a F


the support of (g), 

 , a function of g , is contained in ZF . If AF is the group of diagonal matrices
GF = GL(2, OF ) AF GL(2, OF ). Since and
 are both invariant under subgroups of finite index in
GL(2, OF ) it is enough to show that the function (b), 
 on AF has support in a set ZF with
compact. Since



 

a 0
0 a

b , 
 = (a)(b), 


it is enough to show that the function



a 0
0 1


, 


has compact support in F . This matrix element is equal to


F

(ax)(x)

d x.

Since and
 are functions with compact support the result is clear.
Choose
0 = 0 in V and set

(1 , 2 ) =

ZF \GF

(g)1 ,
0 (g)2,
0  dg.

This is a positive invariant hermitian form on V .

We have incidentally shown that is square-integrable. Observe that even if the absolutely
cuspidal representation is not unitary one can choose a quasi-character such that is unitary.
If is unitary there is a conjugate linear map A : V V defined by

(1 , 2 ) = 1 , A2 .
Clearly A (b) = (b)A for all b in BF . The map A0 defined by

A0 (a) = (a)
has the same properties. We claim that

A = A0
with in C . To see this we have only to apply the following lemma to A1
0 A.

Chapter 1

44

Lemma 2.21.1. Let T be a linear operator on S(F ) which commutes with (b) for all b in BF .

Then T is a scalar.
Since is irreducible it will be enough to show that T has an eigenvector. Let p be the largest
ideal on which is trivial. Let be a non-trivial charcter of UF and let pn be its conductor. T commutes
with the operator








@ 0
0 1

1 (@)

S=
UF

1 C n
0
1

d@

and it leaves the range of the restriction of S to the functions invariant under UF invariant. If is such
a function


1 (@)(a@C n ) d@.

S(a) = (a)
UF

The Gaussian sum is 0 unless a lies in UF . Therefore S is equal to (1) times the function which is
zero outside of UF and equals on UF . Since T leaves a one-dimensional space invariant it has an
eigenvector.
Since A = A0 the hermitian form (1 , 2 ) is equal to

1 (a)2 (a) d a.

Proposition 2.21.2. Let be an absolutely cuspidal representation of G F for which the quasi-character

dened by

a
0

0
a


= (a)I

is a character.
(i) If is in the Kirillov form the hermitian form

1 (a)2 (a) d a
F

is invariant.
(ii) If |z| = 1 then |C(, z)| = 1 and if Res = 1/2
|@(s, , )| = 1.
Since |z0 | = 1 the second relation of part (ii) follows from the first and the relation

@(s, , ) = C(01 , q s1/2 z01 ).

If n is in Z and is a character of UF let

(@C m ) = n,m (@)0 (@)

for m in Z and @ in UF . Then


F

|(a)|2 da = 1.

If  = (w) and C(, t) = C ()t then

 (@C m ) = n,m C ()z0n 1 (@).

Since |z0 | = 1


F

| (a)|2 da = |C ()|2 .

Applying the first part of the lemma we see that, if |z| = 1, both |C ()|2 and |C(, z)|2 = |C ()|2 |z|2
are 1.

Chapter 1

45

Proposition 2.22. Let be an irreducible representation of GF . It is absolutely cuspidal if and only

if for every vector v there is an ideal a in F such that

1
0

x
1


v dx = 0.

It is clear that the condition cannot be satisfied by a finite dimensional representation. Suppose
that is infinite dimensional and in the Kirillov form. If is in V then

if and only if

1 x
0 1


dx = 0


(a)

(ax) dx = 0
a

for all a. If this is so the character x (ax) must be non-trivial on a for all a in the support of . This
happens if and only if is in S(F ). The proposition follows.
Proposition 2.23. Let be an absolutely cuspidal representation and assume the largest ideal on

which is trivial is OF . Then, for all characters , Cn () = 0 if n 1.

Take a character and choose n1 such that Cn1 () = 0. Then Cn () = 0 for n = n1 . If
 = 1 01 then, as we have seen,

C(, t)C(
, t1 z01 ) = 0 (1)
so that

Cn (
) = 0
for n = n1 and

Cn1 ()Cn1 (
) = 0 (1)z0n1 .

In the second part of Proposition 2.11 take n = p = n1 + 1 to obtain

( 1 , C n1 +1 )( 1 , C n1 +1 )C2n1 +2 () = z0n1 +1 0 (1) + (|C| 1)1 z0 Cn1 ()Cn1 (

).

The right side is equal to

z0n1 +1 0 (1)

|C|
.
|C| 1

Assume n1 1. Then (1 , C n1 +1 ) is 0 unless = and (1 , C n1 +1 ) is 0 unless = . Thus

the left side is 0 unless = . However if =  the left side equals C2n1 +2 (). Since this cannot be
zero 2n1 + 2 must be equal n1 so that n1 = 2. This is a contradiction.

Chapter 1

46

3. The principal series for non-archimedean fields. In order to complete the discussion of the previous
pragraph we have to consider representations which are not absolutely cuspidal. This we shall now
do. We recall that if 1 , 2 is a pair of quasi-characters of F the space B(1 , 2 ) consists of all locally
constant functions f on GF which satisfy



f

x
a2

a1
0

 1/2
 
 a1 
g = 1 (a1 )2 (a2 )   f (g)
a2

(3.1)

for all g in GF , a1 , a2 , in F , and x in F . (1 , 2 ) is the representation of GF on B(a , 2 ).

Because of the Iwasawa decomposition GF = PF GL(2, OF ) the functions in B(1 , 2 ) are
determined by their restrictions to GL(2, OF ). The restriction can be any locally constant function on
GL(2, OF ) satisfying
 

a1
0

x
a2

= 1 (a1 )2 (a2 )f (g)

for all g in GL(2, OF ), a1 , a2 in UF , and x in OF . If U is an open subgroup of GL(2, OF ) the restriction

of any function invariant under U is a function on GL(2, OF )/U which is a finite set. Thus the space
of all such functions is finite dimensional and as observed before (1 , 2 ) is admissible.
Let F be the space of continuous functions f on GF which satisfy

f

x
a2

a1
0

   
 a1 
g =   f (g)
a2
1/2

for all g in GF , a1 , a2 in F , and x in F . We observe that F contains B(F , F

The Haar measure on GF if suitably normalized satisfies
1/2

f (g) dg =
GF

NF

AF

 1
 a1 
  f (nak) dn da dk
 
GL(2,OF ) a2

if

a=
It follows easily from this that

). GF acts on F.

a1
0

0
a2


.


f (k) dk
GL(2,OF )

is a GF -invariant linear form on F. There is also a positive constant c such that

f (g) dg = c
GF

Consequently

NF

AF

NF

 1


 
 a1 
  f na 0 1 n1 dn da dn1 .
 a2 
1 0


f (k) dk = c
GL(2,OF )

f
F

0 1
1 0



1 x
0 1


dx.

1
If 1 belongs to B(1 , 2 ) and 2 belongs to B(1
1 , 2 ) then 1 2 belongs to F and we set


1 , 2  =

1 (k) 2(k) dk.

GL(2,OF )

Chapter 1

47

Clearly

(g)1, (g)2  = 1 , 2 

so that this bilinear form is invariant. Since both 1 and 2 are determined by their restrictions to
1
GL(2, OF ) it is also non-degenerate. Thus (1
1 , 2 ) is equivalent to the contragredient of (1 , 2 ).
In Proposition 1.6 we introduced a representation r of GF and then we introduced a representation r = r1 ,2 . Both representations acted on S(F 2 ). If

(a, b) =

(a, y)(by) dy
F

is the partial Fourier transform

and

[r(g)] = (g)

(3.1.1)

r1 ,2 (g) = 1 (detg) |detg|1/2 r(g).

(3.1.2)

We also introduced the integral

(1 , 2 ; ) =

1
1 (t)1
) d t
2 (t)(t, t

and we set

W (g) = (1 , 2 ; r1 ,2 (g)).

(3.1.3)

The space of functions W is denoted W (1 , 2 ; ).

If is a quasi-character of F and if |(C)| = |C|s with s > 0 the integral

z(, ) =

(0, t)(t) d t

F
s
is defined for all in S(F 2 ). In particular if |1 (C)1
1 (C)| = |C| with s > 1 we can consider the
function

f (g) = 1 (detg) |detg|1/2 z(F 1 1

2 , (g))

on GF . Recall that F (a) = |a|. Clearly

(h)f = f
if

(3.1.4)

= 1 (deth) |deth|1/2 (h).

We claim that f belongs to B(1 , 2 ). Since the stabilizer of every under the representation
g 1 (detg) |detg|1/2 (g) is an open subgroup of GF the functions f are locally constant. Since the
space of functions f is invariant under right translations we need verify (3.1) only for g = e.






 



a1 x
a1 x
1
= z 1 2 F ,
(a1 a2 ) |a1 a2 |1/2 .
f
0 a2
0 a2
By definition the right side is equal to

1 (a1 a2 )|a1 a2 |1/2

1 (t)1
2 (t) |t| (0, a2 t) d t.

Changing variables we obtain

 1/2 
 a1 

1 (t)1
1 (a1 )2 (a2 )  
2 (t) |t| (0, t) d t.
a2

The integral is equal to f (e). Hence our assertion.

Chapter 1

48

s
Proposition 3.2. Assume |1 (C)1
2 (C)| = |C| with s > 1.

(i) There is a linear transformation A of W (1 , 2 ; ) into B(1 , 2 ) which for all in S(F 2 ),

sends W to f.
(ii) A is bijective and commutes with right translations.
To

establish
 the first part of the proposition we have to show that f is 0 if W is. Since

NF AF

0 1
1 0

NF is a dense subset of GF this will be a consequence of the following lemma.

Lemma 3.2.1. If the assumptions of the proposition are satised then, for all in S(F 2 ), the function

1/2
1
W
2 (a) |a|

a
0

0
1



is integrable with respect to the additive Haar measure on F and


W

a
0

0
1



1/2
1
(ax) da = f
2 (a) |a|

0
1

1
0



1
0

x
1


.

By definition

f

0 1
1 0



1 x
0 1



(t, tx)1 (t)1

2 (t) |t| d t

while

W

a 0
0 1



1/2
= 1 (a)1
1
2 (a) |a|
2 (a)

(at, t1 ) 1 (t) 1
2 (t) d t.

(3.2.2)

After a change of variable the right side becomes

(t, at1 )1 (t)1

2 (t) d t.

Computing formally we see that


W

a 0
0 1



1/2
(ax) da
1
2 (a) |a|

is equal to



(ax)

(t, at
F

)1 (t)1
2 (t) d t


da =
F

1 (t)1
2 (t)



(t, at


)(ax) da d t

which in turn equals


F

1 (t)1
2 (t) |t|


F




(t, a)(axt) da d t =
F

(t, xt)1 (t)1

2 (t) |t| d t.

Chapter 1

49

Our computation will be justified and the lemma proved if we show that the integral

is convergent. It equals

|(t, at1 )1 (t)| d t da


F

|(t, a)| |t|s+1 d t da

F

which is finite because s is greater than 1.

To show that A is surjective we show that every function f in B(1 , 2 ) is of the form f for
some in S(F 2 ). Given f let (x, y) be 0 if (x, y) is not of the form (0, 1)g for some g in GL(2, OF )
but if (x, y) is of this form let (x, y) = 1
1 (detg)f (g). It is easy to see that is well-defined and
2
belongs to S(F ). To show that f = f we need only show that f (g) = f (g) for all g in GL(2, OF ).
If g belongs to GL(2, OF ) then ((0, t)g) = 0 unless t belongs to UF so that

((0, t)g)1(t)1
2 (t) dt.

f (g) = 1 (detg)
UF

Since

((0, t)g) =

1
1
1 (t)2 (detg)f

1 0
0 t

 
1
g = 1
1 (t)2 (t)2 (detg)f (g)

the required equality follows.

Formulae (3.1.2) to (3.1.4) show that A commutes with right translations. Thus to show that A is
injective we have to show that W (e) = 0 if f is 0. It follows from the previous lemma that


a 0
W
0 1




a 0
is zero for almost all a if f is 0. Since W
is a locally constant function on F it must
0 1
vanish everywhere.
We have incidentally proved the following lemma.
s
Lemma 3.2.3 Suppose |1 (C)1
2 (C)| = |C| with s > 1 and W belongs to W (1 , 2 ; ). If

a
0

0
1



=0

for all a in F then W is 0.

Theorem 3.3 Let 1 and 2 be two quasi-characters of F .
1
(i) If neither 1 1
2 nor 1 2 is F the representations (1 , 2 ) and (2 , 2 ) are equivalent

and irreducible.
1/2
1/2
(ii) If 1 1
. Then B(1 , 2 ) contains a unique proper
2 = F write 1 = F , 2 = F
invariant subspace Bs (1 , 2 ) which is irreducible. B(2 , 1 ) also contains a unique proper
invariant subspace Bf (2 , 1 ). It is one-dimensional and contains the function (detg).
Moreover the GF -modules Bs (1 , 2 ) and B(2 , 1 )/Bf (2 , 1 ) are equivalent as are the
modules B(1 , 2 )/Bs (1 , 2 ) and Bf (2 , 1 ).
We start with a simple lemma.

Chapter 1

50

Lemma 3.3.1 Suppose there is a non-zero function f in B(1 , 2 ) invariant under right translations
1/2

1/2

by elements of NF . Then there is a quasi-character such that 1 = F

and 2 = F and
f is a multiple of .


Since NF AF 01 1
0 NF is an open subset of GF the function f is determined by its value at
 0 1 
1 0 . Thus if 1 and 2 have the indicated form it must be a multiple of .
If c belongs to F then

1 0
c 1

=

c1
0

1
c



0 1
1 0



1 c1
0 1


.

1
Thus if f exists and = 2 1
1 F

f

1 0
c 1



= (c)f

0 1
1 0


.

Since f is locally constant there is an ideal a in F such that is constant on a {0}. It follows
immediately that is identically 1 and that 1 and 2 have the desired form.
The next lemma is the key to the theorem.
Lemma 3.3.2. If | 1 2 (C)| = |C|s with s > 1 there is a minimal non-zero invariant subspace X

of B(1 , 2 ). For all f in B(1 , 2 ) and all n in NF the dierence f (n)f belongs to X.

By Proposition 3.2 it is enough to prove the lemma when B(1 , 2 ) is replaced by

W (1 , 2 ; ). Associate to each function W in W (1 , 2 ; ) a function

(a) = W

a 0
0 1



on F . If is 0 so is W . We may regard = (1 , 2 ) as acting on the space V of such functions. If b

is in BF

(b) = (b).
Appealing to (3.2.2) we see that every function in V has its support in a set of the form


{a F  |a| c}
where c = c() is a constant. As in the second paragraph the difference (n) = (n) is
in S(F ) for all n in NF . Thus V S(F ) is not 0. Since the representation of BF on S(F ) is
irreducible, V and every non-trivial invariant subspace of V contains S(F ). Taking the intersection
of all such spaces we obtain the subspace of the lemma.
s
We first prove the theorem assuming that |1 (C)1
2 (C)| = |C| with s > 1. We have
1
1
defined a non-degenerate pairing between B(1 , 2 ) and B(1 , 2 ). All elements of the orthogonal
complement of X are invariant under NF . Thus if 1 1
2 is not F the orthogonal complement is 0 and
1
X is B(1 , 2 ) so that the representation is irreducible. The contragredient representation (1
1 , 2 )
is also irreducible.
1/2
1/2
. In this case X is the space of the functions
If 1 1
2 = F we write 1 = F , 2 = F
1
1
1
1
orthogonal to the function in B(1 , 2 ). We set Bs (1 , 2 ) = X and we let Bf (1
1 , 2 ) be
1
the space of scalar multiples of . The representation of GF on Bs (1 , 2 ) is irreducible. Since

Chapter 1

51

Bs (1 , 2 ) is of codimension one it is the only proper invariant subspace of B(1 , 2 ). Therefore

1
1
1
Bf (1
1 , 2 ) is the only proper invariant subspace of B(1 , 2 ).
1
1
If |1 (C)2 (C)| = |C|s then |1 (C)2 (C)| = |C|s and either s > 1 or s > 1. Thus if
1
1 2 is neither F nor 1
F the representation = (1 , 1 ) is irreducible. If = 1 2 then

a 0
0 a


= (a)I

1
1
1
so that is equivalent to
or to (1
1 , 2 ). It is easily seen that (1 , 2 ) is equivalent
1
1
to (1 , 1 ) = (2 , 1 ).
= F and is the restriction of to Bs (1 , 2 ) then
is the representation on
If 1 1
2
1
1
1
1
1
B(1
,

)/B
(
,

)
defined
by
(
,

)
.
Thus

is
equivalent
to the tensor product of
f
1
2
1
2
1
2
= 1 2 and this representation. The tensor product is of course equivalent to the representation on
1/2
1/2
B(2 , 1 )/Bf (2 , 1 ). If 1 = F and 2 = F
the representations on B(1 , 2 )/Bs(1 , 2 )
and Bf (2 , 1 ) are both equivalent to the representations g (detg).
The space W (1 , 2 ; ) has been defined for all pairs 1 , 2 .

Proposition 3.4 (i) For all pairs 1 , 2

W (1 , 2 ; ) = W (2 , 1 ; )
(ii) In particular if 1 1
= 1
2
F the representation of GF on W (1 , 2 ; ) is equivalent to

(1 , 2 ).
If is a function on F 2 define by

(x, y) = (y, x).

To prove the proposition we show that, if is in S(F 2 ),





1 (detg) |detg|1/2 1 , 2 ; r(g) = 2 (detg) |detg|1/2 2 , 1 ; r(g) .
If g is the identity this relation follows upon inspection of the definition of (1 , 2 ; ). It is also easily
seen that

r(g) = [r(g)]
if g is in SL(2, F ) so that it is enough to prove the identity for

g=
It reduces to


1 (a)

(at, t

)1 (t)1
2 (t) d t

The left side equals


1 (a)

a 0
0 1


.


= 2 (a)

(at, t1 )2 (t)1
2 (t) d t.

(t1 , at)1 (t)1

2 (t) d t

which, after changing the variable of integration, one sees is equal to the right side.

Chapter 1

52

1
If 1 1
2 is not F or F so that (1 , 2 ) is irreducible we let (1 , 2 ) be any representation
in the class of (1 , 2 ). If (1 , 2 ) is reducible it has two constituents one finite dimensional and one
infinite dimensional. A representation in the class of the first will be called (1 , 2 ). A representation
in the class of the second will be called (1 , 2 ). Any irreducible representation which is not absolutely
cuspidal is either a (1 , 2 ) or a (1 , 2 ). The representations (1 , 2 ) which are defined only for
certain values of 1 and 2 are called special representations.
Before proceeding to the proof of Theorem 2.18 for representations which are not absolutely
cuspidal we introduce some notation. If is an unramified quasi-character of F the associated
L-function is

L(s, ) =

1
.
1 (C) |C|s

It is independent of the choice of the generator C of p. If is ramified L(s, ) = 1. If belongs to

S(F ) the integral


Z(sF , ) =

()() ||s d

is absolutely convergent in some half-plane Re s > s0 and the quotient

Z(sF , )
L(s, )
can be analytically continued to a function holomorphic in the whole complex plane. Moreover for a
suitable choice of the quotient is 1. If is unramified and

d = 0
UF

one could take the characteristic function of OF . There is a factor (s, , ) which, for a given and
, is of the form abs so that if
 is the Fourier transform of


Z( 1 1s
Z(sF , )
F , )
=
(s,
,
)
.
L(1 s, 1 )
L(s, )
If is unramified and OF is the largest ideal on which is trivial (s, , ) = 1.
Proposition 3.5 Suppose 1 and 2 are two quasi-characters of F such that neither 1
1 2 nor

1 1
2 is F and is (1 , 2 ). Then

W (, ) = W (1 , 2 ; )
and if
L(s, ) = L(s, 1 ) L(s, 2 )
1
L(s,
) = L(s, 1
1 ) L(s, 2 )

(s, , ) = @(s, 1 , ) (s, 2 , )

all assertions of Theorem 2.18 are valid. In particular if |1 (C)| = |C|s1 and |2 (C)| = |C|s2
the integrals dening (g, s, W ) are absolutely convergent if Re s > max{s1 , s2 }. If 1 and 2 are

Chapter 1

53

unramifed and OF is the largest ideal of F on which is trivial there is a unique function W0 in
W (, ) which is invariant under GL(2, OF ) and assumes the value 1 at the identity. If

d = 1
UF

then (e, s, W0 ) = 1.
That W (, ) = W (1 , 2 ; ) is of course a consequence of the previous proposition. As we
observed the various assertions need be established only for g = e. Take in S(F 2 ) and let W = W
be the corresponding element of W (, ). Then

(a) = W

a 0
0 1



belongs to the space of the Kirillov model of . As we saw in the closing pages of the first paragraph


(e, s, W ) =

W
F

a 0
0 1



 F
|a|s1/2 d a = (

s1/2

is equal to

Z(1 sF , 2 sF , )
if the last and therefore all of the integrals are defined.
Also

 s, W ) = Z(1 s , 1 s , ).
(e,
F
F
2
1

Any function in S(F 2 ) is a linear combination of functions of the form

(x, y) = 1 (x)2 (y).

Since the assertions to be proved are all linear we need only consider the functions which are given
as products. Then

Z(1 sF , 2 sF , ) = Z(1 sF , 1 )Z(2 sF , 2 )

so that the integral does converge in the indicated region. Moreover
1 s
1 s
1 s
s
Z(1
2 F , 1 F , ) = Z(2 F , 1 )Z(1 F , 2 )

also converges for Re s sufficiently large. (e, s, W ) is equal to

Z(1 sF , 1 ) Z(2 sF , 2 )
L(s, 1 )
L(s, 2 )
and is holomorphic in the whole complex plane. We can choose 1 and 2 so that both factors are 1.
It follows from the Iwasawa decomposition GF = PF GL(2, OF ) that if both 1 and 2 are
unramified there is a non-zero function on B(1 , 2 ) which is invariant under GL(2, OF ) and that it
is unique up to a scalar factor. If the largest ideal on which is trivial is OF , if 0 is the characteristic

function of OF2 , and if

0 is the partial Fourier transform introduced in Proposition 1.6 then 0 = 0 .
Consequently

r1 ,2 (g)0 = 0

Chapter 1

54

for all g in GL(2, OF ). If W0 = W0 then, since 0 is the product of the characteristic function of OF
with itself, (e, s, W0 ) = 1 if


d = 1.

UF

The only thing left to prove is the local functional equation. Observe that




 e, s, (w)W ,
(w,
s, W ) =
that if W = W then (w)W = Wr(w) , and that r(w)(x, y) =  (y, x) if  is the Fourier transform
of . Thus if (x, y) is a product 1 (x)2 (y)
s
s
1 ) Z(1
2 )
Z(1
1 F ,
2 F ,

.
(w,
s, W ) =
1
1
L(s, 1 )
L(s, 2 )

The functional equation follows immediately.

1
If 1 1
2 is F or F and = (1 , 2 ) we still set

L(s, ) = L(s, 1 ) L(s, 2 )

and

(s, , ) = (s, 1 , ) (s, 2 , ).

1
Since
is equivalent to (1
1 , 2 )
1
L(s,
) = L(s, 1
1 )L(s, 2 ).

Theorem 2.18 is not applicable in this case. It has however yet to be proved for the special representa1/2
1/2
tions. Any special representation is of the form (1 , 2 ) with 1 = F and 2 = F . The
1
1
 is (2 , 1 ). This choice of 1 and 2 is implicit in the following
contragredient representation of
proposition.
Proposition 3.6 W (, ) is the space of functions W = W in W (1 , 2 ; ) for which


(x, 0) dx = 0.
F

Theorem 2.18 will be valid if we set L(s, ) = L(s,

) = 1 and (s, , ) = (s, 1 , ) (s, 2 , )
when is ramied and we set L(s, ) = L(s, 1 ), L(s,
) = L(s, 1
2 ), and
(s, , ) = (s, 1 , ) (s, 2 , )

L(1 s, 1
1 )
L(s, 2 )

when is unramied.
W (, ) is of course the subspace of W (1 , 2 ; ) corresponding to the space Bs (1 , 2 ) under
the transformation A of Proposition 3.2. If W = W then A takes W to the function f = f defined
by



f (g) = z 1 1
1 (detg) |detg|1/2 .
2 F , (g)

Chapter 1

55

f belongs to Bs (1 , 2 ) if and only if



1 (g)f (g) dg = 0.
GL(2,OF )

As we observed this integral is equal to a constant times










1 x
1 x
1 x
w
f w
dx =
f w
dx
0 1
0 1
0 1
F

which equals




 
1 x
2

2
z F , (w)

dx =
(t, tx) |t| d t dx.
0 1

The double integral does converge and equals, apart from a constant factor,



(t, tx) |t| dt dx =

which in turn equals



(t, x) dt dx


(t, 0) dt.

We now verify not only the remainder of the theorem but also the following corollary.
Corollary 3.7 (i) If = ( 1 , 2 ) then

(s, , )

L(1 s,
)
L(1 s,
)
= (s, , )
L(s, )
L(s, )

(ii) The quotient

L(s, )
L(s, )
is holomorphic
(iii) For all such that


(x, 0) dx = 0

the quotient

Z(1 sF , 2 sF , )
L(s, )

is holomorphic and there exists such a for which the quotient is one.
The first and second assertions of the corollary are little more than matters of definition. Although
W (1 , 2 ) is not irreducible we may still, for all W in this space, define the integrals



 

a 0
(g, s, W ) = W
g |a|s1/2 d a
0 1



 

a 0
 s, W ) = W
(g,
g |a|s1/2 1 (a) d a.
0 1

Chapter 1

56

They may be treated in the same way as the integrals appearing in the proof of Proposition 3.5. In
particular they converge to the right of some vertical line and if W = W

(e, s, W ) = Z(1 sF , 2 sF , )
 s, W ) = Z(1 sF , 1 sF , ).
(e,
2

Moreover

(g, s, W )
L(s, )

is a holomorphic function of s and

 1 s, W )
(g,
(g, s, W )
= (s, , )
.
L(1 s,
)
L(s, )
Therefore

(g, s, W ) =
and

(g, s, W )
L(s, )


 s, W ) = (g, s, W )
(g,
L(s,
)

are meromorphic functions of s and satisfy the local functional equation


(wg,
1 s, W ) = (s, , ) (g, s, W ).
To compete the proof of the theorem we have to show that (s, , ) is an exponential function of
s and we have to verify the third part of the corollary. The first point is taken care of by the observation
1
that 1
1 (C) |C| = 2 (C) so that

L(1 s, 1
1 2 (C) |C|s
1 )
= 1 (C) |C|s1 .
=
1s
L(s, 2 )
1 1
(C)
|C|
1
If is ramified so that L(s, ) = L(s, ) the quotient part (iii) of the corollary is holomorphic.
Moreover a in S(F 2 ) for which

Z(1 sF , 2 sF , ) = L(s, ) = 1
can be so chosen that

(x, y) = ()(x, y)
for and in UF . Then


(x, 0) dx = 0.
F

Now take unramified so that (a) = |a|r for some complex number r . We have to show that if


(x, 0) dx = 0
F

Chapter 1

57

then

Z(1 sF , 2 sF , )
L(s, 1 )
is a holomorphic function of s. Replacing s by s r + 1/2 we see that it is enough to show that

s+1
(1 |C| ) (x, y) |x|s+1 |y|s d x d y
is a holomorphic function of s. Without any hypothesis on the integral converges for Re s > 0 and
the product has an analytic continuation whose only poles are at the roots of |C|s = 1. To see that these
poles do not occur we have only to check that there is no pole at s = 0. For a given in S(F 2 ) there is
an ideal a such that

(x, y) = (x, 0)
for y in a. If a is the complement of a



is equal to the sum of

(x, y) |x|s+1 |y|s d x d y

 
a

(x, y) |x|s+1 |y|s d x d y

which has no pole at s = 0 and a constant times



 

s
(x, 0) |x| dx
|y| d y
s

If a = pn the second integral is equal to

|C|ns (1 |C|s )1


If

(x, 0) dx = 0
F

the first term, which defines a holomorphic function of s, vanishes at s = 0 and the product has no
pole there.
If 0 is the characteristic function of OF set



(x, y) = 0 (x) |C|1 0 (C 1 x) 0 (y).


Then

(x, 0) dx = 0
F

and

Z(1 sF , 2 sF , )
is equal to



0 (x) |C|

0 (C

 

s
x) 1 (x) |x| d x
0 (y) 2 (y) |y| d y


The second integral equals L(s, 2 ) and the first equals

(1 1 (C) |C|s1 ) L(s, 1 )

so their product is L(s, 1 ) = L(s, ).
Theorem 2.18 is now completely proved. The properties of the local L-functions L(s, ) and the
factors @(s, , ) described in the next proposition will not be used until the paragraph on extraordinary
representations.

Chapter 1

58

Proposition 3.8 (i) If is an irreducible representation there is an integer m such that if the order

of is greater than m both L(s, ) and L(s,

) are 1.
(ii) Suppose 1 and 2 are two irreducible representations of GF and that there is a quasicharacter such that








a 0
a 0
1
= (a)I
2
= (a)I
0 a
0 a
Then there is an integer m such that if the order of is greater than m
(s, 1 , ) = (s, 2 , )
(iii) Let be an irreducible representation and let be the quasi-character dened by

a
0

0
a


= (a)I

Write in any manner as = 1 2 . Then if the order of is suciently large in comparison

to the orders of 1 and 2
(s, , ) = (s, 1 , ) (s, 2 , ).
It is enough to treat infinite-dimensional representations because if = (1 , 2 ) and =
) = L(s,
), and (s, , ) =
(1 , 2 ) are both defined L(s, ) = L(s, ), L(s,
(s, , ) if the order of is sufficiently large.
If is not absolutely cuspidal the first part of the proposition is a matter of definition. If is
) = 1 for all .
absolutely cuspidal we have shown that L(s, ) = L(s,
According to the relation (2.18.1)

(s, , ) = C(01 11 , z01 z11 q 1/2 z 1 )

if the order of is so large that L(s, ) = L(s, 1
) = 1. Thus to prove the second part we
have only to show that if {C1 (, t)} and {C2 (, t)} are the series associated to 1 and 2 then

C1 (, t) = C2 (, t)
if the order of is sufficiently large. This was proved in Lemma 2.16.6. The third part is also a
consequence of that lemma but we can obtain it by applying the second part to 1 = and to
2 = (1 , 2 ).
We finish up this paragraph with some results which will be used in the Hecke theory to be
developed in the second chapter.

Chapter 1

59

Lemma 3.9 The restriction of the irreducible representation to GL(2, OF ) contains the trivial

representation if and only if there are two unramied characters 1 and 2 such that = (1 , 2 ).
This is clear if is one-dimensional so we may as well suppose that is infinite dimensional.
If = (1 , 2 ) we may let = (1 , 2 ). It is clear that there is a non-zero vector in B(1 , 2 )
invariant under GL(2, OF ) if and only if 1 and 2 are unramified and that if there is such a vector
it is determined up to a scalar factor. If = (1 , 2 ) and 1 1
2 = F we can suppose that is the
restriction of (1 , 2 ) to BS (1 , 2 ). The vectors in B(1 , 2 ) invariant under GL(2, OF ) clearly do
not lie in Bs (1 , 2 ) so that the restriction of to GL(2, OF ) does not contain the trivial representation.
All that we have left to do is to show that the restiction of an absolutely cuspidal representation to
GL(2, OF ) does not contain the trivial representation.
Suppose the infinite-dimensional irreducible representation is given in the Kirillov form with
respect to an additive character such that OF is the largest ideal on which is trivial. Suppose the
non-zero vector is invariant under GL(2, OF ). It is clear that if

a 0
0 a


= (a)I

then is unramified, that (, t) = 0 unless = 1 is the trivial character, and that (, t) has no pole
at t = 0. Suppose is absolutely cuspidal so that belongs to S(F ). Since (w) = and the
restriction of to UF is trivial

(1,
 t) = C(1, t) (1,
 z01 t1 )

if z0 = (C). Since C(1, t) is a constant times a negative power of t the series on the left involves no
negative powers of t and that on the right involves only negative powers. This is a contradiction.
Let H0 be the subalgebra of the Hecke algebra formed by the functions which are invariant under
left and right translations by elements of GL(2, OF ). Suppose the irreducible representation acts on
the space X and there is a non-zero vector x in X invariant under GL(2, OF ). If f is in H0 the vector
(f )x has the same property and is therefore a multiple (f )x of x. The map f (f ) is a non-trivial
homomorphism of H0 into the complex numbers.
Lemma 3.10 Suppose = (1 , 2 ) where 1 and 2 are unramied and is the associated homomorphism of H0 into C. There is a constant c such that


|f (g)| dg

|(f )| c

(3.10.1)

GF
s
for all f in H if and only if 1 2 is a character and |1 (C)1
2 (C)| = |C| with 1 s 1.

 and let x
 be such that x, x
Let
act on X
 in X
 =
 0. Replacing x
 by


(g)
x dg
GL(2,OF )

if necessary we may suppose that x

 is invariant under GL(2, OF ). We may also assume that x, x
 = 1.
 then
If (g) = (g)x, x


(f ) (g) =

(gh) f (h) dh
GF

Chapter 1

60

for all f in H0 . In particular


(f ) =

(h) f (h) dh.

GF

If |(h)| c for all h the inequality (3.10.1) is certainly valid. Conversely, since is invariant under
left and right translations by GL(2, OF ) we can, if the inequality holds, apply it to the characteristic
functions of double cosets of this group to see that |(h)| c for all h. Since

a 0
0 a

 
h = 1 (a) 2 (a) (h)

the function is bounded only if 1 2 is a character as we now assume it to be. The finite dimensional
representations take care of themselves so we now assume is infinite-dimensional.
are irreducible the function (g)x, x
 is bounded for a given pair of non-zero
Since and
vectors if and only if it is bounded for all pairs. Since GF = GL(2, OF ) AF GL(2, OF ) and 1 2 is a
character these functions are bounded if and only if the functions





a 0
0 1

x, x


are bounded on F . Take and

in the Kirillov form. If is in V and
 is in V then



a 0
0 1


,
(w)


is equal to

(w)

a 0
0 1



, 
 = 1 (a) 2 (a) <

a1
0

0
1



1 (w),
>

Thus (g) is bounded if and only if the functions



a 0
0 1


, 


are bounded for all in V and all

 in S(F ).
 in S(F ) but only a set which together with its translates by
It is not necessary to consider all

the diagonal matrices spans S(F ). If is a character of UF let be the function on F which is 0
outside of UF and equals on UF . It will be sufficient to consider the functions
 = and all we
need show is that





Cn 0
0 1

, 

(3.10.2)

is a bounded function of n for all and all . The expression (3.10.2) is equal to
n (). If belongs
n ()} has only finitely many non-zero terms and there is no problem. If
to S(F ) the sequence {
= (w)0 then


n ()tn = C(, t) (t)

Chapter 1

61

where (t) depends on 0 and is an arbitrary finite Laurent series. We conclude that (3.10.1) is valid if
and only if 1 2 is a character and the coefficients of the Laurent series C(, t) are bounded for every
choice of .
It follows from Proposition 3.5 and formula (2.18.1) that, in the present case, the series has only
one term if is ramified but that if is trivial

C , |C|

1/2

1
1
1 (C) 2 (C)t



1 1 (C)t1 1 2 (C)t1

.
=
1
1 1
1 (C) |C| t 1 2 (C) |C| t


The function on the right has zeros at t = 1 (C) and t = 2 (C) and poles at t = 0, t = |C|1 1 (C),
and t = |C|1 2 (C). A zero can cancel a pole only if 2 (C) = |C|1 1 (C) or 1 (C) = |C|1 2 (C).
1
Since 1 and 2 are unramified this would mean that 1
1 2 equals F or F which is impossible
when = (1 , 2 ) is infinite dimensional.
If C(, t) has bounded coefficients and 1 2 is a character the function on the right has no poles
for |t| < |C|1/2 and therfore |1 (C)| |C|1/2 and |2 (C)| |C|1/2 . Since
1
2
2
|1 (C)1
2 (C)| = |1 (C)| = |2 (C)|

where 1 2 is a character these two inequalities are equivalent to that of the lemma. Conversely if these
two inequalities are satisifed the rational function on the right has no pole except that at 0 inside the
circle |t| = |C|1/2 and at most simple poles on the circle itself. Applying, for example, partial fractions
to find its Laurent series expansion about 0 one finds that the coefficients of C(, t) are bounded.
Lemma 3.11 Suppose 1 and 2 are unramied, 1 2 is a character, and = (1 , 2 ) is innite
dimensional. Let |1 (C)| = |C|r where r is real so that |2 (C)| = |C|r . Assume OF is the
largest ideal on which is trivial and let W0 be that element of W (, ) which is invariant under
GL(2, OF ) and takes the value 1 at the identity. If s > |r| then






a
W0

0
F

0
1


 s1/2
1
 |a|
d a

s+r
(1 |C| )(1 |C|sr )

if the Haar measure is so normalized that the measure of UF is one.

If is the characteristic function of OF2 then

W0
and

a 0
0 1




= 1 (a) |a|

1/2
F

(at, t1 ) 1 (t) 1
2 d t








a 0  s1/2
W0
|a|
d
a

(at, t1 ) |a|s+r |t|2r d a d t.



0
1

Changing variables in the left-hand side we obtain


OF


OF

|a|s+r |b|sr d a d b =

1
(1

|C|s+r )(1

|C|sr )

Chapter 1

62

4. Examples of absolutely cuspidal representations In this paragraph we will use the results of the
first paragraph to construct some examples of absolutely cuspidal representations.
First of all let K be a quaternion algebra over F . K is of course unique up to isomorphism.
As in the first paragraph will denote a continuous finite-dimensional representation of K the

and is the reduced norm on K we denote

multiplicative group of K . If is a quasi-character

 of F

the one-dimensional representation g (g) of K by also. If is any representation is

the representation g (g) (g). If is irreducible all operators commuting with the action of K
are scalars. In particular there is a quasi-character of F such that
(a) = (a)I
for all a in F which is of course a subgroup of K . If is replaced by then is replaced by
 will denote the representation contragredient to .
2 .
Suppose is irreducible, acts on V , and the quasi-character is a character. Since K /F is
compact there is a positive definite hermitian form on V invariant under K . When this is so we call
unitary.
It is a consequence of the following lemma that any one-dimensional representation of K is the
representation associated to a quasi-character of F .
Lemma 4.1 Let K1 be the subgroup of K consisting of those x for which (x) = 1. Then K1 is the

commutator subgroup, in the sense of group theory, of K .

K1 certainly contains the commutator subgroup. Suppose x belongs to K1 . If x = x then

x = xx = 1 so that x = 1. Otherwise x determines a separable quadratic extension of F . Thus,
in all cases, if xx = 1 there is a subfield L of K which contains x and is quadratic and separable over
L. By Hilberts Theorem 90 there is a y in L such that x = yy . Moreover there is an element in K
such that z1 = z for all z in L. Thus x = yy1 1 is in the commutator subgroup.
In the first paragraph we associated to a representation r of a group G+ on the space S(K, ).
Since F is now non-archimedean the group G+ is now GF = GL(2, F ).
2

Theorem 4.2 (i)

The representation r is admissible.

(ii) Let d = degree . Then r is equivalent to the direct sum of d copies of an irreducible

representation ().
(iii) If is the representation associated to a quasi-character
1/2
1/2
() = (F , F ).
(iv) If d > 1 the representation () is absolutely cuspidal.

of

then

If n is a natural number we set

Gn = {g GL(2, OF ) | g = I(mod pn )}
We have first to show that if is in S(K, ) there is an n such that r (g) = if g is in Gn and that
for a given n the space of in S(K, ) for which r (g) = for all g in Gn is finite dimensional.
Any




in Gn may be written as

g=

g=

a b
c d

1
ca1

0
1



a b
0 d

Chapter 1

63

and both the matrices on the right are in Gn . Thus Gn is generated by the matrices of the forms

a 0
0 1

1 x
0 1

a 0
0 1

1 x
0 1

w1

with a 1 (mod pn ) and x 0 (mod pn ). It will therefore be enough to verify the following three
assertions.
(4.2.1) Given there is an n > 0 such that

a 0
0 1


=

if a 1 (mod pn )
(4.2.2) Given there is an n > 0 such that

1 x
0 1


=

if x 0 (mod pn ).
(4.2.3) Given n > 0 the space of in S(K, ) such that

and

r (w

for all x in pn is finite-dimensional.

If a = (h) then

1 x
0 1

) r



1 x
0 1


r (w) =



a 0
0 1

1/2

= |h|K (h) (xh).

Since has compact suport in K and is locally constant there is a neighborhood U of 1 in K such that
1/2

(h) (xh) |h|K = (x)

for all h in U and all x in K . The assertion (4.2.1) now follows from the observation that is an open
mapping of K onto F .
We recall that





1 x
0 1



(z) = x(z) (z)

m
Let p be the largest ideal on which is trivial and let pK be the prime ideal of K . Since (pm
K ) = pF

1 x
0 1



n
for all x in pn if and only if the support of is contained in pK
. With this (4.2.2) is established.
satisfies the two conditions of (4.2.3) if and only if both and r(w) have support in pn
K
or, since r(w) =  , if and only if and  , its
have support in this set. There
 Fourier transform,
is certainly a natural number k such that (y) = 1 for all y in pkK . Assertion (4.2.3) is therefore a
consequence of the following simple lemma.

Chapter 1

64

k
Lemma 4.2.4 If the support of is contained in p n
K and (y) = 1 for all y in pK the Fourier

transform of is constant on cosets of pk+n

K .


Since


(x) =

pn
K



(y) (x, y) dy

the lemma is clear.

We prove the second part of the theorem for one-dimensional first. Let be the representation
associated to . S(K, ) is the space of in S(K) such that (xh) = (x) for all h in K1 . Thus to
every in S(K, ) we may associate the function on F defined by
1/2

(a) = |h|K (h) (h)

if a = (h). The map is clearly injective. If belongs to S(F ) the function defined by
1/2

(h) = |h|K



1 (h) (h)

if h = 0 and by

(0) = 0
belongs to S(K, ) and = . Let S0 (K, ) be the space of functions obtained in this way. It is the
space of functions in S(K, ) which vanish at 0 and therefore is of codimension one. If belongs to
S0 (K, ), is non-negative, does not vanish identically and  is its Fourier transform then

(0) =

(x) dx = 0.

Thus r (w) does not belong to S0 (K, ) and S0 (K, ) is not invariant. Since it is of codimension one
there is no proper invariant subspace containing it.
Let V be the image of S(K, ) under the map . We may regard r as acting in V . >From
the original definitions we see that

r (b) = (b)
if b is in BF . If V1 is a non-trivial invariant subspace of V the difference

r

1 x
0 1



is in V0 V1 for all in V1 and all x in F . If is not zero we can certainly find an x for which the
difference is not zero. Consequently V0 V1 is not 0 so that V1 contains V0 and hence all of V .
r is therefore irreducible and when considered as acting on V it is in the Kirillov form. Since
V0 is not V it is not absolutely cuspidal. It is thus a (1 , 2 ) or a (1 , 2 ). To see which we have to
find a linear form on V which is trivial on V0 . The obvious choice is

L() = (0)
if = . Then

a1
L r
0

0
a2

 
 
 a1 

= (a1 a2 )   L().
a2

Chapter 1

65

To see this we have only to recall that

r

and that

a 0
0 a

a 0
0 1


= (a)I = 2 (a)I


1/2

(0) = |h|K (h) (0)

1/2

where a = (h) so that |h|K = |a|2F and (h) = (a)I . Thus if

A(g) = L(r (g))

1/2

1/2

A is an injection of V into an irreducible invariant subspace of B(F , F ). The only such

1/2
1/2
1/2
1/2
subspace is Bs (F , F ) and r is therefore (F , F ).
Suppose now that is not one-dimensional. Let act on U . Since K1 is normal and K/K1 is
abelian there is no non-zero vector in U fixed by every element of K1 . If is in S(K, ) then
(xh) = 1 (h) (x)
for all h in K1 . In particular (0) is fixed by every element in K1 and is therefore 0. Thus all functions
in S(K, ) have compact supports in K and if we associate to every in S(K, ) the function
1/2

(a) = |h|K (h) (h)

where a = (h) we obtain a bijection from S(K, ) to S(F , U ). It is again clear that

1 = (b)
if b is in BF and 1 = r (b).
Lemma 4.2.5 Let be an irreducible representation of K in the complex vector space U . Assume

that U has dimension greater than one.

(i) For any in S(K, U ) the integrals

Z(sF
Z(sF

|a|K (a) (a) d a

s/2

, ) =


|a|K 1 (a) (a) d a

s/2

, ) =
K

are absolutely convergent in some half-plane Re s > s0 .

(ii) The functions Z(sF , ) and Z(sF 1 , ) can be analytically continued to functions
meromorphic in the whole complex plane.
(iii) Given u in U there is a in S(K, U ) such that
Z(sF , ) u.
(iv) There is a scalar function (s, , ) such that for all in S(K, U )
3/2s

Z(F

1 ,  ) = (s, , ) Z(F

s+1/2

, )

Chapter 1

66

if  is the Fourier transform of . Moreover, as a function of s, (s, , ) is a constant

times an exponential.


1/2
There is no need to verify the first part of the lemma. Observe that F (x) = |(x)|F = |x|K
so that
s/2
(sF )(x) = |x|K (x).
If belongs to S(K, U ) set


1 (x) =

(h) (xh).
K1

The integration is taken with respect to the normalized Haar measure on the compact group K1 . 1
clearly belongs to S(K, U ) and

Z(sF , ) = Z(sF , 1 )

(4.2.6)

and the Fourier transform 1 of 1 is given by

1 (x) =

(h1 )  (hx)
K1

The function 1 (x ) belongs to S(K, ) and

Z(sF 1 ,  ) = Z(sF 1 , 1 ).

(4.2.7)

Since 1 and 1 both have compact support in K the second assertion is clear.
If u is in U and we let u be the function which is O outside of UK , the group of units of OK ,
and on UK is given by u (x) = 1 (x)u then

Z(sF , u ) = cu


if

d a.

c=
UK

If belongs to S(K ) let A() and B() be the linear transformations of U defined by
s+1/2

A()U = Z(F

, u )

s+3/2

B()u = Z(F

1 ,  u)

where  is the Fourier transform of . If (h) (h) = (h1 x) and (h) (x) = (xh) then
s/2+1/4

A((h)) = |h|K
and

s/21/4

A((h)) = |h|K

(h) A()

A()1 (h).

the Fourier transform of (h) is |h|K (h) and the Fourier transform of

(h) is |h|1
K (h) , the map B() has the same two properties. Since the kernel of is
Since

Chapter 1

67

open it is easily seen that A() and B() are obtained by integrating against locally constant functions and . They will of course take values in the space of linear transformations of U . We will
have
s/2+1/4

(ha) = |h|K

and

(h) (a)

s/21/4

(ah1 ) = |h|K

(a) 1 (h)

will satisfy similar identities. Thus

s/2+1/4

(h) (1)

s/2+1/4
|h|K

(h) (1)

(h) = |h|K
(h) =

(1) is of course the identity. However (1) must commute with (h) for all h in K and therefore it
is a scalar multiple of the identity. Take this scalar to be (s, , ).
The identity of part (iv) is therefore valid for in S(K , U ) and in particular for in S(K, ).
The general case follows from (4.2.6) and (4.2.7). Since
1
3/2s
(s, , ) = Z(F
1 , u )
c
the function (s, , ) is a finite linear combination of powers |C|s if C is a generator of pF . Exchanging
the roles of u and u we see that 1 (s, , ) has the same property. (s, , ) is therefore a multiple
of some power of |C|s .
We have yet to complete the proof of the theorem. Suppose = belongs to S(F , U ) and
 = r (w) . We saw in the first paragraph that if is a quasi-character of F then

()

= Z(F , )

(4.2.8)

 (1 1 ) = Z(F 1 1 ,  ).

(4.2.9)

and, if (a) = (a)I for a in F ,

Suppose U0 is a subspace of U and takes its values in U0 . Then, by the previous lemma, ()

and
 (1 1 ) also lie in U0 for all choices of . Since  lies in S(F , U ) we may apply Fourier
inversion to the multiplicative group to see that  takes values in U0 .
We may regard r as acting on S(F , U ). Then S(F , U0 ) is invariant under
(w). Since
 r
r (b) = (b) for b in BF it is also invariant under the action of BF . Finally r a0 a0 = (a)
so that S(F , U0 ) is invariant under the action of GF itself. If we take U0 to have dimension one
then S(F , U0 ) may be identified with S(F ) and the representation r restricted to S(F , U0 ) is
irreducible. From (4.2.8) and (4.2.9) we obtain
s1/2

(
 F

s+1/2 1 1

 (F
)

so that

s+1/2 1

 (F

s+1/2

) = Z(F
=

, )

s+3/2 1
Z(F

1 ,  )

1 = (s, , ) (
 F

s1/2

).

Chapter 1

68

Thus if 0 is the restriction of r to S(F , U0 )

(s, 0 , ) = (s, , )
so that 0 = () is, apart from equivalence, independent of U0 . The theorem follows.
Let be any irreducible finite-dimensional representation of K and let act on U . The
 acts on the dual space U
 of U . If u belongs to U and u

contragredient representation
 belongs to U

 u = 1 (h)u, u
.
u, (h)
If belongs to S(K) set

 ; u, u
Z(sF ,
) =
and set

Z(sF

 ; u, u
,
) =


K


K

|(h)|s (h) (h)u, u

 d h

 u d h.
|(h)|s (h) u, (h)

Theorem 4.3 Let be an irreducible representation of K in the space U .

(i) For any quasi-character of F

( ) = ().
(ii) There is a real number s0 such that for all u, u
 and and all s with Re s > s0 the integral

) is absolutely convergent.
dening Z(sF , ; u, u
(iii) There is a unique Euler factor L(s, ) such that the quotient
s+1/2

Z(F

, , u, u
)
L(s, )

is holomorphic for all u, u

, and for some choice of these variables is a non-zero constant.
(iv) There is a functional equation
s+1/2
  , u, u
,
)
, , u, u
)
Z(F
= (s, , )

L(s, )
L(1 s, )

3/2s

Z(F

where (s, , ) is, as a function of s, an exponential.

(v) If (a) = (a)I for a in F and if = () then

a
0

0
a


= (a)I.

 and (s, , ) = (s, , ).

Moreover L(s, ) = L(s, ), L(s,
) = L(s, )
The first assertion is a consequence of the definitions.We have
 just proved all the others when
has a degree greater than one. Suppose then that (h) = (h) where is a quasi-character of F .
1/2

1/2

Then () = (F , F

) and if the last part of the theorem is to hold L(s, ), which is of course

Chapter 1

69


uniquely determined by the conditions of part (iii), must equal L(s, ) = L(s, F ). Also L(s, )
1/2

must equal L(s,

) = L(s, 1 F ).
In the case under consideration U = C and we need only consider
1/2

Z(sF , ; 1, 1) = Z(sF , ).
As before the second part is trivial and

Z(sF , ) = Z(sF , 1 )


if

1 (x) =

(xh).
K1

The Fourier transform of 1 is

1 (x) =

 (hx) =

K1

and

 (xh)
K1

  ) = Z(s ,
  ).
Z(sF ,
F
1

It is therefore enough to consider the functions in S(K, ).

If = is defined as before then lies in the space on which the Kirillov model of acts and
s1/2

(
 F

s+1/2

) = A(F

, ).

The third assertion follows from the properties of L(s, ). The fourth follows from the relation

 (F

1/2s

1 ) = Z(F

3/2s

1 ,  ),

which was proved in the first paragraph, and the relation

 (F
1 )
)
(
 F
= (s, , )
,
L(1 s,
)
L(s, )
1/2s

s1/2


which was proved in the second, if we observe that (h)
= 1 (h).  is of course (w).

Corollary 4.4 If = () then
= ( ).
This is clear if if of degree one so suppose it is of degree greater than one. Combining the
 and of and  we
identity of part (iv) with that obtained upon interchanging the roles of and
find that

 ) = (1).
(s, , ) (1 s, ,

The same considerations show that

(s, , ) (1 s,
, ) = (1).
Consequently

 ).
(s,
, ) = (s, ,

Replacing by we see that



 ) = s, 1 (),

(s, 1
, ) = (s, 1 ,

 are both absolutely cuspidal they are equivalent.

for all quasi-characters . Since
and ()
There is a consequence of the theorem whose significance we do not completely understand.

Chapter 1

70

Proposition 4.5 Let be an irreducible representation of K on the space U and suppose that the

 be the dual space of U . Let be the Kirillov model of

dimension of U is greater than one. Let U
 the function
 belong to U
(), let lie in S(F ), and let  = (w). If u belongs to U and u

on K which vanishes at 0 and on K is dened by



 u
(x) = (x) |(x)|1 u, (x)
is in S(K) and its Fourier transform  vanishes at 0 and on K is given by





 (x) =  (x) |(x)|1 1 (x) (x)u, u
if (a) = (a)I for a in F .
It is clear that belongs not merely to S(K) but in fact to S(K). So does the function 1 which
we are claiming is equal to  . The Schur orthogonality relations for the group K1 show that  (0) = 0
so that  also belongs to S(K ).
We are going to show that for every irreducible representation of  of K


  (x)
(s,  , ) (x)  (x)u , u
1 (x), u ,
u  |(x)|3/2s d x
  |(x)|s+1/2 d x
=
 )
L(s,  )
L(1 s,

for all choices of u and u

 . Applying the theorem we see that

  (x)
{1 (x)  (x)} u ,
u  |(x)|3/2s d x = 0

for all choices of  , u , u

 , and all s. An obvious and easy generalization of the Peter-Weyl theorem,
which we do not even bother to state, shows that 1 =  .
If



u, (hx)
u  (hx)u , u
  dh

(x) =
K1

then


K

(x) (x)u , u
  |(x)|

s+1/2

d x=
K /K1

while

 

1 (x) u (x), u
  |(x)|

3/2s

d x=

K /K1



(x) |(x)|s1/2 (x) d x





 (x) 1 (x) |(x)|1/2s (x1 ) d x

If is 0 for all choice of u and u

 the required identity is certainly true. Suppose then is different
 .
from 0 for some choice u and u
Let U be the intersection of the kernels of  and . It is an open normal subgroup of K and
H = U K1 F is open, normal, and of finite index in K . Suppose that  (a) =  (a)I for a in F . If
h belongs to H

(xh) = 0 (h) (x)

where 0 is a quasi-character of H trivial on U and K1 and equal to  1 on F . Moreover 0 extends
to a quasi-character of K so that

(x)
K /H

(x) =
K /F

(x) 1 (x) = 0

may of course be identified with a quasi-character of F .

Chapter 1

71

Lemma 4.5.1 If


K /F

(x) 1 (x) = 0

then  is equivalent to .
 and agree on F and

K /F


u,
(x)
u  (x)u , u
  =
 0.

The lemma follows from the Schur orthogonality relations.

We have therefore only to prove the identity for  = . Set


u, (hx)
u (hx)u , u
  dh.

F (x) =
K1

 . There is a function f on F such that

u and u
 now belong to the spaces U and U


F (x) = f (x)
The identity we are trying to prove may be written as


 (a) 1 (a) 1 (a) f (a1 ) |a|1/2s d a
(a) (a) f (a) |a|s1/2 d a
=
(s,

,
)
. (4.5.2)
L(1 s, 1
)
L(s, )

Let H be the group constructed as before with U taken as the kernel of . The image F  of H under
is a subgroup of finite index in F and f , which is a function on F /F  , may be written as a sum

f (a) =

p


k i (a)

i=1

where {1 , , p } are the characters of F /F  which are not orthogonal to f . By the lemma is
equivalent to i for 1 i p and therefore is equivalent to i . Consequently

(s, , ) = (s, i , )
and


F

1
 (a) 1 (a) 1
(a) |a|1/2s d a
1 (a)
= (s, , )
L(1 s, 1
)


F

(a) (a) i (a) |a|s1/2 d a

.
L(s, )

The identity (4.5.2) follows.

Now let K be a separable quadratic extension of F . We are going to associate to each quasicharacter of K an irreducible representation () of GF . If G+ is the set of all g in GF whose
determinants belong to (K ) we have already, in the first paragraph, associated to a representation
r of G+ . To emphasize the possible dependence of r on we now denote it by (, ). G+ is of
index 2 in GF . Let () be the representation of GF induced from (, ).

Chapter 1
Theorem 4.6 (i)

72

The representation (, ) is irreducible.

(ii) The representation () is admissible and irreducible and its class does not depend on the

choice of .
(iii) If there is no quasi-character of F such that = 0 the representation () is absolutely
cuspidal.
(iv) If = 0 and is the character of F associated to K by local class eld theory then ()
is (, ).
It is clear what the notion of admissibility for a representation of G+ should be. The proof that
(, ) is admissible proceeds like the proof of the first part of Theorem 4.2 and there is little point in
presenting it.
To every in S(K, ) we associate the function on F+ = (K ) defined by
1/2

(a) = (h) |h|K (h)

if a = (h). Clearly = 0 if and only if = 0. Let V+ be the space of functions on F+ obtained in
this manner. V+ clearly contains the space S(F+ ) of locally constant compactly supported functions on
F+ . In fact if belongs to S(F+ ) and
1/2

(h) = 1 (h) |h|K



(h)

then = . If the restriction of to the group K1 of elements of norm 1 in K is not trivial so that
every element of S(K, ) vanishes at 0 then V+ = S(F+ ). Otherwise S(F+ ) is of codimension one in
V+ .
Let B+ be the group of matrices of the form

a x
0 1

with a in F+ and x in F . In the first paragraph we introduced a representation = of B+ on the

space of functions on F+ . It was defined by

and

a 0
0 1

1 x
0 1


(b) = (ba)


(b) = (bx) (b).

We may regard (, ) as acting on V+ and if we do the restriction of (, ) to B+ is .

Lemma 4.6.1 The representation of BF induced from the representation of B+ on S(F+ ) is the

representation of BF . In particular the representation of B+ is irreducible.

The induced representation is of course obtained by letting BF act by right translations on the
 on BF with values in S(F+ ) which satisfy
space of all functions

(b
 1 b) = (b1 ) (b)


Chapter 1

73

for all b1 in B+ . Let L be the linear functional in S(F+ ) which associates to a function its value at 1.
 the function
Associate to








a 0
a 0
(a) = L

=L
(e)

0 1
0 1
 a x 

The value of


0 1

at in F+ is














1 x
a 0
a x


L

= L
0 1
0 1
0
1





a 0
= (ax)L

= (x)(a).
0 1
Since F /F+ is finite it follows immediately that is in S(F ) and that
 is 0 if is. It also shows



 = (b)
 then = (b) for all b in BF . Since
that can be any function in S(F ) and that if
a representation obtained by induction cannot be irreducible unless the original representation is, the
second assertion follows from Lemma 2.9.1.
If the restriction of to K1 is not trivial the first assertion of the theorem follows immediately. If
it is then, by an argument used a number of times previously, any non-zero invariant subspace of V+
contains S(F+ ) so that to prove the assertion we have only to show that S(F+ ) is not invariant.
As before we observe that if in S(K, ) = S(K) is taken to vanish at 0 but to be non-negative
and not identically 0 then


(x) dx = 0

r (w) (0) =
K

so that is in S(F+ ) but r (w) is not.

The representation () is the representation obtained by letting G+ act to the right on the space
 on G+ with values in V+ which satisfy
of functions

(hg)

= (, )(h) (g)

for h in G+ . Replacing the functions
 by the functions

 (g) =


a 0
0 1

 
g

we obtain an equivalent representation, that induced from the representation

g (, )

a 0
0 1


1
a
g
0

0
1



of G+ . It follows from Lemma 1.4 that this representation is equivalent to (,  ) if  (x) = (ax).
Thus () is, apart from equivalence, independent of .
Since





GF =

a 0
0 1


 g G+ , a F

 is determined by its restrictions to BF . This restriction, which we again call

, is any one of the
functions considered in Lemma 4.6.1. Thus, by the construction used in the proof of that lemma, we

Chapter 1

74

can associate to any

 a function on F . Let V be the space of functions so obtained. We can regard
= () as acting on V . It is clear that, for all in V ,

(b) = (b)
if b is in BF . Every function on F+ can, by setting it equal to 0 outside of F+ , be regarded as a function
F . Since





a 0
0 1

() = (a)

V is the space generated by the translates of the functions in V+ . Thus if V+ = S(F+ ) then V = S(F )
and if S(F+ ) is of codimension one in V+ then S(F ) is of codimension two in V .
It follows immediately that () is irreducible and absolutely cuspidal if the restriction of to
K1 is not trivial.
 which is 0 outside of G+ and on G+ is givne
The function in V+ corresponds to the function
by
(g)

= (, )(g).
It is clear that

()(g) = (, )(g)
if g is in G+ . Any non-trivial invariant subspace of V will have to contain S(F ) and therefore S(F+ ).
Since (, ) is irreducible it will have to contain V+ and therefore will be V itself. Thus () is
irreducible for all .
If the restriction of to K1 is trivial there is a quasi-character of F such that = . To
establish the last assertion of the lemma all we have to do is construct a non-zero linear form L on V
which annihilates S(F ) and satisfies

a1
L
0

0
a2

 1/2
 
 a1 
= (a1 a2 ) (a2 )   L()
a2

if = (). We saw in Proposition 1.5 that

a 0
0 a


= 2 (a) (a)

so will only have to verify that

 
a 0
L
= (a)|a|1/2 L()
0 a
If = is in V+ we set

L() = (0)
so that if a is in F+






 
a 0
a 0
(0) = (a) |a|1/2 L().
L
= r
0 1
0 1

Chapter 1

75

If is in F but not in F+ any function in V can be written uniquely as

= 1 +

0
0 1


2

with 1 and 2 in V+ . We set

L() = L(1 ) + () L(2 ).

If = () then = ( ) if (a) = (a ),
= ( 1 ) and = ( ) if


is a quasi-character of F and = .
(ii) If a is in F then




a 0

= (a) (a)I.
0 a


(iii) L(s, ) = L(s, ) and L(s,
) = L(s, 1 ). Moreover if K (x) = F (x) for x in K and
if (K/F, F ) is the factor introduced in the rst paragraph then

Theorem 4.7 (i)

(s, , F ) = (s, , K ) (K/F, F )

It is clear that (, ) of G+ . However by its very construction (, ) = ( , ).
The relation





a 0
0 a

= (a) (a)I

isa consequence
of part (iii) of Proposition 1.5 and has been used before. Since  = is trivial and


(a) = (a) (a)

= 1 1 = ( )
To complete the proof of the first part of the theorem we have to show that () = ( ). It is enough
to verify that (, ) = ( , ). If belongs to S(K) let (x) = (x ). is a bijection of
S(K, ) with S(K, ) which changes (, ) into ( , ). Observe that here as elsewhere we have
written an equality when we really mean an equivalence.
We saw in the first paragraph that if = is in V+ then
s1/2

(
 F

) = Z(sK , )

and that if  = (w) and  is the Fourier transform of then, if 0 (a) = (a) (a) for a in F ,

 (01 F

s1/2

1s 1
) = Z(K
,  )

if = (K/F, F ). Thus for all in V+ the quotient

s1/2

(
 F
)
L(s, )
has an analytic continuation as a holomorphic function of s and for some it is a non-zero constant.
Also

 (w01 F
)
)
(
 F
= (K/F, F ) (s, , K )
.
L(1 s, 1 )
L(s, )
1/2s

s1/2

Chapter 1

76

To prove the theorem we have merely to check that these assertions remain valid when is
allowed to vary in V . In fact we need only consider functions of the form

=

@ 0
0 1


0

where 0 is in V+ and is not in F+ . Since

s1/2

(
 F

s1/2

) = ||1/2s
0 (F

the quotient
s1/2

(
 F
)
L(s, )
is certainly holomorphic in the whole plane. Since

 (01 F

1/2s

) = 0 () 01 () ||1/2s
0 (01 F

1/2s

) = || 2 s
0 (01 F
1

the functional equation is also satisfied.

Observe that if = then () = (, ) so that

L(s, ) = L(s, ) L(s, )

and

(s, , K ) (K/F, F ) = (s, , F ) (s, , F )

These are special cases of the identities of [19].

1/2s

Chapter 1

77

5. Representations of GL(2, R). We must also prove a local functional equation for the real and
complex fields. In this paragraph we consider the field R of real numbers. The standard maximal
compact subgroup of GL(2, R) is the orthogonal group O(2, R). Neither GL(2, R) nor O(2, R) is
connected.
Let H1 be the space of infinitely differentiable compactly supported functions on GL(2, R) which
are O(2, R) finite on both sides. Once a Haar measure on GR = GL(2, R) has been chosen we may
regard the elements of H1 as measures and it is then an algebra under convolution.


f1 (gh1 ) f2 (h) dh.

f1 f2 (g) =
GR

On O(2, R) we choose the normalized Haar measure. Then every function on O(2, R) which is a finite
sum of matrix elements of irreducible representations of O(2, R) may be identified with a measure on
O(2, R) and therefore on GL(2, R). Under convolution these measures form an algebra H2 . HR will
be the sum of H1 and H2 . It is also an algebra under convolution of measures. In particular if belongs
to H2 and f belongs to H1


(u) f (u1 g) du

f (g) =
O(2,R)

and

f (gu1 ) (u) du.

f (g) =
O(2,R)

If i , 1 i p, is a family of inequivalent irreducible representations of O(2, R) and

i (u) = dim i trace i (u1 )

then

p


i=1

is an idempotent of HR . Such an idempotent is called elementary.

It is a consequence of the definitions that for any f in H1 there is an elementary idempotent
such that

f = f = f.

Moreover for any elementary idempotent

H1 = Cc (GR )
is a closed subspace of Cc (GR ), in the Schwartz topology. We give it the induced topology.
A representation of the algebra HR on the complex vector space V is said to be admissible if
the following conditions are satisfied.
(5.1) Every vector v in V is of the form

v=

r

i=1

with fi in H1 and vi in V .

(fi )vi

Chapter 1

78

(5.2) For every elementary idempotent the range of () is finite dimensional.

(5.3) For every elementary idempotent and every vector v in ()V the map f (f )v of
H1 into the 
finite dimensional space ()V is continuous.
r
If v =
i=1 (fi )vi we can choose an elementary idempotent so that fi = fi = fi for
1 i r . Then ()v = v. Let {} be a sequence in Cc (GR ) which converges, in the space of
distributions, towards the Dirac distribution at the origin. Set n = n . For each i the sequence
{n fi } converges to fi in the space H1 . Thus by (5.3) the sequence {(n )v} converges to v in
the finite dimensional space ()v . Thus v is in the closure of the subspace (H1 )v and therefore
belongs to it.
As in the second paragraph the conditions (5.1) and (5.2) enable us to define the representation

contragredient to . Up to equivalence it is characterized by demanding that it satisfy (5.1) and (5.2)

and that there be a non-degenerate bilinear form on V V satisfying

(f )v, 
v = v, (f)
v
for all f in HR . V is the space on which
acts and f is the image of the measure f under the map
1
g g . Notice that we allow ourselves to use the symbol f for all elements of HR . The condition
(5.3) means that for every v in V and every v in V the linear form

f (f )v, v

is continuous on each of the spaces H1 . Therefore
is also admissible.
v = v. Then for any f in H1
()
Choose so that ()v = v and

(f )v, 
v = (f )v, v.
There is therefore a unique distribution on GR such that

(f ) = (f )v, v

for f in H1 . Choose in H1 so that ()v = v . Then

(f ) = (f ) = (f ) = (f )v, 
v = (f )v, v
so that (f ) = (f ). Consequently the distribution is actually a function and it is not unreasonable
v even though is not a representation of GR . For a fixed g , (g)v, v
to write it as g (g)v, 
depends linearly on v and v. If the roles of and
are reversed we obtain a function v,
(g)
v. It is
clear from the definition that

(g)v, 
v = v,
(g 1 )
v.

Let g be the Lie algebra of GR and let gC = g R C. Let A be the universal enveloping algebra
of gC . If we regard the elements of A as distributions on GR with support at the identity we can take
their convolution product with the elements of Cc (GR ). More precisely if X belongs to g
X f (g) =


d 
f exp(tX) t=0
dt

f X(g) =


d 
f g exp(tX) t=0
dt

and

Chapter 1

79

If f belongs to H1 so do f X and X f .
We want to associate to the representation of HR on V a representation of A on V such that

(X) (f ) = (X f )
and

(f ) (X) = (f X)

for all X in A and all f in H1 . If v =
(fi ) vi we will set
(X)v =

(X fi )vi

and the first condition will be satisfied. However we must first verify that if

(fi ) vi = 0

then

w=

(X fi )vi

is also 0. Choose f so that w = (f )w. Then

w=

(f ) (X fi )vi =

(f X fi )vi = (f X){

(fi )vi } = 0.

>From the same calculation we extract the relation

(f )

= (f X)

(X fi )vi




(fi )vi

for all f so that (f )(X) = (f X).

If g is in GR then (g) f = g f if g is the Dirac function at g . If g is in O(2, R) or in ZR , the
groups of scalar matrices, g f is in H1 if f is, so that the same considerations allow us to associate
to a representation of O(2, R) and a representation of ZR . It is easy to see that if h is in either of
these groups then

(AdhX) = (h) (X) (h1).

To dispel any doubts about possible ambiguities of notation there is a remark we should make. For
any f in H1


(f )v, v =

f (g)(g)v, 
v dg.
GR

Thus if h is in O(2, R) or ZR


(f h )v, v =

f (g) (gh)v, 
v dg
GR

Chapter 1

80

and

(f )(h)v, 
v =

f (g)(g)(h)v, 
v dg
GR

so that

(gh)v, v = (g) (h)v, 

v.

A similar argument shows that

(hg)v, v = (g)v,

(h1 v.
It is easily seen that the function (g)v, 
v takes the value of v, v at g = e. Thus if h belongs to
O(2, R) or ZR the two possible interpretations of (h)v, v give the same result.
It is not possible to construct a representation of GR on V and the representation of A is supposed
to be a substitute. Since GR is not connected, it is not adequate and we introduce instead the notion of
a representation 1 of the system {A, } where

=

1 0
0 1


.

It is a representation 1 of A and an operator 1 () which satisfy the relations

12 () = I
and

1 (AdX) = 1 ()1(X) 1 (1 ).

Combining the representation with A with the operator () we obtain a representation of the system
{A, }.
of A associated to
and it is not difficult to see that
There is also a representation

v
(X)v, v = v,
(X)
is the automorphism of A which sends X in g to X .
if X X
Let
(g) = (g)v, v.

is certainly infinitely differentiable. Integrating by parts we see that




f (g) X(g) dg =
GR

f X(g)
(g) dg
GR

The right side is


v =
(f ) (X)v,

v
f (g)(g) (X)v,
GR

so that

v.
X(g)
= (g)(X)v,

Chapter 1

81

Assume now that the operators (X) are scalar if X is in the centre Z of A. Then the standard
proof, which uses the theory of elliptic operators, shows that the functions are analytic on GR . Since

v
X(e)
= (X)v,

v
X()
= () (X)v,
and GR has only two components, one containing e and the other containing . The function vanishes
v and () (X)v,
v are 0 for all X in A. Any subspace V1 of V invariant under
identically if (X)v,
A and is certainly invariant under O(2, R) and therefore is determined by its annihilator in V . If v is
v is 0 so that
in V1 and v annihilates V1 the function (g)v, 

(f )v, v = 0

for all f in H1 . Thus (f )v is also in V1 . Since H2 clearly leaves V1 invariant this space is left invariant
by all of HR .
By the very construction any subspace of V invariant under HR is invariant under A and so
that we have almost proved the following lemma.
Lemma 5.4 The representation of HR is irreducible if and only if the associated representation
of {A, } is.

To prove it completely we have to show that if the representation of {A, } is irreducible the
operator (X) is a scalar for all X in Z. As (X) has to have a non-zero eigenfunction we have only
to check that (X) commutes with (Y ) for Y in A with (). It certainly commutes with (Y ). X is
invariant under the adjoint action not only of the connected component of GR but also of the connected
component of GL(2, C). Since GL(2, C) is conected and contains



() (X) 1 () = Ad(X) = (X).
Slight modifications, which we do not describe, of the proof of Lemma 5.4 lead to the following
lemma.
Lemma 5.5 Suppose and  are two irreducible admissible representations of HR . and  are

equivalent if and only if the associated representations of {A, } are.

We comment briefly on the relation between representations of GR and representations of HR .

Let V be a complete separable locally convex topological space and a continuous representation of
GR on V . Thus the map (g, v) (g)v of GR V to V is continuous and for f in Cc (GR ) the operator


(f ) =

f (x) (x) dx
GR

is defined. So is (f ) for f in H2 . Thus we have a representation of HR on V . Let V0 be the space

of O(2, R)-finite vectors in V . It is the union of the space ()V as ranges over the elementary
idempotents and is invariant under HR . Assume, as is often the case, that the representation 0 of HR
on V0 is admissible. Then 0 is irreducible if and only if is irreducible in the topological sense.
Suppose  is another continuous representation of GR in a space V  and there is a continuous
non-degenerate bilinear form on V V  such that

(g)v, v  = v,  (g 1 )v .

Chapter 1

82

Then the restriction of this form to V0 V0 is non-degenerate and

(f )v, v  = v,  (f)v 

for all f in HR , v in V0 , and v in V0 . Thus 0 is the contragredient of 0 . Since


f (g) (g)v, v 

0 (f )v, v  =
GR

we have

0 (g)v, v  = (g)v, v .

The special orthogonal group SO(2, R) is abelian and so is its Lie algebra. The one-dimensional
representation




cos sin
sin cos

ein

of SO(2, R) and the associated representation of its Lie algebra will be both denoted by n . A
representation of A or of {A, } will be called admissible if its restrictions to the Lie algebra of
SO(2, R) decomposes into a direct sum of the representations n each occurring with finite multiplicity.
If is an admissible representation of HR the corresponding representation of {A, } is also admissible.
We begin the classification of the irreducible admissible representations of HR and of {A, } with the
introduction of some particular representations.
Let 1 and 2 be two quasi-characters of F . Let B(1 , 2 ) be the space of functions f on GR
which satisfy the following two conditions.
(i)

a1
0

x
a2

 1/2
 
 a1 
g = 1 (a1 ) 2 (a2 )   f (g)
a2

for all g in GR , a1 , a2 in R , and x in R.

(ii) f is SO(2, R) finite on the right.
Because of the Iwasawa decomposition

GR = NR AR SO(2, R)
these functions are complete determined by their restrictions to SO(2, R) and in particular are infinitely
differentiable. Write

m

i (g) = |t|si

t
|t|

where si is a complex number and mi is 0 or 1. Set s = s1 s2 and m = |m1 m2 | so that

s t m
1 1
2 (t) = |t| ( |t| ) . If n has the same parity as m let n be the function in B(1 , 2 ) defined by

n

1 x
0 1



a1
0

0
a2



cos sin
sin cos



 1/2
 a1 
= 1 (a1 ) 2 (a2 )   ein .
a2

The collection {n } is a basis of B(1 , 2 ).

For any infinitely differentiable function f on GR and any compactly supported distribution
we defined ()f by



()f (g) =
(g)f

Chapter 1

83

and ()f by



()f (g) = (g 1 )f .

If, for example, is a measure

f (h1 g) d(h)

()f (g) =
GR

and

()f (g) =

f (gh) d(h).
GR

In all cases ()f and ()f are again infinitely differentiable. For all f in HR the space B(1 , 2 )
is invariant under (f ) so that we have a representation (1 , 2 ) of HR on B(1 , 2 ). It is clearly
admissible and the associated representation (1 , 2 ) of {A, } is also defined by right convolution.
We introduce the following elements of g which is identified with the Lie algebra of 22 matrices.

U=

0 1
1 0

,

J=

X+ =

0 1
0 0

1 0
0 1

,

V+ =


X =

0 0
1 0

1 i
i 1

,

V =

as well as

D = X+ X + X X+ +

Z=

1 0
0 1

1 i
i 1


,


,

Z2
,
2

which belongs to A.
Lemma 5.6 The following relations are valid

(i) (U )n = inn

(ii) ()n = (1)m1 n

(iii) (V+ )n = (s + 1 + n)n+2

(iv) (V )n = (s + 1 n)n2

(v) (D)n =

s2 1
n
2

(vi) (J )n = (s1 + s2 )n

The relations (i), (ii), and (vi) are easily proved. It is also clear that for all in B(1 , 2 )

(Z) (e) = (s + 1) (e)

and

(X+ ) (e) = 0.

The relations

Ad

and

Ad

cos sin
sin cos

cos sin
sin cos





V+ = e2i V+

V = e2i V

Chapter 1

84

show that (V+ )n is a multiple of n+2 and that (V )n is a multiple of n2 . Since

V+ = Z iU + 2iX+
and

V = Z + iU 2iX+

the value of (V+ )n at the identity e is s + 1 + n and that of (V )n = s + 1 n. Relations (iii) and
(iv) follow.
= (D)
It is not difficult to see that D belongs to Z the centre of A. Therefore (D) = (D)
. If we write D as
since D = D

2X X+ + Z +

Z2
2

and observe that (X+ ) = 0 and (Z) = (s + 1) if is in B(1 , 2 ) we see that


(D)n =

(s + 1)2
(s + 1) +
2


=

s2 1
n .
2

If s m is not an odd integer B(1 , 2 ) is irreducible under the action of g.

(ii) If s m is an odd integer and s 0 the only proper subspaces of B(1 , 2 ) invariant under
g are

B1 (1 , 2 ) =
Cn

Lemma 5.7 (i)

ns+1
n=s+1(mod 2)

B2 (1 , 2 ) =

Cn

ns1
n=s+1(mod 2)

and, when it is dierent from B(1 , 2 ),

Bs (1 , 2 ) = B1 (1 , 2 ) + B2 (1 , 2 ).
(iii) If s m is an odd integer and s < 0 the only proper subspaces of B(1 , 2 ) invariant under

g are

B1 (1 , 2 ) =

Cn

ns+1
n=s+1(mod 2)

B2 (1 , 2 ) =

Cn

ns1
n=(s+1)(mod 2)

and
Bf (1 , 2 ) = B1 (1 , 2 ) B2 (1 , 2 ).
Since a subspace of B(1 , 2 ) invariant under g is spanned by those of the vectors n that it
contains this lemma is an easy consequence of the relations of Lemma 5.6.
Before stating the corresponding results for {A, } we state some simple lemmas.

Chapter 1

85

Lemma 5.8 If is an irreducible admissible representation of {A, } there are two possibilities:

(i) The
of to A is irreducible and the representations X (X) and X
 restriction


Ad(X) are equivalent.

(ii) The space V on which acts decomposes into a direct sum V1 V2 where V1 and V2 are both
invariant and irreducible under A. The representations 1 and
 2 of A on V1 and V2 are not
equivalent but 2 is equivalent to the representation X Ad(X) .


If the restriction of to A is irreducible the representations X (X) and X Ad(X) are
certainly equivalent. If it is not irreducible let V1 be a proper subspace invariant under A. If V2 = ()V1
then V1 V2 and V1 + V2 are all invariant under {A, }. Thus V1 V2 = {0} and V = V1 V2 . If V1
had a proper subspace V1 invariant under A the same considerations would show that V = V1 V2
with V2 = ()V1 . Since this is impossible V1 and V2 are irreducible under A.
If v1 is in V1


2 (X) ()v1 = () 1 ad (X) v1


so that the representations X 2 (X) and X 1 Ad(X) are equivalent. If 1 and 2 were
equivalent there would be an invertible linear transformation A from V1 to V2 so that A1 (X) =
2 (X)A. If v1 is in V1




A1 () 1 (X)v1 = A1 2 ad (X) () v1 = 1 Ad(X) A1 () v1

Consequently {A1 ()}2 regarded as a linear transformation of V1 commutes with A and is therefore
a scalar. There is no harm in supposing that it is the identity. The linear transformation

v1 + v2 A1 v2 + Av1
then commutes with the action of {A, }. This is a contradiction.
Let be a quasi-character of R and let (t) = tc for t positive. For any admissible representation
of A and therefore of g we define a representation of g and therefore A by setting

c
trace X + (X)
2
if X is in g. If is a representation of {A, } we extend to {A, } by setting
(X) =

() = (1) ()
If is associated to a representation of HR then is associated to the representation of HR defined
by

(f ) = (f )

if f is the product of the functions and f .

Lemma 5.9 Let 0 be an irreducible
admissible
representation of A. Assume that 0 is equivalent



to the representation X 0 Ad(X) . Then there is an irreducible representation of {A, }

whose restriction to A is 0 . If is the non-trivial quadratic character of R the representations
and are not equivalent but any representatin of {A, } whose restriction to A is equivalent
to 0 is equivalent to one of them.
Let 0 act on V . There is an invertible linear transformation A of V such that A0 (X) =
0 Ad(X) A for all X in A. Then A2 commutes with all 0 (X) and is therefore a scalar. We may
suppose that A2 = I . If we set () = A and (X) = 0 (X) for X in A we obtain the required
representation. If we replace A by A we obtain the representation . and are not
equivalent because any operator giving the equivalence would have to commute with all of the (X)
and would therefore be a scalar. Any representation  of {A, } whose restriction to A is equivalent to
0 can be realized on V0 in such a way that  (X) = 0 (X) for all X . Then  () = A.

Chapter 1

86

Lemma 5.10 Let 1 be an irreducible admissible representation of A. Assume that 1 and 2 , with

2 (X) = 1 Ad(X) , are not equivalent. Then there is an irreducible representation of {A, }
whose restriction to A is the direct sum of 1 and 2 . Every irreducible admissible representation
of {A, } whose restriction to A contains 1 is equivalent to . In particular is equivalent
to .
Let 1 act on V1 . To construct we set V = V1 V2 and we set

(X) (v1 v2 ) = 1 (X)v1 2 (X)v2

and

() (v1 v2 ) = v2 v1 .

The last assertion of the lemma is little more than a restatement of the second half of Lemma 5.8.
Theorem 5.11 Let 1 and 2 be two quasi-characters of F .

(i) If 1 1
is not of the form t tp sgn t with p a non-zero integer the space B(1 , 2 ) is
2

irreducible under the action of {A, } or HR . (1 , 2 ) is any representation equivalent to

(1 , 2 ).
p
(ii) If 1 1
2 (t) = t sgn t, where p is a positive integer, the space B(1 , 2 ) contains exactly
one proper subspace Bs (1 , 2 ) invariant under {A, }. It is innite dimensional and any
representation of {A, } equivalent to the restriction of (1 , 2 ) to Bs (1 , 2 ) will be denoted
by (1 , 2 ). The quotient space
Bf (1 , 2 ) = B(1 , 2 )/Bs(1 , 2 )
is nite-dimensional and (1 , 2 ) will be any representation equivalent to the representation
of {A, } on this quotient space.
p
(iii) If 1 1
2 (t) = t sgn t, where p is a negative integer, the space B(1 , 2 ) contains exactly one
proper subspace Bf (1 , 2 ) invariant under {A, }. It is nite-dimensional and (1 , 2 )
will be any representation equivalent to the restriction of (1 , 2 ) to Bf (1 , 2 ). (1 , 2 )
will be any representation equivalent to the representation on the quotient space
Bs (1 , 2 ) = B(1 , 2 )/Bf (1 , 2 ).
(iv) A representation (1 , 2 ) is never equivalent to a representation (1 , 2 ).
(v) The representations (1 , 2 ) and (1 , 2 ) are equivalent if and only if either (1 , 2 ) =

(1 , 2 ) or (1 , 2 ) = (2 , 1 ).
(vi) The representations (1 , 2 ) and (1 , 2 ) are equivalent if and only if (1 , 2 ) is one of
the four pairs (1 , 2 ), (2 , 1 ), (1 , 2 ), or (2 , 1 ).
(vii) Every irreducible admissible representation of {A, } is either a (1 , 2 ) or a (1 , 2 ).

1
s t m
Let 1 1
2 (t) = |t| ( |t| ) . s m is an odd integer if and only if s is an integer p and 1 2 (t) =
t sgn t. Thus the first three parts of the lemma are consequences of Lemma 5.6 and 5.7. The fourth
follows from the observation that (1 , 2 ) and (1 , 2 ) cannot contain the same representations of
the Lie algebra of SO(2, R).
We suppose first that s m is not an odd integer and construct an invertible transformation T
from B(1 , 2 ) to B(2 , 1 ) which commutes with the action of {A, }. We have introduced a basis
p

Chapter 1

87

{n } of B(1 , 2 ). Let {n } be the analogous basis of B(2 , 1 ). T will have to take n to a multiple
an n of n . Appealing to Lemma 5.6 we see that it commutes with the action of {A, } if and only if
(s + 1 + n)an+2 = (s + 1 + n)an
(s + 1 n)an2 = (s + 1 n)an
and

an = (1)m an .
These relations will be satisfied if we set

an = an (s) =

(s+1+n)
2
(s+1+n)
2

Since n = m(mod 2) and s m 1 is not an even integer all these numbers are defined and different
from 0.
If s 0 and s m is an odd integer we set

an (s) = lim an (z)

zs

The numbers an (s) are still defined although some of them may be 0. The associated operator T
maps B(1 , 2 ) into B(2 , 1 ) and commutes with the action of {A, }. If s = 0 the operator T is
non-singular. If s < 0 its kernel is Bf (1 , 2 ) and it defines an invertible linear transformation from
Bs (1 , 2 ) to Bs (2 , 1 ). If s > 0 and s m is an odd integer the functions an (z) have at most simple
poles at s. Let

bn (s) = lim (z s) an (z)

zs

The operator T associated to the family {bn (s)} maps B(1 , 2 ) into B(2 , 1 ) and commutes with
the action of {A, }. It kernel is Bs (1 , 2 ) so that it defines an invertible linear transformation from
Bf (1 , 2 ) to Bf (2 , 1 ). These considerations together with Lemma 5.10 give us the equivalences of
parts (v) and (vi).
Now we assume that = (1 , 2 ) and  = (1 , 2 ) or = (1 , 2 ) and  = (1 , 2 )

t mi
t mi
are equivalent. Let i (T ) = |t|si ( |t|
) and let i (t) = |t|si ( |t|
) . Let s = s1 s2 , m = |m1 m2 |,






s = s1 s2 , m = |m1 m2 |. Since the two representations must contain the same representations of
the Lie algebra of SO(2, R) the numbers m and m are equal. Since (D) and  (D) must be the same
scalar Lemma 5.6 shows that s = s. (J ) and  (J ) must also be the same scalar so s1 + s2 = s1 + s2 .
Thus if (t) = sgn t the pair (1 , 2 ) must be one of the four pairs (1 , 2 ), (2 , 1 ), (1 , 2 ),
(2 , 1 ). Lemma 5.9 shows that (1 2 ) and (1 , 2 ) are not equivalent. Parts (v) and (vi) of
the theorem follow immediately.
Lemmas 5.8, 5.9, and 5.10 show that to prove the last part of the theorem we need only show
that any irreducible admissible representation of A is, for a suitable choice of 1 and 2 , a constituent
of (1 , 2 ). That is there should be two subspace B1 and B2 of B(1 , 2 ) invariant under A so
that B1 contains B2 and is equivalent to the representation of A on the quotient B1 /B2 . If is
a quasi-character of F then is a constituent of (1 , 2 ) if and only if is a constituent of
(1 , 2 ). Thus we may suppose that (J ) is 0 so that is actually a representation of A0 , the
universal enveloping algebra of the Lie algebra of ZR \ GR . Since this group is semi-simple the desired
result is a consequence of the general theorem of HarishChandra [6].

Chapter 1

88

It is an immediate consequence of the last part of the theorem that every irreducible admissible
representations of {A, } is the representation associated to an irreducible admissible representation of
HR . Thus we have classifed the irreducible admissible representations of {A, } and of HR . We can
write such a representation of HR as (1 , 2 ) or (1 , 2 ).
In the first paragraph we associated to every quasi-character of C a representation of r of
G+ the group of matrices with positive determinant. r acts on the space of functions in S(C) which
satisfy

(xh) = 1 (h) (x)

= 1. All elements of S(C, ) are infinitely differentiable vectors for r so that r

for all h such that hh
also determines a representation, again called r , of A. r depended on the choice of a character of R.
If that character is
(x) = e2uxi
then

r (X+ ) (z) = (2uz zi)(z).

Lemma 5.12 Let S 0 (C, ) be the space of functions in S(C, ) of the form

(z) = e2|u|z zP (z, z)

where P (z, z) is a polynomial in z in z. Then S0 (C, ) is invariant under A and the restriction of
r to S0 (C, ) is admissible and irreducible.
It is well known and easily verified that the function e2|u|z z is its own Fourier transform
provided of course that the transform is taken with respect to the character

C (z) = (z + z)
and the self-dual measure for that character. From the elementary properties of the Fourier transform
one deduces that the Fourier transform of a function

(z) = e2|u|z z P (z, z)

where P is a polynomial in z and z is of the same form. Thus r (w) leaves S0 (C, ) invariant. Recall
that




w=

0 1
1 0

S0 (C, ) is clearly invariant under r (X+ ). Since X = Adw(X+ ) it is also invariant under X . But
X+ X X X+ = Z so that it is also invariant under Z . We saw in the first paragraph that if 0 is
the restriction of to R then




a 0
= (sgn a) 0 (a)I
r
0 a
thus r (J ) = cI if 0 (a) = ac for a positive a. In conclusion S0 (C, ) is invariant under g and therefore
under A.
If
m n

(z) = (z z)r

z z

(z z)

m+n
2

Chapter 1

89

where r is a complex number and m and n are two integers, one 0 and the other non-negative, the
functions

p (z) = e2|u|z z z n+p zm+p ,

with p a non-negative integer, form a basis of S0 (C, ). Suppose as usual that

that

1
2 x

1
2i y .

Then the Fourier transform

p (z) =

p

1
2 x

1
2i y

and

of p is given by

n+p m+p 2|u|z z

1
e
(2iu)m+n+2p z n+p zm+p

which is a function of the form

(i sgn u)m+n+2p e2|u|z z zn+p z m+p +

p1


aq e2|u|z z zn+q z m+q .

q=0

Only the coefficient ap1 interests us. It equals

(i sgn u)m+n+2p1
{p(n + m + 1 + p 1)}.
2iu
Since

r (w) (z) = (i sgn u)  (

z)

and

r (X ) = (1)m+n r (w) r (X+ ) r(w)

while

r (X+ )p = (2ui)p+1
we see that


p1

r (X )p = (2ui)p+1 (i sgn u)(n + m + 2p + 1)p +

bq q .

q=0

Since U = X+ X we have


p1

r (U )p = (i sgn u)(n + m + 2p + 1)p

bq q

1=0

and we can find the functions p , p = 0, 1, , such that


p1

p = p +

apq q

q=0

while

r (U )p = (i sgn u)(n + m + 2p + 1)p .

These functions form a basis of S0 (C, ). Consequently r is admissible.
If it were not irreducible there would be a proper invariant subspace which may or not contain
0 . In any case if S1 is the intersection of all invariant subspaces containing 0 and S2 is the sum of all

Chapter 1

90

invariant subspaces which do not contain 0 both S1 and S2 are invariant and the representation 1 of
A on S1 /S2 S1 is irreducible. If the restriction of 1 to the Lie algebra of SO(2,
p it does
 R) contains

not contain p . Thus 1 is not equivalent to the representation X 1 Ad(X) . Consequently
the irreducible representation of {A, } whose restriction to A is 1 must be one of the special
representations (1 , 2 ) or a representation (1 , 2 ). Examining these we see that since contains
q with q = sgn u(n + m + 1) it contains all the representations q with q = sgn u(n + m + 2p + 1),
p = 0, 1, 2, . Thus S1 contains all the functions p and S2 contains none of them. Since this
contradicts the assumption that S0 (C, ) contains a proper invariant subspace the representation r is
irreducible.
For the reasons just given the representation of {A, } whose restriction to A contains r is
either a (1 , 2 ) or a (1 , 1 ). It is a (1 , 1 ) if and only if n + m = 0. Since

a 0
0 a


= (a) sgn aI = (a) (a)I,

we must have 1 2 = 0 in the first case and 21 = 0 in the second. 0 is the restriction of to R .
Since the two solutions 21 = 0differ by they lead to the same representation. If n + m = 0 then
21 = 0 if and only if (z) = 1 (z) for all z in C . Of course (z) = z z.
t mi
) . Because of Theorem 5.11
Suppose n + m > 0 so that is a (1 , 2 ). Let i (t) = |t|si ( |t|
we can suppose that m1 = 0. Let s = s1 s2 . We can also suppose that s is non-negative. If
m = |m1 m2 | then s m is an odd integer so m and m2 are determined by s. We know what
representations of the Lie algebra of SO(2, R) are contained in . Appealing to Lemma 5.7 we see that
s = n + m. Since 1 2 = 0 we have s1 + s2 = 2r . Thus s1 = r + m+n
and s2 = r n+m
2
2 . In all
cases the representation is determined by alone and does not depend on . We refer to it as ().
Every special representation (1 , 2 ) is a () and () is equivalent to ( ) if and only if = 
or  (z) = (
z ).
We can now take the first step in the proof of the local functional equation.
Theorem 5.13 Let be an innite-dimensional irreducible admissible representation of H R . If is

a non-trivial additive character of R there exists exactly one space W (, ) of functions W on GR

with the following properties
(i) If W is in W (, ) then



 
1 x
W
g = (x) W (g)
0 1
for all x in F .

(ii) The functions W are continuous and W (, ) is invariant under (f ) for all f in HR .

Moreover the representation of HR on W (, ) is equivalent to .

(iii) If W is in W (, ) there is a positive number N such that

W

t
0

0
1


= O(|t|N )

as |t| .
We prove first the existence of such a space. Suppose = () is the representation associated
to some quasi-character of C . An additive character being given the restriction of to A contains
the representation r determined by and . For any in S(C, ) define a function W on G+ by

W (g) = r (g) (1)

Chapter 1

91

Since (g) W = Wr (g) the space of such functions is invariant under right translations. Moreover

W

1 x
0 1

 
g = (x) W (g)

Every vector in S(C, ) is infinitely differentiable for the representation r . Therefore the functions
W are all infinitely differentiable and, if X is in A,

(X)W = Wr (X) .
In particular the space W1 (, ) of those W for which is in S0 (C, ) is invariant under A. We set
W equal to 0 outside of G+ and regard it as a function on GR .
We want to take W (, ) to be the sum of W1 (, ) and its right translate by . If we do it
will be invariant under {A, } and transform according to the representation of {A, }. To verify the
second condition we have to show that it is invariant under HR . For this it is enough to show that
S0 (C, ) is invariant under the elements of HR with support in G+ . The elements certainly leave the
space of functions in S(C, ) spanned by the functions transforming according to a one-dimensional
representation of SO(2, R) invariant. Any function in S(C, ) can be approximated uniformly on
compact sets by a function in S0 (C, ). If in addition it transforms according to the representation
n of SO(2, R) it can be approximated by functions in S0 (C, ) transforming according to the same
representation. In other words it can be approximated by multiples of a single function in S0 (C, ) and
therefore is already in S0 (C, ).
The growth condition need only be checked for the functions W in W1 (, ). If a is negative

W
but if a is positive and
it is equal to

a 0
0 1


=0

(z) = e2|u|z zP (z, z)

e2|u|a P (a1/2 , a1/2 ) (a) |a|1/2 ,

and certainly satisfies the required condition.

We have still to prove the existence of W (, ) when = (1 , 2 ) and is infinite dimensional.
As in the first paragraph we set


(1 , 2 , ) =

1
1 (t) 1
) d t
2 (t) (t, t

for in S(Rs ) and we set



W (g) = 1 (detg) |detg|1/2 1 , 2 , r(g)


= 1 , 2 , r1 ,2 (g) .
r1 ,2 is the representation associated to the quasi-character (a, b) 1 (a) 2 (b) of R R . If X is
in A
(X) W (g) = Wr1 ,2 (X) (g)
Let W (1 , 2 ; ) be the space of those W which are associated to O(2, R)-finite functions . W (1 , 2 ; )
is invariant under {A, } and under HR .

Chapter 1

92

s x m
Lemma 5.13.1 Assume 1 (x) 1
with Re s > 1 and m equal to 0 or 1. Then
2 (x) = |x| ( |x| )

there exists a bijection A of W (1 , 2 ; ) with B(1 , 2 ) which commutes with the action of {A, }.
We have already proved a lemma like this in the non-archimedean case. If is in S(R2 ) and
is a quasi-character of R set


(0, t) (t) d (t)

z(, ) =

The integral converges if (t) = |t|r (sgn t)n with r > 0. In particular under the circumstances of the
lemma

f (g) = 1 (detg) |detg|1/2 z(1 1

2 R , (g))

is defined. As usual R (x) = |x|. A simple calculation shows that

f

a1
0

x
a2

 
 a1 1/2
g = 1 (a1 ) 2 (a2 )   f (g).
a2

If is the partial Fourier transform of introduced in the first paragraph then

(g) f = f
1
if 1 = r1 ,2 (f ). A similar relation will be valid for a function f in HR , that is

(f )f = f
1
if 1 = r1 ,2 (f ). In particular if f is O(2, R)-finite there is an elementary idempotent such that
()f = f . Thus, if 1 = r1 ,2 (), f = f
and
1 is O(2, R) finite. Of course f is
1
O(2, R)-finite if and only if it belongs to B(1 , 2 ).
We next show that given any f in B(1 , 2 ) there is an O(2, R)-finite function in S(R2 ) such
that f = f . According to the preceding observation together with the self-duality of S(R2 ) under
Fourier transforms it will be enough to show that for some in S(R2 ), f = f . In fact, by linearity, it
is sufficient to consider the functions n in B(1 , 2 ) defined earlier by demanding that

n



cos sin
sin cos

= ein

n must be of the same parity as m. If = sgn n set

(x, y) = e(x

Then

+y 2 )

cos sin
sin cos

(x + iy)|n|


= ein

Since (g)f = f(g) when detg = 1 the function f is a multiple of n . Since

f (e) = (i)

|n|

= (i)n

et t|n|+s+1 d t
2

(|n|+s+1)
2

(|n| + s + 1)
2

Chapter 1

93

which is not 0, the function f is not 0.

The map A will transform the function W to f . It will certainly commute with the action of
{A, }. That A exists and is injective follows from a lemma which, together with its proof, is almost
identical to the statement and proof of Lemma 3.2.1.
The same proof as that used in the non-archimedian case also shows that W (1 , 2 ; ) =
W (2 , 1 ; ) for all . To prove the existence of W (, ) when = (1 , 2 ) and is infinitedimensional we need only show that when 1 and 2 satisfy the condition the previous lemma the
functions W in W (1 , 2 ; ) satisfy the growth condition of the theorem. We have seen that we can
take W = W with

(x, y) = e(x

+y 2 )

P (x, y)

where P (x, y) is a polynomial in x and y . Then

(x, y) = e(x

+u2 y 2 )

Q(x, y)

where Q(x, y) is another polynomial. Recall that (x) = e2iux . Then

W

a 0
0 1




= 1 (a) |a|

1/2

e(a

t +u2 t2 )

2 2

Q(at, ut1 ) |t|s (sgn t)m d t

The factor in front certainly causes no harm. If > 0 the integrals from to and from to
decrease rapidly as |a| and we need only consider integrals of the form

e(a

t +u2 t2 ) r

2 2

t dt

0
2

2 2

where r is any real number and u is fixed and positive. If v = u2 then u2 = v 2 + 3u4 and e 4 u t tr
is bounded in the interval [0, ] so we can replace u by v and suppose r is 0. We may also suppose that
a and v are positive and write the integral as

2av

e(at+vt

1 2

dt.

The integrand is bounded by 1 so that the integral is O(1). In any case the growth condition is more
than satisfied.
We have still to prove uniqueness. Suppose W1 (, ) is a space of functions satisfying the first
two conditions of the lemma. Let n be a representation of the Lie algebra of SO(2, R) occurring in
and let W1 be a function in W1 (, ) satisfying


cos sin
W1 g
= ein W1 (g).
sin cos


If

1 (t) = W1

t
|t|1/2

0
1
|t|1/2



Chapter 1

94

the function W1 is completely determined by 1 . It is easily seen that



1
|t|1/2

t
|t|1/2

|t|1/2

(U )W1

(Z)W1

(X+ )W1

1
|t|1/2

t
|t|1/2


= i n 1 (t)

= 2t


1
|t|1/2

d1
dt

= i u t 1 (t).

Thus if +
1 and 1 correspond to (V+ )W1 and (V )W1

+
1 (t) = 2t
and

1 (t) = 2t
Since

D=

d1
(2ut n) 1
dt

d1
+ (2ut n) 1 (t).
dt

1
U2
V V+ iU
2
2

(D)W1 corresponds to
d
2t
dt


d1
d2
t
2t
+ (2nut 2u2 t2 )1 .
dt
dt

Finally ()W1 corresponds to 1 (t).

s
m
Suppose that is either (1 , 2 ) or (1 , 2 ). Let 1 1
2 (t) = |t| (sgn t) . If s m is an
odd integer we can take n = |s| + 1. From Lemma 5.6 we have (V )W1 = 0 so that 1 satisfies the
equation

2t

d1
+ (2ut n)1 = 0.
dt

If the growth condition is to be satisfied 1 must be 0 for ut < 0 and a multiple of |t|n/2 eut for ut > 0.
Thus W1 is determined up to a scalar factor and the space W (, ) is unique.
2
Suppose s m is not an odd integer. Since (D)W1 = s 21 W1 the function 1 satisfies the
equation



d 2 1
nu (1 s2 )
2
+
u
+
+
dt2
t
4t2

1 = 0

We have already constructed a candidate for the space W (, ). Lets call this candidate W2 (, ).
There will be a non-zero function 2 in it satisfying the same equation as 1 . Now 1 and all of
its derivatives go to infinity no faster than some power of |t| as t while as we saw 2 and its
derivations go to 0 at least exponentially as |t| . Thus the Wronskian

d2
d1
2
dt
dt

Chapter 1

95

goes to 0 as |t| . By the form of the equation the Wronskian is constant. Therefore it is identically
0 and 1 (t) = 2 (t) for t > 0 and 1 (t) = 2 (t) for t < 0 where and are two constants. The
uniqueness will follow if we can show that for suitable choice of n we have = . If m = 0 we can
take n = 0. If 1 (t) = |t|s1 (sgn t)m1 then ()W1 = (1)m2 W1 so that 1 (t) = (1)m1 1 (t) and
2 (t) = (1)m2 2 (t). Thus = . If m = 1 we can take n = 1. From Lemma 5.6

(V1 )W1 = (1)m1 s()W1

so that

d1
+ (2ut 1)1 (t) = (1)m1 s1 (t).
dt
Since 2 satisfies the same equation = .
If is a quasi-character of R and is the character of C defined by (z) = (z z) then
() = (, ). We have defined W ((), ) in terms of and also as W (1 , 2 ; ). Because of the
uniqueness the two resulting spaces must be equal.
2t

Corollary 5.14 Let m and n be two integers, one positive and the other 0. Let be a quasi-character
of C of the form
m+n
(z) = (z z)r 2 z m zn

and let 1 and 2 be two quasi-characters of R satisfying 1 2 (x) = |x|2r (sgn x)m+n+1 and
m+n
1 1
sgn x so that () = (1 , 2 ). Then the subspace Bs (1 , 2 ) of B(1 , 2 ) is
2 (x) = x
dened and there is an isomorphism of B(1 , 2 ) with W (1 , 2 ; ) which commutes with the action
of {A, }. The image Ws (1 , 2 ; ) of Bs (1 , 2 ) is W ((), ). If belongs to S(R2 ) and W
belongs to W (1 , 2 ; ) then W belongs to Ws (1 , 2 ; ) if and only if


xi

j
(x, 0) dx = 0
y j

for any two non-negative integers i and j with i + j = m + n 1.

Only the last assertion is not a restatement of previously verified facts. To prove it we have to
show that f belongs to Bs (1 , 2 ) if and only if satisfies the given relations. Let f = f . It is
1
1
1
in Bs (1 , 2 ) if and only if it is orthogonal to the functions in Bf (1
1 , 2 ). Since Bf (1 , 2 ) is
finite-dimensional there is a non-zero vector f0 in it such that (X+ )f0 = 0. Then


1 y
= f0 (w)
f0 w
0 1
and f is orthogonal to f0 if and only if


1 y
f w
dy = 0.
0 1
R

(5.14.1)

2
The dimension of Bf (1
1 , 2 ) is m + n. It follows easily from Lemmas 5.6 and 5.7 that the vectors
p
(X+ ) (w) f0 , 0 p m + n 1 span it. Thus f is in Bs (1 , 2 ) if and only if each of the functions
p
(X+
) (w) f satisfy (5.14.1). For f itself the left side of (5.14.1) is equal to

 



1 x

1 (t) 1
(0, t) w
(t)
|t|
d
t
dx.
2
0 1

Chapter 1

96

Apart from a positive constant which relates the additive and multiplicative Haar measure this equals



(t, tx)tm+n sgn t dt dx



which is

(1)

m+n1

(t, x)tm+n1 dt dx

or, in terms of ,

(1)m+n1
By definition

(t, 0) tm+n1 dt.

(5.14.2)

r1 ,2 (w) (x, y) =  (y, x)

and an easy calculation based on the definition shows that

p
r1 ,2 (X+
) (x, y) = (2iuxy)p (x, y).
p

Thus r1 ,2 (X+ ) r1 ,2 (w) is a non-zero scalar times

2p
 (y, x)
p
p
x y
For this function (5.14.2) is the product of a non-zero scalar and



Integrating by parts we obtain

2p
 (0, x) xm+n1 dx.
xp y p
p 
(0, x) xm+np1 dx
y p

except perhaps for sign. If we again ignore a non-zero scalar this can be expressed in terms of as

m+np1
(x, 0) xp dx.
y m+np1

The proof of the corollary is now complete.

Before stating the local functional equation we recall a few facts from the theory of local zetafunctions. If F is R or C and if belongs to S(F ) we set


Z(sF , )

(a) (a) |a|sF d a.

is a quasi-character. The integral converges in a right half-plane. One defines functions L(s, ) and
(s, , F ) with the following properties:
(a) For every the quotient
Z(sF , )
L(s, )

Chapter 1

97

has an analytic continuation to the whole complex plane as a holomorphic function. Moreover
for a suitable choice of it is an exponential function and in fact a constant.
(b) If  is the Fourier transform of with respect to the character F then

Z( 1 1s
Z(sF , )
F , )
)
=
(s,
,

.
F
L(1 s, 1 )
L(s, )

If F = R and (x) = |x|rR (sgn x)m with m equal to 0 or 1 then

s + r + m
1
L(s, ) = 2 (s+r+m)
2
and if F (x) = e2iux
s+r 12

(s, , F ) = (i sgn u)m |u|R

If F = C and

(x) = |x|rC xm x
n

where m and n are non-negative integers, one of which is 0, then

L(s, ) = 2(2)(s+r+m+n) (s + r + m + n).

Recall that |x|C = x
x. If F (x) = e4i Re(wz)
s1/2

(s, , F ) = im+n (w) |w|C

These facts recalled, let be an irreducible admissible representation of HR . If = (1 , 2 ) we

set

L(s, ) = L(s, 1 ) L(s, 2 )

and

(s, , R ) = (s, 1 , R ) (s, 2 , R )

and if = () where is a character of C we set

L(s, ) = L(s, )
and

(s, , R ) = (C/R, R ) (s, , C/R )

if C/R (z) = R (z + z). The factor (C/R, R ) was defined in the first paragraph. It is of course necessary to check that the two definitions coincide if () = (1 , 2 ). This is an immediate consequence
of the duplication formula.

Chapter 1

98

Theorem 5.15 Let be an innite-dimensional irreducible admissible representation of H R . Let

be the quasi-character of R dened by




a 0

= (a)I
0 a

If W is in W (, ) set


W

(g, s, W ) =
R

 s, W ) =
(g,

W
R

 
g |a|s1/2 d a
 
0
g 1 (a) |a|s1/2 d a
a

a
0

0
1

a
0

and let
(g, s, W ) = L(s, ) (g, s, W )
 s, W ) = L(s,
 s, W ).
(g,
) (g,
 s, W ) are absolutely convergent in some right half(i) The integrals dened (g, s, W ) and (g,
plane.

 s, W ) can be analytically continued to the whole complex

(ii) The functions (g, s, W ) and (g,
plane as meromorphic functions. Moreover there exists a W for which (e, s, W ) is an
exponential function of s.
(iii) The functional equation

(wg,
1 s, W ) = (s, , ) (g, s, W )
is satisifed.
(iv) If W is xed (g, s, W ) remains bounded as g varies in a compact set and s varies in the

region obtained by removing discs centred at the poles of L(s, ) from a vertical strip of nite
width.
We suppose first that = (1 , 2 ). Then W (, ) = W (1 , 2 ; ). Each W in W (1 , 2 ; ) is
of the form W = W where

(x, y) = e(x

+u2 y 2 )

P (x, y)

with P (x, y) a polynomial. However we shall verify the assertions of the theorem not merely for W in
W (, ) but for any function W = W with in S(R2 ). Since this class of functions is invariant under
right translations most of the assertions need then be verified only for g = e.
A computation already performed in the non-archimedean case shows that

(e, s, W ) = Z(1 sR , 2 sR , )
the integrals defining these functions both being absolutely convergent in a right half-plane. Also for
s in some left half-plane
1s
1 1s


(w,
1 s, W ) = Z(1
1 R , 2 R , )

if  is the Fourier transform of .

Since can always be taken to be a function of the form (x, y) = 1 (x) 2 (y) the last assertion
of part (ii) is clear. All other assertions of the theorem except the last are consequence of the following
lemma.

Chapter 1

99

Lemma 5.15.1 For every in S(R 2 ) the quotient

Z(1 sR1 , 2 sR2 , )

L(s, 1 ) L(s, 2 )
is a holomorphic function of (s1 , s2 ) and
1s1
1s2
, 1
,  )
Z(1
1 R
2 R
1
L(1 s1 , 1
1 ) L(1 s2 , 2 )

is equal to

Z(1 sR1 , 2 sR2 , )

.
L(s1 , 1 ) L(s2 , 2 )
We may as well assume that 1 and 2 are characters so that the integrals converge for Re s1 > 0
and Re s2 > 0. We shall show that when 0 < Re s1 < 1 and 0 < Re s2 < 1
(s1 , 1 , ) (s2 , 2 , )

1s1
1s2 
, 1
)
Z(1 sR1 , 2 sR2 , ) Z(1
1 R
2 R

is equal to

1s1
1s2
Z(1
, 1
,  ) Z(1 sR1 , 2 sR2 , )
1 R
2 R

if and belong to S(R2 ).

The first of these expressions is equal to


  s1  s2
x
y  x   y 
(x, y) (u, v)1
d x d y du dv
2
u
v u v 


if we assume, as we may, that d x = |x|1 dx. Changing variables we obtain



s2

(xu, yv) (u, v) du dv

The second expression is equal to

1
1s1
1
1 (x) 2 (y) |x|

which equals


|y|

1s1




1 (x) 2 (y) |x| |y|
s1

s2

1 (x) 2 (y) |x| |y|

s1

|xy|

d x d y


(xu, yv) (u, v) du dv

(x

u, y

d x d y


v) (u, v) du dv

d x d y.

Since the Fourier transform of the function (u, v) (xu, yv) is the function |xy|1  (x1 u, y 1 v)
the Plancherel theorem implies that

(xu, yv)  (u, v) du dv =

|xy|1  (x1 u, y 1 v) (u, v) du dv.

The desired equality follows.

Choose 1 and 2 in S(R) such that

L(s, i ) = Z(i sR , i )
and take (x, y) = 1 (x) 2 (y). The functional equation of the lemma follows immediately if 0 <
s1 < 1 and 0 < s2 < 1. The expression on one side of the equation is holomorphic for 0 < Re s1 and
0 < Re s2 . The expression on the other side is holomorphic for Re s1 < 1 and Re s2 < 1. Standard and
easily proved theorems in the theory of functions of several complex variables show that the function
they define is actually an entire function of s1 and s2 . The lemma is completely proved.
For = (1 , 2 ) the final assertion of the theorem is a consequence of the following lemma.

Chapter 1

100

Lemma 5.15.2 Let be a compact subset of S(R 2 ) and C a domain in C2 obtained by removing balls

about the poles of L(s1 , 1 ) L(s2 , 2 ) from a tube a1 Re s1 b1 , a2 Re s2 b2 . Then

Z(1 sR1 , 2 sR2 , )
remains bounded as varies in and (s1 , s2 ) varies in C.

The theorems in the theory of functions alluded to earlier show that it is enough to prove this
when either both a1 and a2 are greater than 0 or both b1 and b2 are less than 1. On a region of the first
type the functions Z(1 sR , 2 sR , ) is defined by a definite integral. Integrating by parts as in the
theory of Fourier transforms one finds that

Z(1 R1 +i1 , 2 R2 +i2 , ) = O(12 + 22 )n

as 12 + 22 uniformly for in and a1 1 b1 , a2 2 b2 which is a much stronger
estimate than required. For a region of the second type one combines the estimates just obtained with
the functional equation and the known asymptotic behavior of the -function.
Now let be a quasi-character of C which is not of the form (z) = (z z) with a quasicharacter of R and let = (). W (, ) is the sum of W1 (, ) and its right translate by . It is
easily seen that

(g, s, ()W ) = (1) (1g, s, W )

and that

1


g, s, W )
(wg,
s, ()W ) = (1) (w

Thus it will be enough to prove the theorem for W in W1 (, ). Since

(g, s, W ) = (g, s, W )
and



(wg,
s, W ) = (wg,
s, W )

we can also take g in G+ . W1 (, ) consists of the functions W with in S0 (C, ). We prove the
assertions for functions W with in S(C, ). Since this class of functions is invariant under right
translations by elements of G+ we may take g = e.
As we observed in the first paragraph we will have

(e, s, W ) = Z(sC , )


(w,
1 s, W ) = (C/R, ) Z( 1 1s
C , )

in some right half plane and the proof proceeds as before. If (z) = (z z)r z m zn and p q = n m the
function

(z) = e2|u|z zz p zq

belongs to S0 (C, ) and


Z(sC , )

= 2

e2|u|t t2 (s + r + p + m) dt
2

= (2|u|)(s+r+p+m) (s + r + p + m)
Taking p = n we obtain an exponential times L(s, ). The last part of the theorem follows from an
analogue of Lemma 5.15.2.
The local functional equation which we have just proved is central to the Hecke theory. We
complete the paragraph with some results which will be used in the paragraph on extraordinary
representations and the chapter on quaternion algebras.

Chapter 1

101

Lemma 5.16 Suppose 1 and 2 are two quasi-characters for which both = (1 , 2 ) and =

(1 , 2 ) are dened. Then

L(1 s,
) (s, , )
L(1 s,
) (s, , )
=
L(s, )
L(s, )
and the quotient
L(s, )
L(s, )
is an exponential times a polynomial.
s
m
Interchanging 1 and 2 if necessary we may suppose that 1 1
2 (x) = |x| (sgn x) with s > 0.
According to Corollary 5.14, W (, ) is a subspace of W (1 , 2 , ). Although W (1 , 2 , ) is not
 s, W ) when W lies in W (1 , 2 , ) and to
irreducible it is still possible to define (g, s, W ) and (g,
use the method used to prove Theorem 5.15 to show that


(wg,
1 s, W )
L(1 s,
)
is equal to

(s, )

(g, s, W )
L(s, )

Applying the equality to an element of W (, ) we obtain the first assertion of the lemma. The second
is most easily obtained by calculation. Replacing 1 and 2 by 1 tR and 2 tR is equivalent to a
translation in s so we may assume 2 is of the form s (x) = (sgn x)m2 . There is a quasi-character
of C such that = (). If (z) = (z z)r z m zn then 1 (x) = |x|2r+m+n (sgn x)m+n+m2 +1 ,
1 (x) = xm+n (sgn x)m2 +1 so that r = 0. Apart from an exponential factor L(s, ) is equal to
(s + m + n) while L(s, ) is, again apart from an exponential factor,

s + m + n + m1
2

s + m2
2

(5.16.1)

where m1 = m + n + m2 + 1 (mod 2). Since m + n > 0 the number

k=

1
(m + n + 1 + m1 m2 ) 1
2

is a non-negative integer and m2 + 2k = m + n + m1 1. Thus

s + m2
2

1
=
2k+1

(s + m2 + 2j)
j=0

s + m + n + m1 + 1
2

By the duplication formula the product (5.16.1) is a constant times an exponential times

(s + m + n + m1 )
.
!k
j=0 (s + m2 + 2j)
If m1 = 0 the lemma follows immediately. If m1 = 1

(s + m + n + m1 ) = (s + m + n) (s + m + n)
and m2 + 2k = m + n. The lemma again follows.

Chapter 1

102

Lemma 5.17 Suppose (z) = (z z) r z m zn is a quasi-charaacter of C with mn = 0 and m + n > 0.

Suppose 1 and 2 are two quasi-characters of F with 1 2 (x) = |x|2r xm+n sgn x and 1 1
2 (x) =
xm+n sgn x. Then for every in S(R2 ) such that

xi

j
(x, 0) dx = 0
yj

for i > 0, j 0, and i + j + 1 = m + n the quotient

Z(1 sR , 2 sR , )


L s, (w)
is a holomorphic function of s and for some it is an exponential.
If W belongs to W (1 , 2 , ) this is a consequence of Corollary 5.14 and Theorem 5.15. Unfortunately we need the result for all . The observations made during the proof of Lemma 5.16 show
that if = (1 , 2 ) the quotient

Z(1 sR , 2 sR , )
L(s, )

is holomorphic. Since L(s, ) and L(s, ) have no zeros we have only to show that the extra poles of
L(s, ) are not really needed to cancel poles of Z(1 sR , 2 sR , ). As in the proof of Lemma 5.16 we
may take r = 0. We have to show that Z(1 sR , 2 sR , ) is holomorphic at s = m2 2j , 0 j k
if m1 = 0 and at s = m2 2j , 0 j k if m1 = 1. We remark first that if 1 and 2 are two
quasi-characters of R , belongs to S(R2 ), and Re s is sufficiently large then, by a partial integration,

1
1 (x) 2 (y) |x| |y| (x, y) d x d y =
s
s


1 (x) 2 (y) (y) |x|s |y|s+1

(x, y) d x d y
y

if (y) = sgn y . Integrating by parts again we obtain

1
1 (x)2 (y) |x| |y| (x, y) d x d y =
s(s + 1)
s


1 (x) 2 (y) |x|s |y|s+2

2
(x, y) d x d y.
y 2

If belongs to S(R2 ) the function defined by

(x, y) |x|s+1 |y|s d x dy y

is certainly holomorphic for Re s > 0. We have to show that if


(x, 0) dx = 0
it is holomorphic for Re s > 1. Suppose first that (x, 0) 0. Since

(x, y) = y
(x, 0) +
y

(y u)
0

2
(x, u) du
y 2

(5.17.1)

Chapter 1

103

the function

1
(x, y)
y
is dominated by the inverse of any polynomial. Thus (5.17.1) which equals

(x, y) |x|s+1 |y|s+1 (y) d x d y
(x, y) =

is absolutely convergent for Re s > 1. In the general case we set

(x, y) = {(x, y) (x, 0)ey } + (x, 0)ey

2

= 1 (x, y) + 2 (x, y).

Since 1 (x, 0) = 0 we need only consider

2 (x, 0)ey |x|s+1 |y|s d x d y

2

which is the product of a constant and

s
2 (x, 0) |x|s dx.

2
The integral defines a function which is holomorphic for Re s > 1 and, when the assumptions are
satisfied, vanishes at s = 0.
We have to show that if 0 j m + n 1 and j m2 is even then Z(1 sR , 2 sR , ) is
holomorphic at j . Under these circumstances the function Z(1 sR , 2 sR , ) is equal to

(x)m1 (y)m2 |x|m+n |x|s |y|s (x, y) d x d y
which equals


(1)j
j
(x, y) d x d y.
(x)m1 |x|s+m+n |y|s+j
!j1
j
y
i=0 (si )
The factor in front is holomorphic at s = j . If
(x, y) = xm+nj1
the integral itself is equal to

j
(x, y)
y j

|x|s+j+1 |y|s+j (x, y) d x d y.

Since, by assumption,


(x, 0) dx = 0,

it is holomorphic at s = j .
We observe that if m + n is even

(x, y) = e(x

+y 2 )

xy m+n

satisfies the conditions of the lemma and, if r = 0 and m2 = 0, Z(1 sR , 2 sR , ) is equal to

e(x

|x|m+n+s+1 |y|m+n+s d x d y


which differs by an exponential from (s + m + n) and L s, () . If m2 = 1 we take (x, y) =
2
2
e(x +y ) y m+n+1 to obtain the same result. If m + n is odd and m2 = 0 the polynomial factor will
be y m+n+1 but if m + n is odd and m2 = 1 it will again be xym+n .
2

+y 2 )

Chapter 1

104

Proposition 5.18 Suppose and  are two innite-dimensional irreducible admissible representa-

tions of HR such that, for some quasi-character of F ,

If

a
0

0
a


= (a)I

a
0

0
a


= (a)I.

L(1 s, 1
)
)
L(1 s, 1
(s, , ) =
(s,  , )
L(s, )
L(s,  )

for all quasi-characters and and  are equivalent.

Suppose = (1 , 2 ) or (1 , 2 ). From Lemma 5.16 and the definitions the expression on
the left is equal to

(i sgn u)

m1 +m2

2s+s1 +s2 1 2s+s1 +s2 1

|u|

 1sr1 +m1   1sr2 +m2 

2

 s+r2 +m   s+r +m
1
1
2
2

2
2

if is trivial and i (x) = |x|ri (sgn x)m1 . If (x) = sgn x and ni is 0 or 1 while mi + n1 = 1 (mod 2)
the quotient is

(i sgn u)

m1 +m2

2s+s1 +s2 1 2s+s1 +s2 1

|u|

 1sr1 n1   1sr2 +n2 

2 
 s+r2 1 +n1   s+r2 +n
.
2

2
2

If we let  be (1 , 2 ) or (1 , 2 ) we obtain similar formulae with ri replaced by ri and mi by mi .
Consider first the quotients for . The first has an infinite number of zeros of the form r1
m1 2p where p is a non-negative integer and an infinite number of the form r2 m2 2p where p is
a non-negative integer, but no other zeros. Similarly the zeros of the second are at points r1 n1 2p
or r2 n2 2p. Thus if the quotients are equal r1 + m1 r2 + n2 r2 + m2 + 1 (mod 2). Moreover
if r1 + m1 = r2 + m2 + 1 (mod 2) then = (1 , 2 ) and, as we saw in Theorem 5.11, (1 , 2 ) =
(1 , 2 ) so that the two quotients are equal. As a result either r1 + m1 = r2 + m2 + 1 (mod 2) and
r1 + m1 = r2 + m2 + 1 (mod 2) or neither of these congruences hold.
Suppose first that = (1 , 2 ) and  (1 , 2 ). Then the first quotient for has zeros at the
points r1 m1 , r1 m1 2, and r2 m2 , r2 m2 2, while that for  has zeros at
r1 m1 , r1 m1 2, and r2 m2 , r2 m2 2, . Thus either r1 + m1 = r1 + m1 or
r1 + m1 = r2 + m2 . Interchanging 1 and 2 if necessary we may assume that the first of these two
alternatives hold. Then r2 + m2 = r2 + m2 . Moreover r1 + r2 = r1 + r2 and |m1 m2 | = |m1 m2 |.
If m1 = m1 it follows immediately that 1 = 1 and 2 = 2 . Suppose that m1 = m1 . Examining
the second quotient we see that either r1 + n1 = r1 + n1 or r1 + n1 = r2 + n2 . The first equality
is incompatible with the relations r1 + m1 = r1 + m1 and m1 = m1 . Thus r1 + n1 = r2 + n2 . For
the same reason r2 + n2 = r1 + n1 . Interchanging the roles of 1 , 2 and 1 , 2 if necessary we may
suppose that m1 = 0 and m1 = 1. Then r1 = r1 + 1. Since r1 + r2 = r1 + r2 we have r2 = r2 1
so that m2 = 1, m2 = 0. Thus n1 = n2 = 1 and r1 = r2 so that r2 = r1 . It follows that 1 = 2 and
2 = 1 .
Finally we suppose that = (1 , 2 ) and  = (1 , 2 ). Then there are quasi-characters 1

and 1 of C such that = (1 ) and  = (1 ). Replacing 1 by the quasi-character z 1 (
z)
r m

r  m
does not change (1 ) so we may suppose that 1 (z) = (z z) z while 1 (z) = (z z) z . Since

Chapter 1

105

1 and 1 must have the same restriction to R the numbers 2r + m and 2r + m are equal while
m m (mod 2). Apart from a constant and an exponential factor the quotient
L(1 s,
)
L(s, )

(s, , )
is given by

(1 s r)
(s + r + m)

whose pole furthest to the left is at 1 r . Consequently r = r and m = m .

Corollary 5.19 Suppose and  are two irreducible admissible representations of HR . Suppose there

is a quasi-character of R such that


a 0

= (a)I
0 a

a
0

0
a


= (a)I

If for all quasi-characters , L(s, ) = L(s,  ), L(s, 1 ) = L

6 (s, 1  ), and
(s, , ) = (s,  , ) then and  are equivalent.
Combining Lemma 5.16 with the previous propositon we infer that there is a pair of quasicharacters 1 and 2 such that both and  are one of the representations (1 , 2 ) or (
 1 , 2 ).
However the computations made during the proof of Lemma 5.16 show that L s, (1 , 2 ) differs


from L s, (1 , 2 ) for a suitable choice of .
Let K be the quaternion algebra of R. We could proceed along the lies of the fourth paragraph
and associate to every finite-dimensional irreducible representation of K a representation () of
GR . Since we have just classified the representations of GR we can actually proceed in a more direct
manner.
We identify K with the algebra of 2 2 complex matrices of the form

z=

a b
b a

Then

z =

a
b
b a

and (z) = z is the scalar matrix (|a|2 + |b|2 )I while (z) is the scalar matrix (a + a
+ b + b)I . Let

1 be the two dimensional representation of K associated to this identification and let n be the nth
symmetric power of 1 . Any irreducible representation is equivalent to a representation of the form
n where is a quasi- character of R . Thus



( n )(h) = (h) n (h)

Since (h) is always positive we may suppose that is of the form (x) = |x|r .
Let be a finite dimensional representation and let act on U . In the first paragraph we
introduced the space S(K, U ). It is clear that if is in S(K, U ) the integrals

Z(sR

, ) =

and

Z(sR 1 , ) =
converge absolutely in some right half-plane.


K

(h) |(h)|s (h) d h

1 (h) |(h)|s (h) d h

Chapter 1

106

Proposition 5.20 Suppose (x) = |x|r and = n . Let be the character of C dened by

(z) = (z z)r1/2 z n+1 . Set L(s, ) = L(s, ) and

(s, , R ) = (C/R, R ) (s, , C/R )

The quotient
s+1/2

Z(R
, )
L(s, )
can be analytically contained to the whole complex plane as a holomorphic function. Given u in U
there exists a in S(K, U ) such that
s+1/2

, )
Z(R
= as u.
L(s, )
For all the two functions
1 ,  )

L(1 s, )

3/2s

Z(R
and

s+1/2

, )
Z(R
(s, , R )
L(s, )
s+1/2

, ) is bounded in any region obtained by removing discs about the

are equal. Finally Z(R
poles L(s, ) from a vertical strip of nite width.
Suppose K1 is the subgroup of K formed by the elements of reduced norm one. Let 1 be the
function on R defined by


1 (t) =

(h) (th) dh
K1

1 belongs to S(R) and if 0 is the quasi-character of R defined by (t) = 0 (t)I the function
0 (t)1 (t) is even. Moreover if the multiplcative Haar measures are suitably normalized
s+1/2

Z(R

, ) = Z(2s+1
0 , 1 ).
R

Since 0 (t) = |t|2r tn we can integrate by parts as in the proof of Lemma 5.17 to see that for any
non-negative integer m

0 , 1 )
Z(2s+1
R

= !m1
j=0

(1)m
(2s + 2r + n + j + 1)


(t)m+n |t|2s+2r+m+n+1
m

m 1
d t.
tm

1
The integral is holomorphic for Re(2s + 2r + m + n) > 1 and, if tm
vanishes at t = 0, for
Re(2s + 2r + m + n) > 2. Thus the function on the left has an analytic continuation to the whole
complex plane as a meromorphic function with simple poles. Since


1
L(s, ) = 2(2)(s+r+n+1/2) s + r + n +
2

Chapter 1

107
m

1
we have to show that its poles occur at the points s + r + n + 12 + j = 0 with j = 0, 1, 2, . Since tm
vanishes at 0 if m + n is odd its only poles are at the points 2s + 2r + 2n + 2j + 1 = 0 with n + 2j 0.
m
1
= 0 at 0 if m < n. If we expand
To exclude the remaining unwanted poles we have to show that tm
m
1
in a Taylors series about 0 we see that tm = 0 at 0 unless the restriction of n to K1 is contained
in the representation on the polynomials of degree m on K . This can happen only if m n.
 is equivalent to the representation h 1 (h ) the quotient
Since

1 ,  )

L(1 s, )

3/2s

Z(R

is also holomorphic. The argument used to prove Lemma 5.15.1 shows that there is a scalar (s) such
that, for all ,

1 ,  )
, )
Z(R
.
= (s)

L(s,
)
L(1 s, )

3/2s

s+1/2

Z(R

We shall used the following lemma to evaluate (s).

Lemma 5.20.1 Let be a function in S(C) of the form

)
(x) = e2xx P (x, x
where P is a polynomial in x and x
. Suppose (xu) = (x) 1 (u) if u
u = 1. Dene the function

in K by
1
 u
(z) = () () (
) 2 u, (z)
if (z) =
. Then extends to a function in S(K) and its Fourier transform is given by
 (z) = (C/R, R )  () 1 () (
) 2 (z)u, u

1

if  is the Fourier transform of .

By linearity we may assume that is of the form

(x) = e2xx (x
x)p x
n+1
where p is a non-negative integer. We may suppose that the restriction of n to the elements of norm
 . Then
 = rn n .
one is orthogonal and identify the space U on which its acts with its dual U
R
Thus if




z=

a b
b a

the value of at z is

 u = e2(aa+bb) (a
e2(aa+bb) (a
a + bb)r+n+p u, (z)
a + bb)p u, n (z)
u

The expression on the right certainly defines a function in S(K).

We are trying to show that if

F (z) = () () (
) 2 1 (z)
1

Chapter 1

108

when z =
then the Fourier transform of F is given by

F  (z) = (C/R, R )  () 1 () (
) 2 (z).
1

If h1 and h2 have norm one

(5.20.2)

1
F (h1zh2 ) = (h1
2 ) F (z) (h1 )

and therefore

F  (h1 zh2 ) = (h1 ) F  (z) (h2 )

In particular if z is a scalar in K the operator F  (z) commutes with the elements of norm one and is
therefore a scalar operator. The expression F1 (z) on the right of (5.20.2) has the same properties so that
 which are not orthogonal
all we need do is show that for some pair of vectors u and u

F  (z)u, u
 = F1 (z)u, u

for all positive scalars z .
If we only wanted to show that F  (z) = c F1 (z) where c is a positive constant it would be enough
to show that

F  (z)u, u
 = cF1 (z)u, u
.

(5.20.3)

Once this was done we could interchange the roles of and  and and  to show that c2 = 1. To
obviate any fuss with Haar measures we prove (5.20.3).
Recall that if


a() =

ei

0
ei

then, apart from a positive constant,

 u f (k) dk
u, (k)
K1

is equal to

u, u




sin(n + 1) sin f a() d

if f is a class function on K1 , the group of elements of norm one. The equality is of course a consequence
of the Weyl character formula and the Schur orthogonality relations.
If x is a positive scalar in K then, apart from a positive constant,  (x) is given by


K



(z) R (xz) |(z)|2 d z

which is a positive multiple of





 u R xt (k) dk
u, (k)

t (t)
0

d t.

K1

Since (k) is a class function this expression is a positive multiple of


u, u

0


t3 (t)
0


sin(n + 1) sin R (2xt cos ) d d t

Chapter 1

109

Integrating the inner integral by parts we obtain

n+1
u, u

4iux



t (t)

cos(n + 1) R (2xt cos ) d

d t.

On the other hand if x, which is a positive real number, is regarded as an element of C then  (x)
is a positive multiple of

C

or of





(z) R (xz) z z d z

t (t)

Since

i(n+1)


R (xt cos ) d)

d t.

i(n+1)

R (xt cos ) d = 2

cos(n + 1) R (xt cos ) d

0

and (C/R) = i sgn u the identity (5.20.3) follows for any choice of u and u
.
To evaluate (s) we choose as in the lemma and compute
s+ 1
Z(R 2

and

3
2 s

Z(R
The first is equal to

(z) |(z)|s+ 2 (z)v, v d z

1

, v), v =


, v), v =

3
 v d z.
(z) |(z)| 2 s v, (z)



s+ 12

K /K1

|(z)|

d z.

K1

Since

(zk)(zk)v, 
v dk


(zk)v, vu, (zk)
u dk
K1

is, by the Schur orthogonality relations, equal to

1
v, u
u, v
deg
the double integral is equal to

1
v, u
u, v
deg

() () (
)s d z

where
= (z). If the Haar measure on C is suitably chosen the integral here is equal to Z(sC , ).
The same choice of Haar measures lead to the relation
3

Z(R2

1 , v), v =

(C/R, R )

v, u
u, vZ( 1 1s
C , ).
deg

Chapter 1

110

 = L(s, 1 ) we can compare the functional equation for

Since L(s, ) = L(s, ) and L(s, )
s+1/2
Z(sC , ) with that for Z(R
, v) to see that
(s) = (C/R, R ) (s, , C/R )
as asserted.
If

(x) = e2xx
s+1/2

then Z(sC , ) is an exponential times L(s, ) so that Z(R

, v) is, with a suitable choice of
v and u
, a non-zero scalar times an exponential times L(s, )u. The last assertion of the proposition is
proved in the same way as Lemma 5.15.2.
We end this paragraph with the observation that the space W (, ) of Theorem 5.13 cannot exist
on the
when is finite-dimensional. If W = W (, ) did exist the contragredient representation

(X+ ) would be nilpotent. However if is the
dual space W would also be finite dimensional and
(X+ ) = 2ia if (x) = e2iax .
linear functional (e) then

Chapter 1

111

6. Representations of GL(2, C). In this paragraph we have to review the representation theory of
GC = GL(2, C) and prove the local functional equation for the complex field. Many of the definitions
and results of the previous paragraph are applicable, after simple modifications which we do not
always make explicit, to the present situation.
The standard maximal compact subgroup of GL(2, C) is the group U (2, C) of unitary matrices.
H1 will be the space of infinitely differentiable compactly supported functions on GC . H2 will be
the space of functions on U (2, C) which are finite linear combinations of the matrix elements of
finite dimensional representations. HC = H1 H2 can be regarded as a space of measures. Under
convolution it forms an algebra called the Hecke algebra. The notion of an elementary idempotent and
the notion of an admissible representation of HC are defined more or less as before.
Let g be the Lie algebra of the real Lie group of GL(2, C) and let gC = g R C. A will be
the universal enveloping algebra of gC . A representation of A will be said to be admissible if its
restriction to the Lie algebra of U (2, C) decomposes into a direct sum of irreducible finite dimensional
representations each occurring with finite multiplicity. There is a one-to-one correspondence between
classes of irreducible admissible representations of HC and those of A. We do not usually distinguish
contragredient to and the tensor product of with a quasibetween the two. The representation

character of C are defined as before.

If 1 and 2 are two quasi-characters of C we can introduce the space B(1 , 2 ) and the
representation (1 , 2 ) of HC or of A on B(1 , 2 ). In order to study this representation we identify
. If A1 is the
gC with g (2, C) g (2, C) in such a way that g corresponds to the elements of X X
universal enveloping algebra of g (2, C) we may then identify A with A1 A1 .
In the previous paragraph we introduced the elements D and J of A1 . Set D1 = D 1,
D2 = 1 D , J1 = J 1, and J2 = 1 J . These four elements lie in the centre of A. A representation
of A is admissible if its restriction to the Lie algebra of the group SU (2, C) of unitary matrices of
determinant one decomposes into the direct sum of irreducible finite dimensional representations each
occurring with finite multiplicity.
The first part of the next lemma is verified by calculations like those used in the proof of
Lemma 5.6. The second is a consequence of the Frobenius reciprocity law applied to the pair SU (2, C)
and its subgroup of diagonal matrices.
Lemma 6.1. Let

and

1 (z) = (z z)si 2 (ai +bi ) z ai zbi

1

1 1
)s 2 (a+b) z a zb
2 (z) = (z) = (z z
1

where ai , bi , a, and b are non-negative integers and ai bi = ab = 0.

(i) On B(1 , 2 ) we have the following four relations

a b 2
1 
s+
1 I
2
2

1 
b a 2
(D2 ) =
1 I
s+
2
2

a1 b1 + a2 b2 
(J1 ) = (s1 + s2 ) +
I
2


b1 a1 + b2 a2
I
(J2 ) = (s1 + s2 ) +
2
(D1 ) =

Chapter 1

112

(ii) (1 , 2 ) is admissible and contains the representation n of the Lie algebra of SU (2, C) if

and only if n a + b and n a + b (mod 2) and then it contains it just once.

n is the unique irreducible representation of SU (2, C) of degree n + 1. Let B(1 , 2 , n ) be the

space of functions in B(1 , 2 ) transforming according to n .
Theorem 6.2. (i) If is not of the form z z p zq or z z p zq with p 1 and q 1 then

(2 , 2 ) is irreducible. A representation equivalent to (1 , 2 ) will be denoted by (1 , 2 ),

(ii) If (z) = z p zq with p 1, q 1 then


Bs (1 , 2 ) =

B(1 , 2 , n )

np+q
np+q (mod 2)

is the only proper invariant subspace of B(1 , 2 ). (1 , 2 ) will be any representation equivalent to the representation on Bs (1 , 2 ) and (1 , 2 ) will be any representation equivalent
to the representation on the quotient space
Bf (1 , 2 ) = B(1 , 2 )/Bs(1 , 2 )
(iii) If (z) = z p zq with p 1, q 1 then

Bf (1 , 2 ) =

B(1 , 2 , n )

|pq|n<p+q
np+q (mod 2)

is the only proper invariant subspace of B(1 , 2 ). (1 , 2 ) will be any representation

equivalent to the representation Bf (1 , 2 ) and (1 , 2 ) will be any representation equivalent
to the representation on the quotient space
Bs (1 , 2 ) = B(1 , 2 )/Bf (1 , 2 ).
(iv) (1 , 2 ) is equivalent to (1 , 2 ) if and only if (1 , 2 ) = (1 , 2 ) or (1 , 2 ) = (2 , 1 ).
(v) If (1 , 2 ) and (1 , 2 ) are dened they are equivalent if and only if (1 , 2 ) = (1 , 2 ) or

(1 , 2 ) = (2 , 1 ).
(vi) If (z) = z p zq with p 1, q 1 there is a pair of characters 1 , 2 such that 1 2 = 1 2
and 1 21 = z p zq and (1 , 2 ) is equivalent to (1 , 2 ).
(vii) Every irreducible admissible representation of HC or A is a (1 , 2 ) for some choice of 1
and 2 .
The proofs of the first three assertions will be based on two lemmas.

Lemma 6.2.1. If there exists a proper invariant subspace V of B( 1 , 2 ) which is nite dimensional
p q
then 1 1
z with p 1, q 1 and V = Bf (1 , 2 ).
2 (z) = z

Lemma 6.2.2. Let V be a proper invariant subspace of B( 1 , 2 ) and let n0 be the smallest integer

such that some subspace of V transforms according to the representation n0 of the Lie algebra of
SU (2, C). Either

B(1 , 2 , n )
V =
nn0

Chapter 1

113

or V contains a nite-dimensional invariant subspace.

Grant these lemmas for a moment and let V be a proper invariant subspae of B(1 , 2 ). As in
the case of the non-archimedean and real fields there is an invariant non-degenerate bilinear form on
1
1

B(1 , 2 ) B(1
of V in B(1
1 , 2 ). The orthogonal complement V
1 , 2 ) is a proper invariant
subspace. By Lemma 6.2.1 they cannot both contain an invariant finite dimensional subspace. Therefore
by Lemma 6.2.2 one of them is of finite codimension. The other must be of finite dimension. If V
p q
is finite-dimensional then 1 1
z and V = Bf (1 , 2 ). If V is finite dimensional
2 (z) = z
1
p q
then 1 2 (z) = z z . Since the orthogonal complement of Bf (1 , 2 ) is Bs (1 , 2 ) we must have
V = Bs (1 , 2 ).
p 1
z . It will follow from
We shall now show that Bf (1 , 2 ) is invariant when 1 1
2 (z) = z
1
p q
duality that Bs (1 , 2 ) is invariant when 1 2 (z) = z z . Every irreducible finite-dimensional
representation of A determines a representation of GC . If acts on X there is a nonzero vector v0
in X such that


z x
0 z 1

v0 = z m zn v0

for all z in C and all x in C. v0 is determined up to a scalar factor and m and n are non-negative
integers. Moreover there is a quasi-character 0 of C such that

Thus

z1
0

x
z2

a 0
0 a


= 0 (a)I


v0 = 1 (z1 ) 2 (z2 ) v0

where 1 21 (z) = z m zn . is determined up to equivalence by 1 and 2 so we write = (1 , 2 ).

As long as 1 21 (z) = z m zn with non-negative integers m and n the representation (1 , 2 ) exists.
By the ClebschGordan formula the restriction of (1 , 2 ) or its contragredient to SU (2, C) breaks up
into the direct sum of the representations i with |m n| i m + n and 1 m + n (mod 2). Let
 . To each vector v in X
 we associate
be (1 , 2 ) and let
, the contragredient representation, act on X
the function

(g) = v0 ,
(g)
v

on GC . The map v is linear and injective. Moreover

(g)
v (g) while

1/2

z1
0

x
z2

 
g = 11 (z1 ) 21 (z2 ) (g)

so that if 1 = 11 C
and 2 = 21 C the function belongs to B(1 , 2 ). As we vary 1 and
p q
2 the quasi-characters 1 and 2 vary over all pairs such that 1 1
z with p 1 and
2 (z) = z
q 1.
We have still to prove the two lemmas. Suppose V is a proper finite-dimensional subspace of
B(1 , 2 ). The representation of A on V is certainly a direct sum of irreducible representations each
occurring with multiplicity one. Let V  be an irreducible subspace of V and let V  be the dual space of
V  . Let be the linear functional : (e) on V  . If is the representation of A or of GC on V 
then


1/2

z1
0

x
z2

1
= 1
1 z21 z21 ) 2
1 (z1 ) 2 (z2 ) (z1 z
1

Chapter 1

114
1/2

Thus if 1 = 1
and 2 = 1
is (1 , 2 ). It follows immediately
1 C
2 C the representation
1
p q
(z)
=
z
z

with
p

1
and
q 1 and that V  and therefore V is
that 1 2 is of the form 1 1
2
Bf (1 , 2 ).
To prove the second lemma we regard g as the real Lie algebra of 2 2 complex matrices. Then
1/2



a=
is the centre of g and



u=

a 0
0 a

ia
b

b ia


a C


 a R, b C

is the Lie algebra of SU (2, C). If



g=

a b
b a


 a R, b C

then u g is the Cartan decomposition of the Lie algebra of the special linear group. The space
gC = g R C is invariant under the adjoint action of u on gC . Moreover u acts on gC according to the
representation 2 . One knows that 2 n is equivalent to n+2 n n2 if n 2, that 2 1 is
equivalent to 3 1 and, of course, that 2 0 is equivalent to 2 . The map of gC B(1 , 2 , n )
into B(1 , 2 ) which sends X f to (X)f commutes with the action of u. Thus (X)f is contained
in

B(1 , 2 , n+2 ) B(1 , 2 , n ) B(1 , 2 , n2 ).

It is understood that B(1 , 2 , ) = 0 if  < 0.

Now let V be a proper invariant subspace of B(1 , 2 ). Let n0 be the smallest non-negative
integer n for which V contains B(1 , 2 , n ). If n n0 set

V (n) =

B(1 , 2 , k )

nkn0
kn0 (mod 2)

If V contains every V (n) there is nothing to prove so assume that there is a largest integer n1 for which
V contains V (n1 ). All we need do is show that V (n1 ) is invariant under g. It is invariant under a and
u by construction so we need only verify that if X lies in gC then (X) takes V (n1 ) into itself. It is
clear that (X) takes V (n1 2) into V (n1 ) so we have only to show that it takes B(1 , 2 , n1 ) into
V (n1 ). Take f in B(1 , 2 , n1 ) and let (X)f = f1 + f2 with f1 in V (n1 ) and f2 in B(1 , 2 , n1 +2 ).
Certainly f2 lies in V . Since

V B(1 , 2 , n1 +2 )

is either 0 or B(1 , 2 , n1 +2 ) and since, by construction, it is not B(1 , 2 , n1 +2 ) the function f2 is 0.

The first three assertions of the theorem are now proved and we consider the remaining ones.
We make use of the fact that D1 , D2 , J1 and J2 generate the centre of A as well as a result of HarishChandra to be quoted later. Suppose and  are two irreducible representations of A which are
constituents of (1 , 2 ) and (1 , 2 ) respectively. Assume and  contain the same representations
of the Lie algebra of SU (2, C) and are associated to the same homomorphism of the centre of A into
C. Comparing the scalars (J1 ) and (J1 ) with  (J1 ) and  (J2 ) we find that 1 2 = 1 2 . Let

Chapter 1

115
a+b

1 1
)s 2 z a zb and let 1 2
2 (z) = (z z
with  (D1 ) and  (D2 ) we see that

s+
and

(z) = (z z)s

a +b
2

z a zb . Comparing (D1 ) and (2 )

a b 2   a b 2
= s +
2
2


b a 2   b a 2
s+
= s +
.
2
2
1

 
 
These relations will hold if 1 1
or 1
and therefore, when 1 2 = 1 2 ,
2 = 1 2
1 2 = 1 2




(1 , 2 ) = (1 , 2 ) or (1 , 2 ) = (2 , 1 ). If neither of these alternatives hold we must have

s=

a b
,
2

s =

ab
,
2

s=

b a
,
2

s =

ba
.
2

or

Since 1 2 = 1 2 the integers a + b and a + b must have the same parity. Let = 1 1
2 and
1
1
 = 1 2 . In the first case  is of the form  (z) = z 2p and  is of the form z2q and in the
1
second  (z) = z2p while  (z) = z 2q . Since {1 , 2 } is not {1 , 2 } neither p nor q is 0. In the
p q

p q
first case = z z and = z z and in the second = zq zp while  = z q zp .
In conclusion we see that and  contain the same representations of the Lie algebra of SU (2, C)
and are associated to the same homomorphism of the centre of A into C if and only if one of the following
alternatives holds.
(i) For some pair of quasi-characters 1 and 2 we have {,  } = {(1 , 2 ), (1 , 2 )} or {,  } =
{(1 , 2 ), (2 , 1 )}.
(ii) For some pair of quasi-characters 1 and 2 we have {,  } = {(1 , 2 ), (1 , 2 )} or {,  } =
{(1 , 2 ), (2 , 1 )}.
(iii) For some pair of quasi-characters 1 and 2 with 1 21 (z) = z p zq where p 1, q 1 we have
1
{,  } = {(1 , 2 ), (1 , 2 )} where 1 2 = 1 2 and  2 (z) is either z p z1 or z p zq .
1
(iv) For some pair of quasi-characters 1 and 2 with 1 2 (z) = z p zq where p 1, q 1 we
1
have {,  } = {(1 , 2 ), (1 , 2 )} where 1 2 = 1 2 and 1 2 (z) is either z p zq or z p zq .
The remaining assertions are now all consequences of a theorem of Harish-Chandra which, in
the special case of interest to us, we may state in the following manner.
Lemma 6.2.3. If is an irreducible admissible representation of A there exists a pair of quasicharacters 1 and 2 such that (1 , 2 ) and contain at least one irreducible representation of
the Lie algebra of SU (2, C) in common and are associated to the same homomorphism of the centre
of A into C. When this is so is a constituent of (1 , 2 ).

As before (1 , 2 ) is (1 , 2 ) and (1 , 2 ) is (1 , 2 ). If

then
= 01 .

a 0
0 a


= 0 (a)I

Chapter 1

116

Theorem 6.3. Let be an innite dimensional irreducible admissible representation of H C and let

be a non-trivial additive character of C. There is exactly one space W (, ) of functions on GC

which satises the following three conditions.
(i) Every function W in W (, ) satisies



 
1 x
W
g = (x) W (g).
0 1

(ii) The functions in W (, ) are continuous and W (, ) is invariant under the operators (f )

for f in HC . Moreover the representation of HC on W (, ) is equivalent to .

(iii) If W is in W (, ) there is a positive number N such that

W

t
0

0
1


= O(|t|N )

as |t| .
Since every is of the form = (1 , 2 ) the existence is rather easy to prove. If is in S(C2 )

let

(1 , 2 , ) =

(t, t1 ) 1 (t) 1
2 (t) d t

We let W (1 , 2 , ) be the space of functions on GC of the form

1/2

W (g) = W (g) = 1 (detg) |detg|C (1 , 2 , r(g))

where in S(C2 ) is SU (2, C)-finite under the action defined by r . It is clear that W (1 , 2 , ) =
W (2 , 1 , ) and that W (1 , 2 , ) is invariant under right translations by elements of HC and of A.
The existence of W (, ) will, as before, be a consequence of the following analogue of Lemma 5.13.1.

s
Lemma 6.3.1 Suppose 1 1
2 (t) = (tt)

a+b
2

ta tb with Re s > 1. Then there is a bijection A of

W (1 , 2 , ) with B(1 , 2 ) which commutes with the action of HC .
As before A associates to W the function


1/2 

f (g) = 1 (detg) |detg|C z 1 1

2 C , (g)
The proof of course proceeds as before. However we should check that A is surjective. Theorem 6.2
shows that, under the present circumstances, there is no proper invariant subspace of B(1 , 2 ) containing B(1 , 2 , a+b ) so that we need only show that at least one nonzero function in B(1 , 2 , a+b )
is of the form f where is in S(C2 ) and SU (2, C)-finite under right translations.
If

(x, y) = e2(xx+yy) ya y b

then, since a + b = 0, transforms under right translations by SU (2, C) according to a+b so we need
only check that f is not 0. Proceeding according to the definition we see that


f (e) =

(0, t)(tt)s

a+b
2

ta tb d t


=

e2tt (tt)1+s+

a+b
2

d t.

Chapter 1

117

Apart from a constant which depends on the choice of Haar measure this is

(2)s

a+b
2


a + b
1+s+
2

and is thus not 0.

Just as in the previous paragraph W (1 , 2 , ) is spanned by functions W where is of the
form

(x, y) = e2(xx+uuyy) xp x
q y m yn

where p, q , m, and n are integers. u is determined by the relation (z) = e4i Re uz . We can show that

W

t 0
0 1



decreases exponentially as |t| .

To prove the uniqueness we will use a differential equation as in the previous chapter. This time
the equations are a little more complicated. Suppose W1 (, ) is a space of functions satisfying the
first two conditions of the theorem. We regard n as acting on the space Vn of binary forms of degree
n according to the rule





a b
c d

(x, y) = (ax + cy, bx + dy)

If

(x, y) =

k x 2 +k x 2 k

|k|n
n kZ
2

then k is called the k th coordinate of . On the dual space Vn we introduce the dual coordinates.
If n is contained in there is an injection A of Vn into W1 (, ) which commutes with the action
of SU (2, C). Let (g) be the function on GC with values in Vn defined by

, (g) = A(g).
It is clear that W1 (, ) is determined by which is in turn determined by W1 (, ) up to a scalar
factor. The function (g) is determined by the function

(t) =

t2
0



t 2
1

on the positive real numbers. If k (t) is the k th coordinate of (t) and if is a constituent of (1 , 2 )
the differential equations


a b 2
1 
s+
1
2
2

b a 2
1 
s+
(D2 ) =
1
2
2
may, if our calculations are correct, be written as
(D1 ) =

#2

1" d
|u|2 k  n
1
a b 2 k

t + k 1 k t2
+
+ k tiuk1 =
s+
2 dt
2
2
2
2
#2

b a 2 k
1" d
|u|2 k  n
1
t k 1 k t2
s+
.

k ti
uk+1 =
2 dt
2
2
2
2

Chapter 1

118

We have set k = 0 if |k| n/2. Recall that (z) = e4i Re uz . These equations allow one to solve for
all k in terms of n/2 or n/2 .
For k = n2 the second equation may be written as







( n2 + 1)2
1 d2 n/2
1 n 1 d n/2
|u|2
b a 2 n/2
n/2

=
s

+
2 dt2
2
2 t dt
2
2t2
2

()

If we have two independent solutions of this equation their Wronskian W (t) is a non-trivial solution
of the equation

dW
(n + 1)
=
W
dt
t
and therefore a non-zero multiple of tn+1 . Since we already have shown the existence of a solution of
() which decreases exponentially we see that there cannot be another solution which is bounded by a
power of t as t . The uniqueness of the space W (, ) follows
Every irreducible admissible representation of HC is of the form = (1 , 2 ). Moreover
(1 , 2 ) = (1 , 2 ) if and only if {1 , 2 } = {1 , 2 }. Thus we may set
L(s, ) = L(s, 1 ) L(s, 2 )
and

@(s, , ) = @(s, 1 , ) @(s, 2 , ).

Then

1
L(s,
) = L(s, 1
1 ) L(s, 2 ).

The local functional equation which is proved just as in the real case read as follows.
Theorem 6.4. Let be an innite-dimensional irreducible admissible representation of H C . Let

be the quasi-character of C dened by

a
0

0
a



for a in C . If W is in W (, ) the integrals


a
(g, s, W ) =
W
0
C




a
 s, W ) =
(g,
W
0

= (a)I
 
s1/2
g |a|C
d a,
 
0
s1/2 1
g |a|C
(a) d a
1
0
1

converge absolutely in some right half-plane. Set

(g, s, W ) = L(s, ) (g, s, W ),
 s, W ) = L(s,
 s, W ).
(g,
) (g,
 s, W ) can be analytically continued to the whole complex plane
The functions (g, s, W ) and (g,
as holomorphic functions of s. For a suitable choice of W the function (e, s, W ) is an exponential
function of s. The functional equation

(wg,
1 s, W ) = @(s, , ) (g, s, W )
is satised. Moreover, if W is xed |(g, s, W )| remains bounded as g varies over a compact subset
of GC and s varies in a vertical strip of nite width from which discs about the poles of L(s, )
have been removed.
The following lemma can be verified by an explicit computation. The first assertion may also be
proved by the method of Lemma 5.16.

Chapter 1

119

Lemma 6.5 If = (1 , 2 ) and = (1 , 2 ) are dened then

L(1 s,
) @(s, , )
L(1 s,
) @(s, , )
=
L(s, )
L(s, )
and the quotient

L(s, )
L(s, )

is the product of a constant, a polynomial, and an exponential. Moreover the polynomial is of

positive degree for some choice of the quasi-character .
We verify the last assertion. There is no harm in supposing that = (1 , 2 ) and that 1 (z) =
z a+p zb+q , 2 (z) = z a zb , 1 (z) = z a+p zb , and 2 (z) = z a z b+q . p 1 and q 1 are integers.
Varying is equivalent to varying a and b through all the integers. If m1 is the largest of a + p and
b + q and m2 is the largest of a and b while n1 is the largest of a + p and b and n2 is the largest of a and
b + q the quotient
L(s, )
L(s, )
differs from

(s + n1 ) (s + n2 )
(s + m1 ) (s + m2 )

by a constant times an exponential. It is clear that n1 and n2 are both greater than or equal to m2 and
that either n1 or n2 is greater than or equal to m1 . Thus the quotient is a polynomial. If p q choose a
and b so that b + q > a b. Then n1 = m + 1 and n2 > m2 so that the quotient is of positive degree.
If q p choose a and b so that a + p > b a. Then n2 = m1 and n1 > m2 .
Lemma 6.6. Let and  be two innite-dimensional irreducible representations of HC . Suppose

there is a quasi-character of C such that

and

a
0

0
a

a
0

0
a


= (a)I

= (a)I

for all a in C . If
@(s,  , )

)
)
L(1 s, 1
L(1 s, 1
=
@(s,

,
)

L(s, )
L(s, )

for all quasi-characters then and  are equivalent.

Let = (1 , 2 ) and let  = (1 , 2 ). We let


i (z) = (z z)

si

z
(z z)1/2

ai

Chapter 1

120

and

i (z)

= (z z)

si

z
(z z)1/2

ai

with ai and ai in Z. By assumptions s1 + s2 = s1 + s2 and a1 + a2 = a1 + a2 . Choose


(z) =

z
(z z)1/2

n

with n in Z. The quotient on the right has the same zeros and poles as
1
2
(1 s s1 + | n+a
|) (1 s s2 + | n+a
|)
2
2

.
n+a1
n+a2
(s + s1 + | 2 |)
(s + s2 + | 2 |)

A pole of the numerator can cancel a pole of the denominator if and only if there are two
non-negative integers  and m such that

 n + a1   n + a2 
+

s1 s2 = 1 +  + m + 
2
2
or

 n + a1   n + a2 
+
.
s2 s1 = 1 +  + m + 
2
2

1
p q
p q
This can happen only if 1 1
or 1 1
z where p 1
2 is of the form 1 2 (z) = z z
2 (z) = z
and q 1 are integers. Since (1 , 2 ) is infinite-dimensional it cannot be of either these forms and no
poles cancel.
n+a

n+a

1
2
|, s2 + | n+a
|} = {s1 + | 2 1 |, s2 + | 2 2 |}. This
Consequently for every integer n, {s1 + | n+a
2
2




can happen only if s1 = s1 , a1 = a1 , s2 = s2 , and a2 = a2 or s1 = s2 , a1 = a2 , s2 = s1 , and a2 = a1 .
Thus and  are equivalent.
The following proposition is an easy consequence of these two lemmas.

Proposition 6.7. Suppose and  are two irreducible admissible representations of HC . Suppose

there is a quasi-character of C such that

and

a
0

0
a

a
0

0
a


= (a)I

= (a)I.

If L(s, ) = L(s,  ), L(s, 1

) = L(s, 1
 ) and
@(s, , ) = @(s,  , )
for all quasi-characters the representations and  are equivalent.

Chapter 1

121

7. Characters. If F is a non-archimedean local field and is an admissible representation of GF the

operator (f ) is of finite rank for every f in HF and therefore ha a trace Tr (f ). In this paragraph we
prove that if is irreducible there is a locally integrable function on Gf such that

Tr (f ) =

f (g) (g) dg.

GF

Although Tr (f ) depends on the choice of the Haar measure the function does not.
The following simple lemma shows that determines the class of .
Lemma 7.1 If {1 , , p } is a set of inequivalent irreducible admissible representations of HF the

set of linear forms Tr 1 , Tr 2 , , Tr p is linearly independent.

Let i act on Vi and let be an elementary idempotent such that none of the spaces i ()Vi ,
1 i p, are 0. Let
i be the representaion of HF on the finite dimensional space i ()Vi = Vi ().
i and
j are equivalent. Then there is an invertible linear map A from Vi () to Vj () which
Suppose
commutes with the action of HF . Choose a non-zero vector vi in Vi () and let vj = Avi . We are
going to show that i and j are equivalent. It is enough to show that, for any f in HF , i (f )vi = 0
if and only if j (f )vj = 0. But i (f )vi = 0 if and only if i ( h)i (f )vi = 0 for all h in HF . Since
i ( h)i (f )vi = i ( h f )vi and h f is in HF the assertion follows.
1 , ,
p are inequivalent. Using this we shall show that the linear
Thus the representations
1 , , Tr
p on HF are linearly independent. The lemma will then be proved. Take h in
forms Tr
HF . Since
i is irreducible and finite-dimensional Tr
i (hf ) = 0 for all f in HF if and only if

i (h) = 0. Suppose we had h1 , , hp in HF so that for at least one i the operator

i (hi ) was not 0
while
p

Tr
i (hi f ) = 0
i=1

for all f in HF . There must then be at least two integers j and k such that
j (hj ) = 0 and

k (hk ) = 0. Since
j and
k are not equivalent we can find an h in HF such that
j (h) = 0 while

k (h) is invertible. Replacing hi by hi h we obtain a relation of the same type in which the number of i
for which
i (hi ) = 0 has been increased. By induction we see that no such relation is possible. Since
HF contains a unit the required independence follows.
For most of these notes the existence of is irrelevant. It is used only toward the end. The
reader who is more interested in automorphic forms than in group representations will probably want
to take the existence of for granted and, for the moment at least, skip this paragraph. To do so will
cause no harm. However he will eventually have to turn back to read the first few pages in order to
review the definition of the Tamagawa measure.
Choose a non-trivial additive character of F . If X is an analytic manifold over F and is a
differential form of highest degree on X we can associate to a measure on X which is denoted by
||F or sometimes simply by . If X = F and = dx is the differential of the identity application
the measure ||F = dx is by definition the Haar measure on F which is self-dual with respect to . In
general if p belongs to X and x1 , , xn are local coordinates near p so that

= a(x1 , , xn ) dx1 dxn

then, if f is a continuous real-valued functions with support in a small neighborhood of p,


f ||F =
X

f (x1 , , xb ) |a(x1 , , xb )| dx1 dxn .

Chapter 1

122

The absolute value |a(x1 , , xn )| is the normalized absolute value in the field F . To prove the existence
of the measure one has to estabish the usual formula for a change of variable in a multiple integral.
For this and other facts about these measures we refer to the notes of Weil [12].
If G is an algebraic group over F then GF is an analytic space. If is a left-invariant form of
highest degree on GF the measure ||F is a Haar measure on GF . It is called the Tamagawa measure.
It depends on and .
If M is the algebra of 2 2 matrices over F the additive group of M is an algebraic group. If a
typical element of M is




x=
then

a b
c d

= da db dc dd

is an invariant form of highest degree and

 || =
 dx is the additive Haar measure which is self-dual
with respect to the character M (x) = F (x) if is the trace of x.
On the multiplicative group G of M we take the form (x) = (detx)2 (x). The associated
Haar measure is
1
|(x)| = |detx|2
F dx = |x|M dx.

An element of G is said to be regular if its eigenvalues are distinct. The centralizer in GF of a

regular element in GF is a Cartan subgroup of GF . Such a Cartan subgroup BF is of course abelian.
There seems to be no canonical choice for the invariant forms on BF . However the centralizer of BF in
MF is an algebra E of degree two over F . It is either isomorphic to the direct sum of F with itself or it
is a separable quadratic of F . BF is the multiplicative group of E . In the first paragraph we introduced
a map from E to F . Once a form E on E which is invariant for the additive group has been chosen
we can set B (x) = (x)1 E (x). b is an invariant form on BF . The associated measure is invariant
under all automorphisms of E over F . We should also recall at this point that two Cartan subgroups
BF and BF are conjugate in GF if and only if the corresponding algebras are isomorphic.
Once E and therefore B has been chosen we can introduce on BF \ GF which is also an
analytic manifold the form B which is the quotient of by B . Then



f (g) (g) =
BF \GF

GF

f (bg) B (b)

B (g).

BF

The centre of the algebra of MF is isomorphic to F and the centre ZF of GF is isomorphic to F . On

F we have the form x1 dx. We take Z to be the corresponding form on ZF . 0B will be the quotient
of B by Z and 0 will be the quotient of by Z . The corresponding integration formulae are


f (b) B (b) =



ZF \GF

0B (b)

ZF



f (G) (g) =
GF


f (zb) Z (z)

Zf \BF

BF

and



f (zg) Z (z)

0 (g).

ZF

If g belongs to GF its eigenvalues 1 and 2 are the roots of the equation

X 2 (g)X + (g) = 0

Chapter 1

123

and

{ (g)}2 4(g)
(1 2 )2
=
1 2
(g)

belongs to F . Set



 (1 2 )2 

 .
(g) = 
1 2 F

 F of regular elements is open in GF and its complement

Since g is regular if and only if (g) = 0 the set G
has measure zero.
There are two more integration formulae that we shall need. Their proof proceeds as for
archimedean fields. Choose a system S of representatives of the conjugacy classes of Cartan subgroups of GF . Then


 1
f (g) (g) =
(b)
f (g 1 bg) B (g) B (b)
2 BF
GF
B
\G
F
F
BF S




 1
0
1
f (g) (g) =
(b)
f (g bg) B (g) 0B (b)
2
ZF \GF
ZF \BF
BF \GF


(7.2.1)

(7.2.2)

BF S

if f is an integrable function on GF or ZF \ GF . Notice that the sum on the right is not necessarily
 F and let
F = BF G
finite. Let B


F , g GF }.
 G = {g 1 bg  b B
B
F
F is the disjoint union
Then G

FG .
B

BF S

There is a simple lemma to be verified.

 G is open.
For any Cartan subgroup BF the set B
F
 F is open.
(ii) The set G
 F of g in GF whose eigenvalues do not belong to F is open.
(iii) The set G

Lemma 7.2 (i)

The second statement is a consequence of the first. If BF corresponds to the separable quadratic
 G is the set of matrices with distinct eigenvalues in E and if BF splits and therefore
extension E then B
F
 G is the set of matrices with distinct eigenvalues in F .
corresponds to the direct sum of F with itself, B
F
Thus the first assertion is a consequence of the following lemma which is a form of Hensels lemma or
of the implicit function theorem.
Lemma 7.2.1 Let E be a separable extension of F . Assume the equation

X p + a1 X p1 + + ap = 0
with coecients in F has a simple root in E. Given > 0 there is a > 0 such that whenever
b1 , , bp are in F and |bi ai |F < for 1 i p the equation
X p + b1 X p1 + + bp = 0

Chapter 1

124

has a root in E for which | |E < .

There is no need to prove this lemma. To prove the third assertion we have to show that the set
of matrices with eigenvalues in F is closed. Suppose gn g and gn has eigenvalues n and n in
F . Then n + n (g) and n n (g). If n and n did not remain in a compact subset of F
then, since their product does, we would have, after passing to a subsequence, |n | 0, |n | or
|n | , |n | 0. In either case n + n could not converge. Thus, again passing to a subsequence,
we have n and n . and are the eigenvalues of g .
 F and G
 F are the same. We now introduce a
If the characteristic of F is not two the sets G
function on GF which plays an important role in the discussion of characters. If BF is a split Cartan
subgroup we set c(BF ) = 1 but if BF is not split and corresponds to the quadratic extension E we set
t+1

c(BF ) = |C|F2

 F belongs to the Cartan

C is a generator of pF and pt+1
is the discriminant of E over F . If g in G
F
subgroup BF we set
(g) = c(BF ) 1/2 (g).
If g is singular we set (g) = . The factor c(BF ) is important only in characteristic two when there
are an infinite number of conjugacy classes of Cartan subgroups.

 F and bounded away from zero on any compact

Lemma 7.3 The function is locally constant on G
subset of GF . It is locally integrable on ZF \ GF and on GF .

It is of course implicit in the satement of the lemma that is constant on cosets of ZF . The two
 F . To prove the remaining assertions we recall
previous lemmas show that is locally constant on G
some facts about orders and modules in separable quadratic extensions of non-archimedean fields.
If E is a separable quadratic extension of F an order R of E is a subring of OE which contains
OF and a basis of E . A module I in E is a finitely generated OF submodule of E which contains a
basis of E . If I is a module the set

{ E | I I}

is an order RI . It is clear that an order is a module and that RR = R. Two modules I and J are said to
be equivalent if there is an in E so that J = I . Then RI = RJ .
Suppose the module I is contained on OE and contains 1. Since I/OF is a torsion-free OF
module the module I has a basis of the form {1, }. Since is integral 2 belongs to I . Therefore I is
an order and RI = I . Since any module is equivalent to a module which contains 1 and lies in OE the
collection of modules I for which RI = R forms, for a given order R, a single equivalence class.
As observed any order has a basis, over OF , of the form {1, }. The absolute values of the
numbers occurring in such bases are bounded below. A basis {1, } is said to be normal if has
the smallest possible absolute value. It is easily seen, by considering the ramified and unramified
extensions separately, that if {1, } is normal

R = OF + OE .
Thus R determines and is determined by ||E . It is easily seen that if E/F is unramified ||E is any
number of the form |CE |n
E with n 0. CE is a generator of pE . We set n = (R). If E/F is ramified,
||E is any number of the form |CE |2n+1
with n 0. We set (R) = n. In the ramified case
E
(R)

[E : F (UE R)] = 2|CF |F

Chapter 1

125

In the unramified case

unless (R) = 0 and then

(R)

[E : F (UE R)] = |CF |F

(1 + |CF |F )

[E : F (UE R)] = 1.

It is clear that R contains R if and only if (R ) (R). Thus (R) + 1 is the number of orders
R
which contain

 . If belongs to OE but not to OF let R() be the order with basis {1, } and let
() = R() .
Lemma 7.3.1 Let
be the conjugate of in E and let
1/2

|( )2 |F

m()

= |CF |F

If pt+1
is the discriminant of E and belongs to OK but not to OF then
E
m() = () +

t+1
.
2

Let {1, } be a normal basis of R(). Then = a + b with a and b in OF . Moreover = c + d

with c and d in OF . Thus = (a + bc) + bd so that a + bc = 0. and bd = 1. Therefore b is a unit and
. We can thus replace by . Suppose first that E/F is unramified so that t + 1 = 0.
| | = | |


We take = CFn where n = R() and is a unit of OE . Since

= ( )CFn
we have only to show that is a unit. is not congruent to an element of OF modulo pE and
therefore {1, } determines a basis of OE /pE . Since the Galois group acts faithfully on OE /pE the
number is not in pE .
If E/F is ramified we may take = CFn CE with n = (). It is well-known that

|CE CE | = |CE |t+1

E
Thus

t+1

2 |1/2 = | |
1/2 = |CF |n |CK | 2 = |CF |n+
|( )
F
K
F
F
E

t+1
2

The lemma follows.

There are two more lemmas to be proved before we return to the proof of Lemma 7.3.
Lemma 7.3.2 Let C be a compact subset of ZF \ GF and let C be the characteristic function of

C and of its inverse image in GF . There is a constant c such that for every b in GF which is
contained in an anisotropic Cartan subgroup

C (g 1 bg) 0 (g) c(b).
ZF \GF

The assertion is trivial unless b is regular. Then the assumption is that its eigenvalues are distinct
and do not lie in F . Any h in GF can be written as

g1

CFp


CFq

g2

Chapter 1

126
p

where g1 and g2 belongs to GL(2, OF ) and p q . The numbers CF and CF are the elementary
divisors of h. Let Tr be the set of all those h for which q p r . Tr is the inverse image of a compact
subset Tr of ZF \ GF . If r is sufficiently large C is contained in Tr . Thus we may replace C and r
the characteristic function of Tr .
If h belongs to GL(2, OF ) then h1 g 1 bgh belongs to Tr if and only if g1 bg belongs to Tr . Thus
the integral is the product of the measure of GL(2, OF ) ZF \ GL(2, OF ) by the number of right cosets
of ZF GL(2, OF ) whose elements g are such that g1 bg belong to Tr . If H is such a coset and BF is
the Cartan subgroup containing b then for any b in BF the coset b H has the same property. Thus the
integral equals



measure GL(2, OF ) ZF \ GL(2, OF )
[BF g GL(2, OF ) : ZF GL(2, OF )].
The sum is over a set of representatives of the cosets in BF \ GF /GL(2, OF ).
Let BF correspond to the separable quadratic extension E . Choose a basis of OE over OF . It
will also be a basis of E over F . By means of this basis we identify GF with the group of invertible
linear transformations of E over F . GL(2, OF ) is the stabilizer of OE . Every in E determines a
linear transformation b : x x of E . The set of all such linear transformations is a Cartan subgroup
conjugate to BF and with no loss of generality we may assume that it is BF . Choose so that b = b .
Every module is of the form gOE with g in GF . Moreover g1 OE and g2 OE are equivalent if
and only if g1 and g2 belong to same double coset in BF \ GF /GL(2, OF ). Thus there is a one-toone correspondence between the collection of double cosets and the collection of orders of E . Let
BF g GL(2, OF ) correspond to the order R. The index

[BF g GL(2, OF ) : ZF GL(2, OF )]

is equal to

[BF : BF ZF g GL(2, OF )g 1 ]

Two elements b1 and b2 in BF belong to the same coset of BF ZF g GL(2, OF )g 1 if and only if there
is a z in ZF and an h in GL(2, OF ) such that

b1 g = b2 zgh
This can happen if and only if

b1 gOE = b2 zgOE .
Let I = gOE and let bi = bi . If we identify ZF and F so that z may be regarded as an element of
F the last relation is equivalent to

1 I = 2 zI
or 11 2 z R UE . Thus

[BF g GL(2, OF ) : ZF GL(2, OF )] = [E : F (R UE )].

p
q
1
Let |detb|F = ||K = |CF |m
bg .
F . Let CF and CF with p q be the elementary divisors of g
1
Certainly p + q = m. The matrix g bg belongs to Tr if and only if q p = m 2p r . If s is the
s 1
bg has integral coefficients, that is if and only if
integral part of rm
2 this is so if and only if CF g

CFs g 1 bg OE OE

Chapter 1

127

or CFs R.
Our integral is therefore equal to


 
measure GL(2, OF ) ZF \ GL(2, OF )
[E : F (R UE )].

()

s R
=F

The sum is over all orders of E which contains CFs . CFs does not lie in F . If it does not lie in OK
the sume is zero. If it lies in OK then CFs belongs to R if and only if (R) (CFs ). In this case the
expression () is bounded by



2 measure GL(2, OF ) ZF \ GL(2, OF )

|CF |k
F .

s )
0k(=F

This in turn is bounded by a constant, which is independent of BF and r , times

s
(=F
)

|CF |F
t+1

We have c(BF ) = |CF |F2 , m(CFs ) = s + m()

1/2

(b)1/2 =

|( )2 |F
1/2
|
|F

rm
2

+ m(), and
m/2

= |CF |F

m()

|CF |F

To prove the lemma we have only to show that

m() +

m t+1
+
+ (CFs )
2
2

is bounded above by a constant which depends only on r . By the previous lemma

(CFs ) = m(CFs )
so that

t+1
2

m t+1
rm m
r
+
+ (CFs )
+
= .
2
2
2
2
2
Suppose the Cartan subalgebra BF corresponds to the algebra E . Once the measure E on E
has been chosen we can form the measure B on BF and the measure B on BF \ GF . Once E and
therefore B and B are chosen
 we let n(BF ) be that constant which makes n(BF )E self-dual with
respect to the character x (x) on E .
m() +

Chapter 1

128

Lemma 7.3.3 If r is a non-negative integer there is a constant d r such that for any Cartan subgroup

BF and any b in BF


BF \GF

r (g 1 bg) B (g) dr n(BF ) (b)1/2 .

F . If BF is anisotropic the left side is equal to

We may again suppose that b belongs to B
1
measure(ZF \ BF )


ZF \GF

r (g 1 bg) 0 (g).

Suppose BF corresponds to the quadratic extension E . If E/F is unramified

measure(Zf \ BF ) =

1
(1 + |CF |)
n(BF )

because n(BF )E assigns the measure 1 to OE . If E/F is ramified n(BF )E assigns the measure
t+1
|CF | 2 to OE and

measure(ZF \ BF ) =

t+1
2
2
|CF | 2 =
c(BF )
n(BF )
n(BF )

In these cases the assertion is therefore a consequence of the previous lemma.

If the inequality of the lemma is true for one Cartan subgroup it is true for all conjugate subgroups.
To complete the proof we have to verify it when BF is the group AF of diagonal matrices. Since we are
now dealing with a fixed Cartan subgroup the choice of Haar measure on BF \ GF is not important.
Moreover GL(2, OF ) Tr GL(2, OF ) = Tr so that, using the Iwasawa decomposition and the associated
decomposition of measures, we may take the itnegral to be


r
F

1 x
0 1



if

b=

0
0
0
0



1 x
0 1


dx

The argument in the integrand is

0
0



1
0

 

1
x
1

Changing the variables in the integral we obtain

1


1 


r
F

0
0



1 x
0 1


dx.

Let || = CF | , || = |CF |m , and |x| = |CF |n . With no loss of generality we may suppose || ||.
If n 0 the elementary divisors of






0
0

1 x
0 1

Chapter 1

129

are CF and CFm so that it is in Tr if and only if m  r . If n < 0 its elementary divisors are CF +n
and CFmn so that it is in Tr if and only if m  2n r . Thus the integral is at most


mr
measure{x  |x| |CF | 2 }
which is, apart from a factor depending on the choice of the Haar measure, |CF |

|CF |
and

mr
2

mr
2

. Since

 1
r
=   2 |CF | 2

  12
 

  = (b) 2
1 

the lemma follows.

We return to Lemma 7.3 and prove first that is bounded away from zero on each compact
subset C . In other words we show that there is a positive constant c such that (h) c on C . There is
a z in ZF such that every matrix in zC has integral entries. Since (zh) = (h) we may as well assume
that every matrix in C itself has integral entires. There is a constant c1 > 0 such that
1/2

|deth|F
on C and a constant c2 such that

c1

| (h)2 4(h)|1/2 c2

on C . and are the trace and determinant of h. Thus

1/2 (h)

c1
c2

on C . (h) is certainly bounded away from 0 on the singular elements and the preceding inequality
shows that it is bounded away from 0 on the regular elements in C which lie in a split Cartan subalgebra.
Suppose h is regular and lies in the anisotropic Cartan subgroup BF . Let BF correspond to the field E
and let h have eigenvalues and in E . Then
1

|( )2 |F 2 c(BF ) = |CF |m() |CF |

t+1
2

= |CF |()

Since () 0 we have (h) c1 .

The function is certainly measurable. It is locally integrable in GF if and only if it is locally
integrable on ZF \ GF . Let C be a compact set in ZF \ GF . We have to show


ZF \GF

C (g) (g) 0(g)

is finite. As usual it will be enough to show that


ZF \GF

r (g) (g) 0(g)

Chapter 1

130

is finite for every non-negative integer r . According to formula (7.2.2) this integral is the sum of

1
2

(a) (a)
ZF \AF

AF \GF

r (g 1 ag) A (g)

0A (a)


1 
(b) (b)
r (g 1 bg) B (g) 0B (b).
2
Z
\B
B
\G

F
F
F
F

and

BF S

It is easy to see that there is a compact set C0 in ZF \ AF such that r (g 1 ag) = 0 for all g unless
the projection of a lies in C0 . Thus the first integral need only be taken over C0 . The inner integral is
at most dr n(AF ) (a)1/2 . Since (a) (a) (a)1/2 = 1 on AF the first integral causes no trouble. We
can also use Lemma 7.3.3 to see that the sum over S  , which is by the way a set of representatives for
the conjugacy classes of anisotropic Cartan subgroups, is less than or equal to


1 
dr n(BF )c(BF )
0B (b).
2
ZF \BF

BF S

If the characteristic is not two this sum is finite and there is no problem.
In general if BF corresponds to the field E and ptFE +1 is the discriminant of E we have c(BF ) =

|CF |

tE +1
2

and


n(BF )

ZF \BF

0B (b) 2|CF |(tE +1)/2

To complete the proof we have to show that

|CF |tE +1

is finite if F has characteristic 2. The sum is over all separable quadratic extensions of F . Let M (t) be
the number of extensions E for which tE = t. Associated to any such E is a quadratic character of F
with conductor pt+1
F . Thus
t+1
2
M (t) [F : (F )2 (1 + pt+1
F )] = 2[UF : UF (1 + pF )]

if t 0. Of course M (1) = 1. Any element of UF is congruent modulo 1 + pt+1

to an element of the
F
form

a0 + a1 CF + + at CFt .

Such a number is a square if ai = 0 for i odd. Thus

M (t) = 0(|CF |
and the series converges.
We can now begin the study of characters.

t+1
2

Chapter 1

131

Proposition 7.4 The character of an absolutely cuspidal representation exists as a locally integrable
F .
F G
function whose absolute value is bounded by a multiple of . It is continuous on G

If the character of exists and is a quasi-character of F then the character of  =

also exists and  (g) = (detg) (g). Thus the proposition has only to be proved for unitary
representations . is then square integrable and we can make use of the following lemma for which,
although it is well-known, we provide a proof.
Lemma 7.4.1 Let f belong to H F and let u be a vector of length one in the space on which the
absolutely cuspidal unitary representation acts. Then





f (h) (g 1 hg)u, u dh

Tr (f ) = d()
ZF \GF


dg

GF

if d() is the formal degree of .

Let Q be the operator


(f ) =

f (h) (h) dh.

GF

Let {vi } be an orthonormal basis of the space on which acts. All but a finite number of the coefficients

Qij = (Qvi , vj )
are zero. We have

 

(g 1 ) Q(g)u, u = Q(g)u, (g)u

The right side equals





Q(g)u, vi vi , (g)u

which in turn equals


i




(g)u, vj Qji vi , (g)u

In both series there are only a finite number of non-zero terms. Thus


ZF \GF



(g 1 ) Q(g)u, u dg =
Qji
i,j


ZF \GF

(g)u, vj




vi , (g)u dg

The integrals on the right exist because the representation is square-integrable. Applying the Schur
orthogonality relations we see that the right side is equal to

1 
1 
1
Qij (vi , vj ) =
Qii =
Tr (f ).
d() i,j
d() i
d()
Since

(g


) Q(g)u, u =


GF

the lemma follows.



f (h) (g 1 ) (h) (g)u, u dh

Chapter 1

132

Observe that the integral of the lemma is an iterated and not a double integral. It is the limit as
r approaches infinity of

 


1
f (h) (g hg)u, u dh dg
Tr

Since

Tr

GF

is compact this integral is absolutely convergent and equals

f (h)

(g


hg)u, u dg

dh.

Tr

GF

To prove the first part of the proposition we show that the sequence of functions


r (h) =

((g 1 hg)u, u) dg
Tr

is dominated locally by a multiple of and converges almost everywhere on GF . We shall set

(h) = d() lim r (h)

r

whenever the limit exists.

When proving the second part of the proposition we shall make use of the following lemma.

 F and let C2 be a compact set in GF . The image in

Lemma 7.4.2 Let C1 be a compact subset of G
ZF \ GF of


{g GF  g 1 C1 g ZF C2 = 0}

is compact.
The set is clearly closed so we have only to show that it is contained in some compact set. We
may suppose that GL(2, OF ) C2 GL(2, OF ) = C2 . Let

g=
with h in GL(2, OF ). Then
if and only if

x
0 1




h

g 1 C1 g ZF C2 = 0

x
0 1

1

C1

x
0 1


ZF C2 = 0.

We have to show that this condition forces to lie in a compact subset of F and x to lie in a compact
subset of F . Since

det(g 1 cg) = detc

we may replace ZF C2 by the compact set


C3 = {h ZF C2  deth detC1 }.

Chapter 1

133

Let

a b
c d

be a typical element of C1 . The entry c is never 0 on C1 and therefore its absolute value is bounded
below,

x
0 1

1

a b
c d



x
0 1

a xc
y
c
cx + d

The number y is of no interest. The matrix on the right cannot lie in C3 unless |cx + d| is bounded
above by some number depending on C3 . Since |d| is bounded above and |c| is bounded below x is
forced to lie in some compact set of F . If C4 is the compact set

1 x
{
0 1




1 x 
h
x , h C1 }
0 1

we have finally to show that if

1
0

0
1


C4

0
0 1


C3 = 0

then is forced to lie in a compact subset of F . We now let

a b
c d

be a typical element of C4 . On C4 both |b| and |c| are bounded blow. Since

0
0 1



a b
c d



1
0

0
1

=

a b/
c d

and all matrix entries are bounded above in absolute value on C3 the absolute value || must indeed
be bounded above and below.
If acts on V then for any u in V the support of the function ((g)u, u) has been shown, in the
second paragraph, to be compact modulo ZF . Let C be its compact image in ZF \ GF . Let C1 be a
compact subset of GF . By the previous lemma the set of g in GF such that



(g 1 hg)u, u = 0
for some h in C1 has an image in ZF \ GF which is contained in a compact set C2 . Therefore the integral


(g 1 hg)u, u dg =

ZF \GF



(g 1 hg)u, u dg
C2

is convergent for h in C1 . Moreover if r is large enough Tr contains C2 and

r (h) =


(g 1 hg)u, u dg.

Zf \GF

F and its limit d1 () (h)

Therefore the sequence {r } converges uniformly on any compact subset of G
 F . We may state the following proposition.
is continuous on G

Chapter 1

134

 F then
Proposition 7.5 If h belongs to G



(g 1 hg)u, u dg

ZF \GF

exists and is equal to d1 () (h).

Since



| (g)u, u | C (g)

it follows from Lemma 7.3.2 that, for some constant c,

|r (h)| c(h)
 F is A
 F . The set G
F G
G which is open. To complete the proof of Proposition 7.4 we show that
on G
F
G
 with a compact subset of GF the sequence {r } is dominated by a multiple
on the intersection of A
F
G .
of and that it converges uniformly in a compact subset of A
F
F may be written in the form
Let C3 be a compact subset of GF . Any h in A
h = h1
1

1 x
0 1



0
0



1 x
0 1


h1

G the absolute values of and are bounded

where h1 belongs to GL(2, OF ) and = . In C3 A
F
G
 the absolute value of 1 is also bounded above and below
above and below. If C3 is contained in A
F

on C3 . Since











 
1 x
0 1

0
0

1 x
0 1

0
0

1
0

1
x
1

the absolute value of x will be bounded above.

Since GL(2, OF ) Tr GL(2, OF ) = Tr the integral which defines r (h) is equal to


(g 1 h g)u, u dg

Tr

if


h =

1 x
0 1



0
0



1 x
0 1

and we may as well assume that h itself is of this form. We are going to show that there is a constant c
such that


1
|r (h)| c1 

for all r and all such h and that the sequence {r } converges uniformly if x remains in a compact subset

of F and , and 1
remain in a compact subset of F . Then the proof of the proposition will be
complete.
The stabilizer of u is some open subgroup U of GL(2, OF ). Let h1 , , hp be a set of coset
representatives for GL(2, OF )/U and let ui = (hi )u. apart from an unimportant factor coming from
the Haar measure r (h) is given by
p

ir (h)
i=1

Chapter 1

135

with

ir (h)









x1
x1
=

h
ui , ui dx1 d .
0
1
0
1

The integral is taken over the set of all those and x1 for which

x1
0
1

belongs to Tr . Since

x1
0
1

1






 1


x1
0
1 1
( x + x1 )
h
=
0
1
0
0
1

we can change variables in the integral to obtain



1 1











0
1 1 1 x + x1
ui , ui dx1 d .

0
1
0

(7.4.3)

The integration is now taken over all those x1 and for which




1
1
x1
0
1

(7.4.4)

is in Tr .

| = |CF |t , || = |CF |m , and |x| = |CF |n . Let CFp and CFq be the elementary divisors
Let |1
of the matrix (7.4.4). We now list the possibilities for p and q together with the condition that the matrix
belong to Tr , that is that q p be at most r .
(a) m 0, t + m + n 0, p = 0, q = m : 0 m r
(b) m 0, t + m + n 0, p = t + m + n, q = n t : r m + 2n 2t
(c) m 0, t + m + n m, p = t + m + n, q = n t : r m + 2n 2t
(d) m 0, t + m + n m, p = m, q = 0 : r m 0.
These conditions amount to the demand r m r and that 2n 2t r m. On the other hand
we know that there is an integer s such that

|x||=F |j

1 x1
0 1


ui dx = 0

for 1 i p if j s.
Thus if || = |CF |m the integral









x+x1 
0
1 1 1 1

ui , ui dx1
0
0
1
taken over all x1 for which


 

1
x1
0
1

is in Tr is zero if 2t r m 2s. Therefore in (7.4.3) we need only take the integral over those and x
m+r
for which || = |CF |m with 0 mr 2(t s) and |x| |CF |t 2 . We should also have m r but

Chapter 1

136

since we are about to replace the integrand by its absolute value that does not matter. For each such
the integration with respect to x gives a result which is bounded in absolute value by a constant times
m+r
|CF |t 2 . Integrating with respect to we obtain a result which is bounded in absolute value by a
constant times

2(ts)1

|CF |

|CF | 2 |Cf |s

k=0

|CF | 2

k=0

The right side depends on neither r nor t.

|ir (h) is
The value of |1










0
1 CFr 1 1
x + x1

ui , ui dx1 d .
0
0
1
The integration is taken over those and x1 for which || = |CF |m with 0 m < 2(t s) and
m
|x| |CF |t 2 . Of course |1 | = |CF |t . Since we are now interested in a set of and on which
t takes only a finite number of values we may as well assume it is constant. Then the integral is taken
over a fixed compact subset of F F . The integrand converges uniformly on this set uniformly in
the , and x under consideration as r approaches infinity.
We have still to prove the existence of the character of a representation which is not absolutely
cuspidal. Most of them are taken care of by the next proposition.
Proposition 7.6 Let 1 and 2 be a pair of quasi-characters of F . Let 1 ,2 be the function which

F G
 F , undened on the singular elements, and equal to
is 0 on G



 1/2

{1 () 2 () + 2 () 1 ()} 
( )2 
G with eigenvalues and . Then , is continuous on G
 F and is
at an element of g of A
1
2
F
dominated in absolute value by some multiple of . Moreover if = (1 , 2 )

Tr (f ) =

1 ,2 (g) f (g) dg
GF

for all f in HF .
Only the last assertion requires verification. Since the absolute value of 1 ,2 is bounded by a
multiple of the function 1 ,2 is locally integrable. Suppose f belongs to HF . When applied to the
function 1 ,2 f the relation (7.2.1) shows that


1 ,2 (g) f (g) dg

(7.6.1)

GF

is equal to

1
2

(a)
AF

AF \GF

1 ,2 (g

ag) f (g

ag) dg

da.

Chapter 1

137

Since 1 ,2 is a class function this may be written as

1
2






 ( )2 1/2

0
1

g) dg da
{1 () 2 () + 2 () 1 ()} 
f (g
0

AF
AF \GF

if

a=
Since a is conjugate to

we have

AF \GF

0
0

0
0


.

 


 

0
0
1
1
f g
f g
g dg =
g dg.
0
0
AF \GF

Thus (7.6.1) is equal to


AF


 


 

 ( )2 1/2

0
1

g dg da.
1 () 2 () 
f g
0

AF \GF

(7.6.2)

As long as the measure on AF \ GF is the quotient of the measure on GF by that on AF the

choice of Haar measure on AF and GF is not relevant. Thus we may write (7.6.2) as


AF








 

 ( )2 1/2
0
1 1


f k n
nk dk dn da.
1 () 2 () 
0


The inner integral is taken over GL(2, OF ) NF . If

n=
then

0
0

1 x
0 1

0
0

n=



1
0

 
1 x
1

Changing variables in the last integral we obtain


AF

 1/2
1 () 2 ()   )



f

0
0


nk


dk dn da.

(7.6.3)

To evaluate Tr (f ) we observe that if belongs to B(1 , 2 ) then, if k1 is in GL(2, OF )


(f ) (k1) =

(k1 g) f (g) dg.

GF

Replacing g by k11 g and writing the integral out in terms of the Haar measure we have chosen we
obtain







(k1)
GL(2,OF )

f (k11

0
0

nk2 )1 () 2 () | |1/2 da dn

dk2 .

Chapter 1

138

The inner integral is taken over AF NF . We have of course used the relation

If

K(k1, k2 ) =

0
0


nk2

= 1 ()2 () | |1/2 (k2 ).





 1/2
0
1
nk2 1 () 2 ()   da dn
f k1
0

then

(f ) (k1) =

K(k1, k2 ) (k2 ) dk2.

GL(2,OF )

B(1 , 2 ) may be regarded as a space of functions on GL(2, OF ). Then (f ) is the integral

operator with kernel K(k1 , k2 ). It is easily seen that this operator, when allowed to act on the space
of all GL(2, OF )-finite functions on GL(2, OF ), has range in B(1 , 2 ). Thus the trace of (f ) is the
same as the trace of the integral operator which is of course

K(k, k) dk.
GL(2,OF )

When written out in full this integral becomes (7.6.3).

Theorem 7.7 Let be an irreducible admissible representation of HF . There is a function which

 F by a multiple of such that

is continuous on GF and locally bounded in absolute value of G

Tr (f ) =
(g) f (g) dg
GF

for all f in HF .
The theorem has only to be verified for the one-dimensional and the special representations. If
is a one-dimensional representation associated to the quasi-character we may take (g) = (detg).
is locally bounded and therefore, by Lemma 7.3, locally bounded by a multiple of .
Suppose 1 , 2 and 3 are three admissible representations of F on space V1 , V2 , and V3
respectively. Suppose also that there is an exact sequence

0 V1 V2 V3 0
of HF -modules. If f is in HF all the operators 1 (f ), 2 (f ) and 3 (f ) are of finite rank so that

Tr 2 (f ) = Tr 1 (f ) + Tr 3 (f ).
Thus if 1 and 2 exist so does 3 . Applying this observation to 3 = (1 , 2 ), 2 = (1 , 2 ),
and 1 = (1 , 2 ) we obtain the theorem.
If F is taken to be the real or complex field Theorem 7.7 is a special case of a general and difficult
theorem of Harish-Chandra. The special case is proved rather easily however. In fact Proposition 7.6
is clearly valid for archimedean fields and Theorem 7.7 is clearly valid for archimedean fields if is
finite-dimensional. There remains only the special representations and these are taken care of as before.

Chapter 1

139

8. Odds and ends In this paragraph various facts which will be used in the discussion of the constant
term in the Fourier expansion of an automorphic form are collected together. If H is a locally compact
abelian group a continuous complex-valued function f on H will be called H -finite or simply finite if
the space spanned by the translates of f is finite dimensional.
Let H be a group of the form
H = H 0 Z m Rn

where H0 is compact. We regard Zm Rn as a subgroup of Rm+n . The projection

i : h = (h0 , x1 , , xm+n ) xi
may be regarded as a function on H with values in R. If p1 , , pm+n is a sequence of non-negative
integers and is a quasi-character we may introduce the function
m+n

ipi

i=1

on H .

!m+n

ipi is
continuous and nite. These functions form a basis of the space of continuous nite functions on
H.

Lemma 8.1 For any sequence p1 , , pm+n and any quasi-character the function

i=1

If is a fixed quasi-character of H and p is a non-negative integer let V (, p) be the space

!m+n p
spanned by the functions i=1 i i with 0 pi p. Since it is finite-dimensional and invariant
under translations the first assertion of the lemma is clear.
To show that these functions are linearly independent we shall use the following simple lemma.
Lemma 8.1.1 Suppose E1 , , Er are r sets and F1 , Fr are linearly independent sets of complexvalued functions on E1 , , Er respectively. Let F be the set of functions

(x1 , , xr ) f1 (x1 ) f2 (x2 ) fr (xr )

on E1 Er . Here fi belongs to Fi . Then F is also linearly independent.
Any relation

a(f1 , , fr ) f1 (x1 ) fr (xr ) 0

f1 ,,fr

leads to


fr

a(f1 , , fr ) f1 (x1 ) fr1 (xr1 )

f1 ,,fr1

fr (xr ) 0

As Fr is linearly independent this implies that

a(f1 , , fr ) f1 (x1 ) fr1 (xr1 ) 0

f1 ,fr1

and the lemma follows by induction.

!m+n p
To show that the functions i=1 i i span the space of continuous finite functions we use
another simple lemma.

Chapter 1

140

Lemma 8.1.2. Let H 1 and H2 be two locally compact abelian groups and let H = H1 H2 . Then

every continuous nite function f on H is a nite linear combination of the form

f (x, y) =

i i (x) i (y)

where the i and i are continuous functions on H1 and H2 respectively.

Let V be any finite dimensional space of continuous functions on H . We associate to any point
in H the linear functional f f () on V . Since no function but zero is annihilated by all these
functionals we can choose 1 , , p so that the corresponding functionals form a basis of the dual of
V . Then we can choose a basis f1 , , fp of V so that fi (j ) = ij .
Now suppose V is invariant under translations. It could for example be the space spanned by
the translates of a single finite continuous function. The space V1 of functions on H1 defined by
(x) = f (x, 0) with f in V is finite dimensional and translation invariant. Therefore the functions
in it are finite and of course continuous. We define V2 in a similar manner. If f is in V the function
h f (g + h) is, for any g in H , also in V . Thus

f (g + h) =

i (g) fi (h).

Since

i (g) = f (g + i )
the function i belongs to V . If i (x) = i (x, 0) and i (y) = fi (0, y) then

f (x, y) =

i (x) i (y)

as required.
These two lemmas show that we need prove the final assertions of Lemma 8.1 only for H
compact, H = Z, or H = R.
Suppose H is compact. If we have a non-trivial relation
r


ai i (h) 0

i=1

we may replace h be g + h to obtain

r


ai i (g) i (h) 0.

i=1

If such a relation holds we must have r 2 and at least two coefficients say a1 and a2 must be different
from zero. Choose g so that 1 (g) = 2 (g). Multiplying the first relation by 1 (g) and subtracting the
second relation from the result we obtain a relation
r

i=2

bi i (h) 0.

Chapter 1

141

Since b2 = {1 (g) 2 (g)}a2 the new relation is non-trivial. The independence of the quasi-characters
can therefore be proved by induction on r .
To prove that when H is compact the quasi-characters span the space of finite continuous
functions we have just to show that any finite-dimensional space V of continuous functions which is
translation invariant is spanned by the quasi-characters it contains. Choose a basis {fi } of V as before
and let


(g) fi =

ij (g) fj .

We saw that the functions ij (g) are continuous. Thus the action of H on V by right translations is
continuous and V is the direct sum of one-dimensional translation invariant spaces. Each such space
is easily seen to contain a character.
When applied to a locally compact abelian group the argument of the previous paragraph leads
to weaker conclusions. We can then find subspaces V1 , , Vr of V and quasi-characters 1 , , r of
H such that

V =

r


Vi

i=1

and, for every h in H ,

{(h) i (h)}dim Vi

annihilates Vi . Now we want to take H equal to Z or R. Then H is not the union of a finite number
of proper closed subgroups. Suppose 1 , , 2 are quasi-characters of H and for every h in H the
operator
s

{(h) i (h)}

(8.1.3)

i=1

on V is singular. Then for every h in H there is an i and a j such that i (h) = j (h). If

Hij = {h | i (h) = j (h)}

then Hij is a closed subgroup of H . Since the union of these closed subgroups is H there must be an
i and a j such that Hij = H and i = j . If the operator (8.1.3) were zero the same argument would
show that for every j there is an i such that i = j .
If is a quasi-character of H , now taken to be Z or R, we let V (, p) be the space spanned by the
functions i , with 0 i p. is the coordinate function on H . It is clear that V (, p) is annihilated
by {(h) (h)}p+1 for all h in H . Suppose , 1 , , 2 are distinct and

V = V (, p)

s


V (i , pi )

i=1

is not zero. Decomposing V as above we see that 1 , , r must all be equal to on one hand and
on the other that every i is a j . This is a contradiction. Thus if there is any non-trivial relation at all
between the functions i where is any quasi-character and i is a non-negative integer there is one
of the form
p


Since the polynomial

p
i=0

ai i = 0.

i=0

ai i would then have an infinite number of zeros this is impossible.

Chapter 1

142

To prove the functions i span the space of finite continuous functions we have only to show
that if is a given quasi-character and V is a finite-dimensional space of continuous functions which is
invariant under translations and annihilated by {(h) (h)}dim V for all h in H then every function
in V is the product of and a polynomial. Since we can always multiply the functions in V by 1
we may as well suppose that is trivial. We have only to observe that any function f annihilated by
the operator {(h) 1}n for all h in H is a polynomial of degree at most n. This is clear if n = 1 so
n1
by induction we can assume that (h)f f is a polynomial i=0 ai (h) i . We can certainly find a
polynomial f  of degree n such that

(1)f  f  =

n1


ai (1) i

i=0


and we may as well replace f by f f . The new f satisfies (1)f = f . It is therefore bounded.
Moreover (h)f f is a bounded polynomial function and therefore a constant c(h). c(h) is a bounded
function of h and satisfies c(h1 + h2 ) = c(h1 ) + c(h2 ). It is therefore zero and the new f is a constant.
Lemma 8.1 is now completely proved. Although it is trivial it is important to the notes and we
thought it best to provide a proof. We might as well prove Lemma 2.16.4 at the same time. Let B be
the space of all functions f on Z such that for some n0 depending on f we have f (n) = 0 for n < n0 .
Let A0 be the space of functions on Z which vanish outside a finite set. Z acts on B and on A0 by
= B/A0 . In particular let D = (1). We have merely
right translations and therefore it also acts on B
is finite
to show that if P is a polynomial with leading coefficient 1 then the null space of P (D) in B
dimensional. If
r

(X i )pi

P (X) =

i=1

the null space of P (D) is the direct sum of the null spaces of the operators (D i )pi . The null space
of the functions in B which are zero to the left of 0 and of the form
of (D )p is the image in B

n n Q(n)
to the right of 0. Q is a polynomial of degree at most p.
Lemma 8.1 is certainly applicable to the direct product of a finite number of copies of the
multiplicative group of a local field F . If H = (F )n any finite continuous function on H is a linear
combination of functions of the form
n

f (x1 , , xn ) =

{i (xi ) (log |xi |F )ni }.

i=1

Let B = BF be the space of continuous functions f on GF which satisfy the following three
conditions.
(i) f is finite on the right under the standard maximal compact subgroup K of GF .
(ii) f is invariant on the left under NF .
(iii) f is AF -finite on the left.
BF is invariant under left translations by elements of AF . If f is in BF let V be the finite-dimensional
space generated by these left translates. Choose g1 , , gp in GF so that the linear functions (gi )
are a basis of the dual of V and let f1 , , fp be the dual basis. If a is AF we may write

f (a, g) =

p

i=1

i (a) fi (g).

Chapter 1

143

Then

i (a) = f (agi )
so that

i (ab) =

p


j (a) fj (bgi ).

j=1

Thus the functions i are continuous and finite. We may write them in the form

i (a) =

cim,n,, (1 ) (2 ) (log |1 |)m (log |2 |)n

if

a=

1
0

0
2


.

The sum is over all quasi-characters and of F and all non-negative integers m and n. Of course
only a finite number of the coefficients cim,n,, are different from zero.
1/2

1/2

We may replace by F and by F

fm,n,, =

in the sum. Thus if

p


cim,n,, fi

i=1

we have

 1/2
 1  
(1 ) (2 ) (log |1 |)m (log |2 |)n fm,n,, (g).
f (ag) =  
2

(8.2)

Let M be a non-negative integer and S a finite set of pairs of quasi-characters of F . B(S, M ) will be
the sset of f in for which the sum in (8.2) need only be taken over those m, n, , for which m + n M
and (, ) belong to S . Observe that the functions fm,n,, are determined by f . B is the union of the
spaces B(S, M ); if S consists of the single pair (1 , 2 ) we write B(1 , 2 , M ) instead of B(S, M ). If
f is in (1 , 2 , M )

 1/2

 1 
f (ag) =   1 (1 ) 2 (2 )
(log |a1 |)m (log |a2 |)n fm,n (g).
2
The space B(1 , 2 , 0) is just B(1 , 2 ).
The functions fm,n,, are uniquely determined and by their construction belong to the space
spanned by left translates of f by elements of AF . Thus if f belongs to B(S, M ) so do the functions
fm,n,, . We want to verify that f0,0,, belongs to B(, , M ). If

b=

1
0

0
2

and we replace a by ab in the relation (8.2) we find that

 1/2
 1  
 
(1 ) (2 ) (log |1 |)m (log |2 |)n ) fm,n,,
 2 

Chapter 1

144

is equal to



 1 1 1/2 


(1 1 ) (2 2 )(log |1 | + log |2 |)m (log |2 | + log |2 |)n fm,n,, (g).
 2 2 
Fix b and g are regard this equality as an identity in the variable a. Because of Lemma 8.1 we can
compare the coefficients of the basic finite functions. The coefficient of (1 )(2 ) on one side is
f0,0,, (bg). On the other it is

 1/2
 1 
 
 2 

(1 ) (2 ) (log |1 |)m (log |2 |)n fm,n,, (g).

m+nM

The resulting identity is the one we wanted to verify.

Taking a = 1 in (8.2) we see that

f (g) =

f0,0,, (g).

(,)S

Therefore

B(S, M ) =

B(, , M ).

(,)S

The sum is direct.

It is fortunately possible to give a simple characterization of B.
Proposition 8.3 Let be a continuous function on GF . Assume is K-nite on the right and
invariant under NF on the left. Then belongs to B if and only if the space

{(f )  f HF }

is nite dimensional for every elementary idempotent in HF .

We have first to show that if belongs to B


{(f )  f HF }
is finite dimensional. belongs to some B(S, M ). Both B and B(S, M ) are invariant under right
translations by elements of HF . Thus we have only to show that the range of () as an operator on
B(S, M ) is finite dimensional. This is tantamount to showing that any irreducible representation of K
occurs with finite multiplicity in the representation of B(S, M ).
Let be such a representation and let V be the space of continuous functions on K which
transform according to under right translations. V is finite dimensional. If f is in B(S, M ) we may
write
 

 1 1/2 
f (ag)  
(1 ) (2 ) (log |1 |)m ) (log |2 |)n fm,n,, (g)
2

The restriction of fm,n,, to K lies in V . Call this restriction fm,n,, . Moreover f is determined by its
restriction to AF K . Thus


fm,n,,

(,)S
m+nM

Chapter 1

145

is an injection of the space of functions under consideration into the direct sum of a finite number of
copies of V .
The converse is more complicated. Suppose is K -finite on the right, invariant under NF on
the left, and the space


{(f )  f HF }

is finite dimensional for every elementary idempotent . Choose so that () = . There is actually
a function f in HF such that (f ) = . If F is non-archimedean is itself a function so this is clear.
If F is archimedean we observe that if f1 is an approximation to the -function then (f1 ) is close to
. Then if f1 = f1 the function f1 is in HF and (f1 ) is also close to . The existence of
f then follows from the fact that (HF ) is finite dimensional. This argument was used before in
Paragraph 5.
Take F to be archimedean. Then must be an infinitely differentiable function on GF . Let Z be
the centre of the universal enveloping algebra of the Lie algebra of GF . If Z is in Z then

(Z) = (Z) (f ) = (Z f )
and Z f is still in HF . Thus is also Z-finite. For the rest of the proof in the archimedean case we
refer to Chapter I of [11].
Now take F non-archimedean. We may replace by any elementary idempotent  for which

= . In particular if we choose n to be a sufficiently large positive integer and let K be the elements
of K which are congruent to the identity modulo pn we may take

where the sum is over all elementary idempotents corresponding to irreducible representations of K
whose kernel contains K  . Notice that n is at least 1. Then HF is the space of functions on GF which
are constant on double cosets of K  .
Let V be the space spanned by the functions (k) with k in K . It is finite dimensional and all
the functions in V satisfy the same conditions as . Let i , 1 i p, be a basis of V . If k belongs to
K we may write

(gk) =

p


i (k) k (g)

i=1

and is determined by the functions i and the restrictions of the functions i to AF . To show that
is AF -finite on the left we have merely to show that the restriction of each i to AF is finite. We may
as well just show that the restriction of to AF is finite.
Suppose f is in HF and (f ) = . If a is in ZF then

(a) = (a1 ) = (a1 f )

if a1 is the -function at a1 . Since a1 f is still in HF the function is certainly ZF -finite and
to AF . If and are units and 1(mod pn ) then
so is its restriction

0
0


= .

Chapter 1

146

Thus the translate of

by the elements of Af K span a finite dimensional space and if C is a generator
by the group
of p we have only to show that the translates of



Cp 0
0 1

H=




p

Z


span a finite-dimensional space. Suppose the span W of




 

Cp 0


p 0
0 1

is finite dimensional. Then

C 1 0
0
1



maps W into itself and annihilates no vector but zero so that it has an inverse on W which must be

C 0
0 1


.

Thus W is invariant under H and

is finite.
To show that W is finite dimensional we show that if

a=

C p 0
0
1

with p > 0 there is a function fa in HF such that

(a) = 
if  = (fa ). There is an f in HF such that


(g) =

(gh) f (h) dh
GF

for all g in GF . Thus if b belongs to AF

1

(a) (b)
= (a


b) =

(ba1 h) f (h) dh.

GF

If f1 (h) = f (ah) the integral is equal to


(bh) f1(h) dh.
GF

If f1 were in HF we would be done. Unfortunately this may not be so. However f1 (hk) =
f1 (h) if k belongs to K  . If




k=

Chapter 1

147

then

f1 (kh) = f

C p

Cp


ah .

Thus f1 (kh) = f1 (h) if 1 (mod pn ), 0 (mod pn ), and 0 (mod pn+p ). Set


f2 (h) =

pn /pn+p

1 x
0 1

f1

 
h dx

where the Haar measure is so chosen that the measure of the underlying space pn /pn+p is 1. Since
(bnh) = (bh) for all n in NF


(a) (b)
=

(bh) f2(h) dh.

GF

We show that f2 lies in HF .

Certainly f2 (hk) = f2 (h) if k is in K  . Moreover, because of its construction, f2 (kh) = f2 (h) if

k=

with 1 (mod pn ) and 0 (mod pn ). Since every element of K  is a product

1 0
1



where both terms lie in K  we have only to show that f2 is invariant under the first factor. If

k=
with 0 (mod pn ) and

k1 (x) =

then

k1 (x)
Moreover if x is in OF

1
0
0 1 + x

1 x
0 1



k=

1
0

1
1+x

1+x

0
1

x
1+x



f1 k1 (x)g = f1 (g).

Thus f2 (kg) which is given by


pn /pn+p

is equal to

1 0
1

f1


pn /pn+p

f1

1
0

1 x
0 1

x
1+x


kh

dx

 
h dx.

Chapter 1

148

x
Since the map x 1+x
is a one-to-one map of the finite set pn /pn+p onto itself it is measure preserving
and the above integral is equal to f2 (h).
are
Analyzing the above proof one sees that in the non-archimedean case the left translates of
contained in the space X obtained by restricting the functions in (HF ) to AF . Thus if Y is the
space of the functions on K/K  the left translates of by elements of AF are contained in the space of
functions on NF \ GF of the form


 (ak) =

i (k) i(a)

with i in Y and i in X .
In the archimedean case Y is the space of continuous functions on K for which = = .
It is again finite dimensional. X is defined in the same way. In this case there are a finite number of
invariant differential operators D1 , , Dr on AF such that the left translates of by elements of AF
are contained in the space of functions NF \ GF of the form

 (ak) =

i (k) i(a)

r

with i in Y and i in j=1 Dj X .

There is a corollary of these observations. Let F1 , , Fn be a finite collection of local fields. Let
Gi = !
GFi , Ni = NFi ,!
Ai = AFi , and let Ki be the standard maximal compact subgroup of Gi . We set
n
n
G = i=1 Gi , N = i=1 Ni and so on. If Hi = HFi we let H = i Hi . H may be regarded as an
algebra of measures on G.
Corollary 8.4 Let be a continuous function on N \ G which is K-nite on the right. If for every
elementary idempotent in H the space

{(f ) | f H}
is nite dimensional is A-nite on the left.
If satisfies the conditions of the lemma so does any left translate by an element of A. Thus we
need only show that is Ai -finite on the left for each i. If g is in G we write g = (gi , gi ) where gi is in
 i = ! Gj . We may suppose that there is a  of the form  = i  where  is an
Gi and gi is in G
i
j=i
! i
elementary idempotent of Hi such that ( ) = . By means of the imbedding f f j=i j the
algebra Hi becomes a subalgebra of H. The left translates of by Ai all lie in the space of functions of
the form


(ai ki , g1 ) =

j (ki ) (ai , g1 )

where j lie in a certain finite dimensional space determined by i and the j lie in the space obtained
 i or, in the archimedean case, the space obtained from
by restricting the functions in (i Hi ) to Ai G
this space by applying certain invariant differential operators. i is a certain elementary idempotent
which may be different from i .
With the odds taken care of we come to the ends.

Chapter 1

149

Proposition 8.5 Let B(, , ) =

M 0 B(, , M ). If an irreducible admissible representation of

HF is a constituent of the representation (, , ) on B(, , ) it is a constituent of (, ).

There are two invariant subspaces V1 and V2 of B(, , ) such that V1 contains V2 and is
equivalent to the representation on HF on V1 /V2 . Choose M so that V1 B(, , M ) is not contained
in V2 . Since is irreducible



V1 = V2 + V1 B(, , M )

and



V1 /V2 = {V2 + V1 B(, , M ) }/V2

is isomorphic as an HF module to

V1 B(, , M )/V2 B(, , M )

so that we may as well suppose that V1 is contained in B(, , M ).
Given we choose M as small as possible. If M = 0 there is nothing to prove so assume M is
positive. If is in B(, , M ) we can express

as

 1/2
 1 
  (1 ) (2 )
 2 

We can express

0
2

1
0

 
g

(log |1 |)m (log |2 |)n m,n (g)

m+nM

0
2

1
0



1
0

0
2

 
g

in two ways because the second factor can be absorbed into the first or the third. One way we obtain

 1/2
 1 
  (1 ) (2 )
 2 

(log |1 |) (log |2 |) m,n

m

m+nM

1 0
0 2

 
g

and the other way we obtain



 1 1 1/2


 2 2  (1 1 ) (2 2 )

(log |1 | + log |1 |)m (log |2 | + log |2 |)n m,n (g).

m+nM

On comparing coefficients we see that if m + n = M

m,n

1
0

0
2

   1/2
 1 
g =   (1 ) (2 ) m,n (g)
2

so that m,n is in B(, ). Consider the map

&

m,n

m+n=M

of V1 into

&

B(, ).

m+n=M

Its kernel is V1 B(, , M 1). Since V2 + (V1 B(, , M 1)) cannot be V1 the image of V2 is not
the same as the image
'of V1 . Since the map clearly commutes with the action of HF the representation
is a constituent of m+n=M (, ).
Proposition 8.5 is now a consequence of the following simple lemma.

Chapter 1

150

Lemma 8.6 Suppose is an irreducible representation of an algebra H. Suppose is a representation

of H of which is a constituent and that is the direct sum of the representations , .
Then is a constituent of at least one of the .

Let act on X and let act on X the direct sum of X . Supose that Y1 and Y2 are invariant
subspaces of X and that the representation on the quotient Y1 /Y2 is equivalent to . There is a finite
subset 0 of such that


Y1 (

X )

is not contained in Y2 . We may as well replace Y1 by Y1 ( 0 X ) and Y2 by Y2 ( 0 X ) and

suppose that is finite. If = {1 , , p } we have only to show that is a constituent of 1 or of
2 p for we can then use induction. Thus we may as well take p = 2. If the projections of Y1
and Y2 on X1 are not equal we can replace Y1 and Y2 by these projections to see that is a constituent
of 1 . If they are equal Y1 = Y2 + (Y1 X2 ) and we can replace Y1 and Y2 by Y1 X2 and Y2 X2
to see that is a constituent of 2 .

Chapter 1

151
References
for Chapter I

The Weil representation is constructed in:

1.

erateurs unitaires, Acta Math., t. 111, 1964.

Weil, A., Sur certains groupes d'op
One of the first to study representations of groups over non-archimedean local fields was F. Mautner

in:
2.

Mautner, F., Spherical functions over p-adic elds, I. Amer. Jour. Math., vol LXXX, 1958.

Absolutely cuspidal representations were first constructed by Gelfand and Graev. References to
their work and that of Kirillov will be found in
3.

Gelfand, I.M., M.I. Graev, and I.I. PyatetskiiShapiro, Representation Theory and Automorphic
Functions, W.B. Saunders Co., 1966.

These representations were constructed in terms of the Weil representation by Shalika and by
Tanaka.
4.

Shalika, J. Representations of the two-by-two unimodular group over local elds, Notes, Institute
for Advanced Study.

5.

Tanaka, S., On irreducible unitary representations of some special linear groups of the second
order, Osaka Jour. Math., 1966.

To classify the representations over an archimedean field we have used a theorem of Harish
Chandra which may be found in
6.

HarishChandra, Representations of semisimple Lie groups, II, T.A.M.S., vol 76, 1954.
Our discussion of characters owes much to:

7.

Sally, P.J. and J.A. Shalika, Characters of the discrete series of representations of SL(2) over a
local eld, P.N.A.S., 1968.
Three standard references to the theory of L-functions are:

8.

Lang, S., Algebraic numbers, AddisonWesley, 1964.

9.

Tate, J., Fourier analysis in number elds and Heckes Zeta-functions in algebraic number
theory, Thompson Book Co., 1967.

10. Weil, A., Basic number theory, SpringerVerlag, 1967.

In Paragraph 8 we have used a result from:
11. HarishChandra, Automorphic forms on semisimple Lie groups, SpringerVerlag, 1968.
Tamagawa measures are discussed in:
12. Weil, A., Ad`
eles and algebraic groups, Institute for Advanced Study, 1961.

Chapter II: Global Theory

9. The global Hecke algebra. Let F be a global field, that is, an algebraic number field of finite degree
over the rationals or a function field in one variable over a finite field. A will be the ad`ele ring of F .
Before studying the representations of GL(2, A) or, more precisely, the representations of a suitable
group algebra of GL(2, A) we introduce some simple algebraic notions.
a finite number
Let {V | } be a family of complex vector spaces. Suppose that for
! all but!
of we are given a non-zero vector e in V . Let V 0 be the set of all x = x in V such that
x = e for al but a finite number of . Let C be the free vector space with complex coefficients over
V 0 and let D be the subspace generated by vectors of the form

(aY + bZ )



x a y
=



x b z
=


x .
=

a and b belong to C and is any element of . The quotient of C by D is called the tensor product of
the V with respect to the family e and is written
V = e V
or simply!V . It has an obvious universal property which characterizes it up to isomorphism. The
image of x in V is written x .
If  is a subset of with finite complement we may form the ordinary tensor product

 V
and we may form

 V

with respect to the family e . Then V is canonically isomorphic to


If S is a finite subset of let

 V

VS = S V

If S is so large that e is defined for not in S let S be the map of VS into V which sends S x to
{S x } {S e }. If S  contains S there is a unique map S,S  of VS into VS  which makes
S,S 

VS

VS

S

V
commutative. If we use these maps to form the inductive limit of the spaces VS we obtain a space
which the layman is unable to distinguish from V .
Suppose that for every we are given a linear map B of V into itself. If B e = e for all but a
finite number of there is exactly one linear transformation B of V such that

B : x B x

Chapter 2

190

B is denoted by B .
For example if A , is a family of associative algebras, which may or may not have a unit,
and if, for almost all , is a given idempotent of A one may turn
A = A
into an algebra in such a way that

(a )(b ) = (a b ).

Let V , , be an A module. If for almost all a vector e such that e = e is given we

may turn V = e V into an A = A module in such a way that

(a )(x ) = (a x )
Suppose the family {e } is replaced by a family {e } but that, for all but a finite number of , e = e
where is a non-zero scalar. Suppose for example that e = e if is not in the finite set S . There
is a unique map of e V to e V which sends


to

 

S x S x

 

S x S x

It is invertible and commutes with the action of A. Moreover apart from a scalar factor it is independent
of S .
Now suppose F is a global field. A place of F is an equivalence class of injections, with dense
image, of F into a local field. If 1 takes F into F1 and 2 takes F into F2 they are equivalent if there
is a topological isomorphism of F1 with F2 such that 2 = 1 . The symbol for a place will be v .
If v contains the imbedding 1 and a belongs to F we set |a|v = |1 (a)|. To be definite we let Fv be
the completion of F with respect to the absolute value a |a|v . v is archimedean or non-archimedean
according to the nature of Fv . Non-archimedean places will sometimes be denoted by p.
If GF = GL(2, F ) we set

Gv = GFv = GL(2, Fv ).
Kv will be the standard maximal compact subgroup of Gv . then GA = GL(2, A) is the restricted direct
product of the groups Gv with respect to the subgroups Kv .
If v is non-archimedean we set Ov = OFv and Uv = UFv . Ov is the ring of integers of Fv and Uv
is the group of units of Ov . Suppose M  is a quaternion algebra over F . Let Mv = MF v = M  F Fv .
For almost all v the algebra Mv is split, that is, there is an isomorphism
v : Mv M (2, Fv )
where M (2, Fv ) is the algebra of 2 2 matrices over Fv . For every place v at which Mv is split we want
to fix such an isomorphism v . Let B be a basis of M over F and let Lv be the Ov module generated in
Mv by B . We may and do choose v so that for almost all v

v (Lv ) = M (2, Ov ).

Chapter 2

191

If B  is another basis and {v } a family of isomorphism associated to B then for every place v at
which Mv splits there is a gv in GL(2, Fv ) such that

v v1 = gv a gv1
for all a in M (2, Fv ). Moreover gv belongs to Kv for all but a finite number of v .
Suppose the family of isomorphisms v has been chosen. If Mv is split we define a maximal
compact subgroup Kv of Gv , the group of invertible elements of Mv , by the condition

v (Kv ) = Kv .
If Mv is not split we set


Kv = {x Mv  |(x)|v = 1}.

This group is compact. In any case Kv is defined for all v . Since many of the constructions to be made
depend on the family Kv which in turn depends on the family of v it is very unfortunate that the family
of v is not unique. We should really check at every stage of the discussion that the constructions are,
apart from some kind of equivalence, independent of the initial choice of v . We prefer to pretend that
the difficulty does not exist. As a matter of fact for anyone lucky enough not to have been indoctrinated
in the functorial point of view it doesnt. We do however remark that any two choices of the family of
Kv lead to the same result for almost all v. GA is the restricted direct product of the groups Gv with
respect to the subgroups Kv .
We have now to introduce the Hecke algebras H and H of GA and GA . Let Hv be HFv . If

Mv is split Gv isomorphic, by means of v , to Gv and we let Hv be the algebra of measures on Gv
corresponding to Hv . Suppose Mv is not split. If v is non-archimedean Hv is the algebra of measures
defined by the locally constant compactly supported functions on Gv . If v is archimedean Hv will
be the sum of two subspaces, the space of measuures defined by infinitely differentiable compactly
supported functions on Gv which are Kv -finite on both sides and the space of measures on Kv defined
by the matrix coeffients of finite-dimensional representations of Kv .
Let v and v be the normalized Haar measures on Kv and Kv . v is an elementary idempotent of
Hv and v is an elementary idempotent of Hv . We set

H = v Hv
and

H = v Hv

If S is the finite set of places at which Mv does not split we may write

H=
and

H =

 

s
vS Hv vS Hv = Hs H

 


vS Hv vS Hv = Hs H
s

By construction, if Mv is split, Hv and Hv are isomorphic in such a way that v and v correspond.
 s and H
  . We may also write
Using these isomorphism we may construct an isomorphism of H
S

GA =

 
Gv


vS


S
Gv = GS G
vS

Chapter 2
and

192

GA =

 
Gv


 .
Gv = GS G
S
vS

vS

 G
s
The second factor is in both cases a restricted direct product. There is an isomorphism : G
S
defined by


gv =

vS

v (gv )
vS

 S and H
  as algebras of measures on G
S and G
  and then the isomorphism between
We will interpret H
S
S
them will be that associated to .
We can also interpret the elements of H and H as measures on GA and GA . For example any
element of H is a linear combination of elements of the form f = v fv . Let T be a finite set of places
and suppose that fv = v for v not in T . If T  contains T , on the group
GA(T  ) =

 
Gv


vT 

Kv

vT 

we can introduce the product of the measures fv . Since GA is the union of these groups and the
measures on them are consistent we can put the measure together to form a measure f on GA . If each
fv is the measure associated to a function then f is also. Such measures form a subalgebra H1 and H.
The notion of an elementary idempotent of H or H is defined in the obvious way. If is an
elementary idempotent of H there is another elementary idempotent 1 of the form 1 = v v where
v is an elementary idempotent of Hv and v = v for almost all v so that 1 = .
We shall now discuss the representations of H. A representation of H on the vector space V
over C will be called admissible if the following conditions
 are satisifed
(i) Every w in V is a linear combination of the form (fi ) wi with fi in H1 .
(ii) If is an elementary idempotent the range of () is finite dimensional.
(iii) Let v0 be an archimedean place. Suppose that for each v an elementary idempotent v is given
and that v = v for almost all v . Let = v v . If w is in V the map




fv0 fv0 v=v0 v w
of v0 Hv0 v0 into the finite dimensional space ()V is continuous.
Suppose that an admissible representation v of HV on Vv is given for each v . Assume that for
almost all v the range of v (v ) is not zero. Assume also that the range of v (v ) has dimension one
when it is not zero. As we saw in the first chapter this supplementary condition is satisified if the
representations v are irreducible. Choosing for almost all v a vector ev such that v (v ) ev = ev we
may form V = ev Vv . Let be the representation v v on V . Because of the supplementary condition
it is, apart from equivalence, independent of the choice of the ev .
will be admissible. To see this observe first of all that condition (i) has only to be verified for
vectors of the form w = v wv . Suppose wv = ev when v is not in the finite set T which we suppose
contains all archimedean places. If v is not in T let fv = v so that wv = (fv ) wv . If v is in T let

wv =
Then

w=

vT

v (fvi ) wvi .

 

v (fvi ) wvi vT (fv ) wv .

Chapter 2

193

Expanding the right hand side we obtain the desired relation. The second condition has only to be
verified for elementary idempotents of the form = v v . Then

() V = (v ) Vv
Since (v )Vv is finite dimensional for all v and (v )Vv = (v )Vv , which has dimension one, for
almost all v the right side is finite dimensional. The last condition results from the admissibility of v0 .
cannot be irreducible unless each v is. Suppose however that each v is irreducible. If v is an
elementary idempotent of Hv and if v (v ) = 0 we have a representation v of v Hv v on v (v ) Vv .
Since it is irreducible v determines a surjective map

v : v Hv v L(v )
if L(v ) is the ring of linear transformations of V (v ) = v (v )Vv . To show that is irreducible we
have only to show that for every elementary idempotent of the form = v v the representation of
H on V () = ()V is irreducible. Suppose that v = v if v is not in T . Then

V () = v V (v )
is isomorphic to vT V (v ). The full ring of linear transformations of this space is

vT L(v )
and therefore the full ring of linear transformations of V () is


This is the image under of

 

vT L(v ) vT v (v ) .

 

vT v Hv v vT v

which is contained in H .
An admissible representation equivalent to one constructed by tensor products is said to be
factorizable.
Proposition 9.1 Every irreducible admissible representation of H is factorizable. The factors are

unique up to equivalence.
Suppose is such a representation. Let I be the set of elementary idempotents of the form = v
for which () is not 0. I is certainly not empty. Let V () = ()V if V is the space on which acts. If
and  are elementary idempotents we write  if  = . Then  will also equal . If = v
and  = v then  if and only if v v = v v = v for all v . If  and belongs to I so does
 . H is a subalgebra of  H  . Let (  , ) be the corresponding injection and let L() and L( ) be
the spaces of linear transformations of V () and V ( ). There is exactly one map

(  , ) : L() L(  )
which makes

( ,)

 H 

 )
( ,)

L()

L(  )

Chapter 2

194

commutative.
There is a map of v Hv v into H which sends fv to fv {w=v w }. Composing this map with
we obtain a map v of v Hv v onto a subalgebra Lv () of L(). L() and Lv () have the same unit,
namely (). If v = w the elements of Lv () commute with those of Lw (). If we form the tensor
product of the algebras
! Lv () with respect to the family of units there is a map from v Lv () to L()
which sends v v to v v . Moreover we may identify v v Hv v and H . Since the diagram

v v Hv v

v v )
v Lv ()

L()

is commutative the bottom arrow is surjective.

Lemma 9.1.1 The algebras Lv () are simple and the map v Lv () L() is an isomorphism.

To show that Lv () is simple we need only show that the faithful Lv ()-module V () is spanned
by a family of equivalent irreducible submodules. Let M be any irreducible submodule. Then the
family {T M } where T runs over the image of 1v {w=v Lw ()} spans V () and each T M is 0 or
equivalent to M because T commutes with the elements of Lv (). 1v is the unit of Lv (). We have only
to show that v Lv () R L(). Since v Lv () is the inductive limit of vT Lv (), where T is a finite
set, we have only to show that the map is injective on these subalgebras. As they are tensor products
of simple algebras they are simple and the map is certainly injective on them.
If  there is a commutative diagram

v vHv v

)
v L
v ()

( ,)

v vHv v

)
v Lv (  )

)
( ,)

L()

L(  )

Moreover if v (  , ) is the imbedding of v Hv v into v Hv v then (  , ) = v v (  , ). We want

to verify that a horizontal arrow v v (  , ) can be inserted in the middle without destroying the
commutativity. To do this we have only to show that if fv is in v Hv v and therefore in v Hv v then
v (fv ) = 0 if and only if v (fv ) = 0. Let U = v (fv ) and let T = v (fv ). If




E =  v w=v w
then




T E =  fv w=v w

is determined by its restriction to V () and that restriction is U .

Chapter 2

195

It is clear that if S is a sufficiently large finite set the map wS Lw (  ) L(  ) is an isomorphism.

We suppose that S contains v . E belongs to the image M of 1v {w=v Lw (  )}. Since M is simple
and E is not 0 there are Ai , Bi 1 i r in M such that
r


Ai E Bi = 1

i=1

Thus

T =


i

T Ai E Bi =

Ai T E Bi

and T = 0 if and only if U = 0.

Since the necessary compatibility conditions are satisifed we can take inductive limits, over I , to
the left and right. The inductive limit of the H is H and that of the v Hv v is Hv . Let Lv be that of
Lv () and L that of L(). There is a map v : Hv Lv and, for almost all v, v (v ) = v is not zero.
We have a commutative diagram

Hv H

v )
)
Lv L
in which the rows are isomophisms. Moreover L acts faithfully on V and the representation of H on V
can be factored through L.
If A is an algebra with a minimal left ideal J then any faithful irreducible representation of A on a
vector space X is equivalent to the representation on J . In fact we can choose x0 in X so that J x0 = 0.
The map j jx0 of J to X gives the equivalence. Thus to prove that is factorizable it will be enough
to show that L has a minimal left ideal, that the representation of L on this minimal left ideal is a tensor
product of representations v of Lv , and that v v is admissible.
Suppose A is a simple algebra and J is a left ideal in A. If a in A is not 0 and aJ = 0 then
Aa AJ = AJ = 0. If J is not 0 this is impossible. Suppose e is an idempotent of A and A1 = eAe. Let
J1 be a minimal left ideal of A1 and let J = AJ1 . If J were not minimal it would properly contain a
non-zero ideal J  . J  A1 would have to be 0. Since J e = J we must have eJ = eJ e = 0. Since this
is a contradiction J is minimal. Suppose for example that A is the union of a family {A } of matrix
algebras. Suppose that for each there is an idempotent e in A such that A = e Ae and that given
1 and 2 there is a 3 such that A3 contains A1 and A2 . Then A is certainly simple and, by the
preceding discussion, contains a minimal left ideal.
The algebras L and Lv satisfy these conditions. In fact, speaking a little loosely, L is the union of
the L() and Lv is the union of Lv (). Choose so that V () = 0 and let Jv be a minimal left ideal
in Lv (). Since Lv () is one-dimensional for almost all v the ideal Jv = Lv () for almost all v . Thus
J = Jv exists and is a minimal left ideal of L(). Thus LJ = Lv Jv . LJ is a minimal left ideal of L
and Lv JV is a minimal left ideal of Lv . The representation of L on LJ is clearly the tensor product of
the representations v of Lv on Lv JV .
Thus is equivalent to the tensor product of the representations v = v v . The representations
v are irreducible. Since it is easily seen that a tensor product v is admissible only if each factor is
admissible we may regard the first assertion of the proposition as proved.
If is an admissible representation of H on V and v is a place we may also introduce a representation of Hv on V which we still call . If u is in V we choose = w w so that ()u = u. Then if f
belongs to Hv we set



(f )u = f v w=v w u
The second part of the proposition is a consequence of the following lemma whose proof is immediate.

Chapter 2

196

Lemma 9.1.2 Suppose = w w . Then the representation of Hv is the direct sum of representa-

tions equivalent to v .
Let Sa be the set of archimedean primes. One can also associate to an admissible representation
S , the group formed by the elements of GA whose components at
of H on V a representation of G
a
every archimedean place are 1, on V . If v is archimedean one can associate to a representation of Av ,
the universal enveloping algebra of the Lie algebra of Gv , on V . Finally determines a representation
of the group ZA of scalar matrices in GL(2, A). If is irreducible there is a quasi-character of I the
group of id`eles such that





a 0
0 a

= (a) I

for all a in I . If v is associated to v and = v v then is associated to the quasi-character defined

by

(a) =

v (av ).
v

One may define the contragredient of and the tensor product of with a quasi-character of I . All the
is equivalent to 1 if is irreducible.
expected formal relations hold. In particular
The above discussion applies, mutatis mutandis, to the algebra H . The next proposition, which
brings us a step closer to the theory of automorphic forms, applies to H alone.
Proposition 9.2 Let = v be an irreducible admissible representation of H. Suppose that v is

innite dimensional for all v. Let be a non-trivial character A/F . There is exactly one space
W (, ) of continuous functions on GA with the following properties:
(i) If W is in W (, ) then for all g in GA and all x in A



W

1
0

x
1

 
g = (x) W (g)

(ii) W (, ) is invariant under the operators (f ), f H, and transforms according to the repre-

sentation of H. In particular it is irreducible under the action of H.

(iii) If F is a number eld and v an archimedean place then for each W in W (, ) there is a real

number N such that

W

a
0

0
1





= O |a|N

as a in Fv .
In the last assertion Fv is regarded as a subgroup of I . Fv is a subgroup of A and the restriction
v of to Fv is non-trivial. Thus for each place v the space W (v , v ) is defined and we may suppose
that v acts on it. Moreover for almost all v the largest ideal of Fv on which v if trivial is Ov and
v contains the trivial representation of Kv . Thus by Proposition 3.5 there is a unique function 0v in
W (v , v ) such that 0v (gv kv ) = 0v (gv ) for all kv in Kv and 0v (I) = 1. Then 0v (kv ) = 1 for all kv in
Kv . acts on

0v W (v , v )

If g is in GA and v belongs to this space then v (gv ) = 1 for almost all v so that we can define a
function on GA by

(g) =

v (gv ).
v

Chapter 2

197

The map v extends to a map of W (v , v ) into a space W (, ) of functions on GA . W (, )

certainly has the required properties. We have to show that it is characterized by these properties.
Suppose M is another space with these properties. There is an isomorphism T of W (v , v )
and M which commutes with the action of H. All we have to do is show that there is a constant c such
that if = v then

T (g) = c

v (gv ).
v

Let S be a finite set of places and let

WS = vS W (v , v )
and

*S = v=S W (v , v ).
W

Then

*S .
W (v , v ) = WS W

*S such that if
S W
We first show that if S is given there is a function cS on G
f =T





vS v

*S then
with in W
f (gh) = cS (h, )

v (gv )
vS

S .
if g is in GS and h is in G
*S and h belongs to G
S associate to
Suppose that S consists of the single place v . If belongs to W
every function v in W (v , v ) the function
v (gv ) = f (gv h)
on Gv . f is T (v ). By construction, if v is replaced by (fv )v with fv in Hv the function v is
replaced by (fv )v . Moreover if x is in Fv

v

1 x
0 1


gv

= v (x) v (gv ).

Since any conditions on rates of growth can easily be verified we see that the functions v are either all
zero or they fill up the space W (v , v ). In both cases the map v v is a map of W (v , v ) into
itself which commutes with the action of Hv and therefore consists merely of multiplication by a scalar
cS (h, ).
Now suppose that S  is obtained by adjoining the place w to S and that our assertion is true for S .
 S  and in W
*S  . If
Take h in G



f =T


vS  v

then, for g in GS , and gw in Gw ,

f (g gw h) = cS (gw h, w )

v (gv ).
vS

Chapter 2

198

The argument used before shows that for a given h and the function

gw cS (gw h, w )
is a multiple cS  (h, ) of w .
To prove the existence of c we obseve first that if S is the disjoint union of S1 and S2 we may write
S as h1 = h !

*
any h1 in G
1
vS2 hv with h in GS . Suppose 1 = {vS2 v } with in WS is in

*S . Then
W
1

cS1 (h1 , 1 ) =


v (hv ) cS (h, )

(9.2.1)

vS2

because the right hand side has all the properties demanded of the left. If S1 is large enough that 0v
exists for v not in S1 then, by its definition, cS1 (h, vS1 0v ) has a constant c(S1 ) on

Kv
vS1

The formula (9.2.1) shows that c(S) = c(S!

1 ) if S contains S1 . We take c to be the common value of
these constants. Given = v and g = gv we choose S so that v = 0v and gv Kv for v not in
S . Then



gv , vS v

T (g) = c

vS

=c

v (gv )

vS

v (gv ).
v

We observed that if v is finite dimensional the space W (v , v ) cannot exist if v is nonarchimedean or real. Although we neglected to mention it, the argument used for real field also
shows that W (v , v ) cannot exist if v is complex. The proof of Proposition 9.2 can therefore be used,
with minor changes, to verify the next proposition.
Proposition 9.3 If = v is given and if one of the representations v is nite dimensional there

can exist no space W (, ) satisfying the rst two conditions of the previous proposition.
An admissible representation of H on the space V is said to be unitary if there is a positive
definite hermitian form (v1 , v2 ) on V such that, if f (g) = f(g 1 ),

 

(f )v1 , v2 = v1 , (f )v2

for all f in H.
Lemma 9.4 If is unitary and admissible then V is the direct sum of mutually orthogonal invariant

irreducible subspaces.
The direct sum of the lemma is to be taken in the algebraic sense. We first verify that if V1 is an
invariant subspace and V2 is its orthogonal complement then V = V1 V2 . Certainly V1 V2 = 0. Let
be an elementary idempotent and let V (), V1 (), V2 () be the ranges of () in V , V1 , and V2 . let
V1 () be the range of 1 () acting on V1 . Then V () and V1 () are orthogonal and

V1 = V1 () V1 ().
Thus V2 () is just the orthogonal complement of V1 () in V (). Since V () is finite dimensional

V () = V1 () V2 ().
Since every element of V is contained in some V () we have V = V1 + V2 .
To complete the proof we shall use the following lemma.

Chapter 2

199

Lemma 9.4.1 If is a unitary admissible representation of H on the space V then V contains a

minimal non-zero invariant subspace.

Choose an idempotent so that V () = ()V = 0. Since V () is finite dimensional amongst

all the non-zero subspaces of it obtained by intersecting it with an invariant subspace of V there is
a minimal one N . Let M be the intersection of all invariant subspaces containing N . If M is not
irreducible it is the direct sum of two orthogonal invariant subspaces M1 and M2 . Then

N = M V () = () M = () M1 () M2
The right side is

M1 V ()} {M2 V ()

so that one of M1 V () and M2 V () is N . Then M1 or M2 contains M . This is a contradiction.

Let A be the set consisting of families of mutually orthogonal invariant, and irreducible subspaces
of V . Each member of the family is to be non-zero. Let {V } be a maximal family. Then V = V . If
not let V1 = V . The orthogonal complement of V1 would be different from zero and therfore would
contain a minimal non-zero invariant subspace which when added to the family {V } would make it
larger.
If T is a finite set of places most of the results of this paragraph are valid for representations of

T .
HT . For example is factorizable and W (, ) exists as a space of functions on G

Chapter 2

200

10. Automorphic Forms. In this paragraph F is still a global field. We shall begin by recalling a
simple result from reduction theory. If v is a place of A and a is in A then |a|v is the absolute value of
av the vth component of a. If a is in I
|a| =

|a|v
v

Lemma 10.1 There is a constant c 0 such that if g belongs to GA there is a in GF for which
1

max{|c|v , |d|v } c0 |detg| 2

v

if

a
c

g =

b
d

If F is a number field let OF be the ring of integers in F and if F is a function field take any
transcendental element x of F over which F is separable and let OF be the integral closure in F of the
ring generated by 1 and x. A place v will be called finite if |a|v 1 for all a in OF ; otherwise it will be
called infinite. If S is a finite set of places which contains all the infinite places let


A(S) = {a A  |a|v 1 if v = S}

I(S) = {a I  |a|v = 1 if v = S}
Then A = F + A(S) and if S is sufficiently large I = F I(S). We first verify that if I = F I(S) then

GA = GF GA(S)


where GA(S) = GL 2, A(S) . If v is not in S then v is non-archimedean and we can speak of ideals of
Fv . Any element of GA may be written as a product

g=



a b
c d

in which the second factor belongs to

K=

Kv
v

and therefore to GA(S) . It will be sufficient to show that the first factor is in GF GA(S) . If = 1 2 and
= 1 2 with 1 and 1 in F and 2 and 2 in I(S)

=

1
0

0
1



1 /1 2
0
1



2
0

0
2

The first factor is in GF and the third in GA(S) . Since 12 belongs to F + A(S) the second factor is in
GF GA(S) and the assertion follows.
There is certainly a u in OF such that |u|v < 1 at all finite places in S . Enlarging S if necessary we
may assume that a finite place v belongs to S if and only if |u|v < 1. Then

F A(S) =

 x 

 x OF , m Z .
m
u

Chapter 2

201

We identify the prime ideals of OF with the places corresponding to them. By the theory of rings of
quotients the proper ideals of F A(S) are the ideals of the form

F A(S)

p mp
pS

Since I = F I(S) every such ideal is principal. Thus F A(S) is a principal ideal domain.
To prove the lemma we show that there is a constant c0 such that if g belongs to GA(S) there is a
in GF A(S) such that
1

max{|c|v , |d|v } c0 |detg| 2

vS

if

g =

a b
c d


.

Fix a Haar measure

on theadditive group
A(S). This determines a measure on A(S) A(S).


The group L = F A(S) F A(S) is a discrete subgroup of A(S) A(S) and the quotient
A(S) A(S)/L is compact and has finite measure c1 . If g belongs to GA(S) the lattice Lg is also discrete
and the quotient A(S) A(S)/Lg has measure c1 |detg|.
Suppose (m, n) = (, )g belongs to Lg . If a = 0 belongs to F A(S) then

max{|am|v , |an|v } =

vS

|a|v =

1=
the product

vS



vS

Since

|a|v


max{|c|v , |d|v } .
vS

|a|v



vS

|a|v

v=S

|a|v is at least 1 and

max{|am|v , |an|v }
vS

max{|m|v , |n|v }.
vS

Let R be a positive number and consider the set



E = (m, n) Lg 


max{|m|v , |n|v } R .
vS

The previous inequality shows that if I contains a non-zero element of Lg it contains one (m, n) =
(, )g for which and are relatively prime. Then we may choose and in F A(S) so that
= 1. If




=

then belongs to GF A(S) and if

g =

a b
c d

Chapter 2

202

then c = m and d = n so that

max{|c|v , |d|v } R.
vS

To prove the lemma we have to show that there is a constant c0 such that if g is in GA(S) and
1
R = c0 |detg| 2 the set E is not reduced to {0}. We will show in fact that there is a constant c2 such that
for all g there is a non-zero vector (m, n) in Lg with
1

sup max{|m|v , |n|v } c2 |detg| 2s

vS

if s is the number of elements in S . There is certainly a positive constant c3 such that the measure of




(m, n) A(S) A(S)  sup max{|m|v , |n|v } R
vS

is, for any choice of R, at least c3 R2s . Choose c2 so that

c2 > 2

 c1  2s1
c3

If Lg contained no non-zero vector satisfying the desired inequality the set



1 
c2
(m, n) As AS  sup max{|m|v , |n|v } |detg| 2s
2
vS
would intersect none of its translates by the elements of Lg . Therefore its measure would not be
changed by projection on A(S) A(S)/Lg and we would have

c1 c3

 c2 2s
2

which is impossible.
Choose some place v of F which is to be archimedean if F is a number field. If c is any positive
constant there is a compact set C in I such that


{a I  |a| c}
is contained in


{ab  a Fv , |a| c, b C}

If 1 is a compact subset of A, 2 a compact subset of I , and c a positive constant we may introduce

the Siegel domain S = S(1 , 2 , c, v) consisting of all

g=

1 x
0 1



a 0
0 a



bb1
0

0
1


k

with x in 1 , a in I , b in 2 , b1 in Fv with |b1 | c, and k in K . Then ZA S = S. If we use the Iwasawa

decomposition of GA to calculate integrals we easily see that the projection of S on ZA \ GA has finite
measure. Moreover it follows readily from the previous lemma that, for a suitable choice of 1 , 2 , and
c,

GA = GF S.

Chapter 2

203

Thus ZA GF \ GA has finite measure.

Let be a continuous function on GF \ GA . If it is ZA -finite the space V spanned by the functions
(a) , a ZA , is finite dimensional. We may choose a finite set of points g1 , , gp and a basis
1 , , p of V so that i (gj ) = ij . Then

(a) =

p


i (a) i .

i=1

Since i (a) = (agi ) the function i are continuous and finite as functions on ZA or ZF \ ZA . Since
ZF \ ZA is isomorphic to F \ I it satisfies the hypothesis of Lemma 8.1 and i is a finite linear
combination of functions of the form

0
0



= (a) (log ||)m

where is a quasi-character of F \ I .
A continuous function on GF \ GA which is ZA -finite will be called slowly increasing if for any
compact set in GA and any c > 0 there are constants M1 and M2 such that




a 0
0 1

 

g  M2 |a|M1

for g in , a in I , and |a| c. If such an inequality is valid, with suitable choice of M2 , for any M1 we
will say, for lack of a better terminology, that is rapidly decreasing.
Suppose is a continuous function on GF \ GA . Assume it is K -finite on the right and that for
every elementary idempotent in H the space


{(f )  f H}

is finite dimensional. An argument used more than once already shows that there is a and an f in
H1 such that (f ) = . If a belongs to ZA

(a) = (a f )
so that is ZA -finite. Thus we can make the following definition.
Definition 10.2 A continuous function on GF \ GA is said to be an automorphic form if

(i) It is K -finite on the right

(ii) For every elementary idempotent in H the space


{(f )  f H}

is finite dimensional.
(iii) If F is a number field is slowly increasing.
We observe, with regret, in passing that there has been a tendency of late to confuse the terms
automorphic form and automorphic function. If not the result it is certainly the cause of much misunderstanding and is to be deplored.
Let A be the vector space of automorphic forms. If is in A and f is in H then (f ) is in A so
that H operates on A. A continuous function on on GF \ GA is said to be cuspidal if

F \A

1 x
0 1

 
g dx = 0

for all g inGA . An automorphic form which is cuspidal is called a cusp form. The space A0 of cusp
forms is stable under the action of H.

Chapter 2

204

Proposition 10.3 Let F be a function eld and let be a function on GF \ GA . If satises the
following three conditions it is a cusp form.
(i) is K-nite on the right.
(ii) is cuspidal.
(iii) There is a quasi-character of F \ I such that

a
0

0
a

 
g = (a) (g)

for all a in I.
If is an elementary idempotent of H there is an open subgroup K of K such that is invariant
under translations on either side by the elements of K  . Therefore the functions (f ) are invariant
under right translations. To prove the proposition we show that if K is a given open subgroup of K
and is a given quasi-character of F \ I then the space V of all continuous functions on GF \ GA
which are cuspidal and satisfy (gk) = (g) for all k in K as well as

a 0
0 a

 
g = (a) (g)

for all a in F \ I is finite-dimensional.

We shall show that there is a compact set C in GA such that the support of every in V is contained
in GF ZZ C . Then the functions in V will be determined by their restrictions to C . Since C is contained
in the union of a finite number of left translates of K they will actually be determined by their values
on a finite set and V will be finite dimensional.
Choose a Siegel domain S = S(1 , 2 , c, v) so that GA = GF S. If


S =



1 x
0 1



bb1
0

0
1


k  x 1 , b 2 , b1 Fv , |b1 | c, k K

we have just to show that the support in S of every in V is contained in a certain compact set which
is independent of . In fact we have to show the existence of a constant c1 such that vanishes on

1 x
0 1



bb1
0

0
1


k

as soon as |b1 | c1 . Let k1 , , kn be a set of representatives of the cosets of K/K  and let i (g) =
(gki ). If k belongs to ki K  then (gk) = i (g) and it will be enough to show that there is a constant
c2 such that, for 1 i n,







1 x
0 1

a 0
0 1

=0

if x belongs to A and |a| > c2 . It is enough to show this for a single, but arbitrary, i . Since i satisifes
the same hypothesis as , perhaps with a different group K , we just prove the corresponding fact for
.
We use the following lemma which is an immediate consequnce of the theorem of RiemannRoch
as described in reference [10] of Chapter I.

Chapter 2

205

Lemma 10.3.1 Let X be an open subgroup of A. There is a constant c 2 such that A = F + aX if a

belongs to I and |a| > c2 .

Let X be the set of all y for which

1 y
0 1

belongs to K  . Since

1 ay
0 1

we have



1 z
0 1

a x
0 1




a x
0 1

a x
0 1



=



a x
0 1

1 y
0 1





if z is in aX . The equation also holds for z in F and therefore for all z in A if |a| > c2 . Then

a x
0 1



1
=
measure(F \ A)

F \A

1 z
0 1



a x
0 1


dz

which by assumption is zero.

There is a corollary.
Proposition 10.4 Suppose is a cusp form and for some quasi-character of F \ I

a
0

0
a

 
g = (a) (g)

for all a in I. Then is compactly supported modulo GF \ ZA . Moreover the function

a

a
0

0
1



on F \ I is compactly supported.
The first assertion has just been verified. We know moreover that there is a constant c such that

is 0 for |a| c. If

a 0
0 1

w=



0 1
1 0

and  (g) = (gw) then  is also a cusp form. Since

a 0
0 1



 


1


1 0
a
1
w = (a)
= w
0 a
0

there is also a constant c1 such that it vanishes for |a| c1 .

0
1



Chapter 2

206

Proposition 10.5 Let F be a function eld and a quasi-character of F \ I. Let A0 () be the space

of cusp forms for which

a
0

0
a

 
g = (a) (g)

for all a in I. The representation of H on A0 () is the direct sum of irreducible admissible

representations each occurring with nite multiplicity.
The proof of Proposition 10.3 showed that the representation of H on A0 () is admissible. Let
 () = |()|1 (). The map  is an isomorphism of A0 () with A0 ( ) which replaces by
1 if 1 () = |()|1/2 . Thus we may as well suppose that is a character. Then if 1 and 2
belong to A0 () the function 1 2 is a function on GF ZA \ GA . Since it has compact support we may
set

(1 , 2 ) =
1 (g) 2 (g) dg.
GF ZA \GA

It is easily seen that


 

(f )1 , 2 = 1 , (f )2

so that, by Lemma 9.4, is the direct sum of irreducible admissible representations. Since is admissible
the range of () is finite dimensional for all so that no irreducible representation occurs an infinite
number of times.
The analogue of this proposition for a number field is somewhat more complicated. If is a
continuous function on GA , if v is a place of F , and if fv belongs to Hv we set


(fv ) =

(ghv ) fv (hv ) dhv .

Gv

Since fv may be a measure the expression on the right is not always to be taken literally. If v is
archimedean and if the function (hgv ) on Gv is infinitely differentiable for any h in GA then for any
X in Av the universal enveloping algebra of Gv , we can also define (X). If S is a finite set of places
we can in a similar fashion let the elements of

Hs = vS Hv
or, if every place in S is archimedean,

AS = vS Av

act on . It is clear what an elementary idempotent in HS is to be. If S = Sa is the set of archimedean

places we set Ha = HS .
Proposition 10.6 Suppose F is a number eld. A continuous function on GF \ GA is a cusp form

if it
(i)
(ii)
(iii)

satises the following ve conditions.

is K-nite on the right.
is cuspidal.
There is a quasi-character of F \ I such that

for all a in I.

a
0

0
a

 
g = (a) (g)

Chapter 2

207

(iv) For any elementary idempotent in Ha the space


{(f )  f Ha }

is nite dimensional.
(v) is slowly increasing.
There is a in Ha such that () = . Because of the fourth condition transforms according
to a finite dimensional representation of Ha and the usual argument shows that there is a function
f in Ha such that (f ) = .
!
Since is invariant under right translations by the elements of an open subgroup of vSa Kv
this implies in turn the existence of another function f in H such that (f ) = . From Theorem 2 of
[14] one infers that is rapidly decreasing.
As before we may assume that is a character. Then is bounded and therefore its absolute value
is square integrable on GF ZA \ GA which has finite measure. Let L2 () be the space of measurable
functions h on GF \ GA such that

h
for all g in GA and all a in I and

a 0
0 a

 
g = (a) h(g)


GF ZA \GA

|h(g)|2 dg < .

According to a theorem of Godement (see reference [11] to Chapter I) any closed subspace of L2 ()
which consists entirely of bounded functions is finite dimensional.
What we show now is that if is an elementary idempotent of H the space


V = {(f )  f H}

is contained in such a closed subspace. The functions in V itself certainly satisfy the five conditions
of the proposition and therefore are bounded and in L2 (). Replacing by a larger idempotent if
necessary we may suppose that = a a where a is an elementary idempotent in Ha . There is
a two-sided ideal a in a Ha a such that (f ) = 0 if f belongs to a. The elements of a continue
to annihilate V and its closure in L2 (). Approximating the -function as usual we see that there is
a function f1 in Ha and a polynomial P with non-zero constant term such that P (f1 ) belongs to a.
Therefore there is a function f2 in Ha such that f2 1 belongs to a. To complete the proof of the
proposition we have merely to refer to Theorem 2 of [14] once again.
For a number field the analogue to Proposition 10.4 is the following.
Propositon 10.7 Suppose is a cusp form and for some quasi-character of F \ I

a
0

0
a

 
g = (a) (g)

for all a in I. Then for any real number M1 there is a real number M2 such that





a 0 

M2 |a|M1

0 a 
for all a in I. Moreover the absolute value of is a square integrable on GF ZA \ GA .
We need another corollary of Proposition 10.6. To prove it one has just to explain the relation
between automorphic forms on GA and GR , which is usually assumed to be universally known, and
then refer to the first chapter of reference [11] to Chapter I. It is perhaps best to dispense with any
pretence of a proof and to rely entirely on the readers initiative. We do not however go so far as to
leave the proposition itself unstated.

Chapter 2

208

Proposition 10.8 Let Zv be the centre of Av and let I be an ideal of nite codimension in Z =
vSa Zv . Let be an elementary idempotent of H and a quasi-character of F \ I. Then the
space of innitely dierentiable functions on GF \ GA which satisfy the following ve conditions
is nite dimensional.
(i) is cuspidal.
(ii) () = .
(iii) If a is in I then



 
a 0

g = (a) (g).
0 a

(iv) (X) = 0 for all X in I

(v) is slowly increasing.
Proposition 10.9 Let be a quasi-charcter of F \ I and let A0 () be the space of cusp forms for

which

a
0

0
a

 
g = (g)

for all a in I. The representation of H on A0 () is the direct sum of irreducible admissible

representations each occurring with nite multiplicity.
Every element of A0 () is annihilated by some ideal of finite codimension in Z. If I is such an
ideal let A0 (, I) be the space of functions in A0 () annihilated by I. It is enough to prove the first
part of the proposition for the space A0 (, I). Then one may use the previous proposition and argue
as in the proof of Proposition 10.5. To show that every representation occurs with finite multiplicity
one combines the previous proposition with the observation that two functions transforming under the
same representation of H are annihilated by the same ideal in Z.
H acts on the space A. An irreducible admissible representation of H is a constituent of the
representation on A or, more briefly, a constituent of A if there are two invariant subspaces U and V of
A such that U contains V and the action on the quotient space U/V is equivalent to . A constituent
of A0 is defined in a similar fashion. The constituents of A0 are more interesting than the constituents
of A which are not constituents of A0 .
Theorem 10.10 Let = v be an irreducible admissible representation of H which is a constituent

of A but not of A0 . Then there are two quasi-characters and of F \ I such that for each place
v the representation v is a constituent of (v , v ).

v is the restriction of to Fv . Let B be the space of all continuous functions on GA satisfying

the following conditions.
(i) For all x in A



 
1 x

g = (g).
0 1
(ii) For all and in F

0
0

 
g = (g).

(iii) is K -finite on the right.

(iv) For every elementary idempotent in H the space


{(f )  f H}

is finite dimensional.

Chapter 2

209

Lemma 10.10.1 A continuous function on G A which satises the rst three of these conditions
satises the fourth if and only if it is AA -nite on the left.

A is the group of diagonal matrices. Since is a function on AF \ GA it is AA finite if and only if

it is AF \ AA finite. If it is AF \ AA finite there is a relation of the form
(ag) =

i (a) i (g)

where the i are finite continuous functions on AF \ AA . Since AF \ AA is isomorphic to the direct
product of F \ I with itself it is a group to which Lemma 8.1 can be applied. Thus there is a unique
family m,n,, of functions on GA such that

a1
0

0
a2

   
 a1 
g =  
(a1 ) (a2 ) (log |a1 |)m (log |a2 |)n m,n,, (g)
a2

The functions m,n,, also satisfy the first three conditions. Moreover there is a finite set S of pairs
(, ) and a non-negative integer M such that m,n,, is 0 if (, ) does not belong to S or m + n > M .
Given S and M let B(S, M ) be the space of continuous functions f on GA which satisfy the first
three conditions and for which
 

a1
0

0
a2

can be expanded in the form

  12
 a1  
 
(a1 ) (a2 ) (log |a1 |)m (log |a2 |)n fm,n,, (g)
 a2 
where the sum is taken only over the pairs (, ) in S the pairs (m, n) for which m + n M . B(S, M )
is invariant under H. To show that if is AF \ AA finite it satisfies the fourth condition we show that
the range of () on B(S, M ) is finite dimensional.
A function f in B(S, M ) is determined by the restriction of the finitely many functions fm,n,, to
K . If f is in the range of () these resrictions lie in the range of () acting on the continuous functions
on K . That range is finite dimensional.
We have also to show that if satisifes the fourth condition it is AA finite. The space V spanned
by the right translates of by the elements of K is finite dimensional and each element in it satisfies
all four conditions. Let 1 , , p be a basis of V . We can express (gk) as
p


i (k) i (g).

i=1

Because of the Iwasawa decomposition GA = NA AA K it is enough to show that the restriction of each
i to AA is finite. Since i satisfies the same conditions as we need only consider the restriction of .
Since is K finite
! there is a finite set S of places such that is invariant under right translations
by the elements of vS Kv . Let

Fv .

IS =
vS

Chapter 2

210

We regard IS as a subgroup of I . If we choose S so large that I = F I(S) then every element of I is

a product of = 1 2 3 with 1 in F , 2 in IS , and 3 in I(S) such that its component at any place
in S is 1. If in I is factored in a similar fashion

0
0



=

2
0

0
2


.

Thus we need only show that the restriction of to

AS =

0
0



 , IS

is finite. This is a consequence of Corollary 8.4 since the restriction of to GS clearly satisfies the
conditions of the corollary.
The next lemma explains the introduction of B.
Lemma 10.10.2 If is a constituent of A but not of A 0 then it is a constituent of B.

If belongs to A the functions

1
0 (g) =
measure(F \ A)

F \A

1 x
0 1

 
g dx

belongs to B. The map 0 commutes with the action of H and its kernel is A0 . Suppose U and V
are two invariant subspaces of A and occurs on the quotient of U by V . Let U0 be the image of U and
V0 be the image of V in B. Since is irreducible there are two possibilities. Either U0 = V0 in which
case is equivalent to the representation on U0 /V0 and is a constituent of B or U0 = V0 . In the latter
case

U = V + U A0

and is equivalent to the representation on

U A0 /V A0
which is precisely the possibility we have excluded.
Lemma 10.10.3 If is a constituent of B then there is a pair of quasi-characters , and a nonnegative integer M such that is a constituent of B(, , M ).

If S consists of the single pair (, ) then, by definition, B(, , M ) = B(S, M ). Suppose occurs
on the quotient of U by V . Choose the finite set S of pairs of quasi-characters and the non-negative
integer M so that U B(S, M ) is different from V B(S, M ). Then occurs on the quotient of
U B(S, M ) by V B(S, M ) and we may as well assume that U is contained in B(S, M ). The
argument used in the eighth paragraph in an almost identical context shows that

B(S, M ) = (,)S B(, , M )

so that the lemma is a consequence of Lemma 8.6.
The next lemma is proved in exactly the same way as Proposition 8.5

Chapter 2

211

Lemma 10.10.4 If is a constituent of B(, , M ) for some M then it is a constituent of B(, ) =

B(, , 0).

Let v and v be the restrictions of and to Fv . For almost all v the quasi-characters v and v
are unramifed and there is a unique function 0v in B(v , v ) such that 0v (gv kv ) = 0v (gv ) for all kv in
Kv while 0v (e) = 1. We can form

0v B(v , v )

There is clearly a linear map of this space into B(, ) which sends v to the function

(g) =

v (gv )
v

It is easily seen to be surjective and is in fact, although this is irrelevant to our purposes, an isomorphism.
In any case an irreducible constituent of B(, ) is a constituent of v (v , v ).
With the following lemma the proof of Theorem 10.10 is complete.
Lemma 10.10.5 If the irreducible admissible representation = v v is a constituent of = v ,

the tensor product of admissible representations which are not necessarily irreducible, then, for each
v, v is a constituent of v .
As in the ninth paragraph and determine representations and of Hv . The new will be a
constituent of the new . By Lemma 9.12 the representation of Hv is the direct sum of representations
equivalent to v . Thus v is a constituent of and therefore of . Since is the direct sum of
representations equivalent to v , Lemma 8.6 shows that v is a constituent of v .
The considerations which led to Proposition 8.5 and its proof will also prove the following proposition.
Proposition 10.11 If is an irreducible constituent of the space A0 then for some quasi-character

it is a constituent of A0 ().

Observe that if is a constituent of A0 () then

a 0
0 a


= (a) I

for all a in I . There are two more lemmas to be proved to complete the preparations for the Hecke
theory.
Lemma 10.12 Suppose there is a continuous function on G A with the following properties.

(i) is K nite on the right.

(ii) For all and in F and all x in A

 
g = (g).

(iii) There is a quasi-character of F \ I such that

for all a in I.

a
0

0
a

 
g = (a) (g)

Chapter 2

212

(iv) There is a nite set S of non-archimedean places such that the space

 S )
V = (H
 S according to the irreducible admissible representation = vS v .
transforms under H
Then V is a subspace of B and there are two quasi-characters and of F \ I such that v
is a constituent of (v , v ) for all v not in S.
If one observes that there is a finite set T of places which is disjoint from S such that I = F IT
one can proceed as in Lemma 10.10.1 to show that is A-finite on the right. Thus there is a finite set R
of pairs of quasi-characters and a non-negative integer M such that V is contained in B(R, M ). The
 S on some B(, ) and
same reduction as before shows that is a constituent of the representation of H
that v is a constituent of (v , v ) if v is not in S .
Lemma 10.13 Let be a continuous function on G F \GA . If satises the four following conditions
it is an automorphic form.
(i) is K nite on the right.
(ii) There is a quasi-character of F \ I such that

a
0

0
a

 
g = (a) (g)

for all a in I.

 S ) transforms according to
(iii) There is a nite set S of non-archimedean places such that (H
 S.
an irreducible admissible representation of H
(iv) If F is a number eld is slowly increasing.

We have to show that for every elementary idempotent in H the space (H) is finite dimensional. If f is a continuous function on GF \ GA let

1
f0 (g) =
measure(F \ A)


f
F \A

1 x
0 1

 
g dx.

 S . Consquently 0 satisfies the conditions

The map f f0 commutes with the action of H or of H
of the previous lemma and belongs to a space B(R, M ) invariant under H on which () has a finite
dimensional range.
We need only show that

V = {f (H)  f0 = 0}

is finite dimensional. If F is a function field then, by Proposition 10.3, V is contained in A0 ().

More precisely it is contained in the range of (), as an operator on A0 (), which we know is finite
dimensional. Suppose F is a number field. Since every place of S is non-archimedean the third
condition guarantees that is an eigenfunction of every element of Z. In particular there is an ideal I of
finite codimension in Z which annihilates and therefore every element of (H). By Proposition 10.6
the space V is contained in A0 () and therefore in A0 (, I). By Proposition 10.8 the range of () in
A0 (, I) is finite dimensional.

Chapter 2

213

11. Hecke theory. The preliminaries are now complete and we can broach the central topic of these
notes. Let be a non-trivial character of F \ A. For each place v the restriction v of to Fv is
non-trivial. Let = v v be an irreducible admissible representation of H. The local L-functions
L(s, v ) and the factors (s, v , v ) have all been defined. Since for almost all v the representation v
contains the trivial representation of Kv and Ov is the largest ideal on which v is trivial almost all of
the factor (s, v , v ) are identically 1 and we can form the product
(s, ) =

(s, v , v ).
v

In general it depends on . Suppose however that

a 0
0 a


= (a)I

and that is trivial on F . If is replaced by the character x (x) with in F then (s, v , v )
is multiplied b v ()||v2s1 so that (s, ) is multiplied by

v () ||v2s1 = ()||2s1 = 1
v

The product

L(s, v )
v

does not converge and define a function L(s, ) unless satisfies some further conditions.
Theorem 11.1 Suppose the irreducible admissible representation = v is a constituent of A.
Then the innite products dening L(s, ) and L(s,
) converge absolutely in a right half-plane and
the functions L(s, ) and L(s,
) themselves can be analytically continued to the whole complex
plane as meromorphic functions of s. If is a constituent of A0 they are entire. If F is a number
eld they have only a nite number of poles and are bounded at innity in any vertical strip of
nite width. If F is a function eld with eld of constants Fq they are rational functions of q s .
Finally they satisfy the functional equation

L(s, ) = (s, ) L(1 s,

).
Observe that if = v v then
= v
v . Consider first a representation which is a constituent
of A but not of A0 . There are quasi-characters and of F \ I such that v is a constituent of (v , v )
for all v . Since v has to contain the trivial representation of Kv for all but a finite number of v it is
equal to (v , v ) for almost all v .
Consider first the representation  = v (v , v ). Recall that



L s, (v , v ) = L(s, v ) L(s, v )


1
L s,
(v , v ) = L(s, 1
v ) L(s, v )
and



s, (v , v ), v = (s, v , v ) (s, v , v )

Chapter 2

214

If is any quasi-character of F \ I the product

L(s, v )
v

is known to converge in a right half plane and the funtion L(s, ) it defines is known to be analytically
continuable to the whole plane as a meromorphic function. Moreover if

(s) =

(s, v , v )
v

the functional equation

is satisifed. Since
and

L(s, ) = (s, ) L(1 s, 1 )

L(s,  ) = L(s, ) L(s, )
L(s,
 ) = L(s, 1 ) L(s, 1 )

they too are defined and meromorphic in the whole plane and satisfy the functional equation

L(s,  ) = (s,  ) L(1 s,

 ).
The other properties of L(s,  ) demanded by the lemma, at least when  is a constituent of A, can be
inferred from the corresponding properties of L(s, ) and L(s, ) which are well known.
When v is not (v , v ) it is (v , v ). We saw in the first chapter that



L s, (v , v )


L s, (v , v )
is the product of a polynomial and an exponential. In particular it is entire. If we replace (v , v ) by
v we change only a finite number of the local factors and do not disturb the converge of the infinite
product. If S is the finite set of places v at which v = (v , v ) then


L(s, ) = L(s, )
vS



L s, (v , v )


L s, (v , v )

and therefore is meromorphic with no more poles that L(s,  ). For L(s,
) the corresponding equation
is



L(s,
) = L(s,
)
vS

1
L s, (1
v , v )


1 .
L s, (1
v , v )

The functional equation of L(s, ) is a consequence of the relations

 




1
L s, (v , v )
s, (v , v ), v L 1 s, (1
v , v )

= 
 

1
L s, (v , v )
s, (v , v ), v L 1 s, (1
v , v )

Chapter 2

215

which were verified in the first chapter. It also follows from the form of the local factors that L(s, )
and L(s,
) are rational functions of qs when F is a function field. If F is a number field L(s, ) is
bounded in vertical strips of finite width in a right half-plane and, because of the functional equation,
in vertical strips in a left half-plane. Its expression in terms of L(s,  ) prevents it from growing very
fast at infinity in any vertical strip of finite width. The PhragmenLindelof principle implies that it is
bounded at infinity in any such strip.
Now suppose is a constituent of A0 . It is then a constituent of A0 () if

a 0
0 a



= (a) I

for a in I . Since the representation of H in A0 () is the direct sum of invariant irreducible subspaces
there is an invariant subspace U of A0 () which transforms according ot . Let belong to U . If g is
in GA



 

1 x
0 1

g (x) =

is a function on F \ A. Since g is continuous it is determined by its Fourier series. The constant term
is



 


1
measure F \ A

1 x
0 1

F \A

dx

which is 0 because is a cusp form. If is a given non-trivial character of F \ A the other non-trivial
characters are the functions x (x) with in F . Set

1
1 (g) =
measure F \ A

0
0 1

 
g =

F \A

Since is a function on GF \ GA .

1 x
0 1

1
measure F \ A

F \A

if belongs to F . Thus, formally at least,

(g) = g (e) =

1

 
g (x) dx.

1 x
0 1

0
0 1

 
g (x) dx

 
g .

In any case it is clear that 1 is not 0 unless is.

Let


U1 = {1  U }.

Since the map 1 commutes with the action of H the space U1 is invariant and transforms
according to under right translation by H. Moreover

1 x
0 1

 
g = (x) 1 (g)

if x is in A. If F is a number field is slowly increasing. Therefore if is a compact subset of GA there

is a real number M such that



 

a 0
0 1

= O(|a|M )

as |a| for all g in . Propositions 9.2 and 9.3 imply that all v are infinite dimensional and that
U1 is W (, ). Therefore U1 is completely determined by and and U is completely determined by
. We have therefore proved the following curious proposition.

Chapter 2

216

Proposition 11.1.1 If an irreducible represntation of H is contained in A0 () it is contained with

multiplicity one.
For almost all v there is in W (v , v ) a function 0v such that 0v (gv kv ) = 0v (gv ) for all kv in Kv
while 0v (e) = 1. W (, ) is spanned by functions of the form

1 (g) =

v (gv )

(11.1.2)

where v is in W (v , v ) for all v and equal to 0v for almost all v .

Suppose corresponds to a function 1 of the form (11.1.2). Suppose = 0v so that v contains the
trivial representation of Kv . If v is the normalized Haar measure on Kv let v be the homomorphism
of v Hv v into C associated to v . If fv is in v Hv v then


v (fv ) (g) =

(gh) fv (h) dh
Gv

and if v is the homomorphism associated to |v |1/2 v

v (fv ) |(detg)| 2 (g)

1

is equal to

|(detgh)| 2 (gh) fv (h) dh.

1

Gv

Since is a cusp form the function |(detg)|1/2 (g) is bounded and v satisifies the conditions of
Lemma 3.10. Thus if v = (v , v ) both v and v are unramified and

|(Cv )| 2 |Cv | 2 |v (Cv )| |(Cv )| 2 |Cv | 2

1

|(Cv )| 2 |Cv | 2 |v (Cv )| |(Cv )| 2 |Cv | 2

1

if Cv is the generator of the maximal ideal of Ov . Consequently the infinite products defining L(s, )
and L(s,
) converge absolutely for Re s sufficiently large.
We know that for any v and any v in W (v , v ) the integral


Fv

av
0

0
1


gv

|av |s 2 d av
1

converges absolutely for Re s large enough. Suppose that, for all a in I , |(a)| = |a|r with r real.
Applying Lemma 3.11 we see that if s + r > 12 and 0v is defined




 0
av
v

0
Fv

is, for gv in Kv , at most

0
1



1
gv  |av |s 2 d av

1
1

1 |Cv |s+r 2

2 .

Chapter 2

217

Thus if 1 is of the form (11.1.2) the integral


(g, s, 1 ) =

a 0
0 1

 
1
g |a|s 2 d a

is absolutely convergent and equal to

(gv , s, v )
v

for Re s sufficiently large. Since (gv , s, v ) is, by Proposition 3.5, equal to 1 for almost all v we can set

(g, s, 1 ) =

(gv , s, v )
v

so that

(g, s, 1 ) = L(s, ) (g, s, 1 ).

We can also introduce

 s, 1 ) =
(g,


1
I

and show that

if

a 0
0 1

 
1
g 1 (a) |a|s 2 d a

 s, 1 )
 s, 1 ) = L(s,
) (g,
(g,
 s, 1 ) =
(g,

 v , s, v ).
(g
v

Lemma 11.1.3 There is a real number s 0 such that for all 1 in W (, ) the integrals


(g, s, 1 ) =
 s, 1 ) =
(g,

1


1

a
0
a
0

 
1
g |a|s 2 d a
 
1
0
g 1 (a) |a|s 2 d a
1
0
1

 s, 1 ) can both be
are absolutely convergent for Re s > s0 . The functions (g, s, 1 ) and (g,
extended to entire functions of s. If F is a number eld they are bounded in vertical strips and if
F is function eld they are rational functions of q s . Moreover

(wg,
1 s, 1 ) = (g, s, 1 ).
We have seen that the first assertion is true for functions of the form (11.1.2). Since they form a
basis of W (, ) it is true in general. To show that

(g) =


F

1

0
0 1

 
g

Chapter 2

218

we need only show that the series on the right is absolutely convergent. We will do this later on in this
paragraph. At the moment we take the equality for granted. Then, for all 1 , (g, s, 1 ) which equals

 
F \I

is equal to

F \I

a 0
0 1

a 0
0 1

 
 s 1
g
|a| 2 d a

 
1
g |a|s 2 d a

 s, 1 ) is equal to
for Re s sufficiently large. Also (g,

F \I

a 0
0 1

 
1
g 1 (a) |a|s 2 d a.

We saw in the previous paragraph that, for a given g and any real number M ,




 


a 0

g  = O(|a|M )

0 1
as |a| approaches 0 or . Thus the two intregals define entire functions of s which are bounded in
vertical strips. If F is a function field the function

a

a 0
0 1

 
g

has compact support on F \ I so that the integral can be expressed as a finite Laurent series in qs .

1 s, 1 ) is equal to
The function (wg,

a 0
0 1


wg

1 (a) |a| 2 s d a.
1

Since w is in GF the equality (wh) = (h) holds for all h in GA and this integral is equal to

Since

1 0
0 a

1 0
0 a

 
1
g 1 (a) |a| 2 s d a.

=

a 0
0 a



a1
0

0
1

we can change variables in the integral to obtain

a 0
0 1

 
1
g |a|s 2 d a

which is (g, s, 1 ).
If we choose 1 of the form (11.1.2) we see that L(s, ) (g, s, 1 ) is entire and bounded in vertical
strips of finite width. For almost all v the value of (gv , s, 0v ) at the identity e is 1 and for such v we

Chapter 2

219

choose v = 0v . At the other places we choose v so that (e, s, v ) is an exponential eav s with real
av . Then (e, s, 1 ) is an exponential. Consequently L(s, ) is also entire and bounded in vertical
strips of finite width. If F is a number field (e, s, 1 ) will be a power of qs so that L(s, ) will be a
).
finite Laurent series in qs . Similar considerations apply to L(s,
To prove the functional equation we start with the relation


(w,
1 s, v ).

(e, s, v ) = L(1 s,
)

L(s, )
v

By the local functional equation the right hand side is

L(1 s,
)


(s, v , v ) (e, s, v ) .

Cancelling the term

!
v

(e, s, v ) we obtain
L(s, ) = (s, ) L(1 s,
).

Corollary 11.2 Suppose = v v is a constituent of A. For any quasi-character of F \ I the

products
L(s, v v )
v

and

L(s, v1
v )
v

are absolutely convergent for Re s suciently large. The functions L(s, ) and L(s, 1
)
they dene can be analytically continued to the whole complex plane as meromorphic functions
which are bounded at innity in vertical strips of nite width and have only a nite number of
poles. If F is a function eld they are rational functions of q s . If is a constituent of A0 they
are entire. In all cases they satisfy the functional equation
L(s, ) = (s, ) L(1 s, 1
)
if
(s, ) =

(s, v v , v ).
v

If = v v is a constituent of A or A0 and is a quasi-character of F \ I so is . Moreover

= v (v v ).
The converses to the corollary can take various forms. We consider only the simplest of these. In
particular, as far as possible, we restrict ourselves to cusp forms.

Chapter 2

220

Theorem 11.3 Let = v be a given irreducible representation of H. Suppose that the quasicharacter of I dened by




a 0

= (a) I
0 a

is trivial on F . Suppose there is a real number r such that whenever v = (v , v ) the inequalities
|Cv |r |v (Cv )| |Cv |r
and

|Cv |r |v (Cv )| Cv |r

are satised. Then for any quasi-character of F \ I the innite products

L(s, ) =

L(s, v v )
v

and

) =
L(s, 1

L(s, v1
v )
v

are absolutely convergent for Re s large enough. Suppose L(s, ) and L(s, 1
) are, for all
, entire functions of s which are bounded in vertical strips and satisfy the functional equation
L(s, ) = (s, ) L(1 s, 1
)
If the v are all innite dimensional is a constituent of A0 .
The absolute convergence of the infinite products is clear. We have to construct a subspace U of
A0 which is invariant under H and transforms according to the representation . W (, ) transforms
according to . If 1 belongs to W (, ) set



 

0
g
(g) =
1
0 1

We shall see later that this series converges absolutely and uniformly on compact subsets of GA . Thus
is a continuous function on GA . Since the map 1 commutes with right translations by elements
of H we have to show that, for all 1 , is in A0 and that is not zero unless 1 is.
Since is a character of F \ A

for all in F . Thus, for each g ,

1
0 1

 
g = (g)

1 x
0 a

 
g

is a function on F \ A. The constant term of its Fourier expansion is

1
measure F \ A

F \A

1 x
0 1

 
g dx.

Chapter 2

221

The integral is equal to



F \A

1 x
0 1



0
0 1

 
g dx.

A typical term of this sum is

 
g

0
0 1

(x) dx = 0.
F \A

In particular is cuspidal. Another simple calculation shows that if belongs to F

1
measure F \ A

F \A

is equal to

1

Thus 1 is zero if is.

By construction

1 x
0 1

0
0 1

0
0 1

 
g (x) dx

 
g .

 
g = (g)

if is in F . Moreover, for all a in I ,

a 0
0 a

 
g = (a) (g).

If a is in F the right side is just (g). Thus is invariant under left translations by elements
 0 1  of PF ,
the group of super-triangular matrices in GF . Since GF is generated by PF and w = 1 0 all we
need do to show that is a function on GF \ GA is to show that

(wg) = (g).
By linearity we need only establish this when 1 has the form (11.1.2). The hypothesis implies as
in the direct theorem that the integrals


(g, s, 1 ) =

1
I

and

 s, 1 ) =
(g,


1
I

a 0
0 1

a 0
0 1

 
1
g |a|s 2 d a

 
1
g 1 (a) |a|s 2 d a

converge absolutely for Re s sufficiently large. Moreover

(g, s, 1 ) =

(gv , s, v ) = L(s, )
v

(gv , s, v ).
v

Chapter 2

222

Almost all factors in the product on the right are identically one so that the product, and therefore
(g, s, 1 ), is an entire function of s. In the same way

 s, 1 ) = L(s,
)
(g,

 v , s, v )
(g
v

and is entire. Since


(wg
v , 1 s, v ) = (s, v , v ) (gv , s, v )


the function (wg,
1 s, 1 ) is equal to
L(1 s,
) (s, )

(gv , s, v )
v

which, because of the functional equation assumed for L(s, ), is equal to (g, s, 1 ).
>From its integral representation the function (g, s, 1 ) is bounded in any vertical strip of finite
width contained in a certain right half-plane. The equation just established shows that it is also bounded
in vertical strips of a left half-plane. To verify that it is bounded in any vertical strip we just have to
check that it grows sufficiently slowly that the PhragmenLindelof principle can be applied,

(g, s, 1 ) = L(s, )

(gv , s, v ).
v

The first term is bounded in any vertical strip by hypothesis. Almost all factors in the infinite product
are identically 1. If v is non-archimedean (gv , s, v ) is a function of |Cv |s and is therefore bounded
in any vertical strip. If v is archimedean

(gv , s, v ) =

(gv , s, v )
L(s, v )

We have shown that the numerator is bounded at infinity in vertical strips. The denominator is, apart
from an exponential factor, a -function. Stirlings formula shows that it goes to 0 sufficiently slowly
at infinity.
If Re s is sufficiently large


(g, s, 1 ) =

F \I F

which is

F \I

a 0
0 1

a 0
0 1

 
1
g |a|s 2 d a

 
1
g |a|s 2 d a.

This integral converges absolutely when Re s is sufficiently large. If Re s is large and negative


(wg,
1 s, 1 ) =
which equals

F \I

a 0
0 1


wg

1 (a) |a| 2 s d a

 
1
1 0
w
g 1 (a) |a| 2 s d a.
0 a
F \I

Chapter 2

223

Using the relation

1 0
0 a

=

a 0
0 a



a1
0

0
1

and changing variables we see that this integral is equal to

 
1
a 0
w
g |a|s 2 d a.
0
1
F \I

Set

f1 (a) =
and

a 0
0 a

 
g

 
a 0
f2 (s) = w
g .
0 a

We are trying to show that for any g the functions f1 and f2 are equal. The previous discussion applies
to as well as to . If 1 is in W (, ) the function

1 (g) = (detg) 1 (g)

is in W ( , ). When 1 is replaced by 1 the function is replaced by

 (g) = (detg) (g)

and fi is replaced by

fi (a) = (detg) (a) fi (a).

Thus for any quasi-character of F \ I the integral

f1 (a) (a) |a|s 2 d a

1

F \I

is absolutely convergent for Re s sufficiently large and the integral

f2 (s) (a) |a|s 2 d a

1

F \I

is absolutely convergent for Re s large and negative. Both integrals represent functions which can be
analytically continued to the same entire function. This entire function is bounded in vertical strips of
finite width.
The equality of f1 and f2 is a result of the following lemma.

Chapter 2

224

Lemma 11.3.1 Let f 1 and f2 be two continuous functions on F \ I. Assume that there is a constant

c such that for all characters of of F \ I the integral


F \I

f1 (a) (a) |a|s d a

is absolutely convergent for Re s > c and the integral


F \I

f2 (a) (a) |a|s d a

is absolutely convergent for Re s < c. Assume that the functions represented by these integrals
can be analytically continued to the same entire function and that this entire function is bounded
in vertical strips of nite width. Then f1 and f2 are equal.
Let I0 be the group of id`eles of norm 1. Then F \ I0 is compact. It will be enough to show that
for each b in I the functions f1 (ab) and f2 (ab) on F \ I0 are equal. They are equal if they have the
same Fourier expansions. Since any character of F \ I0 can be extended to a character of F \ I we
have just to show that for every character of F \ I


f1 (, b) = (b)
is equal to

F \I0


f2 (, b) = (b)

F \I0

f1 (ab) (a) d a

f2 (ab) (a) d a.

These two functions are functions on I0 \ I which is isomorphic to Z if F is a number field and to R if
F is a function field.
If F is a function field we have only to verify the following lemma.

Lemma 11.3.2 Suppose {a 1 (n)  n Z} and {a2 (n)  n Z} are two sequences and q > 1 is a real

number. Suppose

a1 (n) q ns

converges for Re s suciently large and



a2 (n) q ns

converges absolutely for Re s large and negative. If the functions they represent can be analytically
continued to the same entire function of s the two sequences are equal.
Once stated the lemma is seen to amount to the uniqueness of the Laurent expansion. If F is a
number field the lemma to be proved is a little more complicated.

Chapter 2

225

Lemma 11.3.3 Suppose g 1 and g2 are two continuous functions on R. Suppose there is a constant c

such that
g1 (s) =

g1 (x) esx dx

converges absolutely for Re s > c and


g2 (s) =

g2 (x) esx dx

converges absolutely for Re s < c. If g1 and g2 represent the same entire function and this
function is bounded in vertical strips then g1 = g2 .
All we need do is show that for every compactly supported infinitely differentiable function g the
functions g g1 and g g2 are equal. If


g(x) esx dx

g(s) =
R

is the Laplace transform of g the Laplace transform of g gi is g(s) gi (s). By the inversion formula

1
g gi (x) =
2i

b+i

g(s) gi (s) exs ds

bi

where b > c if i = 1 and b < c if i = 2. The integral converges because g goes to 0 faster than the
inverse of any polynomial in a vertical strip. Cauchys integral theorem implies that the integral is
independent of b. The lemma follows.
To complete the proof of Theorem 11.3, and Theorem 11.1, we have to show that for any 1 in
W (, ) the series



 

0
0 1

is uniformly absolutely convergent for g in a compact subset of GA and that if (g) is its sum then, if F
is a number field, for any compact subset of GA and any c > 0 there are constants M1 and M2 such
that



 





a 0
0 1


g  M1 |a|M2

for g in and |a| c. We prefer to prove these facts in a more general context which will now be
described.
For us a divisor is just a formal product of the form

D=

p mp .

It is taken over all non-archimedean places. The integers mp are non-negative and all but a finite
number of them are 0. Let S be a finite set of non-archimedean places containing all the divisors of D ,
that is, all places p for which mp > 0.
S where the components of
If a belongs to I we can write a in a unique manner as a product a!
Sa
S
aS outisde S are 1 and those of a
S inside S are 1. aS belongs to Is = vS Fv . Let ID
be the set of

Chapter 2

226

S
id`eles a such that, for any p in S , ap is a unit which satisfies ap 1(mod pmp ). Then I = F ID
and

S
S
F \ I is isomorphic to F ID \ ID .
If p is in S let KpD be the subgroup of all

a b
c d

 D be the subgroup of such matrices for which a d

in Kp for which c 0 (mod pmp ). Let K
p
mp
1(mod p ). Set
KSD =

KpD
pS

and set

D =
K
S

D
K
p
pS

 D is abelian.
 D is a normal subgroup of K D and the quotient K D /K
K
S
S
S
S
S
Let GD be the set for all g in GA such that gp is in the group KpD for all p in S . Any g in GA may
be written as a product gS gS where gS has component 1 outside of S and gS has component 1 inside
S is the set of gS . In particular
S . GS is the set of gS and G
S .
GSD = KSD G
It is easily seen that

GA = GF GSD .
In addition to D and S we suppose we are given a non-trivial character of F \ A, two characters
 D , two complex valued functions a and a
and of KSD /K
on F , an irreducible
S
 S = vS Hv , and a quasi-character of F \ I .
representation of H
There are a number of conditions to be satisifed. If

belongs to KSD then

a 0
0 b

a 0
0 b



=

b 0
0 a



S
If belongs to F and belongs to F ID

a =

and

a
=

S
0

0
1

S
0

0
1


a

a
.

Chapter 2

227

The functions a and a

are bounded. Moreover a = a
= 0 if for some v in S the number
regarded as an element of Fv does not lie in the largest ideal on which v is trivial. If v belongs to S
and a is a unit in Ov





a 0
0 a

= v (a).

Let = vS v . Then for a in Fv

v

a 0
0 a


= v (a) I.

Because of these two conditions is determined by and . There is a real number r such that if

v = (v , v )

|Cv |r |v (Cv )| |Cv |r

and

|Cv |r |v (Cv )| |Cv |r .

Finally we suppose that v is infinite dimensional for all v not in S .

These conditions are rather complicated. Nonetheless in the next paragraph we shall find ourselves
= 1 for all they reduce
in a situation in which they are satisifed. When S is empty, D = 0, a = a
to those of Theorem 11.3. In particular with the next lemma the proof of that theorem will be complete.
We shall use the conditions to construct a space U of automorphic forms on GA such that U transforms
 S according to while each in U satisifes
under H

(gh) = (h) (g)

for h in KSD . If U is such a space then for any in U and any a in I

a 0
0 a

 
g = (a) (g).

S
S
This is clear if a belongs ot ID
and follows in general from the relation I = F ID
.
 S spanned by functions of the form
Recall that W (, ) is the space of functions on G

1 (g) =

v (gv )
vS

where v belongs to W (v , v ) for all v and is equal to 0v for almost all v .

Lemma 11.4 Suppose 1 belongs to W (, ).

(i) For any g in GS

D the series

(g) =


F

a (gS ) 1

S
0

0
1


gS

converges absolutely. The convergence is uniform on compact subsets of GSD .

Chapter 2

228

(ii) The function dened by this series is invariant under left translation by the matrices in

GF GSD of the form


.

(iii) Suppose F is a number eld. Let be a compact subset of GS

D . Then there are positive

constants M1 and M2 such that


M2
|(g)| M1 |a| + |a|1
if

g=

S
with h in , a in ID
, and

1 x
01

1
0

x
1



a
0

0
1


h

in GSD .

It is enough to prove these assertions when 1 has the form

1 (g) =

v (gv ).
vS

To establish the first and third assertions we need only consider the series


F

()
vS




 


0
 v
gv 

0 1

(11.4.1)

where () = 0 if for some v in S the number regarded as an element of Fv is not the largest ideal on
which v is trivial and () = 1 otherwise.
We need only consider compact sets of the form

= KSD

(11.4.2)

vS

where v is a compact subset of Gv and v = Kv for almost all v .

Lemma 11.4.3 Suppose is of the form (11.4.2). There is a positive number such that for each
non-archimedean place v which is not in S there is a constant Mv such that





a
v

0

0
1

 

h  Mv |a|

for a in Fv and h in v and a constant cv such that





a
 v

0

0
1

 

h  = 0

Chapter 2

229

if |a| > cv and h is in v . Moroever one may take Mv = cv = 1 for almost all v.
Since v is invariant under an open subgroup of Kv for all v and is invariant under Kv for almost
all v while v = Kv for almost all v it is enough to prove the existence of Mv , cv , and such that these
relations are satisifed when h = 1. Since the function

a v

a 0
0 1



belongs to the space of the Kirillov model the existence of cv is clear. cv can be taken to be 1 when Ov
is the largest ideal of Fv on which v is trivial and v = 0v .
The existence of Mv , for a given v and sufficiently large , is a result of the absolute convergence
of the integral defining (e, s, v ). Thus all we need do is show the existence of a fixed such that the
inequality






 v


a 0
0 1


 |a|


is valid for almost all v . For almost all v the representation v is of the form (v , v ) with v and v
unramified, Ov is the largest ideal of Fv on which v is trivial, and v = 0v . Thus, for such v ,

v
if is a unit in Ov and

v

a 0
0 1

Cvn 0
0 1



= v



a 0
0 1





1
|Cv |n s 2 = L(s, v ).

If v = v (Cv ) and v = v (Cv )

L(s, v ) = 

1
1
 
.
1 v |Cv |s 1 v |Cv |s

Since |v | |Cv |r and |v | |Cv |r



n

  n+1

  v vm+1 
C
0
v
 v
 (n + 1) |Cv |rn .
=

0 1   v v 
Since |Cv |

1
2

there is a constant > 0 such that

(n + 1) |Cv |n
for all v and all n 0.
If v is archimedean the integral representations of the functions in W (v , v ) show that there are
positive constants cv , dv , and Mv such that




 




a 0
v
 Mv |a|cv exp dv |a|vv
h


0 1
for a in Fv and h in v . v is 1 if v is real and

1
2

if v is complex.

Chapter 2

230

Since we want to prove not only the first assertion but also the third we consider the sum

f

b 0
0 1

 

()
g =
vS




 


bv 0
v
gv 

0 1

where g lies in the set (11.4.2) and b is an id`ele such that bv = 1 for all non-archimedean v . We also
suppose that there is a positive number t such that bv = t for all archimedean v . If is a set of in F
for which ||v cv for all non-archimedean v not in S and () = 0 then

f
is bounded by



=0

b 0
0 1

 
g

v
Mv |t|c
exp(dv t ||vv )
v


Mv |v | .


vSSa

vSa

If F is a function field is a finite set and there is nothing more to prove. If it is a number field
choose for each v in S a constant cv such that () = 0 unless ||v cv . Since

||v = 1,
v

||
v
vSSa

cv



vS


||v .
vSa

Thus our sum is bounded by a constant times the product of

=0

!
vSa

tcv /v and



v
||c
exp dv t||vv .
v

vSa

The product !
vSa ||v is bounded below on {0}. Multiplying each term by the same sufficiently
high power of vSa ||v we dominate the series by another series

 v

||v exp(dv t ||vv )

vSa

in which the exponents v are non-negative. This in turn is dominated by


vSa

may
be regarded as a lattice in
ai i
if =

!
vSa

exp

 dv
2

!
vSa

tv /v times


t ||vv .

Fv . if 1 , , n is a basis of there is a constant d such that


 dv
|ai |.
||vv d
2

vSa

Chapter 2

231

Thus

b 0
0 1

 
g

is dominated by some power of t times a multiple of

 

edt|a|

n

a=

n

which is bounded by a multiple of 1 + 1t .

The first assertion is now proved and the third will now follow from the second and the observation
S
is the product of an element of F , an id`ele whose components are 1 at all
that every element of ID
non-archimedean places and equal to the same positive number at all archimedean places, and an id`ele
which lies in a certain compact set.
Suppose is in F and




1
0 1

S
belong to GS
D . Then is integral at each prime of S and v () = 1 if a = 0. If g belongs to GD and

h=

1
0 1


g

then (hS ) = (gS ) and if v is not in S

v

0
0 1


hv

= v () v

0
0 1


gv .

If a = 0

v () =
vS

Consequently

S
If b belongs to ID
then

and

bS
0

bS
S
0

0
bS

1
0 1

0
bS


so that

v () = 1.
v

 
g = (g).


gS

= (bS ) (gS )


gS

b 0
0 b

= (bS ) 1

 
g = (b) (g).

S
In particular if belongs to F ID

S
0

0
0

 
g = (g).

0
1


gS

Chapter 2

232

S
If belongs to F ID
and

h=

then

(hS ) =
and (h) is equal to

a

S
0

0
1

0
0 1


g
0
1

S
0


(gS )


(gS ) 1

Since

a =

S
0

0
1

S S
0

0
1

gS .


a

we can change variables in the summation to see that (h) = (g).

The lemma is now proved. The function

(g)

a
(gS ) 1

S
0

0
1


gS

can be treated in the same fashion.

Theorem 11.5 If is a quasi-characters of F \ I such that

v (av )

av
0

0
1


=1

for all units av of Ov set

(s, ) =

a (S ) |S |s 2

L(s, v v ).
vS

S \F
F ID

If

v (av )

av
0

0
1


=1

for all units av in Ov set

 ) =
(s,

a
(S ) |S |s 2

L(s, v v ).
vS

S \F
F ID

 ) are dened for Re s suciently large. Suppose that whenever is such

Then (s, ) and (s,
 ) is dened they can be analytically continued to entire functions which are
that (s, ) or (s,
bounded in vertical strips. Assume also that there is an A in F such that |A|p = |Cp |mp for any
p in S and
(s, ) =

v (A) |A|s1/2
v
vS




 s, 1 1 )
(s, v v , v ) (1
vS

Chapter 2

233

whenever (s, ) is dened. Then for any 1 in W (, ) there is an automorphic form on GA

such that



 


S 0
gS
(g) =
a (gS ) 1
0 1
on GSD .
 ) certainly converge for Re s
The infinite products occurring in the definition of (s, ) and (s,
sufficiently large. To check that the other factors converge one has to check that


|S |s 2

converges for Re s sufficiently large if the sum is taken over those elements of a system of coset
S
\ F for which ||v cv for v in S . This is easily done.
representatives of F ID
AS is the id`ele whose components are 1 outside of S and A in S . Since

0
AS

1
0



a b
c d



the matrix

0 A1
S
1
0
0
AS

1
0

=

d
cA1
S
AS b
a

normalizes KSD . In particular if g belongs to GS

D so does

0 1
A 0




0 A1
S
.
g
1
0

Lemma 11.5.1 If 1 is in W (, ) and g is in GS

D then, under the hypotheses of the theorem,

0
A

1
0



0
g
1

A1
S
0


= (g).

Let  (g) be the function on the left. As before all we need do is show that for every character
S
S
of F ID
\ ID
and every g in GS
D the integral


S \I S
F ID
D

a 0
0 1

 
1
g (a) |a|s 2 d a

(11.5.2)

is absolutely convergent for Re s large and positive. The integral


S \I S
F ID
D

a 0
0 1

 
1
g (a) |a|s 2 d a

(11.5.3)

is absolutely convergent for Re s large and negative, and they can be analytically continued to the same
entire function which is bounded in vertical strips.
If for any v in S the character





a v (a)

a 0
0 a

Chapter 2

234

on the group of units of Ov is not trivial the integrals are 0 when they are convergent. We may thus
assume that





a 0
0 1

v (a)

=1

for all units in Ov if v is in S .

We discuss the first integral in a formal manner. The manipulations will be justified by the final
result. The integrand may be written as a double sum



a

0
1

aS
0

1

gS

a
S
S S
0

0
1


gS

(a) |a|s 2 .

S
S
The inner sum is over in F ID
and the outer over a set of coset representatives of F ID
\ F .
Since









and

0
1

aS
0

S aS
0

= a
1

0
1

(a) |a|s 2 = (a) |a|s 2

the integral is equal to (gS ) times the sum over of


a

S
ID

S a
S
0

0
1




aS
gS
0

0
1



(a) |a|s 2 d a.
1

S
Since ID
is the direct product of


IS = {a I  aS = 1}
and a compact group under which the integrand is invariant the previous expression is equal to


a

IS

S a 0
0
1

(a) |a|s 2 d a.
1

gS

Changing variables to rid ourselves of the

S in the integrand and taking into account the relation
1

1 = () ||s 2 = (S ) (
S ) |S |s 2 |
S |s 2
we can see the original integral is equal to

(gS )

a (S ) |S |s 2


IS

a 0
0 1

There is no harm in supposing that 1 is of the form

1 (
gS ) =

v (gv ).
vS


gS

(a) |a|s 2 d a.
1

Chapter 2

235

We have already seen that, in this case,


1

IS

a 0
0 1

(a) |a|s 2 d a
1

gS

is convergent for Re s large and positive and is equal to


Fv

vS

If v is the function

av
0

0
1

v (av ) |av |s 2 d av .
1

gv

v (h) = v (h) v (h)

in W (v v , v ) this product is



L(s, v v ) (gv , s, v ) v1 (detgv ) .
vS

Thus the integral (11.5.2) is absolutely convergent for Re s large and positive and is equal to

(gv , s, v ).

(gS ) (det
gS ) (s, )
vS

The argument used in the proof of Theorem 11.3 shows that this function is entire.
If







0 1
A 0

h=

a 0
0 1

then

(hS ) =

0 A1
S
1
0

0
1

aS
0

gS .

Thus the integrand in (11.5.3) is equal to

a

aS
0

0
1

1

gS

S
0

0
1



0
AS



1
0

a
S
0

0
1


gS

(a) |a|s 2 .

The sum can again be written as a double sum over and . Since

a
=

aS
0

0
1


=a

which equals

a
(S )

and

S S
0

0
1



1 0
0 S

S1 aS
0

0
AS

1
0






aS

0
1



a
S
0

0
1


gS

0
1



Chapter 2

236

is equal to

(
S ) 1

S
0

0
1



AS



1
0

S
S1 a
0

0
1


gS

S
S
S
we can put the sum over F ID
and the integration over F ID
\ ID
together to obtain (gS ) times

the sum over F ID \ F of


a

S
ID

0
1

aS
0


1

S
0

0
1



0
AS

1
0



0
1

a
S
0

(a) |a|s 2 d a.
1

gS

We write

S
0

0
1



0
AS

1
0



a
S
0

0
1

=

0 1
1 0



S
0



1
S AS a
0

0
1

and then change variables in the integration to obtain the product of (AS ) |AS |s 2 and

(S )

(S ) |S |

1
2 s


IS

0 1
1 0



a 0
0 1

(a) |a|s 2 d a.
1

gS

Replacing a by a1 and making some simple changes we see that the integral is equal to


IS

a 0
0 1



0 1
1 0


gS

1 (s) 1 (a) |a| 2 s d a

1

which converges for Re s large and negative and is equal to



L(1 s, v1 v1 v ) (wgv , 1 s, v ) v (detgv ) .
vS

Thus the integral 11.5.3 is equal to

 s, 1 1 )
(gS ) (det
gS ) (AS ) |AS |s 2 (1

(wg, 1 s, v )

vS

which is entire.
Since

(wgv , 1 s, v ) = (s, v v , ) (gv , s, v )

the analytic continuations of (11.5.2) and (11.5.3) are equal. We show as in the proof of Theorem 11.3
that the resultant entire function is bounded in vertical strips of finite width.
There is now a simple lemma to be proved.

Chapter 2

237

Lemma 11.5.3 The group G F GS

D is generated by the matrices in it of the form

and

a
0


.

This is clear if S is empty. Suppose that S is not empty. If

g=

belongs to GF GS
D and ||v = 1 for all v in S then

g=



1
0

S
and both matrices belong to GF GS
D . In general if g is in GF GD then, for each v in S , ||v 1,
||v 1 and either ||v or ||v is 1. Choose in F so that, for every v in S , |v | = 1 if ||v < 1 and
|v | < 1 if ||v = 1. Then

1
0 1



=

and | + |v = 1 for all v in S . The lemma follows.

We know that if

h=

belongs to GF GS

= (g)
. Suppose
D then (hg) = (g) and (hg)

is in GF GS
D . Then


 







0 1
0
0 A1
S
.
g =

g
1
0
A 0

Since the argument on the right can be written

A1
0



0 1
A 0




0 A1
S
g
1
0

and the first term of this product lies in GF GS

D the right side is equal to

0 1
A 0




0 A1
S
= (g).
g
1
0

Chapter 2

238

S
Thus is invariant under GF GS
D . Since GA = GF GD the function extends in a unique
manner to a function, still denoted , on GF \ GA . It is clear that is K -finite and continuous and that

a 0
0 a

 
g = (a) (g)

for all a in I . It is not quite so clear that is slowly increasing. If is a compact subset of GA there is a
finite set 1 , , in GF such that

= i=1 i1 GSD .
What we have to show then is that if belongs to GF and c > 0 is given there are constants M1 and
M2 such that for all g in 1 GSD and all a in I for which |a| c




 


a
0

g  M1 |a|M2 .

0 1

If v is a place of F , which is not in S and is archimedean if F is a number field, there is a compact

set C in I such that



{a I  |a| c} F {a Fv  |a| c}C

Thus the inequality has only to be verified for a in Fv of course at the cost of enlarging . If

=

then

a 0
0 1

0

g=

a x
0 1


g

with x = (1 ) and the conclusion results from Lemma 11.4 and the relation

BA = (BA GF ) (BA GSD )

if

BA =

Otherwise we write

=

Then

a 0
0 1

g=

1
1

1
0

1
0

1
1

b y
0 1






GA .

0 1
1 0

1
0

0
1 a





1 2
0 1

0 1
1 0


.


1
0

2
a


g.




0 1
1 a2
g
1 0
0 1
lies in a certain compact set which depends on , c, and . The required inequality again follows from
Lemma 11.4
 S according to
The space U of functions corresponding to 1 in W (, ) transforms under H
. Lemma 10.13 implies that every element of U is an automorphic form. If it is not contained in A0 ,
Lemma 10.12 applied to the functions



 

1
1 x
0 (g) =

g dx
0 1
measure F \ A F \A
The matrix

with in U shows that there are two quasi-characters and on F \ I such that v = (v , v ) for
almost all v .

Chapter 2

239

Corollary 11.6 Suppose there does not exist a pair , of quasi-characters of F \ I such that

v = (v , v ) for almost all v. Then there is a constituent  = v of A0 such that v = v for

all v not in S.

 S according to it is, if v is not in S , the direct sum of subspaces

Since U transforms under H
transforming under Hv according to v . By assumption U is contained in A0 and therefore in A0 ().
A0 () is the direct sum of subspaces invariant and irreducible under H. Choose one of these summands
V so that the projection of U on V is not 0. If  = v is the representation of H on V it is clear that
v = v if v is not in S .
Another way to guarantee that U lies in the space of cusp forms and therefore that the conclusion
of the corollary holds is to assume that for at least one v not in S the representation v is absolutely
cuspidal.

Chapter 2

240

12. Some extraordinary representations. In [18] Weil has introduced a generalization of the Artin
L-functions. To define these it is necessary to introduce the Weil groups. These groups are discussed
very clearly in the notes of ArtinTate but we remind the reader of their most important properties. If
F is a local field let CF be the multiplicative group of F and if F is a global field let CF be the id`ele class
group F \ I . If K is a finite Galois extension of F the Weil group WK/F is an extension of G(K/F ),
the Galois group of K/F , by CK . Thus there is an exact sequence
1 CK WK/F G(K/F ) 1.
If L/F is also Galois and L contains K there is a continous homomorphism L/F,K/F of WL/F onto
WK/F . It is determined up to an inner automorphism of WK/F by an element of CK . In particular
WF/F = CF and the kernel of K/F,F/F is the commutator subgroup of WK/F . Also if F E K
we may regard WK/F as a subgroup of WK/F . If F is global and v a place of F we also denote by v
any extension of v to K . There is a homomorphism v of WKv /Fv into WK/F which is determined up
to an inner automorphism by an element of CK .
A representation of WK/F is a continuous homomorphism of WK/F into the group of invertible
linear transformations of a finite-dimensional complex vector space such that (w) is diagonalizable for
all w in WK/F . If K is contained in L then L/F,K/F is a representation of WL/F whose equivalence
class is determined by that of . In particular if is a generalized character of CF then K/F,F/F is
a one-dimensional representation of WK/F which we also call . If is any other representation
has the same dimension as . If F E K and is a representation of WK/E on X let Y be the space
of functions on WK/F with values in X which satisfy

(uw) = (u) (w)

for all u in WK/E . If v WK/F and Y let (v) be the function

W (wv)
(v) also belongs to Y and v (v) is a representation of WK/F . We write
= Ind(WK/F , WK/E , ).
If F is global and is a representation of WK/F then, for any place v , v = v is a representation
of WKv /Fv whose class is determined by that of .
Now we remind ourselves of the definition of the generalized Artin L-functions. Since we are
going to need a substantial amound of detailed information about these functions the best reference is
probably [19]. In fact to some extent the purpose of [19] is to provide the background for this chapter
and the reader who wants to understand all details will need to be quite familiar with it. If F is a local
field then to every representation of WK/F we can associate a local L-function L(s, ). Moreover if
F is a non-trivial additive character of F we can define a local factor (s, , F ). The L-function and
the factor (s, , F ) depend only on the equivalance class of .
If F is a global field we set

L(s, ) =

L(s, v )
v

Chapter 2

241

The product converges in a right half-plane and L(s, ) can be analytically continued to a function
meromorphic in the whole complex plane. If F is a non-trivial character of F \ A the functions
(s, v , v ) are identically 1 for all but a finite number of v. If

(s, ) =

(s, v , v )
v

and
 is the representation contragredient to the functional equation

L(s, ) = (s, ) L(1 s,

)
is satisfied. For all but finitely many places v the representation v is the direct sum of d, the dimension
of , one-dimensional representations. Thus there are generalized characters 1v , dv of CFv such
that v is equivalent to the direct sum of the one dimensional representations



w iv Kv /Fv ,Fv /Fv (w) .
Moreover, for all but finitely many of these v , 1v , , dv are unramified and there is a constant r , which
does not depend on v , such that

|iv (Cv )| |Cv |r

1 i d.

If F is a global or a local field and is a representation of WK/F then w det(w) is a onedimensional representation and therefore corresponds to a generalized character of CF . We denote this
character by det .
If F is a local field, is a two-dimensional representation of WK/F , and F is a non-trivial
additive character of F then, as we saw in the first chapter, there is at most one irreducible admissible
representation of HF such that

0
0


= det() I

and, for all generalized characters of CF ,

L(s, ) = L(s, )
L(s, 1
) = L(s, 1 )
(s, , F ) = (s, , F ).
If F (x) = F (x) then

(s, , F ) = det () (s, , F )

and, since

one also has

0
0


= det() I

(s, , F ) = det () (s, , F ).

Chapter 2

242

Thus , if it exists at all, is independent of F . We write = ().

There are a number of cases in which the existence of () can be verified simply by comparing
the definitions of the previous chapter with those of [19]. If and are two quasi-characters of CF and
is equivalent to the representation





0
K/F,F/F (w)


w
0
K/F,F/F (w)

then () = (, ). If K/F is a separable quadratic extension, is a quasi-character of CK = WK/K ,

and

= Ind(WK/F , WK/K , )
then () = (). Observe that () is alway infinite-dimensional.
Suppose F is a global field and K is a separable quadratic extension of F . Let be a quasi-character
of CK and let

= Ind(WK/F , WK/K , ).
If v does not split in K

v = Ind(WKv /Fv , WKv /Kv , v ),

but if v splits in K the representation v is the direct sum of two one-dimensional representations
corresponding to quasi-characters v and v such that v v1 is a character. Thus (v ) is defined and
infinite-dimensional for all v .
Proposition 12.1 If there is no quasi-character of CF such that () = (NK/F ) for all in CK

the representation v (v ) is a constituent of A0 .

If is a generalized character of F then

( )v = v v .
Define a generalized charcter K/F of CK by



K/F () = NK/F () .

Then
and

= Ind(WK/F , WK/K , K/F )

L(s, ) = L(s, K/F ).

The L-function on the right is the Hecke L-function associated to the generalized character K/F of
CK . It is entire and bounded in vertical strips unless there is a complex number r such that

K/F () () = ||r = |NK/F |r .

But then

() = 1 (NK/F ) |NK/F |r

which is contrary to assumption. The function

1
L(s, 1
) = L(s, K/F
1 )

is also entire and bounded in vertical strips. It follows immediately that the collection {(v )} satisfies
the conditions of Theorem 11.3.
This proposition has a generalization which is one of the principal results of these notes.

Chapter 2

243

Theorem 12.2 Suppose F is a global eld and is a two-dimensional representation of WK/F .

) are
Suppose also that for every generalized character of CF both L(s, ) and L(s, 1
entire functions which are bounded in vertical strips. Then (v ) exists for every place v and
v (v ) is a constituent of A0 .

We observe that the converse to this theorem is an immediate consequence of Theorem 11.1.
We are going to apply Corollary 11.6. There are a large number of conditions which must be
verified. We know that (v ) is defined for all but a finite number of v . In particular it is defined for
v archimedean for then v is either induced from a quasi-character of a quadratic extension of Fv or is
the direct sum of two one-dimensional representations. If v is equivalent to the direct sum of two onedimensional representations corresponding to quasi-charcters v and v then v v1 is a character so
that (v ) is infinite-dimensional. Let S be the set of places for which (v ) is infinite-dimensional. Let
S be the set of places for which (v ) is not defined or, since this is still conceivable, finite dimensional.
We are going to show that S is empty but at the moment we are at least sure that it is finite. If v is not
in S set v = (v ).
If v is in S the representation v must be irreducible so that

L(s, v v ) = L(s, v1
v ) = 1
v
for every generalized character v of Fv . The Artin conductor pm
v of v is defined in the Appendix to
[19]. There is a constant cv , depending on v , such that if v is unramifed
1

(s, v v , v ) = cv v (Cv )mv +2n |Cv |(mv +2nv )(s 2 )

v
if pn
is the largest ideal on which v is trivial. v is the restriction to Fv of a given non-trivial
v
character of F \ A.
We take

p mp

D=
pS

and = det . We define and by

and

av
0

0
bv

av
0

0
bv


= detv (bv )

= detv (av )

nv
if v belongs to S and av and
for every v in S
! bv are units of Ov . If belongs to F and ||v = |Cv |
we set a = 1 and a
= vS cv detv (); otherwise we set a = a
= 0.
The function (s, ) of Theorem 11.5 is defined only if v is unramified at each place of S and
then it equals

v (Cvnv ) |Cv |nv (s 2 )

1

L(s, v v )
vS

vS

which is



1 
v (Cvnv ) |Cv |nv (s 2 ) L(s, ).

vS

Chapter 2

244

 1 1 ) is also defined if v is unramifed at each place of S and is equal to

(s,


1 
cv v (Cvnv ) |Cv |nv (s 2 ) L(s, 1
 ).

vS

Choose A in F so that |Av | = |Cv |mv for every v in S . Then

s 12

v (A) |A|v

v (Cv )mv |Cv |mv (s 2 ) .

vS

vS

The functional equation asserts that L(s, ) is equal to

(s, v v , v )




(s, v v , v ) L(1 s, 1
).
vS

vS

The first factor is equal to

cv v (Cv )2nv |Cv |2nv (s 2 )

vS



s 12 

v (A) |A|v

vS

Therefore (s, ) is equal to

s 12 


 s, 1 1 ).
(s, v v , v ) (1

v (A) |A|v
vS

vS

The assumptions of Theorem 11.5 are now verified. It remains to verify that of Corollary 11.6. It
will be a consequence of the following lemma.
Lemma 12.3 Suppose F is a global eld, K is a Galois extension of F , and and are two
representations of the Weil group WK/F . If for all but a nite number of places v of F the local
representations v and v are equivalent then and are equivalent.

We set

L0 (s, ) =

L(s, p ).
p

The product is taken over all non-archimedean places. We first prove the following lemma.
Lemma 12.4 If is unitary the order of the pole of L0 (s, ) at s = 1 is equal to the multiplicity with

which the trivial representation is contained in .

There are fields E1 , , Er lying between F and K , characters E1 , , Er , and integers m1 , , mr
such that is equivalent to

ri=1 mi Ind(WK/F , WK/E , Ei )

Let i = 1 if Ei is trivial and 0 otherwise. Since

r

L0 (s, Ei )mi

L0 (s, ) =
i=1

Chapter 2

245

the order of its pole at s = 1 is

r
i=1

mi i . However
Ind(WK/F , WK/E , Ei )

contains
the trivial representation if and only if Ei is trivial and then it contains it exactly once. Thus
r
m
i i is also the number of times the trivial representation occurs in .
i=1
Observe that if T is any finite set of non-archimedean primes the order of the pole of

L(s, p )
pT

at s = 1 is the same as that of L0 (s, ).

The first step of the proof of Lemma 12.3 is to reduce it to the case that both and are unitary.
Then and certainly have the same degree d. Let act on X and let act on Y . Under the restriction
to CK the space X decomposes into the direct sum of invariant one-dimensional subspaces X1 , , Xd
which transform according to quasi-characters 1 , , d of CK . If a is a real number let


M (a) = {i  |i ()| = ||a for all in CK }
and let

X(a) =

Xi

iM (a)


X(a) is invariant under WK/F and X = a X(a). Let (a) be the restriction of to X(a). Replacing
by and X by Y we can define 1 , , d and Y (a) in a similar fashion.
We now claim that if v is equivalent to v then v (a) is equivalent v (a) for each a. To see this
we need only verify that any linear transformation from X to Y which commutes with the action of
WKv /Fv or even of CKv takes X(a) to Y (a). Observe that under the restriction of v to CKv the space
Xi transforms according to the character iv and that |iv ()| = ||a for all in CKv if and only if
|i ()| = ||a for all in CK . Thus X(a) and Y (a) can be defined in terms of v and v alone. The
assertion follows.
Thus we may as well assume that for some real number a
|i ()| = | i ()| = ||a
for all i and all in CK . Replacing by ||a () and by ||a () if necessary we
may even assume that a = 0. Then and will be equivalent to unitary representations and we now
suppose them to be unitary.
If is irreducible and !  and !  then v is equivalent to v whenever v is
equivalent to v . Since we can use induction on d it is enough to show that if is irreducible and
 be the representations contragredient
unitary and contained in then it is contained in . Let  and
)v = v v is equivalent to (
)v for all but a finite number of v.
to and . Certainly (
 contains the identity then, as is wellMoreover  contains  which contains the identity. If
known and easily verified, contains . On the other hand the orders of the poles of L0 (s,  ) and
L0 (s,
 ) at s = 1 are clearly equal so that, by Lemma 12.4,
 contains the trivial representations
if  does.

Chapter 2

246

We return to the proof of Theorem 12.2. It follows from Lemma 12.3 that if the assumptions of
Corollary 11.6 are not satisfied is equivalent to the direct sum of two one-dimensional representations
associated to quasi-characters and of CF . Then

L(s, ) = L(s, ) L(s, ).

The two functions on the right are Hecke L-functions. The function on the left is entire for every choice
of . Taking = 1 and = 1 we see that L(s, 1 ) and L(s, 1 ) have a zero at s = 1. Let
1 () = ||r () where is a character. Then

L(s, 1 ) = L(s + r, )
L(s, 1 ) = L(s r, 1 ).
Now neither L(s, ) nor L(s, 1 ) has a zero in the set Re s 1. Therefore 1 + r < 1 and 1 r < 1.
This is impossible.
!
We can now apply Corollary 11.6 to assert that there is a constituent  = v v of A0 such that
v = (v ) for v not in S . To prove the theorem we need only show that v = (v ) for v in S . Taking
the quotient of the two functional equations

L(s, ) =


(s, v v , v L(1 s, 1
)


v

and

L(s,  ) =


(s, v v , v ) L(1 s, 1
 ),
v

we find that
vS

is equal to

vS

L(s, v v )
L(s, v v )

(s, v v v )
(s, v v , v )

vS

L(1 s, v1
v )
1
L(1 s, v
v )

We need one more lemma. If v is a non-archimedean place and v is a quasi-character of Fv

let m(v ) be the smallest non-negative integer such that v is trivial on the units of Ov congruent to
m( )
1 modulo pv v .
Lemma 12.5 Suppose S is a nite set of non-archimedean places and v0 S. Suppose that we are

given a quasi-character v0 of Fv0 and for each v = v0 in S a non-negative integer mv . Then there
is a quasi-character of CF such that v0 = v0 and m(v ) mv if v = v0 belongs to S.

Suppose v0 () = ||av0 v0 () where v0 is a character. If  is a character of CF and v 0 = v0

while m(v ) mv for v = v0 in S we may take to be the generalized character ||r  ()
of CF . In other words we may assume initially that v0 is a character. Let A be the group of id`eles
whose component at places not in S is 1, whose component of a place v  v0 in S is congruent to

v
1 modulo pm
v , and whose component at v0 is arbitrary. Certainly F A = {1}. We claim that F A
is closed in I . Indeed if I there is a compact neighborhood X of on which the norm is bounded

Chapter 2

247

above by 1/ and below by where is a positive constant. If F and A then || = ||.

Moreover


1
A = { A  || }

is compact. Since F is discrete F A is closed. Since any point has a compact neighborhood whose
intersection with F A is closed the set F A is itself closed.
We can certainly find a character of A which equals v0 on Fv0 and, for any v = v0 in S , is

v
non-trivial on the set of units in Ov congruent to 1 modulo pm
v . Extend this character to F A by

setting it equal to 1 on F . The result can be extended to a character of I which is necessarily 1 on F .

We take to
be thischaracter.
!
= v (v ). If () = v v (v ) then is a quasi-character of F \ I . Since, by
Let v 0v 0v
s
the quasi-characters and det are equal. Therefore v = detv for all
construction, = det on ID
v. We know that if m(v ) is sufficiently large

L(s, v v ) = L(s, v v ) = 1
and

L(1 s, v1
v ) = L(1 s, v1
v ) = 1.

Moreover, by Proposition 3.8

(s, v v , v ) = (s, v v , v ) (s, v , v ).

It is shown in the Appendix of [19] that if m(v ) is sufficiently large

(s, v v , v ) = (s, v detv , v ) (s, v , v ).

Applying Lemma 12.5 and the equality preceding it we see that if v is in S and v is any quasicharacter of Fv

L(s, v v )
=
L(s, v v )
Recalling that
for v in S we see that

(s, v v , v )
(s, v v , v )



)
L(1 s, v1
1
L(1 s, v
v )


.

L(s, v v ) = L(1 s, v1
v ) = 1
L(1 s, v1
v )
(s, v v , v )
=
.

L(s, v v )
(, v v , v )

(12.5.1)

The theorem will follow if we show that

L(s, v v ) = L(1 s, v1 v ) = 1
for all choices of v .
If not, either v is a special representation or there are two quasi-characters v and v of Fv such
that v = (v , v ). According to (12.5.1) the quotient

L(1 s, v1 v )
L(s, v v )

Chapter 2

248

is an entire function of s for every choice of v . If v = (v , v ) and m(1

v v ) is positive

L(1 s, v
v )
1 |Cv |s
=

1 |Cv |1s
L(s, 1
v v )
which has a pole at s = 1. If m(1
v v ) = 0

v )
L(1 s, v
=

L(s, 1
v v )

1 |Cv |s
1 |Cv |1s



s
1 1
v v (Cv ) |Cv |
1 v v1 (Cv ) |Cv |1s

which has a pole at s = 1 unless v v1 (Cv ) = |Cv |. But then it has a pole at s = 2. If v is the special
representation associated to the pair of quasi-characters
1

v () || 2
of Fv then

v () || 2
1

L(1 s, v
v )
1 |Cv |s+ 2
=
1

L(s, 1
1 |Cv | 2 s
v v )
1

which has a pole at s = 12 .

There is a consequence of the theorem which we want to observe.
Proposition 12.6 Suppose E is a global eld and that for every separable extension F of E, every

Galois extension K of F , and every irreducible two-dimensional representation of WK/F the

function L(s, ) is entire and bounded in vertical strips. Then if F1 is the completion of E at some
place, K1 is a Galois extension of F1 , and 1 is a two-dimensional representation of WK1 /F1 the
representation (1 ) exists.
We begin with a simple remark. The restriction of 1 to CK1 is the direct sum of two onedimensional representationscorresponding
to generalized characters

 1 and 2 of CK1 . If belongs
to G = G(K1 /F1 ) either 1 () = 1 () for all in CK or 1 () = 2 () for all in CK . If


the representation 1 is irreducible there is at least one for which 1 () = 2 (). If 1 = 2 , the
fixed field L1 of
 




H = G  1 () 1 ()

is a quadratic extension of F . The restriction of 1 to WK1 /L1 is the direct sum of two one-dimensional
c
which is the kernel of
representations and therefore is trivial on the commutator subgroup WK
1 /L1
K1 /F1 ,L1 /F1 . With no loss of generality we may suppose that K1 equals L1 and is therefore a quadratic
extension of F1 . Then 1 is equivalent to the representation

Ind(WK1 /F, , WK1 /K1 , 1 ).

If 1 is reducible (1 ) is defined. The preceding remarks show that it is defined if 1 is irreducible
and 1 () is not a scalar matrix for some in CK1 . The proposition will therefore follow from
Theorem 12.2 and the next lemma.

Chapter 2

249

Lemma 12.7 Suppose F 1 is the completion of the eld E at some place, K1 is a Galois extension of

F1 , and 1 is an irreducible two-dimensional representation such that 1 () is a scalar matrix for

all in CK1 . Then there is a separable extension F of E, a Galois extension K of F , a place v of
K, and isomorphism of Kv with K1 which takes Fv to F1 , and an irreducible two-dimensional
representation of WK/F such that v is equivalent to 1 .
Observe that the existence of 1 forces F1 to be non-archimedean. We establish a further sequence
of lemmas.
Lemma 12.8 Suppose V is a nite dimensional real vector space, G is a nite group of linear

transformations of V , and L is a lattice in V invariant under G. If is a quasi-character of L

invariant under G there is a quasi-character  of V invariant under G and a positive integer m
such that the restrictions of  and to mL are equal.
Let V be the dual of V and VC its complexification. There is a y in VC such that (x) = e2ix,y
for all x in L. If z belongs to VC the generalized character x e2ix,z is trivial on L if and only if z

. L
L
 is the lattice
belongs to m


{v V  x, v Z for all x in L}.


 be the group contragredient to G. We have to establish the existence of an m and a z in L
Let G
m such
. If belongs to G
 then y y = w belongs to L
 . Clearly w + w = w . Set
that y z is fixed by G
z=

1 
w .
[G : 1]

If m is taken to be [G : 1] this is the required element.

Lemma 12.9 Suppose F is a global eld, K is a Galois extension of it, and v is a place of K.
Suppose also that [Kv : Fv ] = [K : F ] and let v be a quasi-character of CKv invariant under
G = G(Kv /Fv ) = G(K/F ). There is a closed subgroup A of nite index in CK which is invariant
under G and contains CKv and a quasi-character of A invariant under G whose restriction to
CKv is v .

Suppose first that the fields have positive characteristic. We can choose a set of non-negative
integers nw , w = v , all but a finite number of which are zero, so that the group

B = CKv

nw
UK
w
w=v

nw
is invariant under G and contains no element of K except 1. UK
is the group of units of OKw which
w
nw
are congruent to 1 modulo pKw . We extend v to B by setting it equal to 1 on
nw
UK
w
w=v

and then to A = K B/K by setting it equal to 1 on K .

Chapter 2

250

Now let the fields have characteristic 0. Divide places of K different from v into two sets, S ,
consisting of the archimedean places, and T , consisting of the non-archimedean ones. Choose a

, w T , all but a finite number of which are zero, so that
collection of non-negative integers w

B  = CKv

n

CKw
wS

UKww .
wT

is invariant under G and contains no roots of unity in K except 1. If w is archimedean let UKw be the
elements of norm 1 in Kw and set

Ba =

n

UKw
wS

UKww .
wT

B  /Ba is isomorphic to the product of CKv and

V =

CKw /UKw
wS

which is a vector group. The projection L of

M = B1 (B  K )/B1
on V is a lattice in V and the projection is an isomorphism. Define the quasi-character of L so that if
m in M projects to m1 in CKv and to m2 in V then

v (m1 ) (m2 ) = 1.
is invariant under G. Choose a quasi-character  of V and an integer n so that  and are equal
on nL. Let  be the quasi-character obtained by lifting v  from CKv V to B  . It follows
froma theorem of Chevalley ([20] Theorem 1) that we can choose a collection of non-negative ingegers
{nw  w T } all but a finite number of which are zero so that nw nw for all w in T , so that
B = CKv

CKw
wS

nw
UK
w
wT

is invariant under G, and so that every element of B K is an nth power of some element of B  K .
The restriction of  to B is trivial on B K . We take A = K B/K and let be the quasi-character
which is 1 on K and on B .
Lemma 12.10 Suppose F 1 is a completion of the global eld E, K1 is a nite Galois extension of F1

with Galois group G1 , and K1 is a quasi-character of CK1 invariant under G1 . There is a separable
extension F of E, a Galois extension K of F , a place v of K such that [Kv : Fv ] = [K : F ], an
isomorphism of Kv with K1 which takes Fv to F1 , and a quasi-character of CK invariant
under g(K/F ) such that v = K1 .
We may as well suppose that F1 = Ew , where w is some place of E . It is known ([8], p.31) that
there is a polynomial with coefficients in E such that if is a root of this polynomial Ew ()/Ew is
isomorphic to K1 /F1 . Let L be the splitting field of this polynomial and extend w to a place of L.
The extended place we also call w. Replacing E by the fixed field of the decomposition group of w if

Chapter 2

251

necessary we may suppose that F1 = Ew , K1 = Lw and [Lw : Ew ] = [L : E]. Now set w = K1 and
extend w to a quasi-character  of A as in the previous lemma.
Let K be the abelian extension of L associated to the subgroup A. Since A is invariant under
G(L/E) the extension K/E is Galois. Let v be a place of K dividing the place w of L. Since A
contains CLw the fields Kv and Lw are equal. Let F be the fixed field of the image of G(Kv /Ew ) in
G(K/E). Let v also denote the restriction of v to F . The fields Fv and Ew are the same. The mapping
NK/L : CK CL maps CK into A. Let =  NK/L . Then is clearly invariant under G(K/F ).
Since NK/L restricted to Kv is an isomorphism of Kv with Lw which takes Fv onto Ew the lemma is
proved.
To prove Lemma 12.7 we need only show that if F is a global field, K is a Galois extension of F ,
is a quasi-character of CK invariant under G(K/F ), v is a place of K such that [K : F ] = [Kv : Fv ],
and 1 is an irreducible two-dimensional representation of WKv /Fv such that v () = v ()I for all
in CKv then there is a two-dimensional representation of WK/F such that v is equivalent to 1 .
will be irreducible because 1 is.
Let 1 act on X . Let v be the right regular representation of WKv /Fv on the space Vv of functions
f on WKv /Fv satisfying

f (w) = v () f (w)
for all in CKv and all w in WKv /Fv . If is a non-zero linear functional on X the map from x to the
function (1 (w)x) is a WKv /Fv -invariant isomorphism of X with a subspace Y of Vv .
Let V be the space of all functions f on WK/F satisfying

f (w) = () f (w)
for all in CK and all w in WK/F . Since [K : F ] = [Kv : Fv ] the groups G(K/F ) and G(Kv /Fv ) are
equal. Therefore

WK/F = CK WKv /Fv

Moreover CKv = CK WKv /Fv . Thus the restriction of functions in V to WKv /Fv is an isomorphism
of V with Vv . For simplicity we identify the two spaces. Let be the right regular representation of
WK/F on V . If belongs to CK then

f (w) = (ww1 ) f (w) = () f (w)

because is G(K/F ) invariant. Therefore () = () I and a subspace V is invariant under WK/F
if and only if it is invariant under WKv /Fv . If we take for the restriction of to Y then v will be
equivalent to 1 .

Chapter 2

252
References
for Chapter II

Automorphic forms are discussed in terms of group representations in [3] and [11] as well as:
13. Godement, R., Analyse spectrale des fonctions modulaires, Seminaire Bourbaki, No. 278.

a la theorie de Langlands, Seminaire Bourbaki, No. 321.

14. Godement, R., Introduction `
As its name implies the Hecke theory is a creation of Hecke.
15. Hecke E., Mathematische Werke.
Maass seems to have been the first to consider it outside the classical context.

eine neue Art von nichtanalytischen automorphen Funktionen und die Bes16. Maass, H., Uber
timmung Dirichletscher Reihen durch Funktionalgleichungen, Math. Ann. 121 (1944).
It seems to have been Weil who first used several L-functions to prove a converse theorem.

17. Weil, A., Uber

die Bestimmung Dirichletscher Reihen durch Funktionalgleichungen, Math. Ann.
168 (1967).
His generalizations of the Artin L-functions are introduced in:

eorie du corps de classes, Jour. Math. Soc. Japan, vol. 3 (1951).

18. Weil, A., Sur la th
For various technical facts used in the twelfth paragraph we refer to:
19. Langlands R., On the functional equation of the Artin L-functions, Notes, Yale University (in
preparation).
We have also had occasion to refer to:

eor`emes d'arithmetique, Jour. Math. Soc. Japan, vol. 3 (1951).

20. Chevalley, C., Deux th
A result more or less the same as Proposition 12.1 is proved in:
21. Shalika, J.A. and S. Tanaka, On an explicit construction of a certain class of automorphic forms,
Preprint.

Chapter 3

253
Chapter III: Quaternion Algebras

13. Zeta-functions for M (2, F ). In this paragraph F is again a local field and A = M (2, F ) is the
algebra of 2 2 matrices with entries from F . The multiplicative group A of A is just GF = GL(2, F ).
If g is in GF we set
|g|A = A (g) = |detg|2F .

Let be an admissible representation of HF on the space V . Let the contragredient representation

act on on V . If v belongs to V and v to V the function

(g)v, v = v,
(g 1 )
v
is characterized by the relation


(gh), v, v f (h) dh = (g) (f )v, v

for all f in HF .
If belongs to the Schwartz space S(A) and v belongs to V and v to V we set

(g) (g)v, v d g

Z(, , v, v) =
GF

and

(g) v,
(g)
v  d g

Z(
, , v, v) =
GF

The choice of Haar measure is not important provided that it is the same for both integrals.
If is a quasi-character of F

Z( , , v, v) =

(g) (detg)(g)v, v d g

GF

The purpose of this paragraph is to prove the following theorem.

Theorem 13.1 Let be an irreducible admissible representation of H F and
its contragredient. Let
act on V and
on V .
(i) For every v in V , v in V , and in S(A) the integrals dening Z(sF , , v, v) and Z(sF

, , v, v) converge absolutely for Re s suciently large.

(ii) Both functions can be analytically continued to functions which are meromorphic in the whole
plane and bounded at innity in vertical strips of nite width.
(iii) If
s+ 1
Z(F 2 , , v, v) = L(s, ) (s, , v, v)

and

s+ 12

Z(F

 , v, v)

, , v, v) = L(s,
) (s,

 , v, v) are entire.
then (s, , v, v) and (s,
n
(iv) There exist , v1 , , vn and v1 , , vn such that i=1 (s, , vi , vi ) is of the form a ebs with
a = 0.
(v) If  is the Fourier transform of with respect to the character Z (x) = F (tr x) then
 s,  , v, v) = (s, , F ), (s, , v, v).
(1
We suppose first that F is non-archimedean and is absolutely cuspidal. Then we may take in
the Kirillov form so that V is just S(F ). Since an additive character F = is given we will of course
want to take the Kirillov model with respect to it. The next lemma is, in the case under consideration,
the key to the theorem.

Lemma 13.1.1 If belongs to S(F ), v belongs to V , and v

belongs to V set

(g) = (detg)v,
(g)
v |detg|1
F
if g belongs to GF and set (g) = 0 if g in A is singular. Then belongs to S(A) and its Fourier
transform is given by
1
(detg)
 (g) =  (detg)(g)v, v |detg|1
F
if g belongs to GF and

 (g) = 0

if g is singular. Here is the quasi-character of F dened by

a
0

0
a

and



= (a) I

0
1

1
0


.

This lemma is more easily appreciated if it is compared with the next one which is simpler but
which we do not really need.
Lemma 13.1.2 Let S 0 (A) be the space of all in S(A) that vanish on the singular elements and

satisfy

1
g1
0

x
1


g2

dx = 0

for g1 and g2 in GF . If is in S0 (A) so is its Fourier transform.

Since S0 (A) is stable under left and right translations by the elements of GF it is enough to show
that

for a in F and that

a 0
0 0
1 x
0 1


=0

dx = 0

To verify these relations we just calculate the left sides!

a 0
0 0




=


a 0
(g) A g
dg
0 0
A

The right side is a positive multiple of


GF

which equals


GF /NF


a 0
|detg|2 d g
(g) A g
0 0





 

a 0
1 x
2
|detg|
A g
g
dx d g
0 0
0 1
F

Chapter 3

255

This is 0 because the inner integral vanishes identically.

is equal to

 



1 x
0 1


dx


F ( + + x) d d d d dx

which, by the Fourier inversion fomula, is equal to

which equals


F ( + ) d d d



 



0
1
|| F ( + )
d d d
0
0 1

and this is 0.
(g)
v has compact
We return to the proof of Lemma 12.1.1 for absolutely cuspidal . Since v,
support on GF modulo ZF the function (g) belongs to S(A). Moreover

is equal to

(detgh) |detgh|1
F

 
1 x
g
h dx
0 1

(g

v,

1 x
0 1



(h) v dx.

Since is absolutely cuspidal this integral is 0. Thus belongs to S0 (A) and, in particular,  vanishes
at the singular elements.
Suppose we can show that for all choice of , v , and v

 (e) =  (e) v, v.

(13.1.3)

If h belongs to GF set 1 (g) = (h1 g). If a = deth, 1 (x) = |a| (a1 x), and v1 = (h) v ,

1 (g) = 1 (detg)v1 ,
(g)
v |detg|1
F .
Then 1 (e) is equal to

1 (e) v1 , v.

On the other hand

1

= (w) = |a| (w)

which equals

|a|

a1
0

a1



a1
0

a 0
0 1

0
1




(w) .

Thus  (h), which equals 1 (e) |deth|2 , is

 (deth)(h) v, v 1 (deth) |deth|1 .

The formula (13.1.3) will be a consequence of the next lemma.

Chapter 3

256

Lemma 13.1.4 Let d be the normalized Haar measure on the group U = UF . If is a character of

U set

() (x) d

(, x) =
U

if x is in F . Let dx be the Haar measure on F which is self-dual with respect to . Then


(, xC n ) (ax) dx = 0
F

unless |a| = |C|n but if a = C n with in U



(, xC n ) (ax) dx = () |C|n c1
F

if c is the measure of U with respect to dx.

The general case results from the case n = 0 by a change of variable; so we suppose n = 0. In this
case the formulae amount to a statement of the Fourier inversion formula for the function which is 0
outside of U and equal to c1 () on U .
Suppose we could show that there is a positive constant d which does not depend on such that
for all , v , and v

 (e) = d (e) v, v.

Then we would have

 (g) = d (detg) (g)v, v 1 (detg).

Exchanging and
and recalling that
= 1 we see that  , the Fourier transform of  , is
given by

 (g) = d2  (detg)v,
(g)
v |detg|1
F (detg),

where  =
(w)1 if 1 (a) =  (a) 1 (a). According to the remarks preceding the statement of
Theorem 2.18,  is the product of (w)  = (1) and 1 (detg). Thus

 (g) = (1) d2 (detg) v,

(g)
v |detg|1
F .
Since  = (g) = (1) (g) the numbers d2 and d are both equal to 1. The upshot is that in
the proof of the formula (13.1.3) we may ignore all positive constants and in particular do not need to
worry about the normalizaton of Haar measures.
Moreover it is enough to prove the formula for , v , v in a basis of the spaces in which they
are constrained to lie. Oddly enough the spaces are all the same and equal to S(F ). Assume
1 = v, 2 = v, and are supported respectively by Cn1 U , C n2 U , and C n U and that, for all in U ,
1 (C n1 ) = 11 (), 2 (C n2 ) = 21 () and (Cn ) = 1 (). , 1 and 2 are three characters of U .
1 (, t) = (11 )tn1 ,
2 (, t) =
The formal Mellin transforms of these three functions are
1 n2
1 n
(2 )t , and (,
 t) = ( )t . Recall that, for example,

(,
 t) =


n


tn

(C n ) () d.
U

Chapter 3

257

The scalar product 1 , 2  is equal to

1 (a) 2 (a) d a = (1 2 ) (n1 n2 ) 2 (1).

If (C n ) = 0 () z0n then

 (, t) = C(, t) (
 1 01 , t1 z01 )

which equals

(0 )

Cm ( 1 01 )tmn z0n .

Consequently

 (1) = Cn ( 1 01 ) z0n .

Thus the formula to be proved reads

 (e) = Cn ( 1 01 ) z0n 2 (1) (12 ) (n1 n2 ).

Almost all g in A can be written in the form

g=

b 0
0 b



1 x
0 1



a 0
0 a



0 1
1 0



1 y
0 1

with a and b in F and x and y in F . The additive Haar measure dg on A may be written as

dg = |detg|2F d g = |b4 | d b dx|a| d a dy

and for any g of this form



A (g) = F b(x y)

while (g) is equal to

(b) |b a|
2

2

(b a)

0 1
1 0



a1
0

0
1



1 x
0 1



1 ,

1 y
0 1


2 .

Let f1 and f2 be the two functions which appear in the scalar product. Their formal Mellin
transforms can be calculated by the methods of the second paragraph,

f1 (, t) = 0 (1) C(, t) (1 01 11 , C n1 x) 1 01 () z0rn1 trn2

if a = r and

f2 (, t) = (21 , C n2 y) tn2 .

The scalar product of f1 and f2 is equal to

f1 (a) f2 (a) d a

Chapter 3

258

which, by the Plancherel theorem for F , is equal to

1
(1)
2

f1 (, ei ) f2 (1 , ei ) d.

A typical integral is equal to the product of 0 (1) 1 01 ()z0rn1 and

C(, ei ) ei(r+n1 +n2 ) (1 01 11 , C n1 x) (1 21 , C n2 y) d

which equals
Also if a = Cr

2Cr+n1 +n2 () (1 01 11 , C n1 x) (1 2 , C n2 y).

1 (b) |b2 a|1 (b2 a) = (b2 C r ) 1 () 1 (b) |b2 |1 |a|1 .

If we put all this information together we get a rather complicated formula for (g) which we
have to use to compute  (e).  (e) is expressed as an integral with respect to a, b, x, and y . We will
not try to write down the integrand. The integral with respect to a is an integration over followed
by a sum over r . The integrand is a sum over . The integration over annihilates all but one term,
that for which 0 = 1. We can now attempt to write down the resulting integrand, which has to be
integrated over b, x, and y , and summed over r . It is the product of

1 (b) |b|2 (1) z0rn1 (b2 C r ) Cr+n1 +n2 ( 1 01 )

and



(11 , C n1 x) (0 21 , C n2 y) F b(x y) .

The second expression can be integrated with respect to x and y . Lemma 13.1.4 shows that the
result is 0 unless |b| = |C|n1 = |C|n2 . In particular  (e) = 0 if n1 = n2 . If n1 = n2 the integration
over b need only be taken over Cn1 U . Then the summation over r disappears and only the term for
which r + 2n1 = n remains. Apart from positive constants which depend only on the choices of Haar
measure  (e) is equal to


z0n 1 (1) Cn ( 1 01 )

11 21 () d.

Since

11 21 () d = (1 2 )
U

the proof of Lemma 13.1.1 is complete.

) = 1 if is absolutely cuspidal the first three assertions of the theorem are,
Since L(s, ) = L(s
for such , consequences of the next lemma.

Chapter 3

259

Lemma 13.1.5 Suppose belongs to S(A), v belongs to V , and v

belongs to V . If is absolutely

cuspidal the integral

(g)(g)v, v |detg|s+ 2 d g
1

is absolutely convergent for Re s suciently large and the functions it denes can be analytically
continued to an entire function.
Suppose the integral is convergent for some s. If is an elementary idempotent such that ()v = v
the integral is not changed if is replaced by

(gh1 ) (h) dh.

1 (g) =
GL(2,OF )

Since is absolutely cuspidal it does not contain the trivial representation of GL(2, OF ) and we can
choose to be orthogonal to the constant functions on GL(2, OF ). Then 1 (0) = 0. Thus, when
proving the second assertion of the lemma we can suppose that (0) = 0.
The support of (g)v, v is contained in a set ZF C with C compact. Moreover there is an
open subgroup K  of GL(2, OF ) such that the functions (g) and (g)v, v are invariant under right
translations by the elements of K  . If
p
C Ui=1
gi K 

the integral is equal to

p


(gi )v, v |detgi |s+ 2

i=1

a 0
0 a

 
gi (a) |a|2s+1 d a,

if each of the integrals in this sum converges. They are easily seen to converge if Re s is sufficiently
large and if (0) = 0 they converge for all s. The lemma is proved.
Now we verify a special case of the fifth assertion.
Lemma 13.1.6 Suppose is in S(F ) and

(g) = (detg)v,
(g)
v |detg|1 .
Then for all u in V and all u
in V
 s,  , u, u) = (s, , ) (s, , u, u).
(1
The expression (s, , u, u
) is the integral over GF of
1

|detg|s 2 (detg) (g)u, u

 v,
(g)
v.
The integral


(gh)u, u
 v,
(gh)
v dh
SL(2,F )

depends only on detg . Set it equal to F (detg). Then (s, , u, u

) is equal to

(a) F (a) |a|s 2 d a.

1

Chapter 3
By Lemma 13.1.1

260

 (g) =  (detg) |detg|1 1 (detg) (g)v, v

so that (s,  , u, u
) is equal to

1
 (a) F (a) |a|s 2 1 (a) d a

if

F(a)


u,
(gh)
u (gh)v, v dh
SL(2,F )

whenever a = detg . Since the integrand is not changed when g is replaced by

b 0
0 b


g

we have F(b2 a) = F (a) and F(a) = F(a1 ). The same relations are valid for F . Also F(a) = F (a1 )
so that F = F .
We remind ourselves that we are now trying to show that

 (a) F (a) 1 (a) |a| 2 s d a

1

is equal to

(a) F (a) |a|s 2 d a.

1

(s, , )
F


If U is an open subgroup of UF such that

and

0
0 1
0
0 1


u
=u


v=v

for in U  then F and F are constant on cosets of (F )2 U  which is of finite index in F . Write

F (a) =

p


ci i (a)

i=1

where i are characters of F /(F )2 U  . We may assume that all ci are different from 0. Then
1

F (a

)=

p


ci i (a1 ).

i=1

The factor (s, i , ) was defined so that


F

1
 (a) 1
(a) |a| 2 s d a
i
1

Chapter 3

261

would be equal to


(s, i , )

(a) i (a) |a|s 2 d a.

1

All we need do is show that and i are equivalent, so that

(s, i , ) = (s, , ).
A character is one of the i if and only if is trivial on (F )2 and

F /(F )2

This integral is equal to

F (a) (a) da = 0.


(g)(g)u, u
 v,
(g)
v dg
GF /ZF

which equals


 (g)u, u
 v,
(g)
v dg.
GF /ZF

The integral does not change if is replaced by . Thus the Schur orthogonality relations imply
that it is non-zero only if and are equivalent.
1
If belongs to S0 (A) the functions (g) |detg|s+ 2 belongs to HF and we can form the operator

(g) |detg|s+ 2 (g) d g.

1

T (s, ) =
GF

If has the form of the previous lemma the functional equation may be written as

T (1 s,  ) = (s, , ) T (s, ).
Lemma 13.1.7 Given a non-zero w in V , the set of all u in V such that for some of the form

(g) = (detg) v,

(g)
v |detg|1
F
the vector T (s, )w is of the form ebs u is a set that spans V .
If the function is of this form so is the function  (g) = (hg) and

T (s,  )w = |deth|(s+ 2 ) (h1 ) T (s, ) w

1

Since is irreducible we need only show that there is at least one non-zero vector in the set under
consideration. Moreover there is an r such that rF is unitary and we may as well suppose that
itself is unitary. Let (u, v) be a positive invariant form on V .
Choose v = w and v so that u, v = (u, w) for all u. Let be the characteristic function of UF .
Then



(g) = w, (g)w

if |detg| = 1 and is 0 otherwise. If


H = {g GF  |detg| = 1}


then

T (s, )w =


w, (g)w (g)w d g

is independent of s and is non-zero because

T (s, )w, w =



 (g)w, w 2 d g.

The fourth assertion follows immediately and the fifth will now be a consequence of the following
lemma.

Chapter 3

262

Lemma 13.1.8 Suppose belongs to S(A) and belongs to S 0 (A). There is a vertical strip in which

the integrals





and

(h)
v |detg|s+ 2 |deth| 2 s d g d h
(g)  (h) (g)v,
1

 (g) (h)  1(g)v,

(h1 )
v  |detg| 2 s |deth|s+ 2 d g d h
3

exist and are equal.

A little juggling shows that there is no harm in supposing that the quasi-character defined by

a 0
0 a


= (a) I

is a character. Fix v and v. Let C be a compact subset of GF which contains the support of and  .
The set


{
(h)
v  h C}

is finite. Thus there is a compact set in GF such that for any h in C the function

g (g)v,
(h)
v
has its support in ZF C  . Moreover these functions are uniformly bounded. The first integral is
therefore absolutely convergent for Re s > 12 . The second is convergent for Re s < 32 .
If 12 < Re s < 32 the first integral is equal to

(h) |deth|

3
2 s




1
(g)(g)v,
(h)
v |detg|s+ 2 d g d h.




1
(gh)(g)v, v |detg|s+ 2 d g d h.

Replacing g by hg we obtain

(h) |deth|

If we take the additive Haar measure to be dh = |deth|2 d h this may be written as

s+ 12

(g)v, v |detg|




(hg)  (h) dh d g.

The second integral is

s+ 12

(h) |deth|




3
1 (h)
v |detg| 2 s d g d h.
 (g) 1 (g)v,

After a change of variables this becomes




(g)v, v |detg|

3
2 s




(gh) (h) dh d g.


Chapter 3

263

Replacing g by g1 we obtain

s+ 12

(g)v, v |detg|

Since

is equal to

|detg|2

|detg|


 (g 1 h) (h) dh d g.

(hg)  (h) dh


 (g 1 h) (h) dh

the lemma follows.

The theorem is now proved when is absolutely cuspidal. Suppose that it is a constituent of
= (1 , 2 ). In this case the field may be archimedean. Although is not necessarily irreducible it is
1
admissible and its matrix coefficients are defined. The contragredient representation  is (1
1 , 2 )
1

and the space of is B(1 , 2 ) while that of  is B(1

1 , 2 ). If f belongs to B(1 , 2 ) and f belongs
1
1
to B(1 , 2 ) then


 (g)f, f =

dk
f (kg) f(k)
K

and

f, (g)f =

f (k) f(kg)
dk
K

if K is the standard maximal compact subgroup of GF .

If we set

L(s, ) = L(s, 1 ) L(s, 2 )

1
L(s, ) = L(s, 1
1 ) L(s, 2 )

and

(s, , ) = (s, 1 , ) (s, 2 , )

the theorem may be formulated for the representation . We prove it first for and then for the
irreducible constituents of .
We use a method of R. Godement. If belongs to S(A) then for brevity the function x (gxh)
which also belongs to S(A) will be denoted by hg . Also let


(a1 , a2 ) =

a1
0

x
a2


dx

where dx is the measure which is self dual with respect to . belongs to S(F 2 ). The map
of S(A) into S(F 2 ) is certainly continuous.
We are now going to define a kernel K (h, g, s) on K K . We set

K (e, e, s) = Z(1 sF , 2 sF , ).
Recall that the right-hand side is



(a1 , a2 ) 1 (a1 ) |a1 |s 2 (a2 ) |a2 |s d a1 d a2 .

Chapter 3

264

In general

K (h, g, s) = Kgh1 (e, e, s).

We also set

 (e, e, s) = Z(1 s , 1 s , )
K
F
F
1
2

and

 (h, g, s) = K
 gh1 (e, e, s).
K

The kernels are defined for Re s sufficiently large and are continuous in h, g , and s and, for fixed h and
g , holomorphic in s.
We now make some formal computations which will be justified by the result. The expression
s+ 12

Z(F

, , f, f) is equal to





dk |detg|s+ 12 d g
f (kg) f(k)

(g)
GF

which is

f(k)




1
(g) f (kg) |detg|s+ 2 d g dk.
GF

Changing variables in the inner integral we obtain


f(k)




1
(k 1 g) f (g) |deth|s+ 2 d g dk.
GF

Using the Iwasawa decomposition to evaluate the integral over GF we see that this is equal to


K (k1 , k2 , s) f (k2) f(k1 ) dk1 dk2 .
KK

Since we could have put in absolute values and obtained a similar result all the integals are convergent
and equal for Re s sufficiently large. A similar computation shows that
s+ 12

Z(F
is equal to

, , f, f)

 (k1 , k2 , s) f (k1 ) f(k2 ) dk1 dk2

K

KK

if Re s is large enough.
s+ 12

If is an elementary idempotent such that ()f = f and ()f = f then Z(F

not changed if is replaced by


1 (g) =

, , f, f) is

(k1gk21 ) (k1 ) (k2 ) dk1 dk2 .

Thus, at least when proving the second and third assertions, we may suppose that is K -finite on both
sides and, in fact, transforms according to a fixed finite set of irreducible representations of K . Then,
as s varies, the functions

K (k1 , k2 , s)

Chapter 3

265

stay in some fixed finite dimensional space U of continuous functions on K K . The map


F (k1, k2 ) f (k2) f(k1 ) dk1 dk2

is a linear form on this space and we can find g1 , , gn and h1 , , hn in K such that it can be
represented in the form

n


i F (gi , hi ).

i=1

Thus

s+ 12

Z(F

, , f, f) =

i K (gi , hi , s).

Thus to prove the second and third assertions we need only show that for each g and h in K the
function

K (g, h, s)
L(s, )

is entire and K (g, h, s) itself is bounded at infinity in vertical strips. There is certainly no harm in
supposing that g = h = e so that

K (e, e, s) = Z(1 sF , 2 sF , )
Thus the desired facts are consequences of the results obtained in paragraphs 3, 5, and 6 when proving
the local functional equation for constituents of . Replacing by its contragredient representation we
s+ 1

obtain the same results for Z(F 2 , , f, f).

To prove the functional equation we have to see what happens to the Fourier transform when we
pass from the function to to 1 . The answer is simple:

1 (g)


=

 (k1 gk21 ) (k1 ) (k2) dk1 dk2 .

Thus in proving the functional equation we may suppose that is K -finite on both sides. We may
s+ 12

also suppose that if F (k1 , k2 ) is in U so is F  (k1 , k2 ) = F (k2 , k1 ). Then Z(F


  (hi , gi , s). To prove the functional equation we have to show that
i K

,  , f, f) =

  (h, g, 1 s)
K
K (g, h, s)
= (s, , )
L(1 s, )
L(s, )
for any h and g in K . Since the Fourier transform of gh1 is h g it will be enough to do this for
h = g = e. Then the equality reduces to
1s
1 1s
Z(1
Z(1 sF , 2 sF , )
1 F , 2 F ,  )
= (s, , )
L(1 s, )
L(s, )

and is a result of the facts proved in the first chapter and the next lemma.

Chapter 3

266

Lemma 13.2.1 The Fourier transform of the function is the function  .

The value of  at

is

x y
z t


(x + z + y + t) dx dy dz dt

if dx, dy , dz , and dt are self-dual with respect to . Thus  (, ) is equal to

 

x y
z t




(x + t) (z) dx dy dz dt d

Applying the Fourier inversion formula to the pair of variables and z we see that this is equal to

x y
0 t


(x + t) dx dy dt

which is the value of the Fourier transform of at (, ).

The theorem, with the exception of the fourth assertion, is now proved for the representation .
We will now deduce it, with the exception of the fourth assertion, for the constituents of . We will
return to the fourth assertion later.
If is a constituent of either = (1 , 2 ) or = (1 , 2 ). In the first case there is nothing left
to prove. In the second only the third assertion remains in doubt. If F is the complex field even it is
alright because we can always find another pair of quasi-characters 1 and 2 such that = (1 , 2 ).
We ignore this case and suppose that F is real or non-archimedean.
1

First take F to be non-archimedean. We may suppose that 1 and 2 are the form 1 = F2 and
1

1
2 = F 2 . The one-dimensional representation g (detg) is contained in  = (1
1 , 2 ) and
acts on the function g (detg). The matrix elements for are the functions

g  (g)f, f = (g)f, f
1
where f belongs to B(1
1 , 2 ) and


f (k) (detk) dk = 0.
K

For such an f there is an elementary idempotent such that ()f = f while


(k) dk = 0
K
s+ 12

The value of Z(F

, , f, f) is not changed if we replace by



(gh1 ) (h) dh.

1 (g) =
K

Chapter 3

267

Lemma 13.2.2 If g 1 and g2 belong to GF then



1 g 1

x
0

y
0


g2

dx dy = 0.

It will be enough to prove this when g1 is the identity. Let

(x, y) = 1

x y
0 0


.

If g1 is the identity then, after a change of variables, the integral becomes

1



|detg2 |

(x, y) dx dy

so that we can also assume g2 is the identity. Then the integral equals

 

x y
0 0

 

k dx dy (k 1 ) dk.

Changing variables as before we see that the inner integral does not depend on K . Since

(k 1 ) dk = 0

the lemma follows.

To establish the third assertion for the representation all we need do is show that for any g and
h in K the function

K (g, h, s)
L(s, )

is entire provided



 



x y
g2 dx dy = 0
g1
0 0

for all g1 and g2 in GF . As usual we need only consider the case that g = h = e. Since

(x, 0) dx = 0
and

K (e, e, s) = Z(1 sF , 2 sF , )
we need only refer to Corollary 3.7.
If F is the field of real numbers the proof is going to be basically the same but a little more
2p+1m
(sgn x)m , where p is a non-negative integer
complicated. We may assume that 1 1
2 (x) = |x|
and m is 0 or 1, and that acts on BS (1 , 2 ). The restriction of to SO(2, R) contains only those
representations n for which n 1 m (mod 2) and |n| 2p + 1 m. Let n be the elementary
idempotent corresponding to the representation n . As before we may suppose that

(xk 1 ) n (k) dk = 0
SO(2,R)

if n does not occur in the restriction of to SO(2, R).

(13.2.3)

Chapter 3

268

Lemma 13.2.4 If satises (13.2.3), if g 1 and g2 belong to GF , and = g1 g2 then

j
(x, 0) dx = 0
y j

xi
R

if i 0, j 0 and i + j = 2p m.
We may assume that g2 = e. If = let


xi

L() =
R

j
(x, 0) dx
y j

and let

F (g) = L(g).
We have to show that, under the hypothesis of the lemma, F (g) = 0 for all g . However F is defined
for all in S(A) and if is replaced by h the function F is replaced by F (gh). Thus to establish the
identity



 

ai z
0 a2

= 1 (a2 ) 2 (a2 ) F (g)

1
where 1 (a1 ) = ai
and 2 (a2 ) = a2 |a2 |1 , we need only establish it for g = e.
1 |a1 |
Let



j

a1
0

h=

Then

x u
0 y



z
a2

a1 y xz + a2 u
0
a2 y


.

If = and 1 = h then 1 (x, y), which is given by

is equal to
1

|a2 |
Moreover F (h) is equal to

a1 x xz + a2 u
0
a2 y

a1 x u
0 a2 y

xi




du,

du = |a2 |1 (a1 x, a2 y).

j 1
(x, 0) dx
y j

which equals
1 j
ai
a2 |a2 |1
1 |a1 |

as required.
Finally if

g=


xi

j
(x, 0) dx
y j

Chapter 3

269

and = g then F (g) is equal to


xi

and

(x, y) =

j
(x, 0) dv
y j
x + u x + u
y
y


du.

Since we can interchange the orders of differentiation and integration,


j

j
n jn
(x,
0)
=

n (x + u, x + u) du
n
y j
n=0
where

j
n (x, y) =
n jn

x y
0 0



and the number n are constants. Thus F (g) is a linear combination of the functions


n jn

xi n (x + u, x + u) dx du.

If = 0 we may substitute x

u
2

for x to obtain




u 
(x u)i
n x, x +
dx du
2
2

n jn

where = detg . Substituting u

n jn

for u we obtain

 
x+

u 

u i 
dx du.
n x,
x

2
2

After one more change of variables this becomes

i

||


n jn

(x u)i n (x, u) dx du.

In conclusion F (g) is a function of the form

F



= i ||1 P (, , , )

where P is a polynomial.
Thus the right translates of F by the elements of GF span a finite-dimensional space. In particular it
1

is O(2, R) finite and if 1 = 1 F2 while 2 = 2 F 2 it lies in a finite-dimensional invariant subspace of

1
B(1 , 2 ). Thus it lies in BF (1 , 2 ). Since 1 2 = 1
1 2 no representation of SO(2, R) occurring


in (1 , 2 ) can occur in = (1 , 2 ). If F is not zero then for at least one such representation n


F1 (g) =
SO(2,R)

f (gk 1 ) n (k) dk

Chapter 3

270

is not identically 0. But F1 is the result of replacing by

(xk 1 ) n (k) dk

1 (x) =
SO(2,R)

in the definition of F . In particular if satisfies the conditions of the lemma both 1 and F1 are zero.
Therefore F is also zero and the lemma is proved.
The third assertion can now be verified as in the non-archimedean case by appealing to Lemma 5.17.
The fourth has still to be proved in general.
If F is the real field let S1 (A) be the space of functions of the form

a b
c d





= exp (a2 + b2 + c2 + d2 ) P (a, b, c, d)

where P is a polynomial. If F is the complex field S1 (A) will be the space of functions of the form

a b
c d





P (a, a

= exp (a
a + bb + c
c + dd)
, b, b, c, c, d, d)

where P is again a polynomial. If F is non-archimedean S1 (A) will just be S(A). The space S1 (F 2 ) is
defined in a similar manner.
Lemma 13.2.5 Suppose belongs to S 1 (F 2 ). Then there is a in S1 (A) such that

K (e, e, s) = Z(1 sF , 2 sF , )
1
and f1 , , fn in B(1 , 2 ) together with f1 , , fn in B(1
1 , 2 ) such that
n 

i=1

K (h, g, s) fi (g) fi (h) dg dh = K (e, e, s).

KK

Since there is a in S1 (F 2 ) such that

Z(1 sF , 2 sF , ) = aebs L(s, )

this lemma will imply the fourth assertion for the representation .
Given the existence of such that = and therefore

K (e, e, s) = Z(1 sF , 2 sF , )
is a triviality and we worry only about the existence of f1 , , fn and f1 , , fn .
It is easily seen that if




and

a1
0

x
a2

b1
0

y
b2

Chapter 3

271

belong to K then

K

is equal to

a1
0

x
a2

h,

b1
0

y
b2


g, s

1
1 (a1 ) 2 (a2 ) 1
1 (b1 ) 2 (b2 ) K (h, g).

Also

K (hh1 , gg1 , s) = Kg1 h1 (h, g, s).

1

Since belongs to S1 (A) it is K -finite on the left and right. Thus there is a finite set S of irreducible
representations of K such that if U1 is the space of functions F on K which satisfy

F

x
a2

a1
0

 
h = 1 (a1 ) 2 (a2 ) F (h)

for all

a1
0

x
a2

in K and can be written as a linear combination of matrix elements of representations in S and U2 is

the space of functions F  on K which satisfy

a1
0

x
a2

 
1

h = 1
1 (a1 ) 2 (a2 ) F (h)

and can be written as a linear combination of matrix elements of representations in S then, for every s,
the function

(h, g) K (h, g, s)

belongs to the finite dimensional space U spanned by functions of the form (h, g) F (h) F  (g) with
F in U1 and F  in U2 .
Choose F1 , , Fn and F1 , , Fn so that for every function F in U

F (e, e) =

n

i=1

F (h, g) Fi (h) Fi (g) dh dg.

i
KK

1

Since Fi is the restriction to K of an element of B(1
1 , 2 ) while Fi is the restriction to K of an
element of B(1 , 2 ) the lemma follows.
Unfortunately this lemma does not prove the fourth assertion in all cases. Moreover there is a
supplementary condition to be verified.

Chapter 3

272

Lemma 13.2.6 Suppose F is non-archimedean and is of the form = ( 1 , 2 ) with 1 and 2

unramied. Suppose is the characteristic function of M (2, OF ) in M (2, F ). If v and v are

invariant under K = GL(2, OF ) and if


d g = 1
K

then

s+ 12

, , v, v) = L(s, ) v, v.

Z(F

1
Suppose f belongs to B(1 , 2 ) and is identically 1 on K while f belongs to B(1
1 , 2 ) and is
identically 1 on K . Then


f, f =

dk = 1
f (k) f(k)
K

and if = (1 , 2 ) we are trying to show that

s+ 12

Z(F
The left side is equal to

, , f, f) = L(s, ).


dh dg.
K (h, g, s) f (h) f(g)
KK

Since is invariant on both side under K this is equal to

K (e, e, s) = Z(1 sF , 2 sF , )

if

(x, y) =

x z
0 y


dz.

Since we have so normalized the Haar measure on GF that

 

F (g) dg =
GF

a1
0

0
a2



1 x
0 1

 

k d a1 d a2 dx dk

where dk is the normalized measure on K , dx is the measure on F which assigns the measure 1 to OF ,
and d a is the measure on F which assigns the measure 1 to UF the function is the characteristic
function of OF OF and

Z(1 sF , 2 sF , ) = L(s, 1 ) L(s, 2 )

as required.
This lemma incidentally proves the fourth assertion for the one-dimensional representation g
(detg) if is unramified. If is ramified and corresponds to then = (1 , 2 ) if 1 (a) =
1
1
(a) |a| 2 and 2 (a) = (a) |a| 2 . Thus L(s, ) = 1. If is the restriction of the function 1 to K
then


d g

Z(, , v, v) = v, v
K

and the fourth assertion is verified in this case.

Chapter 3

273

Take 1 and 2 of this form with possibly unramified and suppose that = (1 , 2 ). Suppose
first that is unramified. Let 0 be the characteristic function of OF in F and let

1 (x) = 0 (x) |C 1 | 0 (C 1x).

It has OF for support. Set



a b
c d

= 1 (a) 0 (b) 0 (c) 0 (d).

It has M (2, OF ) for support and depends only on the residues of a, b, c, and d modulo pF . If


K 1 = {k K  k e(mod p)}
then K (h, g, s) depends only on the cosets of h and g modulo K 1 . Also

 
1 x
K e, w
,s = 0
0 1
if x is in OF . To see this we observe first that if


1 x
1 (g) = gw
0 1
then 1 (a1 , a2 ) is equal to

y
a2

which equals

a1 xy
a2 x


dy


1 (y) 0 (a1 xy) dy.

0 (a2 ) 0 (a2 x)
OF

Since x is in OF the function 0 (a1 xy) equals 0 (a1 ) for y in OF and this expression is 0 because


1 (y) dy = 0.
OF

We choose f in BS (1 , 2 ) so that f (gk) = f (g) if k belongs to K1 , f (e) = 1, and

f (e) +

xOF /p


1 x
f w
= 0.
0 1

We choose f in B(1 , 2 ) so that f(gk) = f(g) if k belongs to K 1 , f(e) = 1, and


1 x

f w
=0
0 1
if x belongs to OF . Then


K (h, g, s) f(h) f (g) dh dg
KK

Chapter 3

274

is equal to


K (e, g, s) f (g) dg = K (e, e, s)
K

which equals

Z(1 sF , 2 sF , ).
Moreover

(a1 , a2 ) = 1 (a) 0 (a2 )

so that, as we saw when proving Corollary 3.7, L(s, ) is a constant times Z(1 sF , 2 sF , ).
s+1/2
, , v, v) is equal to
If is ramified L(s, ) = 1. If has support in K then Z(F


(k)(k)v, v dk
K

and we can certainly choose v , v and so that this is not 0.

We are not yet finished. We have yet to take care of the representations not covered by Lemma 13.2.5
when the field is archimedean. If F is the complex field we have only the finite dimensional representations to consider. There is a pair of characters 1 and 2 such that is realized on the subspace
p q
Bf (1 , 2 ) of B(1 , 2 ). There will be positive integers p and q such that 1 1
z . The
2 (z) = z
representations = |qp| of SU (2, C) which is of degree |q p| + 1 is contained in the restriction
of to SU (2, C). In particular Bf (1 , 2 ) contains all functions f in B(1 , 2 ) whose restrictions to
SU (2, C) satisfy



 

a1
0

0
a2

= 1 (a1 ) 2 (a2 ) f (k)

and transform on the right according to .

We are going to use an argument like that used to prove Lemma 13.2.5. Suppose we can find a
function in S1 (A) such that

Z(1 sF , 2 sF , )

differs from L(s, ) by an exponential factor and such that transforms on the right under SU (2, C)
according to the representation . Then K (h, g, s) will satisfy the same conditions as in Lemma 13.2.5.
Moreover the functions F  in the space we called U2 can be supposed to transform on the right under
SU (2, C) according to so that the functions Fi will correspond to functions fi in Bf (1 , 2 ). Then

KK

s+
K (h, g, s) fi (h) fi (g) dh dg = Z(F 2 , , fi , fi )

is equal to
s+ 12

Z(F

, , vi , vi )

if vi = fi and vi is the restriction of fi , regarded as a linear functional, to Bf (1 , 2 ).

There are four possible ways of writing 1 and 2 .
(i) 1 (z) = z m1 (z z)s1 , 2 (z) = z m2 (z z)s2 , m1 m2 = q p.
(ii) 1 (z) = z m1 (z z)s1 , 2 (z) = zm2 (z z)s2 , m1 + m2 = q p.
(iii) 1 (z) = zm1 (z z)s1 , 2 (z) = z m2 (z z)s2 , m1 m2 = q p.
(iv) 1 (z) = zm1 (z z)s1 , 2 (z) = zm2 (z z)s2 , m2 m1 = q p.

Chapter 3

275

In all four cases m1 and m2 are to be non-negative integers. is the product of exp((a
a + bb +

c
c + dd)) and a polynomial. We write down the polynomial in all four cases and leave the verifications
to the reader.
m1 m2 (
ad b
c)m2 .
(i.a) m1 m2 : a
m1 n2 m1

ad b
c) d
.
(i.b) m1 m2 : (
m1 d m2 .
(ii) a
(iii) am1 dm2 .
(iv.a) m1 m2 : am1 m2 (ad bc)m2 .
(iv.b) m2 m1 : (ad bc)m1 dm2 m1 .
For the real field the situation is similar. Suppose first that = (1 , 2 ) is finite dimensional. If
1 2 (1) = 1 then contains the trivial representation of SO(2, R) and if 1 2 (1) = 1 it contains
the representation




cos sin
sin cos

ei .


We list the four possibilities for 1 and 2 and the polynomial P by which exp (a2 + b2 + c2 + d2 )
is to be multiplied to obtain .
(i) 1 (1) = 2 (1) = 1: P (a, b, c, d) = 1.
(ii) 1 (1) = 2 (1) = 1: P (a, b, c, d) = ad bc.
(iii) 1 (1) = 1, 2 (1) = 1: P (a, b, c, d) = c id.
(iv) 1 (1), 2 (1) = 1: P (a, b, c, d) = a ib.
Only the special representations remain to be considered. We may suppose that = (1 , 2 )
q
q
where 1 and 2 are of the form 1 (x) = |x|r+ 2 and 2 (x) = |x|r 2 (sgn t)m with q = 2p + 1 m and
with p a non-negative integer. Moreover m is 0 or 1. L(s, ) differs from

s + r + q  s + r +
2

2
2

q
2

+ 1

by an exponential as does

Z(1 sF , 2 sF , )
if

(a1 , a2 ) = e(a1 +a2 ) aq+1

.
2
2

Since the representation of q+1 occurs in the restriction of to SO(2, R) we may take

a b
c d





= exp (a2 + b2 + c2 + d2 ) (c + id)q+1 .

Chapter 3

276

14. Automorphic forms and quaternion algebras. Let F be a global field and let M  be a quaternion
algebra over F . The multiplicative group G of M  may be regarded as an algebraic group over F .
In the ninth paragraph we have introduced the group GA and the Hecke algebra H . A continuous
function on GF \ GA is said to be an automorphic form if for every elementary idempotent in H
the space

{(f )  f H }
is finite dimensional.
If is an automorphic form it is ZA finite on the left if Z  is the centre of G . Let A be the space
of automorphic forms on GA and if is a quasi-character of F \ I let A () be the space of in A
for which  (ag) = (a)  (g) for all a in ZA which, for convenience, we identify with I . The first
assertion of the following lemma is easily proved by the methods of the eighth paragraph. The second
is proved by the methods of the tenth. The proof is however a little simpler because GF ZA \ GA is
compact. Since, at least in the case of number fields, the proof ultimately rests on general facts from
the theory of automorphic forms nothing is gained by going into details.
Lemma 14.1 (i) If an irreducible admissible representation of H  is a constituent of A then for

some it is a constituent of A ().

(ii) The space A () is the direct sum of subspaces irreducible and invariant under H . The representation of H on each of these subspaces is admissible and no representation occurs more
than a nite number of times in A ().

Now we have to remind ourselves of some facts whose proofs are scattered throughout the
previous paragraphs. Suppose = v v is an irreducible admissible representation of H . For each v
the representation v of Hv is irreducible and admissible. Suppose is a non-trivial additive character
v ), and (s, v , v ). If uv is in
of F \ A and v is its restriction to Fv . We have defined L(s, v ), L(s,
v in the space of
v we have set
the space of v and u


 s+ 1
Z F 2 v , , uv , u
v
equal to

We know that

is entire and equals

(g) v (g)uv , u
v  |(g)|s+ 2 d g.
1

GF
v



Z s+ 12 , , u , u

v
v v
F
hv (s, v , v )

L(s, v )

 1 s
Z F2
v ,  , uv , uv
.
L(1 s,
v )

The factor hv is 1 of GFv is isomorphic to GL(2, Fv ) and is 1 otherwise. The case that GFv is
isomorphic to GL(2, Fv ) was treated in the previous paragraph. The other cases were treated in the
fourth and fifth paragraphs.

Chapter 3

277

Theorem 14.2 Suppose is a constituent of the space of automorphic forms on G A . The innite

products

L(s, v )
v

and
L(s,
v )
v

are absolutely convergent for Re s suciently large. The functions L(s, ) and L(s,
) dened by
them can be analytically continued to the whole complex plane as meromorphic functions. If F is a
number eld they will have only a nite number of poles and will be bounded at innity in vertical
strips of nite width. If
(s, v , v )
(s, ) =
v

the functional equation

L(s, ) = (s, ) L(1 s,
)
will be satised.
We may suppose that acts on the subspace V of A (). Let be a non-zero function in V . For
almost all v the algebra Mv = M  F Fv is split and GFv = Gv is isomorphic to GL(2, Fv ). Moreover

which
for almost all such v , say for all v not in S , is an eigenfunction of the elements of Hv = HF
v

are invariant on both sides under translations by the elements of Kv . Thus if f is such an element and
(g) = 0 the corresponding eigenvalue v (f ) is

v (f ) = (g)


(gh) f (h) dh.
Gv

To prove the absolute convergence of the infinite products we have only to refer to Lemma 3.11 as in
the proof of Theorem 11.1.
contragredient to can be defined. If = v acts on V = u0v Vv then
The representation

=
v acts on V = u0v Vv where u
0v is, for almost all v, fixed by Kv and satisfies u0v , u0v  = 1. The
pairing between V and V is defined by
uv ,
uv  =

uv , u
v .
v

Almost all terms in the product are equal to 1. If u is in V and u

is in V the matrix element (g)u, u


can also be introduced. If f is in H


(f )u, u
 =

GA

f (g) (g)u, u
 d g.

If F (g) is a linear combination of such matrix elements and belongs to the Schwartz space on AA we
set*


 s+ 1

Z F 2 , , F =

(g) F (g) |(g)|s+ 2 d g.

1

GA

* Unfortunately the symbol F plays two quite different roles on this page!

Chapter 3

278

The function F(g) = F (g 1 ) is a linear combination of matrix coefficients for the representation
. We
set


 s+ 1

Z F 2 , , F =

(g) F (g) |(g)|s+ 2 d g.

1

GA

Before stating thenext lemma we observe that if is a quasi-character of F \I the one dimensional
representation g (g) is certainly a constituent of A .
s 12

s 1

, , F ) and Z(F 2 , , F)

are absolutely convergent for Re s large enough. The two functions can be analytically continued
to the whole complex plane

as meromorphic functions with only a nite number of poles. If is
not of the form g (g) they are entire. If F is a number eld they are bounded at innity in
vertical strips of nite width. In all cases they satisfy the functional equation

Lemma 14.2.1 If is a constituent of A the integrals dening Z(F

 s+ 1

 3 s

Z F 2 , , F = Z F2 ,  , F
if  is the Fourier transform of .
There is no harm in assuming that F is of the form

(gv )uv , uv  =

F (g) =
v

Fv (gv )
v

and that is of the form

(x) =

v (xv )
v

where, for almost all v , v is the characteristic function of M (2, Ov ). Recall that for almost all v we
have fixed an isomorphism v of Mv with M (2, Fv ).
We know that each of the integrals

v (gv ) Fv (gv ) |(gv )|s+ 2 d gv

1

Gv

converges absolutely for Re s sufficiently large. Let S be a finite set of primes which contains all
v is u
0v , v is the characteristic
archimedean primes such that outside of S the vector uv is u0v , the vector u
function of M (2, Ov ), and v = v (v , v ) where v and v are unramified. Let v = v (|v |, |v |). If
v is not in S the integral


(gv ) Fv (gv ) |(gv )|s+ 2 d gv = 1

1

Kv

and if = Re s

|v (gv )| |Fv (gv )| |(gv )|+ 2 d gv

1

Gv

is, as we see if we regard v as acting on B(b , v ), at most

v (gv ) v (gv ) fv , fv  |(gv )|+ 2 d gv

1

Gv

Chapter 3

279

if fv and fv are the unique Kv -invariant elements in B(|v |, |v |) and B(|v |1 , |v |1 ) which take the
value 1 at the identity. We suppose that the total measure of Kv is 1 so that fv , fv  = 1. According to
Lemma 13.2.6 the integral is equal to L(, v ). Since

L(, v )
vS
s+ 12

is absolutely convergent for sufficiently large the integral defining Z(F

to
1



, , F ) is also and is equal

s+

Z Fv 2 v , v , uv , uv

and to

L(s, )

(s, v , uv , u
v ).
v
s+ 12

Notice that (s, v , uv , u

v ) is identically 1 for almost all v. Z(F
fashion. If we take to be the trivial representation we see that

, , F ) may be treated in a similar

(g) |(g)|s+ 2 d g
1

GA

is absolutely convergent for Re s sufficiently large.

It will be enough to prove the remaining assertions of the lemma when is a character. We may
also assume that if is of the form (a) = |a|r then r = 0. We have identified V with a subspace of

A (). We may take V to be {  V }. To see this observe that this space is invariant under H and
that


1 , 2  =

GF ZA \GA

1 (g) 2 (g) dg

is a non-degenerate bilinear form. Here 1 belongs to V and

2 belongs to V . The remaining assertions
need only be verified for functions of the form


F (g) =
with in V and
in V .

s+ 12

For such an F the function Z(F




(g)

GF ZA \GA

(hg) (h)

dh

, , F ) is equal to

1
(hg) (h)

dh |(g)|s+ 2 d g.

Since and
are bounded this double integral converges absolutely for Re s sufficiently large.
We first change variables by substituting h1 g for g . The integration with respect to g can then be
carried out in three steps. We first sum over GF  , then we integrate over ZF \ ZA which we identify
with F \ I , and finally we integrate over GF ZA \ GA . Thus if K (h1 , h2 , s) is
s+ 12
|(h1
1 |

s+ 12

|(h2 )|


F \I G
F

2s+1
(h1
d a
1 ah2 ) (a) |a|

Chapter 3

280
s+ 12

the function Z(F

, , F ) is equal to

(h2 ) (h
1 ) K (h1 , h2 , s) dh1 dh2 .

The integrations with respect to h1 and h2 are taken over GF ZA \ GA . A similar result is of course valid
s+ 12

for Z(F

 (h1 , h2 , s) is
, , F). If K
s+ 12
|(h1
1 )|

s+ 12

then Z(F

s+ 12

|(h2 )|


F \I G
F

, , F) is equal to


 (h1 , h2 , s) dh1 dh2 .

2) K
(h2 ) (h

We first study

(s, ) =
F \I

and

1
(h1
(a) |a|2s+1 d a
1 ah2 )

 ) =
(s,

=0


F \I

(a) (a) |a|2s+1

d a
F

(a) 1 (a) |a|2s+1

d a.
F

=0

GF

The sums are taken over

the set of non-zero elements of M  . Choose two non-negative continuous
functions F0 and F1 on the positive real numbers so that F0 (t) + F1 (t) = 1, F1 (t) = F0 (t1 ), and so
that F0 vanishes near zero while F1 vanishes near infinity. If


i (s, ) =

F \I =0

(a ) (a) |a|2s+1 Fi (|a|) d a

we have

(s, ) = 0 (s, ) + 1 (s, ).

In the same way we may write

 ) = 0 (s, ) + 1 (s, )
(s,

0 (s, ) and 0 (s, ) are entire functions of s which are bounded in vertical strips.
Applying the Poisson formula we obtain
(0) +




 (0) +
(a) = |a|4
 (a1 ) .
F

=0

=0

Thus, for Re s sufficiently large, 1 (s, ) is equal to the sum of


F \I


=0

 (a1 ) (a) |a|2s3 F1 (|a|) d a,

Chapter 3

281

which, after the substitution of a1 for a, is seen to equal 0 (1 s,  ), and


F \I

 

(0) |a|4 (0) (a) |a|2s+1 F1 (|a|) d a.


Thus if

(s) =
F \I

|a|s (a) F1 (|a|) d a

the function (s, ) is equal to

0 (s, ) + 0 (1 s,  ) +  (0) (2s 3) (0) (2s + 1).

 ). The function
A similar result is valid for (s,
0 (s, ) + 0 (1 s,  )
is entire and bounded in vertical strips and does not change when s and are replaced by 1 s and
 .
If is not of the form (a) = |a|r the function (s) vanishes identically. If is trivial and I0 is the
group of id`eles of norm 1


(s) =
F \I

|a|2s+1 Fj (|a|) d a.

It is shown in [10] that this function is meromorphic in the whole plane and satisfies (s) + (s) = 0.
If F is a number field, its only pole is at s = 0 and is simple. Moreover it is bounded at infinity in
vertical strips of finite width. If F is a function field its poles are simple and lie at the zeros of 1 qs .
Here q is the number of elements in the field of constants.
 s,  ). If hg is the function
Thus (s, ) is meromorphic in the whole plane and is equal to (1
x (gxh) then
s+ 2
K (h1 , h2 , s) = |(h1
|(h2 )|s+ 2 (s, h2 h1
1 )|
1 )
1

while

 h2 h1 ).
 (h1 , h2 , s) = |(h1 )|s+ 12 |(h2 )|s+ 12 (s,
K
1
1

Since the Fourier transform of h2 h1

1 is

|(h2 )|2 |(h1 )|2 h1  h1

2
we have

  (h2 , h1 , s).
K (h1 , h2 , s) = K

The functional equation of the lemma follows. So do the other assertions except the fact that the
s+ 1

s+ 1

2
2

functions
 Z(
 F , , F ) and Z(F , , F ) are entire when is trivial and is not of the form
g (g) . In this case the functions and are orthogonal to the constant functions and the
 (h1 , h2 , s) may be replaced by
kernels K (h1 , h2 , s) and K

  (h1 , h2 , s) + (0) (2s + 1)  (0) (2s 3)

K  (h1 , h2 , s) = K

Chapter 3
and

282

  (h1 , h2 , s) +  (0) (2s + 1) (0) (2s 3).

   (h1 , h2 , s) = K
K

The functional equation of the kernels is not destroyed but the poles disappear.
The theorem follows easily from the lemma. In fact suppose that the finite set of places S is so
chosen that for v not in S

(s, 0v , u0v , u0v ) = 1

if 0v is the characteristic function of M (2, Ov ). If v is in S choose v , uiv , u
iv , 1 i nv , so that
nv


(s, iv , uiv , u
iv ) = ebvs

i=1

where bv is real. If is a function from S to the integers and, for each v in S , 1 (v) nv , set

(v)
(gv )
v

(g) =



vS

and set




v (gv )u0v , u
0v  .
vS

vS

Then

vS

v (gv )u(v)
, u(v)

v
v

F (g) =

0v (gv )

  s+ 1

Z F 2 , , F = cbs L(s, )

where b is real. The required analytic properties of L(s, ) follow immediately.

v so
To prove the functional equation choose for each v the function v and the vectors uv and u
that

(s, v , uv , u
v )
is not identically 0. We may suppose that, for almost all v , v = 0v , uv = u0v , and u
v = u
0v . Let

(g) =

v (gv )
v

and let

v (gv )uv, uv .

F (g) =
v

Then

 s+ 1

Z F 2 , , F = L(s, )

(s, v , uv , uv )
v

and

 3 s
Z F2 ,  , F ) = L(1 s,
)

 s,  , uv , u
v ).
(1
v
v

Since

 s,  , uv , uv ) = hv (s, v , v ) (s, v , uv , uv )
(1
v

the functional equation of the lemma implies that


).
hv (s, ) L(1 s,

L(s, ) =
v

Since, by a well-known theorem, the algebra M  is split at an even number of places the product
equals 1.

!
v

hv

Chapter 3

283

Corollary 14.3 If is a constituent of A  which is not of the form g (g) then for any quasi-

) are entire and bounded in vertical

character of F the functions L(s, ) and L(s, 1
strips of nite width. Moreover they satisfy the functional equation
L(s, ) = (s, ) L(1 s, 1
).

We have only to observe that if is a constituent of A then is also.

Now we change the notation slightly and let  = v be an irreducible admissible representation

of H . We want to associate to it a representation = v of H the Hecke algebra of GL(2, A). If Mv is
split then v is just the representation corresponding to v by means of the isomorphism v of GFv and
GFv . If Mv is not split v is the representation v (v ) introduced in the fourth and fifth paragraphs.
In both cases v is defined unambiguously by the following relations

L(s, v v ) = L(s, v v )
L(s, v
v ) = L(s, v
v )
(s, v v , v ) = (s, v v , v )
which holds for all quasi-characters v of Fv .
Applying the previous corollary and Theorem 11.3 we obtain the following theorem.
Theorem 14.4 If  is a constituent of A and v is innite dimensional at any place where M  splits

then is a constituent of A0 .

Some comments on the assumptions are necessary. If  is a constituent of A we can always

find a quasi-character of of F \ I such that  is unitary. If  = v the same is true of the
representations v . In particular if M  splits at v the representation v will not be finite dimensional
unless it is one dimensional.

 Various density theorems probably prevent this from happening unless
 is of the form g (g) . If  is of this form then all but a finite number of the representations v
are one dimensional. But if M  does not split at v the representation v is infinite dimensional. Thus
cannot act on a subspace of A. However it can still be a constituent of A. This is in fact extremely
likely. Since the proof we have in mind involves the theory of Eisenstein series we prefer to leave the
question unsettled for now.

Chapter 3

284

15. Some orthogonality relations. It is of some importance to characterize the range of the map 
from the constituents of A to those of A discussed in the last chapter. In this paragraph we take up the
corresponding local question. Suppose F is a local field and M  is the quaternion algebra over F . GF
is the group of invertible elements of M  . We know how to associate to every irreducible admissible

an irreducible admissible representation = ( ) of HF the Hecke algebra
representation  of HF
of GL(2, F ).
Theorem 15.1 Suppose F is non-archimedean. Then the map  is injective and its image is

the collection of special representations together with the absolutely cuspidal representations.
The proof requires some preparation. We need not distinguish between representations of GF and
or between representations of GF and HF . An irreducible admissible representation of GF is
1
said to be square-integrable if for any two vectors u1 and u2 in the space of and any two vectors u
and u
2 in the space of
the integral

HF


ZF \GF

(g)u1 , u
1  u2 ,
(g)
u2  dg

is absolutely convergent. Since

is equivalent to 1 if

a 0
0 a


= (a) I

this is equivalent to demanding that


ZF \GF

|(g)u1 , u
1 |2 |1 (detg)| dg

be finite for every u1 and u2 .

If is square-integrable and is a quasi-character of F then is square integrable. We can
always choose so that 2 is a character. If is a character choose u0 different from 0 in the space V
of . Then


(u1 , u2 ) =

ZF \GF

(g)u1 , u0  (g)u2, u0  dg

is a positive-definite form on the space V of so that is unitary and square-integrable in the usual
sense.
The Schur orthogonality relations when written in the form


ZF \GF

(g)u1 , u
1  u2 ,
(g)
u2  dg =

1
1  u1 , u2 
u2 , u
d()

are valid not only for representations which are square-integrable in the usual sense but also for
represntations which are square-integrable in our sense. The formal degree d() depends on the choice
of Haar measure. Notice that d( ) = d().
The absolutely cuspidal representations are certainly square-integrable because their matrix elements are compactly support modulo ZF .

Chapter 3

285

Lemma 15.2 The special representations are square-integrable.

1

Suppose = (F2 , F 2 ). Since

 1
1
= F2 , F 2
1

it is enough to show that is square-integrable. If belongs to Bs (F2 , F 2 ) and

belongs to
1

B(F 2 , F2 ) then

f (g) = , (g 1 )

is the most general matrix coefficient of . B(F 2 , F2 ) is the space of locally constant functions on
1

NF AF \ GF and Bs (F2 , F 2 ) is the space of locally constant functions on GF that satisfy

and

x
a2

a1
0

   
 a1 
g =   (g)
a2


1 x
w
dx = 0.
0 1

Since

GF =

ZF K

n0

Cn 0
0 1


K

we can choose the Haar measure on ZF \ GF so that


ZF \GF

is equal to


n0

|f (g)|2 dg

 2
 

n


C
0

k2  dk1 dk2
c(n) f k1
0
1

where c(0) = 1 and


1
c(n) = q n 1 +
q

if n > 0. Here q = |C|1 . Since f is K -finite on both sides and its translates are also matrix coefficients
we need only show that

2


n


C
0  n
f
q

0
1 

n=0

is finite. It will be more than enough to show that

(a) = f
as a .

a 0
0 1



= O(|a|1 )

Chapter 3

286

We recall that


(a) =
F







1 x
1 x
a 0
w
w
dx.
0 1
0 1
0 1

The function


1 x
1 (x) = w
0 1

is integrable and the function


1 x
2 (x) = w
0 1

is bounded and locally constant. Moreover

1 (x) 2 (a1 x) dx.

(a) =
F

Suppose 2 (x) = 2 (0) for |x| M . If |a| 1

1 (x) dx +

{x  |x||a|M }

(a) = 2 (0)

1 (x) 2 (a1 x) dx.



{x |x|>|a|M }

Since

1 (x) dx = 0
F

(a) is equal to



2 (a1 x) 2 (0) 1 (x) dx.

{x  |x|>|a|M }

The function 2 is bounded so we need only check that




{x  |x|>|a>}

|1 (x)| dx = O(|a|1 )

as |a| . The absolute value of the function is certainly bounded by some multiple of the function
1

 in B(F2 , F 2 ) defined by

a1
0

x
a2

   
 a1 
k = 
a2

if k is in GL(2, OF ). Since

w

1 x
0 1

=

0
1
1 x

=

x1
0

0
x



1 y
0 1


k

with y in F and k in GL(2, OF ) if |x| > 1


|1 (x)| dx = O

{x  |x|>|=|n }

 
k=n+1


|C|k = O(|C|n ).

Chapter 3

287

Since we need to compare orthogonality relations on the two groups GF = GL(2, F ) and GF
we have to normalize their Haar measure simultaneously. There are two ways of doing this. We first
describe the simplest.
Choose a non-trivial additive character of F . Then M (x) = (tr x) and

M  (x) = (x) are non-trivial additive characters of M = M (2, F ) and M  . Let dx and dx be the
Haar measures on M and M  self-dual with respect to M and M  . Then
2
d x = |x|1
M dx = |detx|F dx

and


 2

d x = |x |1
M  dx = |(x )|F dx

are Haar measure on GF and GF .

The second method takes longer to describe but is more generally applicable and for this reason
well worth mentioning. Suppose G and G are two linear groups defined over F and suppose there is
an isomorphism of G with G defined over the finite Galois extension K . Suppose the differential
form on G is defined over F . In general the form  = on G is not defined over F . Suppose
however that is left and right invariant and under an arbitrary isomorphism it is either fixed or
changes sign. Suppose moreover that for every in G(K/F ) the automorphism ()1 of G is inner.
Then



(  ) = () = () = () 1 = = 

and  is also defined over F . If is another such isomorphism of G with G then

() = (1 ) = = 
and the measures associated to and are the same. Thus a Tamagawa measure on GF determines
one on GF .
We apply this method to the simple case under consideration. If

x=
is a typical element of M then

a b
c d

= da db dc dd

is a differential form invariant under translations and the associated measure is self-dual with respect
to M . If = (detx)2 then is an invariant form on G and the associated measure is d x.
If K is any separable quadratic extension of F we may imbed K in both M and M  . Let be
the non-trivial element of G(K/F ). There is a u in M and a u in M  such that M = K + Ku and
1
2
M  = K + Ku while uxu1 = x and u xu = x for all x in K . u2 is a square in F and u = is
an element of F which is not the norm of any element of K . We may suppose that u2 = 1. If we let K
act to the right the algebra L = K F K is an algebra over K . acts on L through its action on the first
factor. There is an isomorphism L K K which transforms into the involution (x, y) (y, x).
In particular every element of K 1 is of the form with in L. Choose so that = . If

MK
= M  F K = L Lu

and

MK = M F K = L Lu,

Chapter 3

288


let be the linear map from MK
to MK which sends x + yu to x + yu. is easily seen to be an

isomorphism of MK and MK as algebras over K . ()1 takes x + yu to

x + y 1 u = 1 (x + yu)
and is therefore inner. Thus determines an isomorphism of G the multiplicative group of M  with G
the multiplicative group of M . is defined over K and ()1 is inner. Let | | be the Haar measure
on GF associated to the Haar measure || = d x of GF . We want to show that | | is just d x .
Let be an invariant form on K . The obvious projections of M = K Ku on K define differential
forms 1 and 2 on M . Let 1 2 = c. In the same way the projections of M  = K Ku on K
define differential forms 1 and 2 on M  . If we extend the scalars from F to K we can consider the
map x x of L into itself. is a form on L and its inverse image is N () = . Thus

(1 2 ) = 1 2 .
Thus if  = ()

c = 1 2 .

Suppose c1 || is self-dual with respect to the character K (x) = (x) on K . Then

 

and

||2F

 


(a, b) K (ax + by ) |(a)| |(b)| |(x)| |(y)| = c4
1 (0, 0)

(a, b) K (ax + by ) |(a)| |(b)| |(x)| |(y)| = c4
1 (0, 0).

If x + yu belongs to M with x and y in K then, since (u) = 0,

(x + yu) = (x) = TrK/F (x).

In the same way
Thus

(x + yu ) = TrK/F (x).



M (x + yu)(a + bu) = K (xa + yb )


M  (x + yu )(a + bu ) = K (xa + yb ).

Thus c21 |1 2 | is self-dual with respect to M and c21 ||F |1 2 | is self-dual with respect to M  .
Since c21 = |c|F the measure | | is self-dual with respect to M  . Finally  = (x )2 dx so that | |
is just d x . Thus the two normalizations lead to the same result.
If b is in M or M  the eigenvalues of b are the roots 1 and 2 of the equation

X 2 (b) X + (b) = 0.
If b is in GF or GF it is said to be regular of 1 and 2 are distinct; otherwise it is singular. We set



 (1 2 )2 
 .

(b) = 
1 2 F

Chapter 3

289

The set of singular elements is of measure 0. If b is regular the subalgebra of M or M  generated by b is

a separable quadratic extension E of F and the multiplicative group of E is a Cartan subgroup of GF
or GF . To obtain a set of representatives for the conjugacy classes of Cartan subgroups of GF or GF we
choose once and for all a set S  of representatives for the classes of separable quadratic extensions of
F . We also choose for each E in S  an imbedding of E in M and in M  . The multiplicative group of E
may be regarded as a Cartan subgroup BF of either GF or GF . S  will also stand for the collection of
Cartan subgroups obtained in this way. It is a complete set of representatives for the conjugacy classes
of Cartan subgroups of GF . If S is the result of adjoining to S the group AF of diagonal matrices then
S is a complete set of representatives for the conjugacy classes of Cartan subgroups of GF . If BF is in
S  we choose the Tamagawa measure B on BF as in the seventh paragraph. The analogue for GF of
the formula (1.2.2) is

 \G
ZF
F

f (g) 0 (g) =



1
S

(b)

ZF \BF

BF \GF



f (g 1 bg) B
(g) 0B (b).

F be the set of regular elements in BF and let

Let B
C=

F .
ZF \ B

S

F . We introduce on C the measure (c)

We may regard C as the discrete union of the spaces ZF \ B
defined by



f (c) (c) =

1
2

S

1
measure(ZF \ BF )

F
ZF \B

f (b) (b) 0B (b).

Lemma 15.3 Let be a quasi-character of F and let  () be the set of equivalence classes of

irreducible representations of GF such that (a) = (a) for a in ZF , which we identify with F .
If 1 and 2 belong to  () and
f (g) = 1 (g) 2 (g)
where (g) = Tr (g) then


f (c) (c) = 0
C

if 1 and 2 are not equivalent and


f (c) (c) = 1
C

if they are.
Since ZF \ GF is compact we may apply the Schur orthogonality relations for characters to see
that

1
measure ZF \ GF

 \G
ZF
F

f (g) 0 (g)

is 0 if 1 and 2 are not equivalent and is 1 if they are. According to the integration formula remarked
above this expression is equal to

1
1
f (b) (b) (measure BF \ GF ) 0b (b).
measure ZF \ GF  2 ZF \BF
S

Since

measure ZF \ GF = (measure ZF \ BF )(measure BF \ GF )

the lemma follows. Observe that ZF and ZF tend to be confounded.

There is form of this lemma which is valid for GF .

Chapter 3

290

Lemma 15.4 Let be a quasi-character of F . Let 0 () be the set of equivalence classes of ir-

reducible admissible representations of GF which are either special or absolutely cuspidal and
satisfy (a) = (a) for all a in ZF . Suppose 1 and 2 belong to 0 (). Let f = f1 ,2 be the
function
f (b) = 1 (b) 2 (b)
on C. Then f is integrable and


f (c) (c)
C

is 1 if 1 and 2 are equivalent and 0 otherwise.

It is enough to prove the lemma when is a character. Then is the complex conjugate of
and f, is positive. If the functions f, are integrable then by the Schwarz inequality all the functions
f1 ,2 are integrable.
Let () be the set of irreducible admissible representations of GF such that (a) = (a) for a
in ZF . If is a locally constant function on GF such that

(ag) = 1 (a) (g)

for a in ZF and such that the projection of the support of on ZF \ GF is compact then we define (),
if is in (), by


(g) (g) 0(g).

() =
ZF \GF

It is easily seen that () is an operator of finite rank and that the trace of () is given by the convergent
integral

ZF \GF

(g) (g) 0 (g).

In fact this follows from the observation that there is a 1 in HF such that

1 (g) =

1 (ag) (a) Z (a)

ZF

and the results of the seventh paragraph.

Suppose 1 is absolutely cuspidal and unitary and acts on the space V1 . Suppose also that
1 (a) = (a) for a in ZF . Choose a unit vector u1 and V1 and set



(g) = d(1 ) u1 , 1 (g)u1 .
Since 1 is integrable it follows from the Schur orthogonality relations that 2 () = 0 if 2 in () is
not equivalent to 1 but that 2 () is the orthogonal projection on Cu1 if 2 = 1 . In the first case
Tr 1 () = 0 and in the second Tr 2 () = 1.
On the other hand


Tr 2 () =

ZF \GF

2 (g) (g) 0(g).

We apply formula (7.2.2) to the right side to obtain

1
S

ZF \BF


2 (b) (b)

(g
BF \GF

bg) B (g) 0B (g).

Chapter 3

291

If BF belongs to S  the inner integral is equal to

1
d(1 )
measure ZF \ BF

ZF \GF


u1 , 1 (g 1 bg)u1 B (g)

which by Proposition 7.5 is equal to

1
1 (b).
measure ZF \ BF
If BF is AF the group of diagonal matrices the inner integral is, apart from a constant relating Haar
measures, the product of d(1 ) and the integral over GL(2, OF ) of

 
1
F

1 x
0 1





1 x
b
1 (k)u1, 1 (k)u1 dx.
0 1

If

b=
this is

1
0

0
2


1  








1 x
2
1


(b)

(k)u
,

(k)u

1
1
1
1 1
1 dx

0 1
1 
F

which we know is 0. Collecting these facts together we see that f = f2 ,1 is integrable on C if 1 is

absolutely cuspidal and that its integral has the required value.
1

To complete the proof all we need do is show that if = (2 F , F 2 ) is a special representation

then f = f, is integrable on C and


f (c) (c) = 1.
C

If  is the one-dimensional representation g (g) of GF then = ( ). To prove the existence

F then
of we had to show in effect that if BF was in S  and b was in B

(b) =  (b).
Thus f, = f , and the assertion in this case follows from the previous lemma.
The relation just used does not seem to be accidental.
Proposition 15.5 Suppose  is an irreducible admissible representation of GF and = (  ) the

F
corresponding representation of GF . If BF is in S  and b is in B
 (b) = (b).

We may suppose that  is not one-dimensional and that is absolutely cuspidal. We may also
suppose that they are both unitary. We take in Kirillov form with respect to some additive character
. If is in S(F ) the function





 =

is also.

0 1
1 0

Chapter 3

292

Since the measure and  are self-dual with respect to the characters M and M  Lemma 13.1.1
and Proposition 4.5 show us that for any in F



(detg) (g 1 )u, u |detg| M (g) (g)

(15.5.1)

GF

is equal to

 (2 ) 1 () ||2


and that

GF



(detg)  (g 1 )u , u |detg| M  (g)  (g)

is equal to

(15.5.2)

 (2 ) 1 () ||2 .

Here u is a unit vector in the space of and u a unit vector in the space  . In any case (15.5.1) is just
the negative of (15.5.2).
If we use formula (7.2.1) to express the integral (15.5.1) as a sum over S we obtain

1
1
2  measure ZF \ BF


(detb) |detb|
BF

(b1 )
(b) M (b) B (b).
d()

The contribution from AF vanishes as in the previous lemma. The other integrals have been simplified by means of Proposition 7.5. There is of course an obvious analogue for the group GF of the
formula (7.2.1). If we apply it we see that (15.5.2) is equal to

1
1
2  measure ZF \ BF
S


(detb) |(b)|
BF

 (b1 )
(b) M  (b) B (b)
d(  )

if (b) is the reduced norm. Of course on BF the functions (b) and detb are the same. Choose BF0 in
 0 . We shall show that
S  and b0 in B
F

 (b1
(b1
0 )
0 )
=
.
d(  )
d()
The orthogonality relations of the previous two lemmas will show that d() = d( ) and we will
conclude that
1
 (b1
0 ) = (b0 ).

The norm and the trace of b0 are the same whether it is regarded as an element of M or of M  . In
 0 is the multiplicative group of E in S the norm and the trace are in both cases the norm and
fact if B
F
the trace of b0 as an element of E . Since b0 and its conjugate in E are conjugate in GF and GF we can
choose an open set U in E containing both b0 and its conjugate so that

|(b)|  (b1 ) (b) = |(b0 )|  (b1

0 ) (b0 )
 0 . Thus we can also suppose that
if b is in U . Lemma 7.4.2 shows that is locally constant in B
F
|detb| (b1 ) (b) = |detb0 | (b1
0 ) (b0 )

Chapter 3

293

if b is in U . Suppose 0 and 0 are the trace and norm of b0 . We can choose a positive integer m so that
if 0 and 0 belong to pm
F the roots of

X 2 X + b
belong to E and in fact lie in U .
Let () be the expression (15.5.1) regarded as a function of . Keeping in mind the fact that

M (b) = M  (b) = ( tr b),

we compute

1
measure pmn
F


pmn
F

() (0 ) d

(15.5.3)

where pn
F is the largest ideal on which is trivial. Since

1
mn
measure pF


pmn
F



(tr b 0 ) d

is 0, unless tr b 0 belongs to pm
F when it is 1, the integral (15.5.3) is equal to

1
1
2  measure ZF \ BF
S

if


(detb) |detb|
V (BF )

(b1 )
(b) B (b)
d()


V (BF ) = {b BF  tr b 0 pm
F }.

If we take to be the characteristic function of


{ F  0 pm
F}
the summation disappears and we are left with

1
(b1
1
0 )
|detb0 |

(b0 )
2 measure ZF \ BF
d()


(detb) B (b).
0)
V (BF

If we replace () by the expression (15.5.2) the final result will be

1
 (b1
1
0 )
|(b0 )|

(b0 )
2 measure ZF \ BF
d()


(detb) B (b).
0)
V (BF

Since these differ only in sign the proposition follows.

We are now in a position to prove Theorem 15.1. The orthogonality relations and the previous
lemma show that the map  is injective because the map takes  () into 0 (). It is enough to
verify that V is surjective when is unitary. Let L2 () be the space of all measurable functions f on

F
S  B

Chapter 3

294

such that f (ab) = (a) f (b) if a is in ZF and


|f (c)|2 (c)
C

is finite. By the PeterWeyl theorem the set of functions  ,   (), form an orthonormal basis of
L2 (). The family , 0 (), is an orthonormal family in L2 (). By the previous proposition the
image of  () in 0 () is actually an orthonormal basis and must therefore be the whole family.
We observe that it would be surprising if the relation d() = d( ) were not also true when 
is one-dimensional. The facts just discussed are also valid when F is the field of real numbers. They
follow immediately from the classification and the remarks at the end of the seventh paragraph.
We conclude this paragraph with some miscellaneous facts which will be used elsewhere. F is
again a non-archimedean field. Let K = GL(2, OF ) and let K0 be the set of all matrices

a b
c d

in K for which c 0(mod pF ). Suppose is an irreducible admissible representation of GF in the

space V . We are interested in the existence of a non-zero vector v in V such that

a b
c d

for all matrices in K0 while


v = 1 (a) 2 (d) v

0 1
C 0


v = 0 v

0 is a constant and 1 and 2 two characters of UF . C is a generator of pF . Since








0 C 1
a b
1 1
d C 1 c
=
Cb
a
1
0
c d
C 0
such a vector can exist only if 1 = 2 = .
Lemma 15.6 Suppose and 0 are given. Let be (1 , 2 ) which may not be irreducible. There

is a non-zero vector in B(1 , 2 ) satisfying the above conditions if and only if the restrictions of
1 and 2 to UF , the group of units of OF , are equal to and
02 = (C) 2 (C)

Moreover if it exists is unique apart from a scalar factor.

It is easily seen that K is the disjoint union of K0 and

K0

0 1
1 0


K0 = K0 wK0

Let 1 be the function which is 0 on K0 wK0 and on K0 is given by

1

a b
c d


= (ad).

Chapter 3

295

Let 2 be the function which is 0 on K0 and takes the value (a d ad) at

a
c

If in B(1 , 2 ) satisfies



b
d

0 1
1 0

a b
c d



a b
c d


= (ad)

for all matrices in K0 then the restrictions of to K must be a linear combination of 1 and 2 . This
already implies that is the restriction of 1 and 2 to UF . Suppose = a1 + b2 . Since

0 1
C 0

and



1 = |C| 2 1 (C) 2(1) 2

0 1
C 0



2 = |C| 2 2 (C) 1
1

while 1 (1) = 2 (1) = (1), we have

1

0 b = |C| 2 1 (C)a
and

0 a = |C| 2 2 (C)b
1

Apart from scalar factors there is at most one solution of this equation. There is one non-trivial solution
if and only if 02 = 1 (C) 2 (C).
Lemma 15.7 Suppose = (1 , 2 ) is the special representation corresponding to the quasi-chacters
1

1 = F 2 and 2 = F2 . There is a non-zero vector v in the space of such that


a b

v = (ad)v
c d
for all matrices in K0 while

0
C

1
0


v = 0 v

if and only if is the restriction of to UF and 0 = (C). If v exists it is unique apart from
a scalar.
We first let act on Bs (2 , 2 ) a subspace of B(2 , 1 ). The condition on follows from the
previous lemma which also shows that 0 must be (C). If we take the plus sign we see that v
must correspond to the function whose restriction to K is constant. Since this function does not lie in
Bs (2 , 1 ) only the minus sign is possible. To see the existence we let act on

Bs (1 , 2 ) = B(1 , 2 )/Bf (1 , 2 )
In B(1 , 2 ) there are two functions satisfying the conditions of the lemma. One with 0 = (C)
and one with 0 = (C). One of the two, and we know which, must have a non-zero projection on
Bs (1 , 2 ).
The above lemmas together with the next one sometimes allow us to decide whether or not a given
representation is special.

Chapter 3

296

Lemma 15.8 If the absolutely cuspidal representation acts on V there is no non-zero vector v in

V such that

and

0
C

a
c

1
0
b
d


v = 0 v


v = (ad)v

for all matrices in K0 .

We may suppose that is the Kirillov form with respect to an additive character such that OF
is the largest ideal on which is trivial. Then v is a function in S(F ). If a is in UF and b is in F
we must have (ab) = (a) (b). Moreover if b is in F and x is in OF then (b) = (xb) (b). Thus
(b) = 0 if b is not in OF . Consequently (,
t) is 0 if = 1 but (
1 , t) is a polynomial of the
form

am tm + + an tn

with am an = 0. If 1 (b) = (Cb) then

1 ( 1 , t) =

Let

(1)
(
1 , t).
t

a 0
0 a


= (a) I

and let 0 be the restriction of to UF while z0 = (C). The character 0 will have to be equal to 2 .
The relation









0 =
implies that

0 1
C 0

0 1
1 0

0 (
1 , t) = C( 1 , t) (1) z0 t (
1 , z01 t1 ).

By Proposition 2.23, C(1 , t) is of the form ct with  2. Thus the right side has a pole at 0 not
shared by the left. This is a contradiction.

Chapter 3

297

16. An application of the Selberg trace formula. In the

paragraph we saw that if  = v v
 fourteenth



is a constituent of A and is not of the form g (g) where is a quasi-character of F \ I then
= v v , with v = (v ), is a constituent of A0 . Let S be the set of places at which the quaternion
algebra M  does not split. Given the results of the previous paragraph it is tempting to conjecture that
the following theorem is valid.
Theorem 16.1 Suppose = v is a constituent of A0 . If for every v in S the representation v

is special or absolutely cuspidal then for every v there is a representation v such that v = (v )
and  = v is a constituent of A .
The existence of v has been shown. What is not clear is that  is a constituent of A . It seems
to be possible to prove this by means of the Selberg trace formula. Unfortunately a large number of
analytical facts need to be verified. We have not yet verified them. However the theorem and its proof
seem very beautiful to us; so we decided to include a sketch of the proof with a promise to work out the
analytical details and publish them later. We must stress that the sketch is merely a formal argument
so that the theorem must remain, for the moment, conjectural.
We first review some general facts about traces and group representations. Suppose G is a locally
compact unimodular group and Z is a closed subgroup of the centre of G. Let be a character of Z .
We introduce the space L1 () of all measurable functions f on G which satisfy f (ag) = 1 (a) f (g)
for all a in Z and whose absolute values are integrable on Z \ G. If f1 and f2 belong to L1 () so does
their product f1 f2 which is defined by

f1 (gh1 ) f2 (h) dh

f1 f2 (g) =
Z\G

If f belongs to L1 () let f be the function f (g) = f(g 1 ). It also belongs to L1 (). A subalgebra B
of L1 () will be called ample if it is dense and closed under the operation f f .
Let be a unitary representation of G on the Hilbert space H such that (a) = (a) I for all a in
Z . We do not suppose that is irreducible. If f belongs to L1 () we set


(f )

f (g) (g) dg
Z\G

If (f ) is compact for all f in some ample subalgebra B then decomposes into the direct sum of
irreducible representations no one of which occurs more than a finite number of times.
Lemma 16.1.1 Suppose 1 and 2 are two unitary representations of G such that 1 (a) = (a) I and
2 (a) = (a) I for all a in Z. Suppose there is an ample subalgebra B of L1 () such that 1 (f )
and 2 (f ) are of HilbertSchmidt class for all f in B.
(i) If for every f in B
trace 1 (f ) 1 (f ) trace 2 (f ) 2(f )

then 2 is equivalent to a subrepresentation of 1 .

(ii) If for every f in B
trace 1 (f ) 1 (f ) = trace 2 (f ) 2(f )
then 2 is equivalent to 1 .
Let 1 act on H1 and let 2 act on H2 . A simple application of Zorns lemma shows that we
can choose a pair of closed invariant subspaces M1 and M2 , of H1 and H2 respectively, such that the

Chapter 3

298

restrictions of 1 to M1 and 2 to M2 are equivalent and such that the pair M1 , M2 is maximal with
respect to this property. Replacing H1 and H2 by the orthogonal complements of M1 and M2 we may
suppose that M1 = 0 and that M2 = 0. To prove the first assertion of the lemma we have to show that
with this assumption H2 = 0. If the second condition is satisfied we can reverse the roles of 1 and 2
to see that H1 is also 0.
Before beginning the proof we make a simple remark. Suppose is an irreducible unitary representation of G on L and , A, is an irreducible unitary representaion of G on L . Suppose
that (a) = (a) I for all a in Z and (a) = (a) I for all a in Z and all in A. Suppose that is
equivalent to none of the and that a non-zero vector x in L and vectors x in L are given. Finally
suppose that


 (f ) x 2

is finite for every f in B . Then if is any positive number there is an f in B such that

 (f )x2 < (f )x2.

Suppose the contrary and let L be the closure in L of


{ (f ) x  f B}
L is invariant under G and the map
(f )x (f )x
may be extended to a continuous G-invariant map A of L into L. If A were 0 then (f )x = 0 for all
f in B which is impossible. Let A be the linear transformation from L to L which is A on L and 0
on its orthogonal complement. A commutes with G and is not 0. Let A be the restriction of A to L .
A is a G-invariant map of L into L and is therefore 0. Thus A is 0. This is a contradiction.
Suppose H2 is not 0. There is an h in B such that 1 (h) = 0. If f = h h then 2 (f ) is positive
semi-definite and of trace class. It has a positive eigenvalue and with no loss of generality we may

suppose that its largest eigenvalue is 1. Let 2 = 2 , where 2 acts on H2 , be a decomposition of 1

0
into irreducible representations. There is a 0 and a unit vector x in H2 such that 2 (f )x = x. Let

1 = 1 , where 1 acts
decomposition of 1 into irreducible representations. Choose an
 on H1 , be a
, 
} of H1 consisting of eigenvectors of 1 (f ). Since
orthogonal basis {x

trace 1 (f ) trace 2 (f )
the largest eigenvalue of 1 (f ) is positive. Let it be .
If f1 belongs to B ,


1 (f1 ) x, 2

is the HilbertSchmidt norm of 1 (f1 ) and is therefore finite. By assumption 2 0 is not equivalent to
any of the representations 1 so that we can apply our earlier remark to the vector x and the family of
,
representations 1 = 1 together with the family of vectors x, to infer the existence of an f1 in B
such that



1 (f1 ) x, 2 <

1
2 (f1 ) x2 .
2

Chapter 3
Then
is equal to

299

trace 1 (f1 f ) 1 (f1 f ) = trace 1 (f1 f ) 1 (f1 f )



1 (f1 ) 1 (f ) x, 2

1 (f1 ) x, 2 .

The right side is less than

1
1
2 (f1 ) x2 = 2 (f1 f ) x2
2
2
which is at most

1
trace 2 (f1 f ) 2 (f1 f ).
2

This is a contradiction.
The next lemma is a consequence of the results of [35].
Lemma 16.1.2 Suppose is trivial so that L 1 () = L1 (Z \ G). Suppose that B is an ample subal-

gebra of L1 () which is contained in L2 (Z \ G). If there is a positive constant and a unitary

representation of Z \ G such that (f ) is of HilbertSchmidt class for all f in B and
trace (f ) (f ) =


|f (g)|2 dg
Z\G

then Z \ G is compact.
In proving the theorem it is better to deal with representations in the ad`ele groups than to deal
with representations of the global Hecke algebras. We have to assume that the reader is sufficiently
well acquainted with the theory of group representations to pass back and forth unaided between the
two viewpoints.
If F is a global field, A is the ad`ele ring of F , G = GL(2), and is a character of the id`ele class
group F \ I the space A() of all measurable functions on GF \ GA that satisfy

a 0
0 a

 
g = (a) (g)

for all a in I and whose absolute values are square-integrable on GF ZA \ GA is a Hilbert space. If
belongs to this space


(ng) dn
NF \NA

is defined for almost all g . If it is 0 for almost all g the function is said to be a cusp form. The space
A0 () of all such cusp forms is closed and invariant under GA . It is in fact the closure of A0 (). It
decomposes in the same way but now into a direct sum of closed orthogonal subspaces V on which
GA acts according to an irreducible representation = v . Thus V is now isomorphic to a tensor
product of Hilbert spaces. Of course the same representations occur now as occurred before. Similar
remarks apply to the multiplicative group G of quaternion algebra M  over F .
It will be enough to prove the theorem when is a constituent of some A0 () or A0 () and is a
character because we can always take the tensor product of with a suitable quasi-character. Suppose

Chapter 3

300

is given. Let S be the set of places at which M  does not split. Suppose that for each v in S we are
given an irreducible unitary representation v of GFv = Gv such that
v (a) = v (a) I
for all a in Fv which we identify with Zv = ZF v . Let v = (v ) be the representation of Gv
corresponding to v . We may take v unitary. Let v act on Uv and let v act on Uv . Fix a unit vector
uv in Uv and a unit vector uv in Uv which is Kv -finite. uv is automatically Kv -finite.
Write A0 () as the direct sum, in the Hilbert space sense, of mutually orthogonal invariant irreducible subspaces V 1 , V 2 , . Let the factorization of the representation i on V i be vi . Let vi act
on Vvi . For simplicity of notation we identify V i with Vvi . We also suppose that if v is in S and vi is
equivalent to v then Uv = Vvi and vi = v . Let X be the set of all i such that vi = v for all v in S
and if i belongs to X let

M i = {vS uv } {vS Vvi }.

M i is invariant and irreducible under the action of


 S = {g = (gv )  gv = 1 for all v in S}.
G
Let

M = iX M i .

 S acts on M . If at least one of the representations  , v S , is not oneM is a Hilbert space and G
v
dimensional set N = M . If they are all one-dimensional let N be the subspace of A0 () spanned, in
M and the functions g (detg) where is a character of F \ I such that
the Hilbert space sense, by

2 = and v (g) = v (g) for all g in Gv if v is in S . If v is non-archimedean this last condition
determines v uniquely. If v is real it only determines it on the positive numbers.
Let A () be the space of all measurable functions on GF \ GA that satisfy (ag) = (a) (g)
for all a in I and whose absolute values are square integrable on GF ZA \ GA . Replacing v by v and
uv by uv we define N  in the same way as we defined M . If at least one of the representations v ,
v S , is not one-dimensional we set M  = N  . However

 if they are all one-dimensional and is a
character of F \ I such that 2 = and v (g) = v (g) for all G in GV if v is in S then the function


g (g) belongs to N  . We let M  be the orthogonal complement in N  of the set of such functions.
  acts on M  and N  . However by means of the local isomorphisms v we can define an
The group G
S
S and G
  . Thus G
 S acts on M and M  . To prove the theorem we need only show
isomorphism of G
S
that the representations on these two spaces are equivalent. To do this we combine Lemma 16.1.1 with
the Selberg trace formula.
To apply Lemma 16.1.1 we have to introduce an algebra B . B will be the linear span of B0 , the set
 S of the form
of functions f on G
f (g) =
fv (gv )
vS

where the functions fv satisfy the following conditions.

(i) If av belongs to Fv then

fv (av gv ) = v1 (av ) fv (gv ).

(ii) fv is Kv -finite on both sides and the projection of the support of fv on Zv \ Gv is compact.
(iii) If v is archimedean fv is infinitely differentiable.
(iv) If v is non-archimedean fv is locally constant.

Chapter 3

301

(v) For almost all non-archimedean v the functions fv is 0 outside of Zv Kv but on Zv Kv is given by

fv (g) = v1 (detg)
where v is unramified and satisfies v2 = v .
We introduce B  in the same way. We may identify B and B and to verify the conditions of the
lemma we need only show that if f = f1 f2 with f1 and f2 in B0 then

trace (f ) = trace  (f )
if is the representation on M and  that on M  . Let be the representation on N and  that on N  .
Since


trace (f ) = trace (f ) +

S \G
S
Z

and


trace (f ) = trace (f ) +
we need only show that

(g) f (g) dg


S
S \G
Z

(g) f (g) dg

trace (f ) = trace  (f ).

Before beginning the proof we had better describe the relation between the Haar measures on the
groups ZA \ GA and ZA \ GA . Choose a non-trivial character of F \ A. If 0 is any invariant form
of maximal degree on Z \ G defined over F and therefore over each Fv we can associate to 0 and v
!


a
!Haar measure 0 (v) on Zv \ Gv . Then vS 0 (v) determines a Haar measure 0 on ZS \ GS and
v 0 (v) determines a Haar measure 0 on ZA \ GA . The measure on ZA \ GA is independent of and
is called the Tamagawa measure. As in the previous paragraph we can associate to 0 (v) a measure
  or Z  \ G .
0 (v) on Zv \ Gv and therefore to 0 a measure 0 on ZS \ G
S
A
A
We first take f = f1 f2 in B  and find a formula for trace  (f ). Let d(v ) be the formal degree
of v with respect to the measure 0 (v) and let v be the function

v (g) = d(v ) (v (g)uv , uv )

on Gv . Let  = f be the function

 (g) =


v (gv ) f (
gS )
vS

  . If  is the representation of G on A () the restriction of

on GA . Here gS is the projection of g on G
S
A





( ) to N is (f ) and () annihilates the orthogonal complement of N  . Thus
trace  ( ) = trace  (f ).
If is in A () then  ( ) (g) is equal to


ZA \GA

(gh)

(h) 0 (h)


=

 \G
ZA
A

(h)  (g 1 h) 0 (h).

Chapter 3

302

The integration on the right can be performed by first summing over ZF \ GF and then integrating
over ZA GF \ GA . If

 (g, h) =

 (g 1 h)

 \G
ZF
F

the result is

ZA GF \GA

Thus the trace of  () is equal to

(h)  (g, h) 0 (h).


 G \G
ZA
F
A

 (g, g) dg.

If we write out the integrand and perform the usual manipulations (cf [29]) we see that this integral
is



measure ZA GF () \ GA ()


GA ()\GA

{}

 (g 1 g).

(16.1.3)

The sum is over a set of representatives of the conjugacy classes in GF . GA () is the centralizer of in
GA and GF () is its centralizer in GF .
Let Q be a set of representatives for the equivalence classes of quadratic extensions E of F such
that E F Fv is a field for all v in S . For each E in Q fix an imbedding of E in the quaternion algebra
M  . Let BF = BF (E) be the multiplicative group of E , considered as a subalgebra of M  , or what is
the same the centralizer of E in GF . Let BA = BA (E) be the centralizer of E in GA . Let Q1 be the
separable extensions in Q and Q2 the inseparable ones if they exist. Then (16.1.3) is the sum of

measure(ZA GF \ GA )  (e),

(16.1.4)


1 

measure(ZA BF \ BA )
 (g 1 g) B (g)
2  Z  \B
BA \GA

(16.1.5)

if e is the identity,

Q1

and

Z 
F

Q2

Z  \BF
F

ZF

measure(ZA


BF \ BA )

BA \GA

 (g 1 g) B (g).

(16.1.6)

The last sum is deceptive because Q2 has at most one element. B is the quotient of the measure on
ZA \ GA by that on ZA \ BA . The choice of the measure on ZA \ BA is not too important. We do suppose
that it is a product measure.*
The expression (16.1.4) is equal to

measure(ZA GF \ GA )


d(v ) f (e).
vS

* In (16.1.5) the factor 12 is not quite correct. If we want to leave it in, both and its conjugate must
be counted, even if they differ only by an element of F .

Chapter 3

303

The integrals of (16.1.5) and (16.1.6) are equal to the product

vS

v ( 1 )
measure(Zv \ Bv )

and

S \G
S
B

f (g 1 g) B .

Now regard f = f1 f2 as an element of B . We can still introduce for each v in S the function



v (g) = d(v ) v (g)uv uv
on Gv . d(v ) is the formal degree of v with respect to the measure 0 (v). If v is not one-dimensional
v is integrable and we can use it to define a function to which we can hope to apply the trace formula.
When v is one-dimensional the function v is not even integrable so it is of no use to us. However in
this case we can find an integrable function v with the following properties.
(i) For all a in Fv



 

a 0
0 a

= v1 (a) v (g).

(ii) For a suitable choice of uv the operator v (v ) is the orthogonal projection on the space Cuv .
(iii) If v is a character of Fv such that 2v = v then


Zv \Gv

v (detg) v (g) 0 (v)

is 1 if v (h) = v (h) for all h in Gv and is 0 otherwise.

(iv) If v is a unitary infinite-dimensional irreducible admissible representation of Gv which is not
equivalent to v but satisfies





a 0
0 a

= v (a) I

for all a in Fv then

trace v (v ) = 0.
If v is real we cannot describe v without a great deal more explanation than is desirable at present.
However after a few preliminary

 remarks we will be able to describe it when v is non-archimedean.
Suppose v (g) = v (g) for g in Gv and v is a representation of Gv such that

v

a 0
0 a


= v (a) I

for all a in Fv . Applying Lemma 3.9 to 1

v v we see that the restriction of v to Kv contains the
representation k v (detk) if and only if v = (v , v ), v v = v , and the restrictions of v and
v to Uv , the group of units of Fv , are both equal to the restriction of v . Let v be the function on Gv
which is 0 outside of Zv Kv but on Kv is equal to

1
1 (detg).
measure(Zv \ Zv Kv ) v

Chapter 3

304

Let Hv be the group generated by Zv , the matrices

in Kv for which c 0(mod pv ), and

a b
c d

0
Cv

1
0


.

Let v be the character v (a) = (1)n v (a) if |a| = |Cv |n . According to the concluding lemmas of
the previous paragraph there is a non-zero vector u in the space of v such that

v (g) u = v (detg) u
for all g in Hv if and only if v is equivalent to v , v = (v , v ) is infinite-dimensional, v v = v ,
and the restrictions of v and v to Uv are equal to the restriction of v , or v is the one-dimensional
representation

g v (detg).

Let v be the function which is 0 outside of Hv and equal to

1
1 (detg)
measure Zv \ Hv v
on Hv . We may take

v = v v .

There are some consequences of the four conditions on v which we shall need. If v and v are
two characters of Fv such that v v = v , the trace of (v , v , v ) is a multiple of

  12   

 
v () v ()  
v (k 1 ank) dn dk da

Zv \Av
Nv Kv

if

a=

0
0


.

Since this is 0 for all possible choice of v and v


Nv

v (k 1 ank) dk dn = 0
Kv

for all a. We also observe that if v is not one-dimensional then


Nv

for all a.

v (k 1 ank) dk dn = 0
Kv

Chapter 3

305

If v is special or absolutely cuspidal trace v (v ) is therefore equal to

1
2 
S


Zv \Bv

Bv \Gv


v (g 1 bg) b (v) v (b) (b)0B (b).

Since trace v (v ) is 1 if v is equivalent to v and 0 otherwise the orthogonality relations imply that


Bv \Gv

v (g 1 bg) B (v) =

1
(b1 )
measure Zv \ Bv v

for all regular b and therefore, by continuity, for all b whose eigenvalues do not lie in Fv . It probably also
follows from the Plancherel theorem that v (e) = d(v ). We do not need this but we shall eventually
need to know that v (e) = d(v ). For the moment we content ourselves with observing that if v is a
character of Fv and v is replaced by v v the formal degree does not change and v is replaced by
the function g v1 (detg) v (g) so that v (e) does not change. Thus the relation v (e) = d(v ) need
only be proved when v is trivial.
Let S1 be the subset of v in S for which v is one-dimensional and let S2 be the complement of S1
in S . Given f = f1 f2 in B we set



(g) =


v (gv ) f (
gS ).

v (gv )
vS1

vS2

+
Let +
0 be the representation of GA on A0 () the sum, in the Hilbert space sense, of A0 () and
the functions : g (detg) where is a character of F \ I such that 2 = and let be
the representation on A(). If at least one of the representations v is not one-dimensional +
0 ()
annihilates the orthogonal complement of A0 (). If they are all one-dimensional we apply the third
condition on the functions v together

 with the fact that the number of places in S is even to see that

(h)
+
()

=
0
unless

(h)
=

for all h in Gv and all v in S but that if this is so

v
v
0

+
0 () = (f ) .
Recall that A0 () is the direct sum of spaces V i on which GA acts according to representations i = vi .
If at least one of the representations v is not one-dimensional +
0 () is equal to (f ) on M and
annihilates the orthogonal complement of M in A0 (). Suppose they are all one-dimensional. If i
+
i
belongs to X the restrictions of +
0 () and (f ) or (f ) to M are equal and 0 () annihilates the
i
i
i
orthogonal complement of M in V . If i is not in X the trace of the restriction of +
0 () to V is


trace vi (v ) {trace
Si (f )}
vS

in
Si = vS vi . Since vi , v S , are all infinite-dimensional and for at least one such v the
representation vi is not equivalent to v ,

trace vi (v ) = 0.
vS

We conclude that

trace +
0 () = trace (f ).

Chapter 3

306

To show that

trace (f ) = trace  (f )

we have to apply the trace formula to find a suitable expression for trace +
0 (). In order to describe
the formula we need to state some results in the theory of Eisenstein series.
Consider the collection of pairs of character , of F \ I such that = . To such pairs, ,
and  ,  are said to be equivalent if there is a complex number r such that  = rF and  = r
F .
r
r
If a belongs to I then F (a) = |a| . Let P be a set of representatives for these equivalence classes.
s

Suppose (, ) belongs to P . If s is a complex number the space B(F2 , F 2 ) of functions on

NA \ GA is defined as in the tenth paragraph. Since the functions in this space are determined by their
restrictions to K we may think of it as a space of functions on K in which case it is independent of s.
Thus we have isomorphisms
s

Ts : B(F2 , F 2 ) B(, ).
The theory of Eisenstein series provides us with a function (, s) E(, s) from B(, ) C to
A(). E(g, , s) is the value of E(, s) at g . For a given the function E(g, , s) is continuous in g
and meromorphic in s. Moreover there is a discrete set of points in C such that outside of this set it is
s

holomorphic in s for all g and . If s is not in this set the map E(Ts , s) of B(F2 , F 2 ) into
A() commutes with the action of H.
If the total measure of NF \ NA is taken to be 1 the integral


NF \NA

is equal to

E(ng, Ts , s) dn



(g) + M (s) (g).
s

M (s) is a linear transformation from B(F2 , F 2 ) to B(F 2 , F2 ) which commutes with the action
of H. It is mermorphic in the sense that
M (s)Ts11 , Ts1 2 
is meromorphic if 1 belongs to B(, ) and 2 belongs to B( 1 , 1 ). The quotient of M (s) by

L(1 s, 1 )
L(s, 1 )
1
)
=
(1

s,

L(1 + s, 1 )
L(1 + s, 1 )
is holomorphic for Re s 0. Since the analytic behavior of E(g, , s) is controlled by that of M (s) it
should be possible, as we observed before, to use the Eisenstein series to show that a constituent of
s

B(F2 , F 2 ) is also a constituent of A().

To indicate the dependence of M (s) on and we write M (, , s). Then
M (, , s)M (, , s) = I.
If s is purely imaginary we can introduce the inner product


(1 , 2 ) =

1 (k)
s(k) dk
K

Chapter 3
s

307
s

on B(F2 , F 2 ). Let B(F2 , F 2 ) be its completion with respect to this inner product. B(F2 , F 2 )
may be thought of as a function space on GA on which GA acts by right translations. The representation
s

of GA on B(F2 , F 2 ) is unitary. Let g correspond to the operator (g, , , s) and if f is in L1 () let


(f, , , s) =
ZA \GA
s

f (g) (g, , , s) 0 (g)

Ts extends to an isometry, from B(F2 , F 2 ) to B(, ) and M (, , s) extends to an isometry from

s
s
s
s
B(F2 , F 2 ) to B(F 2 , F2 ). In particular
M (, , s) = M (, , s).
r

2
Suppose (, ) is in P and, for some r , = rF and = r
F . Replacing by F and by

F 2 if necessary we may suppose that = . We may also suppose that if (, ) is in P and is not
equivalent to (, ) then (, ) is also in P . Let L be the Hilbert space sum
P B(, )
and let L be the algebraic sum
If we define L(s) to be
and L(s) to be

P B(, ).
 s
s
p B F2 , F 2
 s
s
P B F2 , F 2

we can introduce the map

Ts : L(s) L.

The representation g (g, s) is the representation

g (g, , , s)
on L(s). M (s) will be the operator on L(s) which takes (, ) to 1 (, ) with

1 (, ) = M (, , s) (, ).
It is unitary.
If F has characteristic O let H be the space of all square integrable functions from the imaginary
axis to L such that
1
Ts
(s) = M (s) Ts1(s)

with the norm

i
i

(s)2 d|s|

Chapter 3

308

c is a positive constant relating various Haar measures. It will be defined more precisely later. If F is
q
i and the
a function field with field of constants Fq the functions in H are to be periodic of period log
2
norm is to be
 2
c log u

log q

(s)2 d|s|.

On the whole we shall proceed as though F had characteristic O merely remarking from time to time
the changes to be made when the characteristic is positive.
If = (, ) is in L we set

E(g, , s) =
If in H takes values in L

1
lim
T 2

 

E g, (, ), s .



E g, (s), s d|s| = (g)

iT

iT

exists in A(). The map

extends to an isometry of H with a subspace A1 () of A(). If g is in

GA and is defined by

 (s) = Ts (g, s) Ts1 (s)

then
 is (g).
The orthogonal complement of A1 () is A+
0 (). Thus if E is the orthogonal projection of A() on
+
A1 () the trace of 0 () is the trace of () E() which, according the the Selberg trace formula,
is the sum of the following expressions which we first write out and then explain.
(i)

measure(ZA GF \ GA ) (e).

(ii)


1 
measure(ZA BF \ BA )
(g 1 g) B (g).
2
BA \GA
Z \B
Q1

(iii)

F
F

ZF

Q2

ZF \BF

ZF


measure(ZA BF \ BA )

(iv)


ZF \AF

ZF

(v)

"
c 0

but

log q
4

(0, fv ) + 1
v

(vi) If F is a number field



BA \GA

(g 1 g) b (g).


(, fw ) (, fv ).
w=v


v

#

 (0, fv )

(0, fw )

w=v

1
trace M (0) (, 0),
4


  

trace M (0) (, 0) + trace M
,
log q
log q

Chapter 3

309

if F is a function field.
(vii) If F is a number field

1
4

trace m1 (s) m (s) (, s) d|s|,

but

log q
4

2
log q

trace m1 (s) m (s) (, s) d|s|

if F is a function field.
(viii) The sum over (, ) and v of

1
4

tr R1 (v , v , s) R (v , v , s) (f, v , v , s)




tr (fw , w , w , s) d|s|
w=v

if F is a number field and of

log q
4

2
log q

tr R1 (v , v , s) R (v , v , s) (f, v , v , s)




tr (fw , w , w , s) d|s|
w=v

if F is a function field.
The function is of the form

(g) =

fv (gv ).
v

Let Q be a set of representatives for the equivalence clases of quadratic extensions of F . For each E in
Q fix an imbedding of E in the matrix algebra M = M (2, F ). Let BF = BF (E) be the multiplicative
group of E , considered as a subalgebra of M . BF is the centralizer of E in GF . Let BA = BA (E) be
the centralizer of E in GA . Q1 is the collection of separable extensions in Q and Q2 is the collection of
inseparable extensions. AF is the group of diagonal matrices in GF .
Choose on NA that Haar measure which makes the measure of NF \ NA equal to 1. Choose on K
the normalized Haar measure. On the compact group H obtained by taking the quotient of



0
0




AA || = ||

by ZA AF choose the normalized Haar measure. H is the kernel of the map

0
0

  12
 
log  

of AF ZA \ AA onto R or log q Z. On R one has the standard measure dx and on log q Z one has the
standard measure which assigns the measure 1 to each point. The measures on H and on H \ (AF ZA \
AA ) together with the measure on ZA \ AF ZA which assigns the measure 1 to each point serve to
define a measure da on ZA \ AA . The constant c is defined by demanding that


ZA \GA

f (g) 0 (g)

Chapter 3

310

be equal to

f (ank) da dn dk
ZA \AA

NA

if f is an integrable function on ZA \ GA . We may suppose that the measures on ZA \ AA , NA , and K

are given as product measures and in particular that


dkv = 1
Kv

and

(nv ) dnv = 1
Nv

for almost all v if is the charcteristic function of



1 x
0 1


 x OF


.

The factors (, fv ) and 1 (, fv ) appearing in the fourth expression are defined by

fv (kv1 n1
v nv kv ) dnv dkv

(, fv ) =
Nv

and

fv (kv1 n1
v nv kv ) log (nv ) dnv dkv .

1 (, fv ) =
Nv

If

Kv

Kv

0 1
1 0

then

n=


0

0


n k 

  
 
(n) =   .

Set (s, fv ) equal to

1
L(1 + s, 1v )


Zv \Av

 1s
 v
fv (kv1 a1
n
a
k
)
dav dkv

0 v v 
v
v
Kv

where

av

v
0

and

n0 =

0
v

1 1
0 1


.

1v is the trivial character of Fv . Then (s, fv ) is analytic at least for Re s > 1. It derivative at 0 is
 (0, fv ). If
L(1 + s, 1F ) =
L(1 + s, 1v )
v

Chapter 3

311

the Laurent expansion of L(1 + s, 1F ) about s = 0 is

1
+ 0 + .
s
The operator m(s) is the operator on L(s) which for each (, ) multiplies every element of
s

B(F2 , F 2 ) by

L(1 s, 1 )
.
L(1 + s, 1 )
s

We may represent B(F2 v , F 2 ) as

s

v B(v F2 v , v Fv2 )
s

when s is purely imaginary. If Re s > 0 let R(v , v , s) be the operator from B(v F2 v , v Fv2 ) to
s

B(v Fv2 v F2 v ) defined by setting

R(v , v , s) (g)

equal to

(1

s, 1
v v , v )

L(1 + s, v v1 )
L(s, v v1 )

Nv

0 1
1 0


ng

dn.

These operators can be defined for s purely imaginary by analytic continuation. They are then scalar
multiples of unitary operators and for a given , are in fact unitary for almost all v . Thus R(v , v , s)
s

can be defined as an operator B(v F2 v , v Fv2 ) when s is purely imaginary and

M (s) =



v R(v , v , s)

 L(1 s, 1 )

(,)

Set

L(1 + s, 1 )

N (s) = TS M (s) TS1

and if N  (s) is the deriviative of N (s) set

M  (s) = Ts1 N  (s) Ts

Define R (v , v , s) in a similar fashion. Then

trace M 1 (s) M  (s) (, s)

is the sum of
and

 

trace m1 (s) m (s) (, s)

tr R1 (v , v , s) R (v , v , s) (fv , v , v , s)



(,) v


tr (fw , w , w , s) ,
w=v

where (fv , v , v , s) is the restriction of (fv ) to B(v F2 v , v Fv2 ).

Chapter 3

312
s

If E(, , s) is the projection of L(s) on B(F2 , F 2 ) we can write

m(s) =

a(, , s) E(, , s)

where the a(, , s) are scalars. Thus

trace m1 (s) m (s) (, s)

is equal to


trace (fv , v , v , s) .

 a (, , s) 
a(, , s)

We can also write

M (0) =

a(, ) E(, , 0)

so that

trace M (0) (, 0)
is equal to


trace (fv , v , v , 0) .

a(, )
v

If F is a function field

M
If

  


=
B(, ) E , ,
.
log q
log q



Kv

fv (k 1 ank) dn dk = 0

(16.1.7)

Nv

for all a in Av = AFv then (, fv ) = 0 for all , (0, fv ) = 0, and

trace (fv , v , v , s) = 0
for all v , v , and s. In particular if (16.1.7) is satisifed for at least two v the expressions (iv) to (viii)
vanish and the trace formula simplifies considerably.
We now apply this formula to the function



(g) =


v (gv ) f (
gS )

v (gv )
vS1

vS2

where f = f1 f2 with f1 and f2 in B is of the form

f (
gS ) =

fv (gv ).
vS

Since S has at least two elements and the functions v and v satisfy (16.1.7) only the expressions (i) to
(iii) do not vanish identically. The expression (i) is now equal to




d(v ) f (e)

v (e)
vS1

vS2

Chapter 3

313

We recall that d(v ) = d(v ) if v is in S2 .

We may suppose that Q2 is equal to Q2 and that Q1 is a subset of Q1 . If E is in Q1 or Q2 and is
in BF = BF (E) but not in ZF

BA \GA

(g 1 g) B (g)

is equal to the product of


vS1

Bv \Gv

v (gv1 gv ) B (v)


vS2

and

S \G
S
B

Bv \Gv

v (gv1 gv ) B (v)

f (g 1 g) B .

If v is in S and E F Fv is not a field so that Bv is conjugate to Av the corresponding factor in the first
of these two expressions vanishes. Thus the sum in (ii) need only be taken over Q1 . If E is in Q1 or Q2
the first of these two expressions is equal to

vS

v ( 1 )
.
measure Zv \ Bv

Thus, in the special case under consideration, (ii) is equal to (16.1.5) and (iii) is equal to (16.1.6) so
that

trace (f )




d() measure(ZA GF \ GA ) f (e)

v (e)
vS1

is equal to

trace  (f )

vS2


d(v ) measure(ZA GF \ GA ) f (e).
vS

We may take to be trivial and apply Lemmas 16.1.1 and 16.1.2 to see that, in this case,

trace (f ) = trace  (f )


and

d(v )

measure(ZA GF \ GA )

vS

is equal to



v (e)
vS1

Still taking trivial we choose the

d(v )

measure(ZA GF \ GA ).

vS2

v

so that none of them are one-dimensional and conclude that

measure(ZA GF \ GA ) = measure(ZA GF \ GA ).

(16.1.8)

Then we take exactly one of them to one-dimensional and conclude that v (e) = d(v ). Thus v (e) =
d(v ) and

trace (f ) = trace  (f )

Chapter 3

314

in general.
The relation (16.1.8) is well-known. One can hope however that the proof of it just given can
eventually be used to show that the Tamagawa numbers of two groups which differ only by an inner
twisting are the same or at least differ only by an explicitly given factor. Since the method of [33] can
probably be used to evaluate the Tamagawa numbers of quasi-split groups the problem of evaluating
the Tamagawa numbers of reductive groups would then be solved. However a great deal of work on
the representation theory of groups over local fields remains to be done before this suggestion can be
carried out.
To complete our formal argument we need to sketch a proof of the trace formula itself. One must
use a bootstrap method. The first step, which is all we shall discuss, is to prove it for some simple class
of functions . We take of the form = f  f  with

f  (g) =

fv (gv )
v

and

f  (g) =

fv (gv )
v

fv

fv

where
and
satisfy the five conditions on page 325. The function fv is fv fv .
Suppose is a K -finite compactly supported function in A(). For each purely imaginary s define
(s)
in L by demanding that

1
2c


GF ZA \GA



 , s) 0 (g) = (s),
(g) E(g,

be valid for all  in L. The map (s)

extends to a continuous map of A() onto H . (s)

is the
function in H corresponding to E in A1 ().
For each (, ) in P choose an orthonormal basis {i (, )} of B(, ). We may suppose that any
elementary idemotent in H annihilates all but finitely many elements of this basis. If

(s)


(,)

then

1
ai (, , s) =
2c
Let


GF ZA \GA

(, s) TS1 i (, ) =

ai (, , s) i (, )



g, i (, ), s 0 (g).
(g) E

ji (, , , s) TS1 j (, ).

For all but finitely many , , i and j the functions ji (, , , s) vanish identically. E() is equal to

lim



, i,j

iT
iT



ij (, , , s) aj (, , s) E g, i (, ), s d|s|.

A typical one of these integrals is equal to the integral over GF ZA \ GA of the product of (g) and

iT
iT


 

h, j (, ), s d|s|.
ij (, , , s) E g, i (, ), s E

Chapter 3

315

Thus the kernel of E () is the sum over (, ) and i, j of

1
4c

i
i


 

ij (, , , s) E g, i (, ), s E h, j (, ), s d|s|.

The kernel of () is

(g, h) =

(g 1 h).

ZF \GF

To compute the trace of ()E () we integrate the difference of these two kernels over the diagonal.
The function (g, g) may be written as the sum of

(g 1 1 g),

(16.2.1)

PF \GF NF

=e

where PF is the group of super-triangular matrices in GF ,

1
2

ZF \AF

ZF

AF \GF

(g 1 1 g),

(16.2.2)

where AF is the group of diagonal matrices in GF ,

1 
2
Z \B
Q1

and

F
F

ZF

(g 1 1 g)

(16.2.3)

BF \GF

Q2

ZF \BF

ZF

BF \GF

(g 1 1 g)

(16.2.4)

together with

(e).

(16.2.5)

The constant (e) can be integrated over GF ZA \ GA immediately to give the first term of the
trace formula. The standard manipulations convert (16.2.3) and (16.2.4) into the second and third terms
of the trace formula.
The expressions (16.2.1) and (16.2.2) have to be treated in a more subtle fashion. We can choose a
constant e1 > 0 so that if






1 x
0 1

g=

0
0

with x in A, and in I such that |

| c1 , and k in K and if

g =

1 x
0 1




0

0


k

with in GF , x in A,  and  in I such that |

| c1 , and k  in K then belongs to PF . Let be the

characteristic function of






 

1 x
0 1

0
0

 
k    c1 .

Chapter 3

316

The expression (16.2.2) is the sum of

PF \GF

ZF \PF

ZF NF

PF \GF

ZF \PF

ZF NF

1
2
and

1
2



(g 1 1 g) (g) + (()g)



(g 1 1 g) 1 (g) (()g) .

Here () is any element of GF not in PF such that

() 1 () PF .
There is always at least one such (). The integral of the second sum over GF ZA \ GA converges. It
is equal to


1
2

ZA PF \GA



(g 1 g) 1 (g) (()g) 0 (g).

ZF \PF

ZF NF

Every occurring in the sum can be written as 1 0 with 0 in AF and in PF . Then




1

1 (0 ) ( 1 0 ) 1 (0 )
= 1 (0 )0 1 (0 )

so that we can take () = 1 (0 ) . We take

(0 ) = w =
Since (g) = (g) and
the integrand is

0 1
1 0


.

( 1 wg) = (wg)


ZF \AF

ZF

AF \PF



(g 1 1 g) 1 (g) (wg) .

The integral itself is equal to

1
2


ZF \AF

ZF


ZA AF \GA



(g 1 g) 1 (g) (wg) 0 (g).

All but a finite number of the integrals in this sum are 0.

It is convenient to write each of them in another form. If

g=

0
0


nk

Chapter 3

317

then (g) is 1 if |
| c1 and is 0 if | | < c1 . If

wn =



0

0


n k 

(n)

(n)

and (n) is |
 | then (wg) is 1 if | | c1 and is 0 if | | > c1 . It is easily seen that (n) 1. Thus
if c1 > 1, as we may suppose, one of (g) and (wg) is always 0. The integral


ZA AF \GA

is equal to



(k 1 n1 nk) 2 log c1 log (n) dn dk

c
NA



(g 1 g) 1 (g) (wg) 0 (g)

which we write as the sum of

(k 1 n1 nk) dn dk

2c log c1
NA

and

 

c
NA

(16.2.6)

(k 1 n1 nk) log (nv ) dn dk.

if we express each of the integrals in the second expression as a product of local integrals we obtain the
fourth term of the trace formula. All but a finite number of the integrals are 0 so that the sum is really
finite. We will return to (16.2.6) later. If F is a function field over Fq it is best to take c1 to be a power
of q n of q . Then 2 log c1 is replaced by 2n 1.
The expression (16.2.1) is the sum of

PF \GF

NF

=e

PF \GF

NF

=e

and

(g 1 1 g) (g)



(g 1 1 g) 1 (g) .

The integral of the second expression over GF ZA \ GA converges. It is equal to


PF ZA \GA



(g 1 g) 1 (g) 0 (g).

NF

=e

If

n0 =
the integrand is equal to


NF ZF \PF

1 1
0 1



(g 1 1 n0 g) 1 (g)

Chapter 3

318

so that the integral itself is equal to


NF ZA \GA

which is



(g 1 n0 g) 1 (g) 0 (g)


  1
(k 1 a1 n0 ak) 1 (a)   da dk

c
ZA \AA

if

a=

0
0


.

The integrand vanishes outside of a compact set. Thus the integral is the limit as s approaches 0 from
above of


 


  1s
(k 1 a1 n0 ak) 1 (a)  
da dk,

c
ZA \AA

which is the difference of


c
ZA \AA

and

c
ZA \AA

 1s
 
(k 1 a1 n0 ak) 
da dk

 1s
 
(k 1 a1 n0 ak) 
(a) da dk.

The first of these two expressions is equal to

c
v

Zv \Av

 1s

 v
fv (kv1 a1
dav dkv

v n0 av kv )
v
Kv


which is


(s, fv ) .

c L(1 + s, 1F )

(16.2.7)

Observe that if v is non-archimedean and fv is 0 outside of Zv Kv and is 1 on the elements of Zv Kv of

determinant 1 then


 
Zv \Av

 v 1s
fv (kv1 a1
dav dkv

v n0 av kv )
v
Kv

is the product of the measure of



and

0
0



Zv \ Av  || = ||

|Cvn |1+s = L(1 + s, 1v ),

n=0

so that

(s, fv ) = (s, )
v

Chapter 3

319

is analytic for Re s > 1 and its derivative at 0 is

 (s, fv )


(s, fw ) .
w=v

The function (16.2.7) has a simple pole at s = 0. The constant term in its Laurent expansion is

"
c 0

(0, fv ) + 1



which is the fifth term of the trace formula.

The expression



(k
ZA \AA

is equal to

c
ZA AF \AA

1 1

NF

=e

#

 (0, fv )

(0, fw )
w=v

 1s
 
n0 ak) 
(a) da dk


 1s
(k 1 a1 ak) 
(a) da dk.

Choose a non-trivial character of F \ A and let

 
1 x
1
(y, g) =
g
g (xy) dx.
0 1
A


Then

 1 

 
(y, ag) =   y, g .

Moreover by the Poisson summation formula

(k 1 a1 ak)

NF

=e

is equal to

  1
  1 
 
  y, k +   (0, k) (e).

y=0

The integral

  s
  


c
y, k da dk
  (a)

ZA AF \AA K


y=0

is a holomorphic function of s and its value at s = 0 approaches 0 as c1 approaches . Since we shall

eventually let c1 approach it contributes nothing to the trace formula. If F is a number field


c
ZA AF \AA

 1s
 
(e)  
(a) da dk

is a multiple of

1
1
1+s
1 + s c1

Chapter 3

320

which is defined at s = 0. Its value there approaches 0 as c1 approaches . Finally


c
ZA AF \AA

 s
 
(0, k)   (a) da dk

is equal to

c
s cs1


(0, k) dk.
K

The pole of this function at s = 0 must cancel that of (16.2.7). Consequently


(0, k) dk = 1 (0, ).
K

The constant term in its Laurent expansion about 0 is


c log c1

(0, k) dk.
K

Not this expression but its negative


c log c1

(0, k) dk

(16.2.8)

enters into the integral of the kernel of () E () over the diagonal. If F is a function field s ccs is
1
to be replaced by

cq ns
1 q s

and log c1 by n 12 .
The Poisson summation formula can be used to simplify the remaining part of (16.2.1). We recall
that it is



(g 1 1 g) (g).

PF \GF

NF

=e

We subtract from this

(0, g) (g)

PF \GF

to obtain the difference betwen

(y, g) (g)

PF \GF y=0

and

(e) (g).

PF \GF

The integrals of both these functions over ZA GF \ GA converge and approach 0 as c1 approaches .
They may be ignored.
The remaining part of (16.2.2) is the sum of

1
2

PF \GF

ZF \PF

ZF NF

(g 1 1 g) (g)

Chapter 3
and

321

1
2

PF \GF

ZF \PF

ZF NF



(g 1 1 g) ()g .

These two sums may be written as

1
2
and

1
2

ZF \AF

ZF

AF \GF

ZF \AF

ZF

AF \GF

(g 1 1 g) (g)

(g 1 1 g) (wg).

Replacing by w1 in the second sum we see that the two expressions are equal. Their sum is equal
to twice the first which we write as

1 ZF \AF
1
ZF

(g 1 1 1 2 g) (g).

PF \GF 2 NF

For a given all but finitely many of the sums

(g 1 1 1 2 g) (g)

(16.2.9)

PF \GF 2 NF

are zero. Set

(y, 1 , g) =

1

1 x
0 1

 
g (xy) dx.

The expression (16.2.9) is the sum of

(y, 1 , g) (g)

PF \GF y=0

and

(0, 1 , g) (g).

PF \GF

The first of these two expressions is integrable on GF ZA \ GA and its integral approaches 0 as c1
approaches .
Since (0, g) = (0, e, g) we have expressed (g, g) as the sum of

(0, , g) (g)

(16.2.10)

PF \GF ZF \AF

and a function which can be integrated over GF ZA \ GA to give the first five terms of the trace formula,
the sum of (16.2.8) and one-half of the sum over in ZF \ AF but not in ZF of (16.2.6) which is

c log c1


ZF \AF


NA


K

(k 1 nk) dn dk,

(16.2.11)

Chapter 3

322

and an expression which goes to 0 as c1 approaches .

Now we discuss the kernel of E () in the same way. Set H(g; , , i, j, s) equal to


 

ij (, , , s) E g, i (, ), s E g, j (, ), s .
On the diagonal the kernel of E () is equal to

  1  i
H(g; , , i, j, s) d|s|
4c i
, i,j
if F is a number field and to

  log q 
, i,j

4c

2
log q

H(g; , , i, j, s) d|s|

if F is a function field. We set E1 (g, , s) equal to

 

Ts1 (g) + M (s) TS1 (g) (g)
PF \GF

and let

E2 (g, , s) = E(g, , s) E1 (g, , s).

If, for m = 1, 2, n = 1, 2, Hmn (g; , , i, j, s) is

 


n g, j (, ), s
ij (, , , s) Em g, i (, ), s E
and mn (g) is, at least when F is a number field,

  1  i
Hm,n (g; , , i, j, s) d|s|,
4c i
, i,j
the kernel of E () is

n 
2


mn (g)

m=1 n=1

on the diagonal.
If m or n is 2


GF ZA \GA

is equal to

1
4c

mn (g) 0 (g)



i , i,j

GF ZA \GA


Hm,n (g, , , i, j, s) 0 (g) d|s|.

Take first m = n = 2. If F is a number field a formula for the inner product


GF ZA \GA

2 (g, 2 , s) 0 (g)
E2 (g, 1 , s) E

(16.2.12)

Chapter 3

323

can be inferred from the formulae of [26] and [27]. The result is the sum of



1  2t
c1 (1 , 2 ) c12t N (t + s)1 , N (t + s)2 ,
t0 2t

c lim

1
where N (t + s) = Tt+s M (t + s) Tt+s
, and




1  2s 
c1 1 , N (t + s)2 c12s N (t + s)1 , 2 .
t0 2s

c lim

The second expression is equal to




c  2s 
c1 2 , N (s)2 c12s N (s)1 , 2 .
2s
The first is the sum of

2c log c1 (1 , 2 )
and

 

c  1
N (s) N  (s)1 , 2 + 1 , N 1 (s) N  (s) 2 .
2
If F is a function field over Fq and c1 = q n the inner product is the sum of


c log q

 2(n1)s 1 q s + q s 

1 q s + q s 

q
N (s)1 , 2 q 2(n1)s
,
N
(s)
+
1
2
2s
2s
1q
1q

and

(2n 1)c(1, 2 )

and

Certainly

 

c  1
N (s) N  (s) 1 , 2 + 1 , N 1 (s) N  (s) 2 .
2





ij (, , , s) i (, ), j (, ) = trace (, s)

, i,j

which equals

NA

or

 
c
 
c
,

NA


ZA \AA

  s+1
  2
(k 1 ank) () () 
dn da dk


ZA AF \AA


K Z \A
F
F

Thus if H is the set of all

  s+1
  2
(k 1 ank) () ()  .

0
0

in ZA AF \ AA for which || = ||

1
4c

trace (, s) d|s|
i

(16.2.13)

Chapter 3

324

is equal to

1
2 ,

 
H

which is

 
NA

1
2

(k 1 ank) () () dn dk da

NA

(k 1 nk) dn dk.

When multiplied by 2c log c1 the effect of this is to cancel the term (16.2.11). If F is a function field
(16.2.13) is said to be replaced by

log q
4c
but the conclusion is the same.
The expression


2
log q

trace (, s) d|s|



ij (, , , s) i (, ), N (s)j (, )

, i,j

is equal to

trace M 1 (s) (, s)

when s is purely imaginary and





ij (, , , s) N (s)i (, ), j (, )

, i,j

is equal to

trace M (s) (, s).

Since M (0) = M 1 (0)

1
lim
c1 8

i
i


1  2s
c1 trace M 1 (s) (, s) c12s trace M (s) (, s) d|s|
2

is equal to

1
trace M (0) (, 0).
4
When multiplied by 1 this is the sixth term of the trace formula. For a function field it is to be replaced
by
  
log q 

trace M (0) (, 0) + trace M
,
.
4
log q
log q
When s is purely imaginary


Moreover

 

N 1 (s) N  (s) 1, 2 = 1 , N 1 (s) N  (s) 2 .


, i,j



ij (, , , s) N 1 (s) N  (s) i (, ), j (, )

Chapter 3

325

is equal to

trace M 1 (s) M  (s) (, s).

Thus

1
4

trace M 1 (s) M  (s) (, s) d|s|

is to be added to the trace formula. It gives the seventh and eighth terms.
Next we consider (16.2.12) when m = 2 and n = 1. If 2 = TS1 2 and 2 = M (s) Ts1 2 the
integral

GF ZA \GA

is the sum of

PF ZA \GA

and

PF ZA \GA

Since

2 ,

2

E2 (g, 1 , s) E1 (g, 2 , s) 0 (g)

(16.2.14)

E2 (g, 1 , s) 2 (g) (g) 0(g)

E2 (g, 1 , s)
2 (g) (g) 0(g).

and are all functions on ZA NA PF \ GA while, as is known,


(g)

E2 (ng, 1 , s) dn = 0
NA

when c1 is sufficiently large, the integral (16.2.14) is 0. Thus (16.2.12) is 0 when m = 2 and n = 1 and
also when m = 1 and n = 2.
Set

F (g, , s) = Ts1 (g) + M (s) Ts1 (g)

and set H0 (g, , , i, j, s) equal to


 

ij (, , , s) F g, i (, ), s F g, j (, ), s (g).
If c1 is so large that (1 g) (2 g) = 0 when 1 and 2 do not belong to the same coset of PF the
function 1,1 (g) is equal to

 
, i,j PF \GF

1
4c

i
i

H0 (g, , , i, j, s) d|s|.

If i (g, , ) is the value of TS1 i (, ) at g then

ij (, , , s) i (h, , )
j (g, , )

i,j

is the kernel of (, , , s) which is


c
NA

  s+1
  2
(g 1 anh) 
() () dn da.

ZA \AA

Chapter 3

326

If we set h = g , divide by 4c, integrate from i to i, and then sum over and we obtain

1
2

(0, , g).

ZF \AF

If i (g, , ) is the value of M (s) TS1 i (, ) at g

ij (, , , s) i (h, , )
j (g, , )

i,j

is the kernel of

M (, , s) (, , , s) M (, , s) = (, , , s).

Thus 1,1 (g) is the sum of

(0, , g) (g)

(16.2.15)

PF \GF ZF \AF

and

   (g)  i 

H1 (g, , , i, j, s) + H2 (g, , , i, j, s) d|s|
4c i
, i,j
PF \GF

where H1 (g, , , i, j, s) is
and H2 (g, , , i, j, s) is

ij (, , , s) i (g, , ) j (g, , )

ij (, , , s) i (g, , ) j (g, , ).

The expression (16.2.15) cancels (16.2.10). If g = nak with

a=

0
0


,

H1 (g, , , i, j, s) is equal to
   s+1
 
ij (, , , s)
i (k)
j (k).
 

The functions ij (, , , s) are infinitely differentiable on the imaginary axis. Thus

1
4c


i
i

H1 (g, , , i, j, s) d|s|

M
is, O |
as |
|
| for any real M . Thus if this expression is multiplied by (g) and averaged
over PF \ GF the result is integrable on ZA GF \ GA and its integral approaches 0 as c1 approaches .
Thus it contributes nothing to the trace. Nor do the analogous integrals for H2 (g, , , i, j, s).

Chapter 3

327
References
for Chapter III

The Dirichlet series associated to automorphic forms on a quaternion algebra are discussed in:

ebres simples, I, II Seminaire Bourbaki, 1958/1959

22. Godement, R., Les functions des alg`
23. Shimura, G., On Dirichlet series and abelian varieties attached to automorphic forms, Ann of
Math., vol 76(1962)
24. Tamagawa, T., On -functions of a division algebra, Ann of Math., vol 77(1963)
They are also referred to, but rather obliquely, in:
25. Selberg, A., Discontinuous groups and harmonic analysis, Proc. Int. Cong. Math. (1962)
The theory of Eisenstein series is discussed in [11], [13], [14], and [25] as well as in:
26. Langlands, R.P., On the functional equations satisifed by Eisenstein series, Mimeographed notes
27. Langlands, R.P., Eisenstein series, in Algebraic Groups and Discontinuous Subgroups, Amer.
Math. Soc. (1966)
28. Selberg, A., Harmonic analysis and discontinuous groups in weakly symmetric Riemannian
spaces with applications to Dirichlet series, Jour. Ind. Math. Soc., vol XX (1956)
[25] and [28] are of course the basic references for the Selberg trace formula. Some of its formal
aspects are also described in:
29. Langlands, R.P., Dimension of spaces of automorphic forms, in Algebraic Groups and Discontinuous Subgroups, Amer. Math. Soc. (1966)
The theorem of 16 can still be stated and proved if M is replaced by a quaternion algebra which
splits everywhere that M  does. The proof is in fact rather easier. However these apparently more
general theorems are immediate consequences of the proof of the original theorem. Theorems very
similar to that of 16 and its extensions have been proved by Shimizu. Our methods differ little from
his.
30. Shimizu, H., On discontinuous groups operating on the product of the upper half planes, Ann.
of Math., vol 77(1963)
31. Shimizu, H., On traces of Hecke operators, Jour. Fac. Sci. Univ. Tokyo, vol. 10 (1963)
32. Shimizu, H., On zeta functions of quaternion algebras, Ann. of Math., vol 81(1965)
We have also had occasion to refer to:
33. Langlands, R.P., The volume of the fundamental domain for some arithmetical subgroups of
Chevalley groups, in Algebraic Groups and Discontinuous Subgroups, Amer. Math. Soc. (1966)
Orthogonality relations for the characters of non-compact groups first appeared in:
34. HarishChandra, Discrete series for semisimple Lie groups, II, Acta Math., vol. 116 (1966)
We thank R. Lipsman for the reference to:
35. Rickert, Neil. W., Convolution of L2 -functions, Coll. Math, v. XIX (1968)