Asens 1978 4 11 4 471 0

A NNALES SCIENTIFIQUES DE L’É.N.S.
S TEPHEN G ELBART
H ERVÉ JACQUET
A relation between automorphic representations of GL(2) and GL(3)
Annales scientifiques de l’É.N.S. 4e série, tome 11, no 4 (1978), p. 471-542
<http://www.numdam.org/item?id=ASENS_1978_4_11_4_471_0>
© Gauthier-Villars (Éditions scientifiques et médicales Elsevier), 1978, tous droits réservés.

L’accès aux archives de la revue « Annales scientifiques de l’É.N.S. » (http://www.
elsevier.com/locate/ansens) implique l’accord avec les conditions générales d’utilisation
(http://www.numdam.org/conditions). Toute utilisation commerciale ou impression systé-
matique est constitutive d’une infraction pénale. Toute copie ou impression de ce fi-
chier doit contenir la présente mention de copyright.
Article numérisé dans le cadre du programme

Numérisation de documents anciens mathématiques
http://www.numdam.org/
Ann. scient. 6c. Norm. Sup.,
4s serie, t. 11, 1978, p. 471 a 542.
A RELATION BETWEEN
AUTOMORPHIC REPRESENTATIONS OF GL (2) AND GL (3)'
BY STEPHEN GELBART** AND HERV^ JACQUET
Some notations
We will denote by Gy the group GL (r), by B,. the standard Borel subgroup of GL (r)
(upper triangular matrices), by N,. the unipotent radical of By, and by Zy the center of G,..
On the other hand, we will denote by G the group SL (2), by B its standard Borel subgroup,
by N = N2 the unipotent radical group of B, and by A the subgroup of diagonal matrices
in G. Finally P will be the standard parabolic subgroup of type (2, 1) in G3 and U its
unipotent radical.
If F is a local non archimedean field, we denote by 9lp or ^ ê ring of integers in F,
by ^PF or ^P its maximal ideal, and by £ a generator of ^.Finally we set K,. = GL (r, SRp)?
K = SL (2, 9lp). If F is archimedean the same notations will be used for the standard
maximal compact subgroups. When F is local, we will often write Gy, By, . . . for Gy (F),
B,(F)...
Introduction
The purpose of this paper is to establish a relation between automorphic forms on

GL (2) and GL (3). To formulate our main result more precisely we need first to recall
some basic facts.
Suppose F is an A-field and n is an irreducible unitary representation of GL (r. A).
Then n can be written as an infinite tensor product TT = ® Ky where TTy is an irreducible
v
representation of Gy = GL (r, Fy) for each place v. Moreover, for almost all finite v
the representation Ky is unramified, i. e., contains the trivial representation of a maximal
compact subgroup of Gy. But such representations are well known to be parameterized
by semi-simple conjugacy classes in GL (r, C). Thus one can associate to n a family of
semi-simple conjugacy classes (^y) in GL (r, C), dy being defined for almost all finite v.
* Both authors have been supported by grants from the N.S.F.

** Alfred P. Sloan Fellow.
ANNALES SCEENTIFIQUES DE L'ECOLE NORMALE SUPERIEURE

472 S. GELBART AND H. JACQUET
Now suppose <T is an automorphic representation of GL (2, A), i. e., CT occurs in the
space of automorphic forms on GL (2). If (&„) is the corresponding family of conjugacy
classes in GL (2, C), define a family of conjugacy classes in GL (3, C) by
^mn~ 1 0 0
a^[ 0 1 0 if b^ . ,
\ 0 0 m^nj ^0 n -
Our main result then asserts that there is an essentially unique automorphic represen-
tation of GL (3, A)—the "lift of <r— whose corresponding family is (^y).
To see how this "lifting" appears as a special case of the more general conjectures of
Langlands the reader can consult [Bo], Roughly speaking, if Wy denotes the Weil-
Deligne group of F(,, and a = ® <jy, then each <7y corresponds (sometimes only conjectu-
rally) to a two-dimensional representation (py of Wy. So if p denotes the map
GL(2, C)——p—^GL(3, C)
PGL(2, C)
determined by the adjoint action of PGL (2, C), then (according to Langland's general
philosophy) the resulting three dimensional representation
po(p,: W,-^GL(3,C)
should correspond to an irreducible admissible representation n^ = p* (a,,) of GL (3, Fy).

This is the local "lifting" of <jy to GL (3). Globally, the product n == ® ^ should (and
v
does!) define an automorphic representation of GL(3, A).
The idea of our proof can be explained as follows. Following [GoJa] one can attach
to each irreducible representation n of GL (r. A) an infinite Euler product
L(s,7c)=riL(s,7i,),
with
L^Ti^detO-Ô -i
for almost all v. This product converges in some right half-plane. Moreover, if TT is
automorphic cuspidal, i. e., n occurs in the space of cusp forms, then L (s, n) continues to
a holomorphic function of s and satisfies a functional equation. For r = 2 or 3 there is
also a converse; (c/. [JPSS]). For each character % of the idele class group consider the
representation g i-» (g) % (dot (g)) (also denoted n ® %). Then if L (s, n ® %) satisfies the
above conditions for each %, 7t must be automorphic cuspidal.
4° SfiRIE — TOMB 11 — 1978 — N° 4

RELATION BETWEEN AUTOMORPHIC REPRESENTATIONS OF GL (2) AND GL (3) 473
Now recall that to each pair of irreducible representations (n^ n^) °f GL (2, A) one can
attach an infinite Euler product
L(s, TCiXTia) =^ L ( S » î^x71^)?

v
with
L^Tri^XTC^^detCl-^fli^® fl2.i,)~S
for almost all v; (cf. [Ja]). Ifn^ and 71:2 are automorphic then this infinite product continues
to a meromorphic function of s and satisfies a functional equation In particular, if
TCi = a ® 5C and 71:2 = 5- (the contragredient), then:
L(s, (a ® /) x a) = L^ (s, a, x) L(s, x),
where L (s, %) is the usual Hecke L-function attached to % and L^ (^, o, /) is a new Euler
product of degree 3. If TT is automorphic, then L^ (s, CT, %) is also meromorphic and
satisfies a functional equation. These facts are discussed in Sections 1 and 2.
In Section 3 we introduce a representation n of GL (3, A)—the "lift" of a—by requiring
that
L(s,7t®^)=L2(s,a,/).
At almost all places, <Jy is unramified, and the relation between n^ and Gy is the one already
explained. The situation at the remaining places is discussed in Section 3.
The fact that L^ (s, a, 7) has no poles at all is proved using the method of [Sh]. Our
main theorem then results from the converse theorem for GL (3) already alluded to.
To establish the holomorphy of L^ (s, <J, %) we generalize [Sh] by considering integrals
on the metaplectic group Mp (A). This group is an extension of SL (2, A) by the torus T
which splits over SL (2, F). Thus any form on SL (2, A) may be regarded as a form on
Mp (A). The "main integral" we consider is an integral over T SL (2, F)\(Mp (A)) of
the product a form on GL (2, A) belonging to 71, an Eisenstein series on Mp (A), and a
form on the metaplectic group given by a theta series. In Sections 5 and 6 we show that
this integral decomposes as a product of integrals over local metaplectic groups and equals
La (s, a, 5c) for a suitable choice of functions involved. Thus we need only analytically
continue the Eisenstein series. This is carried out in Section 8 after analyzing the "constant
term" in Section 7.
Actually, our paper deviates a bit from the simple scheme just outlined-for technical
reasons discussed within the text. We also have to complete the results of [JPSS] in
Section 4.
Many of the results of this paper were announced earlier by us in [GeJ 2]. We are
grateful to Arletta Havlik, Joanne Martin, Margeret Reif and Diana Shye for their expedi-
tious typing of the manuscript. The second-named author would also like to thank
R. P. Langlands for pointing out the results of Corollary 1.7, and W. Casselman for
bringing to his attention the results of Theorem 2.2.
ANNALES SCIENTDFIQUES DE L'ECOLE NORMALE SUPERIEURE

1. GL (2) x GL (2): local theory
In this Section (and the next) we complete the theory of automorphic forms on
GL (2) x GL (2) developed in [Ja]. In particular, we describe the poles of L (s, n^ x n^)
both locally and globally. Although some of our results are not used later in this paper
they all seem to be of independent interest.
(1.1) Let F denote a local non-archimedean field, q the cardinality of the residue class
field, Vp or v the normalized valuation on F, and [ [p or | | the corresponding absolue
value. Then | x |p = q~v °°. When convenient, we also let let a? or a denote the absolute
value. By v|/p or \|/ we denote a fixed non-trivial additive character of F.
If (Tii, TT^) is a pair of irreducible admissible infinite-dimensional representations of
GL (2, F) then L {s, n^xn^) will denote the Euler-factor defined in [Ja].
In this section we shall recall the definition of L (s, n^ x n^) and complete the results
of [Ja] by computing this factor in all cases.
Let IT (jr,; \]/p) denote the Whittaker model of TT; (with respect to v|/p). Recall that this
is a space of functions W on GL (2, F) satisfying the following properties :
(1.1.1) IT (TCf; \l/p) is invariant for the right action of G^ (F) and the resulting repre-
sentation of G^ (F) on "W (TC,; \|/p) is equivalent to TT,;
(1.1.2) for each W in ^ (n,; \|/p):
^ ( S Î^F^)^] for each êF and geG^F).
The existence and uniqueness of such a space is proved in Chapter I of [JL].

For Wf e IT (n^ \|/), 0 a Schwartz-Bruhat function on F2, and s e C, set
(1.1.3) y(5,Wi,W2,<D)= f Wi(g)W20lg)0[(0, l)g]|detg|^g,

J N2 (F)\G2 (F)
where
/I 0\
"'(o -1}
Then the factor L (s, n^xn^) arises as the "g. c. d." of these integrals. More precisely,
these integrals converge for Re (s) sufficiently large and define rational functions of ^~ 5 ;
since the subvector space of C (^-s) spanned by them is a fractional ideal of C [y"5, q5]
which contains 1 it has a unique generator of the form P~ 1 (^-s) with P e C [^~5] and
P(0) = 1: this generator is-by definition— L(s, n^xu^).
Recall also that if W is mi^ (n^; vj/p) then the function W defined by
(1.1.4) W(g)=W[u/g- 1 ], ^ = ( ^ ^)
belongs to the Whittaker model 1^ (S,; \|/p) of the representation 7t, contragredient to TC,.
4° SERIE — TOME 11 — 1978 — N° 4

RELATION BETWEEN AUTOMORPfflC REPRESENTATIONS OF GL (2) AND GL (3) 475
Finally the factor s (s, n^ x n^; ^p)— which is a monomial in g^—is defined by the func-
tional equation:
(1.1.5) ^(l-s, Wi, ^2, 6)/L(1-5, ^ x^)

= £(S, T C i X 7 C 2 ; \|/F)^(S,WI, W2,0)/L(S, T^ X^).
Here
(1.1.6) ^(X, }Q == $(M, l;)\|/F(MX+^)rfM^
is the Fourier transform of 0 (with respect to the self-dual Haar measure). Denote by
CD( the central quasi-character of n^
PROPOSITION (1.2). - Suppose 7i2 is super cuspidal. Then the poles of L (s, n^ x n^) are
simple. If a5 ® 71:1 denotes the representation g\-> \ detg ^n^ (g) then SQ is a pole of
L(s, n^xu^) if and only if Vs0 ® n^ is equivalent to n^9
Proof. — It is clear that SQ is a pole of L (s, n^ x n^) if and only if it is a pole
x
? (s, W\, W^, 0) for at least one choice of W^, W^, and 0. Moreover, the multiplicity
of SQ as a pole of L (s, n^ x n^) is the maximum multiplicity of SQ as a pole of each of these
integrals.
By Iwasawa's decomposition,
^W,W,$)=JJ^W,[(;; ?)^]W3[(7 ^fW^d-adk,
with
K = GL(2, R^) and f(g) = | detg| 5 fo[(0, Qg] | t^^^d^.
But 71:2 is supercuspidal. Therefore ([JL], Prop. (2.16)) the function
—'[Co ?)]
has compact support in F". Thus every pole of ^ must come from the integral defining/.
In particular, if SQ is a pole, then SQ is a simple pole, and a250 = (®i 0)2)-1.
So suppose now that SQ satisfies a250 = (©i 0)2) ~1. Then
lim(s-so) p[(0, 0] |f| 2s (o)l0)2)- l (0d^= c0(0, 0),

s-^so J
with c a non-zero constant independent of 0. It follows that
lim(s-so)^(s, Wi, W2, 0) = c0(0, 0)b(Wi, W2),

s-^so
ANNALES SCIENTIFIQUES DE L'ECOLE NORMALE SUPERIEURE 62

with
b (Wi, W2) = f Wi (g) W2 (TI g) I det g Is0 dg.
JZ2(F)N2(F)\G2(F)
Thus .SQ is a pole of L (s, n^ x n^) if and only if the bilinear form b is not identically zero.
Note that integrand defining b is indeed invariant on the left by N3 (F) Z^ (F) and compactly
supported mod N2 (F) Z^ (F). But since TT, acts by right translations in i^ (n^ \[/),
&(7Ci(g)W,, 7r,(g)W,) = Idetgl^^Wi, W,).
Thus if .So is a pole of L (.y, CTI x 02), 6 is non-zero and a50 ® 7Ci is contragredient to n^-
Conversely, suppose (Xs0 ® 71^ ^ ^2. Since we may replace the pair (71:1, n^) by
(Tii ® a^ 7i2 ® oi'O, we may assume without loss of generality that ©2 is unitary. Then
7i2 is preunitary, and a50 ® n^ is imaginary conjugate to TT^. In particular, if W\ is in
fw
'(^ ^p)? Aen the function g^\deig^°W^(g) belongs to ^ (TT^; vj/p). But the
function g i-> W^ [r| g'] also belongs to i^ (n^; il/p). Therefore b is not identically zero,
and SQ is a pole.
Q.E.D.
COROLLARY (1.3). — Suppose n is a supercuspidal representation and T| is the unramified

quadratic character o/F^. Ifn is not equivalent to n ® T|, then:
L(5,7CX?C)=(1-^- S )- 1 ;
otherwise
L(S, 71X71) =(1-^- 25 )^ 1 .
Proof. — The number ô is a pole of L (s, n x TT) if and only if n ® a30 w 71. But this
relation implies a250 = 1, that is a50 = T| or 1. Since the poles ofL are simple, the Corol-
lary follows.
For completeness, and future reference, we now give the values of L and e when n^ is
not supercuspidal.
PROPOSITION (1.4). - Suppose 7i2 = TC (|^, v^) (using the notation o/[JL], § 3). Then:
^ 4 ^ f L(S, ^l X7^) = L(5,-TCi ®|^)L(S, TCl 8)^

[ L(s, î x^) = L(5, Tii ®^2' l )L(s, 5ri ®v;1)
and
8(5, TCi X7t2; \1/) = 8(S, TTi ® ^; ^F)s(s, 71:1 ® V 2 ; vl/p).
Suppose TT^ ^ ^^ special representation 0(^2, V2) m^ 1^2 ^'I1 :== a

^M? î ls not
special
(loc. cit.). Then:
(1.4.2) L(5, Tli X7l2) = L(S, TTi (g) ^2), L(s, TCi ^2) = L(s, ^l ®V2' 1 )
4° StKEB, — TOME 11 — 1978 — N° 4

and
8(S, TCi X7t2; \|/F) = e(s, Tli ®jl2; \MS(S, 7ti ® V 2 ; ^p) - , s> nl ^2 '
L(S, TTi ®V2)
FWfy suppose TC( = CT(|A(, v<), H^v,"1 = ap, i = 1,2. 77^;
^,. 4. ^j; f L(s, î X7C2) = L(5, Hi U2)L(s, Vi ^2).

• • 1 L(S, %i x%2) = L(5, vr^^, nrsr1),
and
8(5, TCi X7C2; V^F) = s(s, TCi ® ^2; ^p)8^ ^1 ® V 2 ; vl/p)-
Here the L ^zrf e factors on the right are those of [JL].

Proof. - For the first two assertions see Theorem 15.1 in [Ja]. We prove the third
one since—contrary to what is implied in [Ja] — Theorem 15.1 does not apply to this case.
Recall that IT (n^; \|/p) is a subspace of codimension one of the space denoted
ifT (^, vi; vl/p) in [JL] (§ 3, p. 94). Although the representation of €2 (F) in this larger
space is reducible the assertions recalled in (1.1) still apply to this space. Similarly (1.4.1)
and (1.4.2) apply with the obvious modifications. In particular the "g. c. d." of the
integrals ^ (s, Wi, W2, 0) with Wi e IT (^, v^; vl/p) and W2 e HT (n^, v^) is
L(S, 7t2 ®Pi)L(s, 7T2 ®Vi) = L(s, ^^^(s, V^^
Similarly the "g. c. d." of the integrals ^ (s, W^, ^2, 0) with Wi and W2 in the same
spaces is
L(s, n^^1)^, ^2 ®vi- 1 ) = L(s, vi^y^)^ v^v^)
and there is a similar assertion for the s-factor.

We conclude that the quotients
L(S, Tti X 712)/L(S, Hi H2)L(5, Vi ^2)
and
L(S, 7liX7C2)/L(s, Vl~ l V2 1 )L(5, Hi' 1 ^ 1 )
are polynomials. To see that they are identically one—as we must- we observe that
appropriate functional equations imply that the quotients L (1 —s, n^xu^)/^ (s, n^xn^)
and
L(l-s, v^v^^L^-s, ^v^1)/!^, Hi^2)L(s, Vi^)
can differ only by a constant times a power of ^-s. But the second quotient has coprime
numerator and denominator. Thus the required identities follow.
Remark (1.4.4). - If n is the one-dimensional representation g h-» p, (det g) then it can
be denoted ^(pa172, ^la"172) as in [JL] (p. 104). Thus one can use (1.4) to define the
factors L and e for all pairs of irreducible admissible representations of G^ (F).
ANNALES SCDBNTIFIQUES DE L'^COLE NORMALE SUP&UEURE

(1.5) Recall that (conjecturally) there is a bijection between the set of classes of 2 dimen-
sional representations of the Weil-Deligne group of F and the set of classes of admissible
irreducible representations of G^ (F). This bijection, denoted
<Jh->7C(CT),
should satisfy the following properties:
(1.5.1) n (oT =71(0), 7i:(CT®5c)=7r(a)®/,
(1.5.2) det (cr) = central quasi-character of n (cr),
(1.5.3) L(s,7i(a)®x)=L(s,CT®x),
(1.5.4) L(s, ?c(or) ®x) = L(s, a ®x),
and
(1-5.5) £(s,7t(a)(x)5C;v|/)=8(s,CT®5C;\|/);
(c/. [De]); the factors on the left are those of [JL] and the ones on the right are those
of [De], [La 2].
According to the conjectures of Langlands one should also have
(1.5.6) L(S, 7r(CT)X7T(T)) = L(5, CT ®T),
and
(1.5.7) E(S, 7I;(CT)X7C(T); V|/F)=£(S, <T®T;^F).
The factors on the left are now the ones described in (1.1) and the ones on the right are
again those of [De], [La 2].
We note that (1.5.6) and (1.5.7) follow from (1.4) except when both or and T are irredu-
cible representations of the Weil-group. In that case the existence of n (or) and n (r) is
perhaps not yet known. Nevertheless we have the following proposition (which completes
Corollary 19.16 of [Ja]):
PROPOSITION (1.6). — Regard F as the completion at some place v of some global

field K. Let a and T be two-dimensional irreducible representations of the Weil-group Wp
satisfying the following properties:
(i) there exist two dimensional representations S and T of the Weil-group WK whose
compositions with the morphism Wp —» W^ are CT and T;
(ii) for each place wofK let Ey, and T^ denote the compositions of I. and T with the mor-
phism Wy, —> WK," then n (2J and n (TJ exist;
(iii) the representations ® n (2^) and ® n (TJ are automorphic cuspidal.
Then (1.5.6) and (1.5.7) hold.
4° S^RIE -~ TOME 11 -- 1978 — N° 4

Proof. - By Corollary 19.16 of [Ja],

L (1 — s, n (<r) x n (r)) s (s, n (a) x n (r), vhQ/L (s, n (a) x n (r))
= L(l—s, a®T)s(s, <J®T; \|/p)/L(5, a®T).
Thus it suffices to prove (1.5.6). Since there is no harm in assuming a and T unitary and
irreducible, n (a) and n (r) can also be taken to be preunitary and super cuspidal.
If SQ is a pole of L (s, CT ® r) then the representation a50 0 a is equivalent to the repre-
sentation ? ([JL], Lemma 12.4); thus SQ is purely imaginary, and the rational fraction
L(5, CT®T)/L(1-S, CT®^)
is irreducible in the ring C [^"s, ^s]. But by Proposition (1.2) the same is true of
L (5, 71 (CT) X K (T))/L (1 - S, ?C (CT) X K (z)),
and since these fraction differ only by a unit C [^"-s, ^s], they have in fact the same nume-
rator.
Q.E.D.
COROLLARY (1.7). — Suppose a and T are as in (1.6). Ifn (a) ^ n (a), then a ^ T.
Proof. — Indeed 0 is a pole of L (*?, 71 (a) x n (?)). Since T? (r) = TT (?) and the pair (a, ?)
satisfies the assumptions of (1.6) we find that 0 is also a pole of L (s, a ® ?). Thus our
conclusion follows again from [JL] (Lemma 12.4).
(1.8) Suppose a and T are "dihedral representations" of the Weil group Wp. This
means that there are separable quadratic extensions K and H of F, quasi-characters %
and 9 of K^ and W so that
(1.8.1) a = Ind(WF, WK, x) and T = Ind(WF, WH, 9).
Then n (a) and n (r) exist and (1.5.6) and (1.5.7) are satisfied. This is clear by (1.4.1)
if either a on T is reducible. Otherwise it follows from (1.6).
In particular, if the residual characteristic of F is not 2, the bijection cr \—> n (c) exists
and the relations [(1.5.6), (1.5.7)] apply to all pairs of representations.
(1.9) Now we complete the results of [Ja] (§ 20) (although this will not be needed later
in the paper). Let K be a separable quadratic extension of F and set \|/K = ij/p o Tr^/p-
Then for any quasi-character 9 of F^,
(1.9.1) L(5, 9oNK/F) = L(5, 9)L(5, 9Q,

and
1.9.2) X(K/F, v]/F)8(5, 9oNK/F ; vk) = s(5, 9; ^(s, 9^; v^).
Here ig is the quadratic character of F^ attached to K and X (K/F, \|/p) is a contant whichs
depends on the extension and the choice of \|/p (cf. [JL], p. 6).
ANNALES SCIENTIFIQUES DE L'ECOLE NORMALE SUPERIEURE

If % is any character of K^ let CT^ be the representation defined by (1.8.1). If a is

any admissible irreducible reprensentation of G^ (F) and n is an admissible irreducible
representation of G^ (K) we way that n is a base change lifting of CT (a strict lifting in the
terminology of [Ja]) if the following conditions are satisfied:
(1.9.3) the central quasi-characters o and co' of CT and n satisfy the relation
<O'=O°NK/F;
(1.9.4) for any quasi-character % of K",
L(s, CT x 7c(CT^)) = L(s, n ® %)
L(5, cr x n(a^)) = L(s, n ® 50)

and
s(s, CTX7c(CT,); v^) = ^(K/F^ê^ n ® x ; ^
We note that if CT exists at all it is unique. Moreover, suppose CT = n (r) where T is a

2-dimensional representation of the Weil-Deligne group of F. Suppose also that (1.5.6)
and (1.5.7) are true for each pair (n(G^), TC(T)). Let T' be the restriction of T to the
Weil-Deligne group of K. If n (r') exists it is clearly a lifting of n (r). Thus the
existence of a lifting is established in all cases except the following: the residual characte-
ristic of F is 2 and n is a supercuspidal representation not of the form n (<jy), ^ a quasi-
character of K^. Nevertheless the following Lemma holds:
LEMMA (1.9.5). — Suppose CT is a supercuspidal representation ofG^ (F) not of the form
K (CT ). Suppose K is an irreducible admissible representation of G^ (K) satisfying (1.9.3)
and the following condition:
E(S, CTXTl(CT^); \|/p)L(l-5, CTX7C(CT^))/L(5, CTXTC(CT^),
= e(5,7i ®%; \|/K)L(I-S, n ®3C~1)/L(5, n ®%).
Then n is a supercuspidal representation of G^ (F) and a lifting of CT.

Proof. - By Proposition (1.2):
L(S, CT X ^t(CT^)) = L(S, CT X CT(CT^) = 1.
Thus
L(l-s, K g)^"1)^, K ®%)
is monomial for all 50. Since this can happen only if n is supercuspidal ([JPSS], § 7),
L(5, TC ®;c) = L(5, ii ®x"'1) = 1,
and (1.9.4) is satisfied.
4° SERIE — TOME 11 — 1978 — N° 4

RELATION BETWEEN AUTOMORPH1C REPRESENTATIONS OF GL (2) AND GL (3) 481
Examples (1.9.6). - In the Table below, the representation on the right is a base change
lifting of the representation on the left.
a = n (î, Ha), ^ == ^ (^ o NK/F . ^2 ° NK/F)»

a = CT(pi, ^)? l 1 !-^ 1 == ^ 7l
= ^(^NK/F^ I^NK/F^
a = n (CT^), x quasi-character of J^, n = n (%, %') where ^/ is the conjugate of x by the

non-trivial element of Gal (K/F).
(1.10) The discussion above applies to the case of an archimedean field too. In this
case the bijection a^-> n (a) is known and the factors L{s, n^ XTI^)- defined by (1.5.6)
and (1.5.7)-are still related to integrals like (1.1.3) (c/. [Ja], § 17, § 18). The notion of
a base-change lifting is also defined and the first and third examples of (1.9.6) now cover
all cases.
Remark (1.10.1). - For all fields F the representation aQ^, ^2) where \ti^21 = ^F
can be defined as the infinite dimensional component of the induced representation:
Ind(G2, B2, Hi, 112).
Of course if F = R or C then CT (î, ^2) = ^ (K)for a suitable ^,
2. GL (2) x GL (2): global theory
(2.1) In this section F will denote a global field and \|/ = ]"[ \K, a non-trivial additive
v
character of A/F. By TC,, i = 1,2, we denote an irreducible admissible representation of
(the Hecke algebra of) G^ (A), with central quasi-character co,. We assume co» is trivial
on F". Then 7 1 , = ® n^ „ and we set:
(2.1.1) L(s, Tii x 712) = n L(s, TCi, „ x 712,»,),
v
(2.1.2) L(s,TCiX7T2)=nM^l.i,><^2.,^
v
(2.1.3) e(s, Tii x 71:2)= II^71!.^ ^2.1;; ^)-

u
For almost all v the representation n^vis unramified and has the form
(2.1.4) ^,=7c(^„v,^),^,(x)=|x|: < *^v^,(x)=|x[ t<>v .
We assume also that there is a constant c such that

(2.1.5) -c^Re(s^)^c, -c^Re((,^)=c
for almost all v. Note that this condition is always satisfied (with c = 1/2) if n^ and n^
are preunitary. In any case (2.1.5) implies (2.1.1) and (2.1.2) converge absolutely in
some half-plane Re (s) > SQ. As for (2.1.3), it is actually independent of \|/ and almost

all its factors are equal to 1; in particular, as a function of s, it is just a constant times an
exponential function of s.
THEOREM (2.2). — (1) Suppose Tii and n^ ^re automorphic. Then the functions
L (s, TCi x n^) and L(s, n^ xn^)—originally defined only in some half-plane—continue as
meromorphic functions to C.
(2) IfF is a number field these functions have only finitely many poles and are bounded at
infinity in vertical strips. If¥ is a function field whose field of constants has Q elements
then they are rational functions o/Q~ 5 .
(3) These functions satisfy the functional equation
L(s, 7ii xn^) = s(5,7Ci xn^)L(l—s, n^ xn^).
(4) If Tti is cuspidal but n^, ls not ^^ L (s, n^ x n^) is entire, ffiti and n^ are cuspidal
then the poles ofL, (s, 711 x 71:2) are simple. Moreover SQ is a pole if and only if
a50®^!^^ or
a1"50 ®7Ti =7C2.
Proof. — Suppose n^ is not cuspidal. Then by Theorem (10.10) of [JL] there are two
quasi-characters p, v of F^/F^ such that for all v, n^,^ is a component of the representation
P (IV ^ of ^2, v induced by the quasi-character of B^ y given by the pair (|y v,,) (loc. cit.).
Actually, if p (|Xy, Vy) is irreducible then n (^, v^) = p (^, v,,); if not, then p (|^, Vy) has
two irreducible components n (\iy, Vy) and CT (\iy, Vy). If ^ly, Vy are unramified then n (|A(,, v,,)
is unramified. In any case, for almost all v,
(2.2.5)
{ ^2.*;='"^^
L(5, 7I;i^ X TT^^) =L(S, 7Ti^,®^)L(s, Tr^^V^),
L ( 5 , ? C i ^ X T C 2 ^ ) = L ( 5 , TCl^®^ l )L(5, 7Ti^®V^ 1 ),
and
S(S, ^.^^^^^(^l^®^ ^S^Tli^^V^; Vl/j.
On the other hand, for all v, the ratios
L (S, TTi ^ X 7C2. t,)/L (5, Ki ^ ® ^) L (5, ^ ^ ® V^),
L(5,-iii^x^.t,)/L(5, icl^®^~ l )L(s, 7Ci^®v<7 1 )
are entire and

8(5, TCi^XT^; ^)L(1-5, 7ri^X%2,,,)/L(5, ^l.t/X^.t;)
= SO, TCi^ ® ^; \|^) L(l -5', ?T^ „ ® ^- ^/L^, TCi, „ ® ^)
x
£ ( ^ ^ l . i , ® V ^ ; \l/„)L(l-5•,7Tl^®V^," l )/L(^7I;l^®V„),
as can be checked directly. Thus our assertion follows from the known analytic properties
of L (s, Tti ® n) and L (^ Tii ® v). (Corollary 11.2 w [JL].)
4° SERBS - TOME 11 - 1978 - N° 4

Suppose now both n^ and n^. are cuspidal. By Theorem 19.14 of [Ja] it suffices to
prove (4). For WÎ^TI,; \|/) (notations of [JL], § 10 and § 11) and 0 a Schwartz-
Bruhat function on A2, define an integralx? (s, W\, W^, 9) by (1.1.3) the integral now
being on N^ (A)\G^ (A). These integrals converge in some right half-space and L (s, n^ x n^)
is a linear combinaison of integralsx?. Conversely, at least for <D in a certain dense subset,
each integral ^¥ is equal to the L-factor times an entire function ofs.
Now set
',\(: °u
^-.rfs ;>]-
Then (p, is a cusp form which belongs to a space realizing the representation n^ Consider
also for each quasi-character co of F^/F^ the Eisenstein series E (g, 0, CD, s) defined by
E(g, 0, co, 5) = 2/(yg), YeB,(F)\G^(F),
/(g)=f^[(0,0g]|^| 2s co(0d^.
J^
Then, for Res large enough,
^(S, Wi, W,, 0) = f<Pi(g)(p2(g)Efe, ^ <OlC02, 5)dg,
geZ,(A)G,(F)\G^(A).
(cf. [Ja], § 19.) This shows the left hand side is meromorphic. Moreover, if SQ is a pole
of the left hand-side, it is a pole of the Eisenstein-series. Thus SQ is a simple pole and
yi -2so ^ ^ ^ Qj. o^2so == coi'1 coj1. In the first case
limCs-^^.Wi.WÔ)
s-»so
= ci$(0, 0) f|detg|so-l(pl(g)(p2(g)dg, geG2(F)Z2(A)\G2(A).
In the second case
limCs-SoW^Wi.W,,^)
S-*SQ
= c,0(0, 0)f|detg|so(p,(g)(p2(g)dg, geG,(F)Z,(A)\G,(A).
Here c^, c^ are non zero constants.

Arguing as in the local case we find first that any pole SQ of L (s, n^ x n^) is simple and
such that a250 = (0)1 co^)"1 or a2"250 = (®i co^)"1. If SQ is any number such that
a250 = (co^ o^)"1 it will be a pole of L if and only if the bilinear form
| det g Is0 (pi (g) (p2 (g) dg

((pi, (p^) ^.f|detg|
J

is not identically zero, that is, a50 (X) 711 is equivalent to n^ - Conversely if SQ is any number
such that a50 (x) n^ ^n^ Aen a50 = (©i 0)2) ~ 1 and the above argument applies. The
other case is handled similarly.
(2.3) We can now complete the global results of [Ja] (§ 20). Suppose n is an admissible
irreducible representation of the (Hecke algebra) of G^ (F^) and K is a separable quadratic
extension of F. Then an admissible irreducible representation II of G^ (K^) is said to
be a base-change lifting of n if the following conditions are satisfied:
(2.3.1) Suppose v is a place of F which splits in K, with w^ w^ the corresponding places
of K, so that F, ^ K^ ^ K,, and G^,, ^ G^, „ ^ G^, ^; then TT, ^ n,, ^ II,,.
(2.3.2) Suppose ms a place of F which does not split in K, and w is the place of K above
v, so that Ky, is a quadratic extension of Fy; then 11^ is a base change lifting of Tiy in the
sense of (1.9) above.
Suppose % is a character of K^/K". Ifi; is a place of F which does not split let
TCy = n (<j^) [using the notation of (2.3.2)]; otherwise let n^ = n (/^, %^) [notation of
(2.3.1)]. Then set n (c^) = ® Tiy. It is clear that n (o^) admits a lifting II; indeed
IL, = 7t (X«?» X'J where 50' is the conjugate of ^ under the action of the Galois group.
One can show that both n (o-^) and 11 are automorphic; ([La 5]). If n is automorphic
cuspidal but not of the form n (c^) then the following is true:
PROPOSITION (2.3.3). — Let n be an automorphic cuspidal representation of G^ (F^)
not of the form n (<7^). Then n admits a base change lifting II to G^ (K^) and H is
automorphic cuspidal.
Proof. — Let S be the set (finite and possibly empty) of places v of F of residual charac-
teristic 2 where n^ is supercuspidal but not of the form n (c^) [here 50 is a quasi-character
of K^ and we use the notations of (2.3.2)]. If v does not split and is not in S then TT(,
admits a lifting 11^. If v is in S (and does not split) then by Theorem (20.6) of [Ja] we
at least know there exists a representation 11^ satisfying the conditions of Lemma (1.9.5).
So by this same Lemma, 11^ is actually a lifting of TC«, and we conclude n has a lifting II.
Since we may assume n to be preunitary it is easily checked that 11 is preunitary and the II y,
are infinite-dimensional. Moreover (cf. (20.6) of [Ja]),
L(5, 11 ® 50) = L(S, 71 X 7T(<7^)),
L(s, n ®x~ 1 ) = L(s, nxn(Gy)),

and
s(s, n ® 50) = s(s, 7i X7r(c^)).
It follows now from (2.2) that L (s, II ® a^ is entire, bounded in vertical strips if F
is a number field, and satisfies
L(s, n ®x) = s(s, n ®5c)L(i-s, n ®x~ 1 ).
Thus our conclusion follows from [JL] (Th. 11.3).
Remark (2.3.4). - From the "strong multiplicity" one theorem for G^ it follows that 11
is indeed the base change lifting of n as defined in [La 4].
4° SERIE — TOME 11 — 1978 — N° 4

3. The notion of lifting
Henceforth, a will be used to denote a typical representation of GL (2), and K a repre-

sentation of GL (3).
In this Section we shall introduce the notion of lifting a representation a of GL (2) to
GL (3). We do this in terms of the L-functions
. L(s,(g®x)xa)
L2(s, a, 50) = —————-——•
L(s, x)
(3.1) Let F denote a local field and \|/F = \|/ a non-trivial additive character of F. For
each irreducible admissible representation a of G^ (F), set
(3.1.1) L^(s, a, x) = L(5, (a®x)x5)/L(s, x)
and
(3.1.2) 82 (s, a, x; ^) = s(s, (<y ®x)xS; Ws(s, X; \k).
We note that the factors on the right side have been defined for all a and all F [(1.4.4)
and (1.10)].
Definition 3.1.3. — Let n be an admissible irreducible representation of
G3(F)=GL(3,F)).
We shall say that n is a lift of a if the following conditions are satisfied:

(i) the central quasi-character of n is trivial;
(ii) n^n;
(m) for any quasi-character x of F^'
L(5,7i ® x) = L2 (s, CT, x), e(5,7i ® %; \[/) = £2 (5, a, 50; \10.
We note that if the last condition is satisfied for one choice of \|/, then by (i) it is satisfied
for all choices of \(/. Moreover if n is a lift of a, it is also a lift of CT ® % f^ any 7 and,
in particular, o f G ^ a ^ c o " 1 , ® denoting the central quasi-character of a.
PROPOSITION 3 . 2 . — Suppose F is archimedean. Then any CT admits a lift n, unique up
to equivalence.
Proof. — Let us denote by Tt-^7i(T) the "natural" bijection between the semi-simple
representations of degree n of the Weil-group Wp and the irreducible admissible repre-

sentations of Gn (F). Write accordingly
(3.2.1) (y=7r(T), deg(r)=2.
Then:
(3.2.2) T®?=X©I, where deg(X)=3.
Set
(3.2.3) IT=IT(X).
We contend that n is a lift of CT. Indeed
(3.2.4) O^^^^T^®?.
Moreover det (r ® ?) = 1. Therefore
(3.2.5) det(X)=l,
and—since the central quasi-character of 7i(X) is det (X)-we see that condition (i) is
satisfied. From (3.2.4) one concludes that
(3.2.6) r'=5i.
So since n (X") == n (X)", we also see that condition (ii) is satisfied. Now note
(3.2.6) L(s,(7C®5c)x7c)=L(s,(T®x)®?),
e(s, (7t ® /) x it; \[/) = e(5, (r ® /) ® ?; \|/).
But
(3.2.7) (T®x)®?=(T®?)(S)X=(^0X)ex.
Thus:
(3.2.8) L(5, (T ®x) ®^) = L(5, X ®/)L(s, x),

e(5, (T ®x) ®?; Y|O = e(s, X ® X ; vl0e(5, x; ^).
So comparing (3.2.6), (3.2.8), (3.1.1), and (3.1.2) we see that condition (iii) is also
satisfied.
The uniqueness follows from the following Lemma, whose tedious but straightforward
proof is left to the reader.
LEMMA (3.2.9), — Suppose that K and V are semi-simple representations of degree 3
o/Wp. Suppose moreover that det (K) = det (X'),
L(5, K ®5c) = L(s, K' ®x), L(s, i ®5c) = L(s, X' ®/),
4® SERIE — TOME 11 — 1978 — N° 4
and
e(s, ^®X;v|0=e(s, ?i'®5c;\|/) for all %.
Then ^ ^ ^
In the non-archimedean case we will eventually manage to prove a result analogous to
Proposition 3.2. For the time being, however, we content ourselves with the following
weaker result.
PROPOSITION 3.3. — Suppose F is non-archimedean.
(1) If a lift ofo exists, it is unique up to equivalence.
(2) If a is not an "extraordinary representation", then a admits a lift.
(3) If <J is extraordinary and admits a lift n, then n is cuspidal.
Proof. — The first assertion follows from Definition (3.1.3), and Lemma (7.5.3) of
[JPSS].
Recall that an extraordinary representation is a cuspidal representation which is not of
the form n (r) where T is a two-dimensional induced representation of Wp. If CT is not
extraordinary, then a = n (r) where T is a two-dimensional representation of the Weil-
Deligne group. Then (3.2.4) to (3.2.5), as well as (3.2.7) and (3.2.8) still hold. In
principle, the lift n of a is given by (3.2.3). But since TT (K) has not yet been defined, we
give now an explicit description of the lift of CT.
Suppose first that T is a representation of the Weil group. Suppose moreover that
(3.3.4) T = H i © U 2 , ^=Xi<A with ^<=R, ^ = 1.
In other words, suppose a = n (jii, ^2)* Then:

1 1
(3.3.5) Hi.^^/i^ ^ ^ ^r^^.xr1^1.
Let us form the induced representation
(3.3.6) ^ == Ind(Gs, B3; ^.^-1, 1, ^.Hi~1).
If ^ — ^ = ^ ^ ls a unitary irreducible (principal series) representation. We contend

then that TT = ^ is a lift of cr. Indeed the central quasi-character of i; is ^4. ^ ^. ^ ^2 2 = 1 •
Now
(3.3.7) | = Ind(G3, Ba; ^.^-\ 1, î.P2 1 ),

<w
^
and this representation has the same character as ^. Thus ^ = ^ or TT = TT. Finally we
have to check that, with the notations of (3.2.2),
(3.3.8) L(s, n ® %) = L(s, ^ ® x) ^d e(s, n ® x; v|^) = e(s, ^ ® ^ ^lO-
Since ^ = ^•PF 1 © z ® l^r^ 1 this ôws from [Go Ja] [Theorem (3.4)].
Now suppose t^—t^ ^ 0- We may assume ^ > ^ and then form the representation
(3.3.6). It may fail to be irreducible, but it always admits a maximal subrepresentation ^'.
ANNALES SCIENTIFIQUES DE L'^COLE NORMALE SUPERIEURE

The quotient representation n = ^ is then irreducible {see [Si] for instance) and
again we contend that n is a lift of CT. As before, (i) is satisfied. Now n can also be
obtained as the unique irreducible quotient of
(3.3.9) p=Ind(G3/B3;^.Hr^l,Hi.^
(Cy. [J 1] for instance). If
/O 0 1\
w== 0 - 1 0
\1 0 O/
then gv->wgw~1 takes ^B to B and p to i;. We conclude that n ^ n. Finally, since

(3.3.8) has also been established in [J 1] [Prop. (3.4)], (iii) is satisfied.
Now suppose that
(3.3.10) T=Ind(WF,WK,x),
where K is a separable quadratic extension of F and x is a quasi-character of K", i. e.,

a = K (o^). Let also x' be the quasi-character of % conjugate to x by the action of the
nontrivial element of the Galois group. Then:
T == Ind(WF, WK, X'), ^ = Ind(WF, WK, X- 1 ), and ? | WK = x- 1 © x ' - 1 .
Thus:
(3.3.11) T ® ^ I n d ( W F , W K , X ® ( ? | WK» = Ind(WF, W^, 1 ©X.X'- 1 )

= Ind(WF, WK, 1) ©Ind(WF, WK, X.x'- 1 )
=l©i1©Ind(WF,WK,X.X" 1 ),
where T| is the quadratic character attached to K, i. e.,
(3.3.12) T ® ^ ^ ® ! , where ^ = ^ © r | and H = Ind(Wp, WK, X.X'" 1 ).
But x.X'" 1 is a character. Thus \i is unitary. Moreover (x.x'"1)' = (X-X'" 1 )" 1 ? and
(3.3.13) il = Ind(WF, WK, (x.X""T1) = Ind(Wp, WK, (x.X'" 1 /) = H.
The representation n (n) is therefore unitary and isomorphic to its contragredient. Its
central character is T| and it is either in the principal series or cuspidal. It follows that
the induced representation
(3.3.14) 7r=Ind(G3,P;7c(H),ii),
where P is the parabolic subgroup of type (2, 1), is unitary irreducible. (C/. for instance
[JPSS], § 6.) We contend it is a lift of a. Indeed its central quasi-character is ri2 = 1,
4° SERIE — TOME 11 — 1978 — N° 4

and since ^ == r\, and n (u) = n (u), we get 71;=%. Finally (c/. [J 1]):
L(s, TC ®x) = L(s, 7i(u) ®x)L(s, r|5c) = L(s, ^ ®x)L(5, i^) = L(s, ?i ®%).
Since a similar relation is true for e, (3.3.8) [and hence condition (iii)] is satisfied.
Of course all our discussion so far applies to the archimedean case as well. Suppose
now T is a 2-dimensional representation of the Weil-Deligne group which is not a represen-
tation of the Weil group. Then a is special, that is of the form
c^(xa 1 / 2 , xoc- 172 ) = X ® cr(a1/2, a-172).
Let 7i be the Steinberg representation of €3 (F), that is, the square integrable component of
Ind(G3,B3;a,l,a- 1 ).
That n satisfies (i) and (ii) is well known. As for condition (iii), it follows at once from
Theorem 7.11 of [Go J] and (1.4.3) of this paper.
Next suppose a is an extraordinary representation. We want to show first that
L2(s,G,x)=l.
For this we appeal to the fact that a O ^ ^ o r i f ^ l . (<y [La 4], Lemma 5.16.) Thus
SQ is a pole of L (s, (a 00 x) x 5) it and only if a50 / = 1, that is, SQ is a pole of L (s, ^).
Our conclusion then follows. This being so, if n is a lift of <7, we get
L(s, 7i 00 x) = L(5, n ® x) = 1,
and it is easily checked that this can happen only if n is cuspidal. (Cf. [JPSS], § 7).
(3.4) Suppose a is unitary, but not one dimensional, the field F being archimedean or
not. For our purposes, it will be important to determine when n is unitary "generic";
this means (c/. [JPSS], § 6 and § 10) that the restriction of n to the subgroup
f / a b «\1
(3.4.1) P^ c d v }\
[\0 0 1 / J
is equivalent to the irreducible representation of P1 induced by the character 9 of N3

defined by
/l . A
(3.4.2) 9 0 1 y ]=^(x+y).
1
\° ° /
If F is non-archimedean, a is extraordinary, and a admits a lift TC, then n is cuspidal and

has a trivial central quasi-character; thus n is unitary generic (cf. [JPSS]). Similarly, if
F is non-archimedean and CT is special, its lift is special and again unitary generic (loc. cit.).

The remaining discussion will apply to the archimedean and non-archimedean case as
well. We write <J = n (r) where T is a two-dimensional representation of Wp. If T has
the form (3.3.10), then n is given by (3.3.14) which is irreducible generic (loc. cit.). If T
has the form (3.3.4) and t^ = t^ then n is given by (3.3.6) which is irreducible generic.
So it remains only to examine the case where TT is in the complementary series, that is T
has the form (3.3.4) where x is a character and ^ = - ^ = t, 0 < t < 1/2. Then (3.3.6)
takes the form
^hKKGa.Ba^l.or 2 ^
Now the maximal subrepresentation ^ / of i; is nothing but the kernel of the intertwining
operator
/^ \f(wng)dn, neNs.
It is then easy to determine ^' and TL One finds that i; is irreducible unless t = 1/4 in
which case
7T=Ind(G3,P;l2,l)
where 1^ is the trivial representation of G^ and 1 the trivial representation of F-^. The
representation n is then unitary but not generic.
If 0 < t < 1/4, then the representation n (a2^ a" 2 ^ is unitary in the complementary
series. Since ^ is irreducible,
n = ^ == Ind(G3, Bs; o^, a- 2 ^ 1) = Ind(G3, P, n(^\ oF^), 1)
and this shows that n unitary generic (Joe. cit.) If 1/4 < t < 1/2 then n = ^ is not unitary.
(3.5) Suppose F is non-archimedean and CT "quasi-unramified". This means that
CT = GO ® % where <7o is unramified, that is contains the trivial representation of
Ka = GL (2, Rp). Then <r = n (a5150, a52 /) = n (r), where T = a51 7 © a52 50. As we
have seen, a admits a lift n = TT (X) where X = a51""52 © 1 © a52"51. More precisely, n
is the only irreducible quotient of
^hKKGsâ^l.a 52 - 51 ),
and 7i contains the trivial representation of K3 = GL (3, Rp), that is, n is unramified.
Note that if % is also unramified then % = o^ and
(3.5.1) L2(5,CT,/)=L(5,7C®X)
=[(l-^-^+^-^(l_^-s-^(^^-s-r+^-^J-l
and if v|/ has exponent zero,
(3.5.2) 8(5, CT, X;^) = S(S, 71 ® X ; V[/) = 1.
4' S6R1B ~ TOME 11 - 1978 - N° 4

(3.6) Let now F be an A-field and a an admissible irreducible representation of G^ (A)

(or rather of its Hecke algebra). Then we say that an irreducible admissible representation
7i of €3 (A) is a lift of a if, for all places v, Uy is a lift of Oy. One of the purposes of this
paper is to show that if a is automorphic cuspidal then a admits a lift n and n is automorphic
(although perhaps not cuspidal).
(3.7) There is already one case we can dispose of immediately. Indeed suppose CT is
automorphic cuspidal and there is a character % of F^/F^ so that ^ 7^ 1 and or ® % 2^ a.
Then %2 = 1, and % determines a quadratic extension H of F. According to the main
result of [LL] there is a quasi-character Q of H^/H^ so that o- is the automorphic represen-
tation attached to Q. More precisely, if ^ ^ 1 then v does not split in H and <7y = n (T(,)
with Ty == Ind (Wp , WH , ^(p) and w the unique place of H above v. Then the lift Tiy
of <7y is given by
^=Ind(G3^,P^<,^),
where
<= ^(T.), T, = Ind(W^, WH,, Q.X-1),
and Qy, is the quasi-character conjugate to Qy,. On the other hand, if ^ = 1, then u
splits and, if w^ w^ are the two places of H above v, we have
a,, = K (^) with Ty = 0^ ® Qy,^
The lift Tiy of <jy is then given by
7^=Ind(G3^, ?„; <, Xu)»

where
< = 7l(T,), T, = Q^.0^1 ®Q,,.^1.
Thus we see that n' = ® n'y is the representation attached to the character Q.Q'"1 of
H^/IF, where Q' is the conjugate of D. If O.Q'"1 = T| o NK/F where TJ is some character
of F^/F-", then
< = ^ (^\v^v%v\
^ = Ind(G3^, B3^; ri^, ^x^ Kv\
and TT = ® Tt,, is automorphic by [La 5]. Otherwise n' = ® TI;^ is automorphic cuspidal
and again TC = ® Tiy is automorphic (loc. cit.).
(3.8) Suppose CT is arbitrary automorphic cuspidal. Set
(3.8.1) L^(5, a, x) = riL2(5, a,, x.)

i;
and
(3.8.2) £2(5,o,)=n£2(^,^,v|/,).
v
ANNALES SCIENTIFIQUES DE L'^COLE NORMALE SUP^RIEURE 64

It follows from (3.5.1) that (3.8.1) converges in some right half-plane. On the other
hand almost all factors in (3.8.2) are one, and their product is an exponential function
of s which does not depend on v|/. This being so, the analytic properties of L (s, (a ® %) x a)
and L (s, %) imply that L^ (s, a, /) continues to a meromorphic function of s and satisfies
the functional equation
^2 (S, 0, X) = S2(5, CT, X)L2(1-S, CT, X~ 1 ).
On the other hand, if n is a lift of a, then

L (.y, TT ® /) = L^ C?, a, /) and e (^ TT ® /) = e^ (^, CT, 7).
4. Complements for GL (3, F) (F archimedean)
This Section collects some technical results about Whittaker models for representations
of GL (3, F) when F is archimedean. Thus F = R or C in this Section.
(4.1) In [JPSS] there is attached to every unitary generic representation n of €3 (F)
a space i^ (n; vl/p). However, in lifting unitary representations from G^ (F) to €3 (F)
irreducible representations arise which are not unitary generic. As we have seen, these
are precisely the representations of the form
(4.1.1) n=n(^) where ^ ( ^ © l e o T ^ and - ^ t < 1.
More explicitly, such a representation is the unique irreducible quotient of the induced
representation
(4.1.2) ^ = I n d (03^3;^, 1,0^).
What we are going to do now is define a space i^ (n; vl/p) for representations of the form
(4.1.1).
Let i^ be the space of C°° functions/on G^ (F) such that
(4.1.3) / ( 0 a, y } g =\a,\yl\a,\-t-l f(g).

AO 0 a,/ _
For / in V set
(4.1.4) W^(g)=[ f(wng)Q(n)dn,

J N3 (F)
where 9 is defined by (3.4.2) and
/O 0 1\
w = 0 -1 0 ;
\1 0 O/
4° SERIE — TOME 11 — 1978 — N° 4
this integral converges absolutely and uniformly for g in a compact set. We let IT (n; \|/)
be the space spanned by the functions (4.1.4) for/e V.
LEMMA (4.1.5). — The map f\-> Wj is a bijection.
Proof, - Suppose W^ = 0. Then from (4.1.4) we get
Ih x 0\ 1
\|/(-ocx)Ac
fv|/(~ax)dxf ,/ w 0 1 0 \ug \Q(u)du==0
J J U (F)
lO 0 1
for a = 1. Changing g to diag (a, 1, 1) g we get the same identity for a ^ 0. By conti-
nuity it is therefore true for a = 0. Thus we get from the uniqueness of the Fourier
transform,
f[wug]Q(u)du =0.
»U(F)
Changing g into
-i
we find
JJ-/ H O 1 y gv|/(ax+P^)dz^=0
^0 0 1
for all a e F, P e F". As before, we find / = 0, which concludes the proof of (4.1.5).
Note that n ^ n. Moreover, as in [JPSS], if W is in ^ (n; \|/), then the function W
defined by
(4.1.6) W(g)=W[Wg- 1 ]
is still in "W (n; \|/) = i^ (n; \|/); this follows from thef act that the automorphism
g\->wtg~lw~l leaves invariant N3, 83, 6, and i;.
We also denote by iô (7T? ^ ^ space of K^-finite elements miT (n; \|/). It is clear
that this space is invariant under right convolution by the elements of the Hecke-algebra
and the representation of the Hecke algebra on iô (n^ ^) ls equivalent to i;; moreover,
since n is the unique irreducible quotient of ^, the representation ^ is semi-simple if and
only if it is irreducible. Then ^ = n and t -^ 1/2. If in addition i; is unitary generic
then 0 < t < (1/2). As in [JPSS] we set
(4.1.7) llgil^gl^ldetgl- 3 / 2 if F = R ,
=(Sg,,.^)(detgg)-3/2 if F = C .
LEMMA (4.1.8). — There is a number r > 0 such that any element W ofi^Q (n, \[/) is
dominated by g\-> \\ g\\1'. Moreover any element ofi^Q (n; \|/) is dominated by a gauge,

Proof, — The notion of gauge was introduced in [JPSS] (§ 8) and the second assertion
follows from the first (loc. cit.). To prove the first we observe that, given fei^ and a
compact set Q of €3 (F), there is a c > 0 such that
|/(g/i)|^c/o(g) for /ieQ;
here fo is the element of 'T such that fo (k) = 1 for k e K^. Note that fo > 0. Thus
|W^(g)| ^ /i(g), f,(g) = Sfo(wng)dn.

Now /i is given by
x z
/î \~
fi 0 a, y }k îr^l^r1/^).
AO 0 03/ _
80 |/i fe) | ^/i (^) || ^ H^ if '• is large enough, and our assertion follows.
We want also to replace n by a representation TC' = TC 00 ^ where / is a character. Then
7i' = TC ® 7~1. We define 'W (n'\ v|/) to be the space spanned by the functions
W ® x : g^W(g)x(detg),We-T(7r;vlO.
PROPOSITION (4.1.9). — Suppose n' is any representation of the form n ® % where n is

as in (4.1.1) and % is a character. Then the assertions of theorems (9.2) and (11.2) of
[JPSS] apply to the space iT^; v|/).
Proof. — Although ^ may fail to be irreducible, the assertions of Theorem (8.7) of
[Go Ja] apply to i; with
L(s, ^ ® Z) = L(s, n ® 7) and e(s, ^ 0 x; \|/) = s(5, TC ® 50; v|0.
We note that if W is in iô ^ ^) and T| is a Ka-finite function on K^ then the function (p
defined by
<P(g)=f il(fe)W(feg)dfe
JK3
is a (bi-K3-finite) matrix coefficient of ^. Indeed for / in Y^, set
^ W - f r\(k)dk\ f(wnk)Q(n)dn.
J K3 J N3
It will suffice to show that this is a continuous linear form on i^. [Note (p (^) = X (^ (^)/)
i f W = Wy.] Of course ||/|| = Sup^ |/| defines the topology of-T. So the function fo
being as in the proof of (4.1.8), we have
and
m^/oii/n
l^(/)|^||/|lJJlnW|/o(«'")rf»^=l|/|l||ill|i/i(e).
Thus X is continuous.
4* SERIE — TOMB 11 — 1978 — N° 4

Combining these remarks with lemmas (4.1.5) and (4.1.8) it becomes clear that the
proof of (9.2) and (11.2) in [JPSS] applies to the case at hand.
5. Product decomposition of the main integral
Our purpose in this section is to decompose into local factors an integral on the meta-
plectic group involving an Eisenstein series, a theta-function and a cusp form on G^.
Accordingly F will denote an A-field of characteristic not equal to 2, and G will denote
the group SL (2) regarded as an algebraic group over F.
(5.1) By Mp (A) we denote the (global) metaplectic group introduced in [We]. This
is a group of unitary operators in L2 (A) which fits into the sequence
(5.1.1) l-^T^Mp(A)^G(A)->l.
Here T = { z e C | z z = = l } i s regarded as the group of unitary scalar operators 01-> K 0

in L2 (A); it is central in Mp (A) and G (A) is (topologically) isomorphic to Mp (A)/T.
In other words, the sequence is exact.
If 0 is L2 (A) and g is in Mp (A) we write g . 0 for the image of 0 by the unitary operator
g. According to paragraph 41 of [We] the sequence (5.1.1) splits over the subgroup
G(F), the splitting homomorphism
(5.1.2) rp: G(F)->Mp(A)
being determined by the condition
(5.1.3) £OF(Y)^)= £^(0 for Oe^(A), yeG(F).

êF êF
The sequence also splits over ^ (A) or, what amounts to the same thing, over ^ (A)
and ^ (A). The corresponding splitting homomorphisms
(5.1.4) t : ^-(A)->Mp(A) and d : ^(A)->Mp(A)
are given by
(5.1.5) t(^ ^.OM^vl/QxÔ^),
and
dk ^^(y^W2^).
Note that t agrees with rp on N (F) and d agrees with rp on A (F). Finally, if
w /\
WQ.^(X) = 0(—^c) where 0 is the Fourier transform of $ with respect to \|/, then
pr (M;o) = w = (i o) and rF (w) = wot
ANNALES SCffiNTIFIQUES DE I/ECOLE NORMALE SUPERIEURE

As in [Ge] we say that a function (p on Mp (A) is genuine if (p (Kg) = ^(p (g) for all
g e Mp (A) and X e T. Thus, if (pi and (p^ are genuine functions on Mp (A), the product
(pi. (p2 is invariant under T, and there is a functions/on G (A) such that (pi. ^ (g) = /(pr (^)).
Similarly | (pi | may be regarded as a function on G (A).
To prove ultimately that cuspidal representations of GL (2) lift to GL (3) we first need
to prove the following:
THEOREM 8.1. — Suppose a is cuspidal representation of GL(2, A) and % is a "highly
ramified" character of F^F^ (see (5.3) below). Then the Euler product of degree 3
defined by
s
L(5,(G(g)x)xa)
^2 W X» ) — ——————————————
L(s, X)
is entire and bounded in vertical strips of finite width.
The proof of this Theorem will be lengthy. Thus it might help to explain it classical
significance as well as the ideas behind its proof. Suppose
/(z)= £a^ 2 -
w==l
is a holomorphic cusp form of weight k and character co. Suppose also that \|/ is a primi-
tive Dirichlet character of Z,
S^n-^nKl-a^-OO-P^- 5 )]- 1 ,
and \|/co (—1) = 1. Then in [Sh] Shimura proves that the Euler product
M^-.—r^r^r^)
x^[(l-*(p)«;p"•)(l-*(?)«.p,ll-•)(l-lKp)p;p-•)]-•
P
is holomorphic everywhere except possibly at s = k or k — 1.

Now suppose that F = Q, a is the cuspidal representation of GL (2, A) generated by/,
and / is the character co\|/. Then:
L,(^fc-l,/,^)=L,(.,a,x)= L(5?(a0x)xa) .
L(s, X)
But in Section 9 we characterize those a whose corresponding L-functions have no poles.

Thus our results refine as well as generalize Shimura's.
The idea behind the proof of Theorem 8.1 is to generalize the Rankin-Selberg method
used in [Sh]. More precisely, Shimura exploits the fact that L^ (s,f, v|/) has an integral
representation of the form
(*) L,(5, /, v|/) = ff /(z)9(z)E(z, 5)^.

JJro(N)\H }T
4' SERIE — TOME 11 — 1978 — N° 4

Here N is an integer depending on the level of/and \|/,
6(z)= £ vK^2"1"22,
n= — oo
and E (z, s) is a certain real analytic Eisenstein series of "half-integral weight". Since
both sides of (*) define holomorphic functions of s for Re (s) sufficiently large, the problem
of analytically continuing L^ (s, /, v|/) is therefore reduced to the problem of analytically
continuing E (z, s).
To generalize Shimura's method we introduce functions on the metaplectic group
Mp (A) which generalize the functions f(z), Q (z), and E (z, s) in (*). We denote these
functions by (p (g), 9y (g), and E (g, s) respectively. The functions (p, Qxp and E are
invariant by r^ (G (F)) and our generalization of the integral in (*) is
I (s, x, F, ^ (p) = f (p (g) 9^ (g) E (g, 5) dg.

Jrp(G(F))\Mp(A)
As already indicated, our purpose in this Section is to prove that this "main integral"
can be expressed as the product of certain local integrals ly (s, Fy, ^y, Wy).
(5.2) Let MK denote the subgroup of Mp (A) which projects to the standard maximal
compact subgroup K of G (A). Let % be a character of F^/F^ and F a continuous function
on C x Mp (A). We will often write Fg (g) in place of F (s, g). We assume that for each
g e Mp (A), s \-> Ps fe) is holomorphic and Fg is MK-finite on the right, uniformly with
respect to s; in other words, there is a MK-finite function ^ on MK such that
jF,(gfe)^(fe)dfe=F,(g),
for all s and all g. Finally we assume that
(5.2.1) F,[^^ ^t(^ ^gj^^^lal-^F^g),
for all geMp(A),?ieT.
The Eisenstein series corresponding to F is

(5.2.2) E(g,5)= E F,[rF(y)g].
B (F)\G (F)
As explained above, the function | F, | may be regarded as a function on G (A). As

such, it is majorized by a function h on G (A) which is K-invariant and such that
'[(^ ^'H"^2^-
Thus (5.2.2) is dominated by the ordinary Eisenstein series
(5.2.3) S /z(yg),
B(F)\G(F)

and hence (5.2.2) converges absolutely for Res > (3/2), uniformly for (Re(^),^) in a
compact set. Consequently this series defines a continuous function on the set of pairs
(s, g) with Re s > (3/2); the resulting function s i-> E (g, s) is holomorphic, and
E(>F(Y)^s)=E(g,5) for yeG(F).
As on p. 329 of [JL] we may introduce the notion of a slowly increasing (resp. rapidly
decreasing) function on G(F)\G(A). If/is a genuine function on Mp (A) which is
invariant under rp (G (F)) on the left, we will say that/is slowly increasing (resp. rapidly
decreasing) if [/I—regarded as a function on G(A)—is slowly increasing (resp. rapidly
decreasing). For Res > (3/2), it is well known that the series (5.2.3) defines a slowly
increasing function on G(F)\G(A). Thus (5.2.2) is, for Re s> (3/2), a slowly
increasing function of g on rp (G (F))\Mp (A). In fact this condition is satisfied uni-
formly for Re (s) in a compact set.
Now suppose x? belongs to the Schwartz-Bruhat space y (A) and set
(5.2.4) 9^(g)= Efe.^)^).
êF
Since (^g).O = ^fe.O) for ?ieT, this "theta-series" is a genuine function on Mp(A).
By the very definition of rp (5.1.3), it is also invariant on the left by rp (G (F)). Finally
it is slowly increasing (c/ [We]). Thus for Re s > (3/2),
g»->9<p(g)E(g,5)
may be regarded as a slowly increasing function on G (F)\G (A).
(5.3) Let CT be an automorphic cuspidal representation of G^ (A) whose central character
(o has module one. In particular, a is unitary. If (p is a (K^-finite) form belonging to
the corresponding space of cusp-forms, then q> is left G^ (F)-invariant, rapidly decreasing,
and
(5.3.1) (p^ ^g|=o)(z)(pfe) for zeF^, geGÂ).
The main integral we wish to consider is
(5.3.2) 1(5, x, F, y, (p) == | (p(A)9y(g)E(g, s)dg, pr(g) = ft.

JG(F)\G(A)
If Re s > (3/2), the series (5.2.2) converges absolutely, and the integral (5.3.2) converges
absolutely precisely because (p is rapidly decreasing and 9y E is slowly increasing. The
convergence is even uniform for Re (s) on a compact set; thus the integral is a holomorphic
function of s in the half-plane Re (s) > (3/2).
Now let S be the finite set of finite places v of F where Oy is not quasi-unramified in the
sense of paragraph (3.5). Let 30 be a character of F^/F" whose ramification at each place v
in S is so high that
(5.3.3) L(s, 5C,) = 1 and L(s, (a, ®x,)xo,) = L(s, (a, ® ^ ') x ^) = 1.
4® S^RIE — TOME 11 — 1978 — N° 4
Given CT and 7 satisfying the above conditions we shall show in paragraph 6 that
(5-3-4) I^F^q^Lôr,^
for F, x?, and (p appropriately chosen, and Re 0) sufficiently large. We shall also show
in paragraphs 7 and 8 that E (g, s) extends to an entire function of s if-in addition—
5C2 is ramified at a least one place. Thus it will easily follow that L^ (s, a, %) itself is
entire when % satisfies the above conditions.
In order to establish (5.3.4) we shall need to write the main integral as a product of
local ones. Although these integrals are taken over the local groups G,, = SL (2, F^),
we still need to introduce the local metaplectic groups before defining them.
(5.4) Let F now be a local field. The (local) metaplectic group Mp (F,,) is a group of
unitary operators on L2 (FJ which fits into the exact sequence
(5.4.1) l^T^Mp(F)- P ^G(F)^L
Here T is again the ordinary torus regarded as a group of operators in L2 (F) and the
sequence (5.4.1) again splits over B (F), N(F), and A(F). I f ^ . O again denotes the
image of ^ in L2 (F) under g in Mp (F), the splitting homomorphisms tp : N (F) —> Mp (F)
and dp : A (F) -^ Mp (F), noted also t and d, are defined by
C 5 - 4 ^) [^ ^].OM=vkF[|^2]o(3.)
and
(5A-^ d(a o
\<S>(y)=\a\l'2ây).
a
\ /
Also we define WQ in Mp (F) and w in G^ (F) by
wo^(x)=^(-x)=^(y)^<i-yx)dy, w=(^ ~~~l\
so that pr (wo) = w.
If F does not have residual characteristic 2, and if the conductor of vj/p is SRp, then the
sequence (5.4.1) splits over K = SL (2,7?p). The corresponding homomorphism
r^ : K —> Mp (F) is determined by the condition
r^(g).0(x)dx= O(JC)AC for Oe^(F);

jMp j9lp
(c/. [We], § 19). In fact if 0° denotes the characteristic function of SRp, then
^(g)(D°=0 0 for all geK,

and it is easy to check that
r^(w)=Wo.
It is also clear that r^ coincide with t and d on N (F) n K and A (F) n K respectively.

500 S. GELfiAkt AND H. JACQUET
(5.5) Now let F be an A-field. For each place v of F we write ^, d^ pr,, for tp , rfp »
prp^; similarly if ^ is finite, does not have characteristic 2, and \|/y has exponent zero, we
write Yy for r^. Then there is a natural injection
(5.5.1) i,: G.^G(A)
of Gy as a pseudo-factor. Similarly there is a homomorphism
(5.5.2) ^: Mp(F,)->Mp(A)
determined by the condition
(5.5.3) ^(g.).^(x) = g..0,(x,). n <l>.OcJ
w ^y
for each ^ <= Mp (F^) and 0 = f] °» ^ ^ (A)- The diagram

w
l^^-^GCA)
(5.5.4) ,T t
Pfv
>
is commutative. ^(^ G^
Let A = (A^) in G(A). For each v choose ^ in Mp (F^) so that pr^ (^) = A^. If h,
is in K.,, and ^ is defined, we may take gy = /•„ (Ay); let us agree to take this this choice for
almost all v. Then there is now exactly one element g of Mp (A) such that
(5.5.5) g.<i>(x) = n^.^(^), o = n<^.
t7 P
Moreover pr (g) = A. We write (^) for ^ [even though Mp (A) is not a restricted product
of the groups Mp (F(,)].
(5.6) We are now ready to return to the main integral (5.3.2). Let us take Re(s)
sufficiently large and compute formally. We can write the integral (5.3.2) as an integral
over T rp (G (F))\Mp (A). Then we can replace E by its expression (5.2.2) to get
1(5, x, F, y, (p) = | (p(pr(g))SF,(r(Y)g)9^(g)dg

jTrF(G(F))\Mp(A)
-L ;Trp(B(F))\Mp(A)
<P(prte))F,(g)9y(g)dg
-I./ B (F)\G (A)

<PWFsfe)Mg)^
where g is any element which projects to A. But by Iwasawa's decomposition,
(5.6.1) I(,,X,F,^)-J,[(; ^ ;-,).]
-•K; ^('0 «°.>

0
^K; 0<s « -)'"]^^^
x 9J t
4® SERBB — TOME 11 — 1978 — N° 4

where m e Mp (A) projects to k, k is integrated over K, x over F^ and a over F^. In view
of (5.2.1), this reads:
(5.6.2) [v^mââ^d-adk
W ^ W-W ^ W-
Now we recall that the function
^\[o i
has no constant Fourier coefficient, i. e.,
[(i M g
(5.6.3)
Moreover, if
UG 0
<P|(1 ' C )M^=0.
(5.6.4)
"J-Ki 04
W ( / 0 = U ( ^ ^)/»k(-x)dx,
then the §-th Fourier coefficient of this function is
<C ^l-J-K; i
w
^ ;)t]-J<(; O*]^-'^-
Taking (5.1.5) and (5.2.4) into account, we get that—for h = pr(g)-
(5.6.0 J,[(; ^].,[(; ^].«
-HO ^.^(-H^^
-E...-i-(oJ,[(; ^]*(|e«)*
1
^ 0
2
- SeF*
I; »-^©w I
0 L
-^
2
0
= E g.y^w
S6F« 0 ^- 1
ANNALES SCIENTIFIQUES DE L'^COLE NORMALE SUPERIEURE

Thus (5.6.2) gives
(5.6.7) Hs,^¥^^)=! ¥,(m)dkf %(a)\a\^312 dx

JK JF^F^
-â 0
S v" a /
x FX
§6F« ^
,(
d ;
d
o .i
e I w V U f l
0
_Jm.^)W
/
\ tT^t-Nxxr
w - ^ ^ W |l 2-
2
/
i ^
-^
v
y _ l _ l l
fe
0 i; a
==fF,(m)dfef x^M^ô Z mâiywn^ 0 0

JK JFS/F* SeF" I r. e-1-1
0 ^'a-
(5.6.8) F^^w^C^W ( 2 ° ()
)fe xWlal'" 1 ^^^
kxFl
0 a-1/ ]
Finally, let co be the central character of o so that W transforms according to co. Then:
x ^ 0
(5.6.9)
.6.9) I = f ]
¥,(m)m. P(a)W\[ 2 \k [^ââ^-1 cfadk, pr(m) = fe.
JKXFI
0 1,
Before justifying our steps, let us specify F,. For each place v we select a function
Pv,s(gv) on Mp(F^)xC such that
(5.6.10) ^{^((i ^-(^ î)^^^^)!^!^172^,^).
For each ^ in Mp (F^) the ratio

F^(g,)/L(25,x2)
will be an entire function of s, in fact a polynomial in ^s, ^^ if v is finite (with module ^)
and a sum of functions as1 e^ if v is infinite. Moreover, let K,, be the standard maximal
compact subgroup of Gy and MK(, its inverse image in Mp (Fy). Then for each s,
fe^^s(W(25,x2)
belongs to a fixed finite dimensional right-invariant subspace of continuous functions

on MK^.
Finally, for almost all v, we select
(5.6.11) F^(r,(fe,)) = L(25, x2) if êK,.

Then if § = (^p) as in (5.5.5), we set
(5-6-12) F,(g)=nFsfe.).
4° SERIE — TOME 11 — 1978 — N° 4

I RELATION BETWEEN AUTOMORPHIC REPRESENTATIONS OF GL (2) AND GL (3) 503
Since almost all these factors are equal to L(2s, ^), the product in (5.6.12) converges
absolutely for Res large enough. In fact it continues to an holomorphic function of s
since f / 2 (by assumption) is not principal (it is ramified at some place v). We observe
also that
F,(fe)/L(25,x2)
is for any k e MK a sum of functions of the form as1 e^ and for a fixed s belongs to a
finite dimensional space of right-invariant functions on MK, independent of s.
Let us recall also that W belongs to the spaced (cr; \[/) spanned by functions of the form
^nw,(g,)
where W^ is miT (^; \[/y) for all v, and for almost all v is the unique element of^ (<jy; \l/y)
invariant under (and equal to one on) K^, y = GL (2, ^).
This being so, for F^ , as above, ^ in y (?„), and W^, ini^(a^ \|/^), we introduce the
local integral
(5.6.13) I,==I(5,F,,^,W,)
|^| s ~ l X.^ l (^)Ft,.s(^)

JKyXF^
^ 2
x m,.^(a,)wJ 2^ 0 }k,\dk,dxa,
L\ 0 1>
pr^m^=k^
Then we have:
PROPOSITION (5.6.14). - Suppose ^ = TP¥^ W = n W,,, and F, = II F,,,, as above.
Then for Re (s) large enough, each one of the integrals (5.6.13) converges absolutely, their
infinite product converges absolutely, and
I(5,F,^,(p)=ni(s,F,,^,W,).
v
Proof. — We can choose the element m in Mp (A) which projects to k = (fey) as in

(5.5). Then m = (Wy) and pr,, (m^) = ky It is then formally clear that (5.6.9) is equal
to the product of the integrals (5.6.13). We need however to justify all our steps.
Fix s in a compact set with Re s large. For each place v there is a number c^ and a
function ^ > 0 in y (Fy) so that
|F,.,(m)(m.^ l )(^)|^cJL(25,7 2 .)|^ l (a).
There is also a function ^ > 0 and t > 0 independent of v such that
ÎYÂ o \ 1 . ]-^2._.
^v I o ^ f^v \ = i \ U v \ fv^v)9

(<y. [JPSS], § (2.3) and (8.3.3).) Furthermore, we may assume ^ = 1 s real, ^ == 1, and
î == x^ = characteristic function of SR,, for almost all v. Then the integral (5.6.13)
is dominated by
c,L(25,x2.) f \a,\s~l-t^<ia)^(la2}dxa^.
JK^XF? \2 /
This shows that when Re s is large enough, the integrals (5.6.13) converge absolutely.
Similarly the integral on the right hand side of (5.6.9) is dominated by
IIc^^f ^-^^(a^P-aâdk
JKXFI \2 /
with x?1 = IPF,1, ^¥2 = y^. Thus it converges absolutely and is equal to the product of
the local integrals for Re s large enough.
Now we have to prove identity (5.6.9), for Re s large enough. Certainly we may
assume
\m.xP\(x)^xyl(x)
for all weMK. Then:
0 2 8
•{•(^H'o. -)'"]^'" ^ -
Going back over our computations we see that all our steps will be justified once we show
that
f -
JA/FXF^XK
HLVo0 ^)(S
HITS J S ^^^ââr^dxd-adk
W^ a
^^^^âr^dxd-c
/ J
-k<=F
êF
is finite for s real and large enough. But there is a c > 0 such that
Syâi^clal- 1 .
êF
Thus it suffices to show that
f, ,r/l x\(
x a u0 \\,"],
\/ a » I i.-3/2 i, «:,.
ls-3/2
jH|lo 1 1AO 0 a - 1 ^ I0! 3

a-^Jl^l
3 //22
^ a <•
^^"<+oo
for s real and large. Since (p is bounded and for any r
<G ^ ^î"^
for | ^ | large, this is clear.
Q.E.D.
4° SERIE — TOME 11 — 1978 — N° 4

6. Local computations for the main integral: identification with L^ (s, a, %)
In this Section we prove that

W !(., x, F. ^ <p) = ^^^^^(^ L,(.. a, x))
L(s, X)
for Re s sufficiently large, % "highly ramified", and F,, ^F and q> appropriately chosen.
But both sides of (*) decompose as products over the places of F (the left side by Propo-
sitions (5.6.14), the right side by definition). Therefore (*) is actually a local assertion.
(6.1) Throughout this section, F will denote a local field of characteristic not two and
\|/F or \|/ will denote a fixed non-trivial character of F. If F = R we take \|/ (x) = exp (2 i n x)
and if F = C we take \|/ (z) = exp [2 f7c(z+z)]. By CT we denote a fixed irreducible
unitary generic representation of G^ (F), co its central character, and 7 a character of F^.
If F is non-archimedean and a is not quasi-unramified we assume that the ramification
of % ls so high that
(6.1.1) L(5, x) = L(5, x'1) = 1 and L^, a, x) = L,(s, a, x-1) = 1.
We will consider functions F (s, g) = Fs fe) of s eC and g e Mp (F) satisfying the follow^
ing conditions:
(6.1.2) F,[^ ^)i(; ^^(^r^cg]
foralloeF^ xeF, geMp(F);
(6.1.3) there is an entire function H (s) with values in a fixed, finite dimensional right
invariant space of continuous functions on MK such that
F, (fe) = L (2 5, x2) H (s) (fe) for all k e MK.
If F = R or C, H (s) is a polynomial in s times an exponential factor. If F is non-
archimedean H (s) is a polynomial in q'^ (f. In fact in all cases but the complex case F,
will have the form
(6.1.4) F,(g)==/,(pr(g))(g.0)(0),
where 0 is in y (F) (and MK finite) and/, is a function of s e C and g e G (F) such that
(6.1.5) fif^ ^gl^^xW/sfe)
and a condition analogous to (6.1.3) holds with MK replaced by K.

For F, as above, W in ^ (a; \|/), and ^F in ^ (F), we set
(6.1.6) I=I(5,F,^,W)=f F,(A)(^)(1)W ( 2 °}g \dg,

JN(F)\G(F) \
where pr (A) == g.
ANNALES SCDENTIFIQUES DE L'ECOLE NORMALE SUPERIEURE

When F has the form (6.1.4) we also write
IÎ^/.O.^W).
By Iwasawa's decomposition, we have then:
(6.1.7) i =JF-«XK
1= H1 - %(a- (a)f,(k)(m.^)(0)
[JF-«XK \a\s l l
xw.^fiOwK^ )fe dkd'a, pi(m)=k.

0
V -
Now we have already observed in (5.6) that such an integral converges for Re (s) sufficiently
large. Our purpose in this section is to prove:
(6.2) PROPOSITION. — For Re (s) sufficiently large, and appropriately chosen F, (orfs, 0
z / F ^ C), ^F, andW, the integral (6.1.6) equals L^ (s, a, %) times an exponential/actor.
The exact form of Fs (orfs and <D) will be needed for later reference. Moreover, a more
precise result will be needed in the "unramified situation" (see below). In the meantime
we make two useful remarks.
(6.2.1) Remark. — Replace a by a (x) T|. Then the L^-factor does not change. The
space IT (o-; \|/), however, does change in general; it is replaced by the space spanned by
the functions
W ® i i : g^W(g)ii(detg), Weâ;^).
However, the functions W enters the integral only through their restrictions to the set
'\ °)G(F,
\0 I/
It follows that in proving (6.2) we may, without loss of generality, replace CT by <r ® r\.
(6.2.2) Remark. — Replace \)/ by i)»' where v|/' (x) = \|/ (ax) and a -f- 0. Then the group
of unitary operators Mp (F) does not change. The projection pr, however, is replaced
by another map pr', and there is a commutative diagram
Mp(F)-^-»-G(F)
(6.2.3) J |
Mp(F)-p'-»G(F)
where the second vertical arrow is the restriction to G (P) of the inner automorphism
a
(f. ••> A\ ( °\ ^ a - l ^
(6 2 4)
•• ^(o iM o i)
4° SERIE — TOME 11 — 1978 — N° 4
of G^ (F). Similarly there is a commutative diagram
N(F)-^Mp(F)
(6.2.5) | L
N(F)-^Mp(F)
where t (resp. t ' ) is the splitting morphism defined by (5.4.3) and \|/p = ^ (resp. \|/F = ^/),
and the first vertical arrow is induced by (6.2.4). Note that (6.2.4) induces the identity
on A (F) and d does not depend on \|/. Moreover the inner automorphism (6.2.4) leaves
a invariant and takes the function W in^ Or; \|/) to the function
-<(s ;M°o-;)]
'a 0\
i^Y o lyj
( a - 1 OM
which is in IT (n; v|/'). Thus it follows that in proving (6.2) we may replace \|/ by \]/'.
Our proof will now proceed case by case.
(6.3) F is non-archimedean, a is quasi-unramified, % is unramified.
By remark (6.2.1) we may assume that <r is actually unramified. Suppose for the time
being the residual characteristic of F is not two. By remark (6.2.2) we may also assume
that the conductor of \|/ is 91^ ê situation at hand is then the "unramified one". Let
Wo be the element ofi^ (a, \[/) which is one on K^ and invariant under K^. Take 0 = ^
to be the characteristic function Oo of SRp. Finally define / by
(6.3.1) /J(^ ^ i ^ ^ l ^ X ^ ^ ^ x 2 )
LEMMA (6.3.2) ("unramified situation"). — Assume the Haar measures of F^ and K

are so chosen that Vol (W) = 1, Vol (K) = 1. Then fs being given by (6.3.1):
I(s,/,Oo,^Wo)=L,(5,a,x).
Proof. — The extension splits over K and Oo ls ^F (K)-invariant. On the other hand
fs and Wo are K-invariant. So from (6.1.7) we get
1
-<
I = L ( 2 s , x 2 ) /f l ^ r l X ( 0 - l W W o 2^
I
o
}dxa.
• \ 0f\ 1/
1.
Now
^ ?)-(;;) for ae9{\
So I is the product of L (2 s, j2) and the series
(6.3.3) S XVx^-^Wof 0 ; 2v" 1°\ X^^-5,

W>
w >--o
0o0 \
\ v 1
/
/
2
and what remains to be shown is that the sum of this series is L^ (s, a, %).L(2s, %2\-1
)

Since a is unramified, we have or = n (\t, v) for some choice of quasi-characters n, v of

F". Then, by Proposition (1.4),
L2(s,a,7)=(l-^lv- l x(c^)^~ a )- l (l-X(co)^" 5 )" l (l-^~ l vx(®)^-T l •

On the other hand ([JL]), Prop. (9.5)):
L^S ^l^r^^^SX-^W^J ^=L(.,a)=L(.,n)L(.,v).
while
L(25,X 2 )=(1+X(^)^~ S )~ 1 (1-X(^)^~T 1 .
Thus our assertion follows from the following Lemma, whose proof is left to the reader:
LEMMA (6.3.4). — Let X denote an indeterminate. If
E^xô-axrî-px)- 1
nÔ
then:
E C2„X"=(l+apX)(l~a 2 X^ l (l-P 2 X^ l .
nÔ
Suppose now F has residual characteristic 2. Let \|/o be a character whose conductor
is 91. We may assume \|/ (x) = vj/o (2 x) [Remark (6.2.2)]. Let/,, Oo. Wo be as before.
LEMMA (6.3.5). - There is c > 0 so that:
(m. Oo) (0) m. Oo W dk = c ^ (a), (pr(m)=k).

JK
Proof. — In the integral above let us change m into
<; ^)-
with ^ e SR and let us integrate over 91. We find that the integral gets multiplied by
f vl/o(xa2)^.
Jw
Thus the integral vanishes unless a is in 91. For a = 0, it has some positive value c. So
it will suffice to show that the integral does not depend on a, provided a is in 91.
So assume a is in 91. Then the integrand is invariant under t (N n K) on the left. On
the other hand, Oo ^ invariant under t (N n K) and rf(A n K), so the integrand too is
invariant on the right for these groups. Thus there are two positive constants c^ c^ so
that the integral is:
(6.3.6) Ci(wo.Oo)(0)tô(a)+cJ (^(^ ô1)-^)
x wot
[ [o ô1).^)^ where p r ( w o ) = ( ^ ~^.
4® SERIE — TOME 11 — 1978 — N° 4

We also have
Wo^(x)^<!>(-x).
Thus:
w^^(x)^<S>^lx)\l\1'2
and the first term in (6.3.6) is indeed independent of a. We also have
'1 z'
(t(1 ô l yôM=ô(2x)vô(x 2 ^)|2| l / 2 .
So the second term can be written as
c^H dz\ Oo(2î)ô0<^)î ^0(2^2) v|/o(^z-2 X2 a) dx^
In this integral x^ e 1/2 M. So 2 ^2 ^ ^ in 9i and the integral does not depend on a.

The Lemma being proved, set
r/2 o'
w(g)=Wo L jg
{G M
Then W is inÔT; v|/) and invariant under K. From (6.1.7) and (6.3.5) we seet hat
the integral l(s,f, <Do, Oo» ^0 is thus e(lual to (6.3.3) and our assertion is proved in the
case at hand just as in the previous case.
(6.4) F is nonarchimedean, a is not quasi-unramified, or % is ramified.
If CT is quasi-unramified and % ramified then
(6.4.1) L(5, (a ®x)xS) = î(s, ^ X) = L(5, x) = 1-

On the other hand recall that if or is not quasi-unramified then we assume that % is so
highly ramified that (6.4.1) holds. Thus Proposition (6.2) amounts to the assertion that
(6.4.2) lOÔ^W)^
for appropriately chosen/, 0, y, and W.
To prove this we introduce the function
(6.4.3) H(fe,s)=(m.0)(0)| \a\s•~l%^l(a)(m^f)(a)

JF<
xW ( 2 ^ |fe â where pr(m)=fe.

A 0 I/ ,
Then:
(6.4.4) HU8 ^ ^ . s ^x'^^H^s), for ee9^ and êW^
ANNALES SCIENTIFIQUES DE L'ECOLE NORMALE SUP^RIEURE

and
(6.4.5) I (5, /, 0, ^ W) = f H (fe, s) f, (k) dk.

jBnK\K
Choose n so large that % identically 1 on 1 + ^ and H (k, s) is right invariant by the

congruence subgroup
(6.4.6) ^"{l êK | a = r f = l , f c = c = O m o d < P " l .

Let also
K"=^ ^ e K | c = O m o d r } .
Then K" = (B n K).K'. Define
a b
AW
/ , ( f e ) = {x(a) if
" k= (ac bdd)eK"'
-~\c
0 otherwise.
For this choice of/,, it follows from (6.4.4) that
I(s, /, 0, y, W) = f H(fe, s)f,(k)dk = cH(^, s), c > 0.

J BnK\K"
Thus it remains to prove we can choose 0,x?, and W so that H (e, s) is a non-zero constant,
So choose y (a) equal to 70-1 (^) if a e ^ and zero otherwise. We assume 0 (0) ^ 0
and we choose W in such a way that
^1 .^
a^Wl^ °
\ 0 1
is the characteristic function of W in Fx (cf. [JL], Lemma (2.16.1)). Then:

/1
H(e, (D(0) ff
H(<?, s) = 0(0) lar-ô'^^y^wlz 0 2 0 ) ^ a = Vol (9T) $ (0),
JF-«
Jpc
\ 0 1
and we are done.
(6.5) Real case.
If F ^ C, Mp (F) reduces to a non-trivial extension of G (F) = SL (2, F) by { 1, -1 }.
We denote the corresponding two sheeted covering of G (F) by Sp^ (F). When F = R,
Sp^ (R) is therefore the unique two-sheeted covering of G (R) given by the theory of
Lie-groups. Its Lie-algebra is the same as the Lie-algebra of G(R): it consists of all
real two by two matrices of trace zero. Recall \|/ (x) = exp (2 i n x).
4" SERIE — TOME 11 — 1978 — N° 4

For our purposes it is convenient to single out the one parameter subgroups of Spa (R)
corresponding to the Lie algebra elements
/O 1\ . I T ( ^ 0\
(6.5.1) X , = ^ ;) and
and tl=
H=(; _^
'^[o o) [o -1}
These are the subgroups
(6.5.2) N , = { ^ ^) XCR,} A,=^ ^) a > o}.
(C/. (5,4.2) and (5.4.3).) Recall also that K = SO (2, R) [resp. MK n Spa (R)] is the
subgroup of G (R) [resp. Sp^ (R)] corresponding to the Lie algebra element
(6.5.3) =(V-i°^
^-ro
o/
As defined, the group Sp^ (R) is provided with a unitary representation on L2 (R); the
elements of y (R) are C°°-vectors and the corresponding action of the Lie-algebra on
them is given by
(6.5.4) X+.ÔO^TCÔOO,
(6.5.5) H.e)(^)=^0 / 00+ 1 00Q,
and
(6.5.6) U.O(^) = fTC/OOQ-f-^ -^000.
4n dy
Suppose now 0 is an eigenvector for U, that is,
(6.5.7) U.O=m.O
or, as we shall say, <I> is of weight n. Then 0 satisfies the second order differential equation
d2^.
(y)=(4n2y-4nn)^(y).
dy2
It follows that an orthogonal basis of eigenvectors for U is provided by the classical
Hermite functions
(6.5.8) 0,(x)=^-^H,(xy27i),
where H^ denotes the classical Hermite polynomial of degree m. The corresponding

weight is
(6.5.9) ?=m+l/2.

512 $. GELBART AND H. JACQUET
Thus if 0 = <]>„ we find that

exp(tU).0=exp( i-n0.
V 2 '
Similarly if (p is a C°°-function on G (R) or G^ (R) let us say it is of weight n if
(6.5.10) — (p [ g exp (t U)] = in (p (g).

A /=o
In proving (6.2) we may — without loss of generality — assume 50 = 1 or 50 = sgn. Indeed
% can be replaced by ^ by changing s io s—t. We may also replace a by CT ® p, and
assume (whenever convenient) that the central character G) of a is either 1 or sgn. Thus
the cases we have to examine are the following:
TABLE (6.5.11)
Case <T L2(s,a,%)
"(Hi. Ha)
1
a...
H2 = a". Hz = a 02 Ô^5^)^541'?)
"(Hi> ^2)
b. ''^ sgn rf^ 1 )^^ 1 ^-^)^ 54 - 1 ^-^)
Hi = a01, Hz = a
"(Hi.
4> Ha)
Hs)
a01 sgn
Hi =-a-sgn 1 r^W^^-^W^1^2-^)
2
a"2
H2 = a"
"(Hi, ^2)
-.4. t»2)
d. Hi = a°1
H,=a-sgn sgn r^)?^^-^)^^^-0^
H2 = a'2
II- — ?"2
CTQ
Z"
(i(2)= 1 rf^V^+m)
(zz)"-/2
0(1
sgn
"^'(zl^) ^l)^5^)
In this Table, the appropriate values of L (s, CT, ^) are given within an exponential
factor (c/. § 3 and [Ja], § 17). The function 0 (resp. ^¥) will be one of the functions (6.5.8)
4° SERDS — TOME 11 — 1978 — N° 4

and so will be determined by its weight. Similarly^ will be given by

6 5 12
' - - ) 4Co ^X-^^)]-^^"^^),
where P is a polynomial; so/, will be determined by its weight and the polynomial P.
Finally, since the restriction of cr to K = SO (2, R) decomposes with multiplicity one,
W will also—up to a scalar factor—be determined by its weight. Thus the appropriate
choice of/, 0, y, W, and P is given by the following Table:
TABLE (6.5.13)
Case wgt(/,) P wgt(O) wgr(T) wgt (W)
a........... 0 1 1/2 1/2 0

A........... -I s 5/2 3/2 0
c........... 0 1 1/2 3/2 1
d. . . . . . . . . . . -1 1 1/2 1/2 1
<?........... 0 1 1/2 w+3/2 m+1
/. . . . . . . . . . . 1 1 1/2 m + 1/2 m +1
In each case we have

wgt (/,) + wgt (0) - wgt W + wgt (W) == 0.
Thus formula (6.1.7) reduces to

(6.5.14) 1(5, /, 0, ^ W) = I = 0(0)L(2s, x2)?^)
1 2
r°° Sl l X
— / n\
x H ' X(o~ (a) F(a)W 2 K^-
Jo
\ 0 I/
Since 0 (0) 7^ O, up to an exponential factor this is
(6.5.15) L(25,z 2 )P(5)f oo |a|< l/2) < s - l) x(o- l (^)^(72^W^ ^Va.
We treat this integral case by case in (6.6) and (6.7).

(6.6) Cases a to d.
We need to compute the Mellin transform of the functions involved in (6.5.15). For
a h-^ ^ (^/2a) this is easy enough. Indeed, up to scalar factors,
(6.6.1) f(^)={ e^-2nx) cases a and d,

v
[ xexp(—2nx) cases b and c.
ANNALES SCIENTIFIQUES DE L'ECOLE NORMALE SUP&UEURE

Thus:
^+(1/2)
( 5^(1/2) \^s-(il2)\
\^^12)\ . _ ,
TI 1 ( ———— jl ( ———— ) cases a, a,
p+°°_ ,_ \ " ~\v 2 / / vV 2 / 7
(6.6.2)
.6.2) f ooxyG/^lar-
¥(^2a)\a\
w /1
s 172
~lf2dx^
a
Jo ' ^,^fs+(312)\^s+(112)\
1 | ———— |1 ( ———— | cases b and c.
\ 2 j \ 2 j
To compute the other Mellin transform, recall that for any infinite dimensional represen-
tation <j of the form n (î, (i^)? the space i^ (a; \|/) can be described in the following way:
for each 0 in <^ (F2) of the form exp (-71 (x2^-^2)) P (x, y)— where P is a polynomial—set
(6.6.3) /(g)=^(g)=Jo[(0,Og]îl^2 1 (0|^^^l(detg)|detg| l / 2 .
Then the elements ofi^ (a; \|/) are exactly the functions of the form
(6.6.4) W(g)=fjL^ ^)g1vK--^.

•) L \ / J
The integral may be divergent but may be given a meaning as in [JL] (§ 5). In particular
(6.6.5) W^ ^fo'C^x-^Hi.^'WHiO)!^ 1 7 2 ^^
where
(6.6.6) <S>\x,y)= \<S>(x,v)^[-yv]dv
(in the notation of [JL], W = W<^).

Thus:
(6.6.7) FW^ ^H5-172^
4[J::w(, ;)l,|-»...J:w(, ;)|.|-^.]
= l fp(x,^)^l(x)|^| s ^(^)|^| s s x xd^
+lffo'(^^)^tlW|x|sxx^,(^)|^|s-3-d^d^.
^J 1^1 M
To get the right weight we take
0 ^ y) ^ J exp ( - -JT (x2 + 3/2)) in cases a, b,

' 9 y } \ exp(-n(x2+y2))(x+iy) in cases c, d.
4° S^RIE — TOME 11 — 1978 — N° 4

DELATION BETWEEN AUTOMORPHIC REPRESENTATIONS OF GL (2) AND GL (3) 515
Then:
J expexp(-7T(x
^ v^' yv)) =\ (- 7r(xl+ 2y+^
2
^2))(x+^)in cases a9 b9
in cases c, d
and, up to scalar factors
FW^ î^| s - l / 2 â=7^-< s+ol)/2 ^(' s ±^^-<^^^^
in cases a, b
_^(s+ai+i)/2p/^+cyi+l\^/5+a2\
V 2 M 2 ;
^^^s+oî^r^+^^r^ 54 "' 724 " 1 ^

V 2 M ^ )
in cases c, d.
To continue we use a Bames Mellin Lemma as formulated on page 77 of [Ja]. The

result then follows immediately.
(6.7) Cases e and /.
This is a new situation, since CT belongs to the "discrete series". More precisely, it
is the representation in the discrete series with central character (sgn^4"1 and lowest
positive weight m+1. In particular ^(a; \[/) contains—up to a scalar factor—exactly
one element W of weight m+1. It is given by
^ °w
. a (A^fe- 2 '"'^'"^ 2
\o0 1V / 1I 0
ifa>0,
iffl<0.
Then:
J;"w(^)i,r"^.)-r(^)
(c/. Section 5 of [JL]).
Now consider case (e) where ^ = 1. Then co = (sgn)"14'1,
0 (x) = e-^2 and ^ (x) = ^""x2 H^+1 (x 727t).
Our main integral reduces to

1
P+oo/ \w+l I 2 r\\ __
a
I=L(2s,l) ( ) lal'^Wn e-^H^â^^^fl.
J-Alîy \o i/
But
H„+l(-a)=(-l)ffi+lH^l(a).
ANNALES SCIENTIFIQUES DE L'ECOLE NORMALE SUPEMEURE 67

516 S. CrELBARt AND H. JACQUET
Therefore:
p+oo /I 2 n\|
u
I=2L(25,1) ^-^2 x^Hâ^)^.
J0 \ 0^ ^l y/
So replacing W by its expression gives
I =L(2s, l)Jî(5),
with
^(s)= f00^-1^12^)^.
Jo
(Recall that we are ignoring exponential factors).

To continue, we need to compute ^ (.?). This is done by integrating by parts using
the Rodriguez formula
^(^2)=(-l)"^2H„(;c).
dx
The end result is that
^(s) = (s+fe-2)(5+fe-3) . . . srf't 1 \

\ 2 /
Therefore:
I = r(s)(s+m-l)(s+m-2) ... sF^54!1^
=r(5+m)r^^=L,(5,a,x).
Case (/) is entirely similar and we leave it to the reader.

(6.8) Complex case.
In this case Mp (C) is a trivial extension of SL (2, C) by T. In other words, there is a
splitting homomorphism r : G (C) —> Mp (C) which provides an (ordinary) representation
(also denoted r) of G (C) on L2 (C). In fact if we set
L^/eL^C) | /(x)=/(-x)},
L^/eL^C) | /(x)=-/(-x)},
then:
L^C^L+OL,
and the spaces L+, L« are invariant and irreducible under G (C). For our purposes it
is convenient to extend the resulting representations r+ and r_ to G^ by taking them to
4' SERIE — TOME 11 — 1978 — N° 4
be trivial on
t>0^
{(o°)
Then:
r^ = 7T(o^4, ac174), r_ = Ji(^ 1), where i;(z) = z(zz)- 172 ;
(c/. [Ge], p. 95).

Suppose 0 is in L+ (resp. L_) and K-finite. The^ 0 is actually in y (C). Set
w.te)-({(^ ;)4.)(,,
Then:
w[(^ ^]-*MW[g].
Moreover
•C ?)
W*^ 2 ° ) = 0 ( 1 ) for |z|-.+oo.
Finally, the space spanned by the W*7 s realizes the representation r+ (resp. r_) of the
Hecke algebra. It follows that W* is arbitrary in ^ (r+; \|/c), [resp. ^ (r_; vj/c)].
Since the extension splits, we may regard Fy as a function on G (C) such that
F,[(; ^^XWW1'2^).
Our local integral may be written then as an integral over N (C)\G (C):
I^s.F^W)
4
^1
8
J.
0\1
F,(g)r(g) .^(1)W 2 '*"»
J N (F)\G (F)
<0
1 0\1
JN(F)\G(F)
It remains to select F,. We take

F,(g)W*
[( 10
)
-\0 I/
g W
[(
.\0 J8-
Fs(g) = L(5, x) f 0[(0, O g j I f r ^ x O ) ^ ^

Jc^
where:
0(Zi, Z 2 ) = e X p ( - 2 7 I ; ( Z i Z i + Z 2 Z 2 ) ) P ( Z i , Zi, Z2, Z2),
and P is a polynomial.
Then, for k e K = SU (2),
F,(fe)=L(s,x)Lf5+,,xlQ(s)(fe)

518 S. GELBAJR.T AND H. JACQUET
where Q is a polynomial with values in the space of K-finite functions on K. Since
L(25,x 2 )=L(5,x)L^+j,^
up to exponential factors, F, has the required analytic properties (6.1.3).

Now we want to write I as an integral over N (C)\GL (2, C). To that end, recall
that without loss of generality we may replace a by <r ® [i. But this replaces the central
character co of a by op.2. Therefore we may assume that we are in one of the following
two cases:
Case (X®-1) Y is in
% =© u
<y —— ©£
$(z)=z(zi)- 1 / 2 L2
Then in case (g) [resp. (A)] W* is in 'W (r+; <)/) [resp. iy (/•_; i|/)] and transforms under
the central character (»+ = 1 of ?•+ (resp. (B_ = t, of r_). Since wo^ = / (resp. (O(B_ = y)
and Ga (C) = C" G (C) we can write
u
I=L(s,x) W* 2 dw 2 g Idetg^^ÔKO.Dgj^g.
'^•J,N(C)\GL(2.C)
<0 1, <0 1.
Finally W* has the form
W*(g) = W,[iig], 11 = diag(-l, 1),
with W^ in i^ (r+; v|/) [resp. i^ (r_; \|/)]. In other words
I = L ( 5 , x ) X^+l,w,w^).
Ff^+lWl,W2,^
\2 4
(Notations of§ 1.) By [Ja], Proposition 17.4 (see also § 1), for appropriate Wi, W^ and 0:
^(s, Wi, W^, 0) = L(s, CTXr+) [resp. L(s, crxr+)].
Taking into account the explicit expansion for all the L-factors concerned, and the dupli-
cation formula, we obtain (6.2). This completes the proof of Proposition (6.2) for = C
(and for that matter for arbitrary F).
4° SERIE — TOME 11 — 1978 — N° 4

7. Local computation of the constant term of the Eisenstein series
Having proved that (for appropriate s, F,, ^V and (p):
L2<s, <y, X) = (pfe)0yfe)Efe s)ds

JG(F)\G(A)
it remains to analytically continue E (g, s). This we do by first analytically continuing
the constant term of E (g, s) (when F, is appropriately chosen).
Conveniently, the "intertwining integral" describing the constant term of E(g, s) is a
product of local integrals. Since we are therefore dealing with a local problem, we can
(and do) keep the notation of Section 6.
(7.1) If F, is the function on Mp (F) defined in paragraph 6, we want to consider the
"intertwining integral"
x
(7.1.1) F;(g)=fFjwo/^ \~\dx,pT(w^=w.
Replacing F^ by the function | F, | [which is a function on G (F)] and applying standard

results for G (F) we find that this integral converges absolutely for Re s > 1/2. Also
a formal computation shows that
(7.1.2) F^(^ ^t(^ ^j^X-^lar-^Cg], XeT,
whenever F^ is defined.
What we want to prove is that F^ satisfies a condition analogous to (6.1.3).
(7.2) PROPOSITION. — Notations being as above, there is an entire function H*, with
values in a fixed finite dimensional right-invariant space of continuous functions on MK,
such that
F,*(fe) = L(2s-l, %W(s)(k), feeMK.
Moreover, ifF== R or C, then H* is a polynomial in s times an exponential factor; if¥ is

non-archimedean of module q, then H* is a polynomial in q'5, q8.
The proof will proceed case by case. In the unramified case, additional information
will be obtained. Recall also that except in the complex case, F^ is given by (6.1.4).
(7.2) F is non-archimedean, % is unramified.
By Remark (6.2.2) we may assume SRp ls ê conductor of \|/p. Assume also—for the
time being—that the residual characteristic of F is not 2. Then we are in the "unramified
situation".
LEMME (7.2.1) ("unramified situation9'). — With the notation as above, suppose the Haar
measure ofF is chosen so that vol (9Ip) = 1- Then:
F;(fe)=L(25-l,x 2 ) for fee^(K).

Since Eg is ^ (K)-invariant, F^* has the same property. Thus it suffices to establish
this formula for k = e. Recall that
(7.2.2) Fsfe)=/,(prg)(g.a>o)(0),
where ô is the characteristic function of SRp- As for the function fs, it will be convenient
to extend it to a function (still denoted by/J on G^ (F) defined by
(7.2.3) \(a A^k

Llo b) \- arî^.x2).
Thus:
F?(e)=J/,[w(; ^)](^(; ^)).0o(l)^
= /S 000
J K^ ^j^ ^!^] ^
since for 0 in y (F),
F).
WO.$(I)=I<I»M^.
Now
J^?)] dx <+oo
for Re 5- > 1, and since \|/ has module one, we see that for Re s > 1 (which is sufficient
for our purposes), F^* (e) is given by the absolutely convergent double integral
(7.2.4) F;00= f f /L( 1 ^loo^^r^^l^^.

J Fj F |__ \ /_J |_ Z, _J
By a simple change of variables this equals
\,\A^ ^l't)]•l'w*[^M-^
or, taking into account the relation
,-2
/I y -^\(y 0 \ /I xy^ 0 \
[o 1 ;-^0 ^- 1 ;^0 lAo y - 1 ) '
^Wl.lô^l.l-2^/,^; ^(^ ;.)]^|.]^.
Now set \|/' (^) = \|/ (-(1/2) x). Then the conductor of \|/ is still 9lp> and for each ^,
(7.2.5)
^Hy-K; ^Ki^
W,W=
4® S^RIE — TOME 11 — 1978 — N° 4

is in the space denoted^ (/a5 ~ 1 , 1; \|/') in paragraph 3 of [JL] (if x^"1 ^ a, this space
is also the space i^(nW1, 1); v|/)).
A description of that space (valid for all local fields) has been recalled in (6.6) (cf. [JL],
Prop. 3.2).
Let 0 be the characteristic function of $R^ in F2. Then let
(7.2.6) /o,,(g) = fo[(0, OglxoÔ^.XO^^detg)
and
(7.2.7) W?(g)=L,L^ ^g]vk'(-x)dx.
This is the unique element of IT (xa5"1? 1, ^') invariant under (and equal to one on) K^.
Note that f^^ ls proportional to /, and
(7.2.8) /^00=L(s,x);
it follows that
Thus:
H"(; ^],^].-^w,.,
^^L(2^r ^^ ÔoWx^l^r 2 ^
0 y
L(s,x) J^ V • /
or, since W^ transforms under the quasi-character co, = â5"1 of the center,
(7.2.9) F:(.) - ^ f W..u^; ?)1^.M,*-

L(5,x) JF L\ VJ l^l
.(^4)o-,-.))f w..(^0 1^,.
\ L(s,x) /JF» V /
Here rf*^ is so chosen that vol (W) == 1 or, what amounts to the same
^(l-,-1)^.
1^1
Also we have used the fact that

if
^Q) ?)=° H^-
But the convergent integral in (7.2.9) is the value at t == s of the integral
Ji^r^cor^^w^J ^W
Since the integrand here vanishes for \y \ > 1, this integral converges for Re(Q > s.

Moreover, for Re (^) sufficiently large, it is given by (6.3.3) with s replaced by r, co by co,,
and a by Tiâ5"1, 1), i. e., its value is
LOx'^-^La, x)LO, a l - s )L(2f, 5C2)-1.

Setting t = s we get
fw*^ 2 OV^.I^-l.^L^La,!)
J ' V O U ^—————L(2.7?)—————
l+X^g" 5
(l-X 2 ^)^ 2 ^ 1 )^--^ 1 )'
Therefore
(7.2.10) F^)=L(25~l,x 2 )
as required. This concludes the proof of (7.2.1).

Suppose now the residual characteristic of F is 2. Then we choose \|/ as in paragraph 6:
\|/ (x) = \|/o (2 x), and \|/' (^) = \[/o (—^), where \|/o has conductor SKp. Again F, is defined
by (7.2.2) and we extend /, to G^ (F) by (7.2.3). We check that for all k in MK:
(7.2.11) F;(fe)/L(25,x2)
is a polynomial in ^~5, ^s. As before
Fs*(fe)=(l+X(®)^~ s )(l-^~ l ).fw?f^ ^fe.0o(^)d^.
For fe = e, this is (7.2.10). Thus it will suffice to show that if
U,(fe)=f W;f^ °)fe.0o00^

Jjy|=l \0 I/
then:
U,(fe)=U,(^)^(fe),
where X is a character of MK.
Note that any element of MK projects to an element of K of the form
/ I Xi\ ( l xAfa 0 \ ., ^
(o i J ^ O l A o a-1)9 xie9t9 ae9t
or
1 x 0 x aeytx
^ ^y ^ ^ a -} ^ ^ -
In the first case we may assume
, / I xA (\ x A J a 0 \
^ ^ O l ^ o ^ O i j ^ O a-1}
4® S^RIE — TOME 11 — 1978 — N° 4
Then, for y e SR,

fe.OoM=|2| l / 2 vl/o[^^]Oo(2^)=|2| l / 2 .
In the second case we may assume

, (\ xA _ i / l X2\,fa 0 \
f e = w o ^ ô1^ Jô a-^}
Then for y e 91,
fe.OoM=|2| l / 2 [oo(2^)vko[^^]+o[-2z^]dz.
However ô (- 2 zy) = 1 if Oo (2 ^) ^ 0. Thus fe.Oo 00 doe3 not depend in y e 91,

and our conclusion follows.
(7.3) F is non archimedean, % is ramified. In this case recall that Fg has the form
(7.3.1) F,(g)=/,(prg)(g.<D)(0),
where fs satisfies (6.1.5) and fs \ K is actually independent of s. Thus it will suffice to

prove the following Lemma:
LEMMA (7.3.2). - Suppose F, is a function of the form (7.3.1) where 0 is in y (F) andfs
is a function on G^ (F) satisfying
fa
^J-H172^!^-!;
4(°o
"L\o ^]°
^ f'xoM-'.w,
w/^re /? ^ a fixed finite function on GL (2, 91). TTz^z (/'F^ is as in (7.1. l),/^ ^.y g
F,*(g)/L(2s-l,x2)
f^ a polynomial in q~3,
Proof. — We may assume pr (g) is in K. Replacing Fs by a translate we may even
assume g = e. As in (7.2) we find that for Re .y large enough,
2
-Jw.[(^
F?00= Wj ?)].(^,
where
^-{/•Ki i'»[»-
In particular (cf. Prop. 3.2 of [JL])
W^ °^-L,|1/2
^ ^-W'W ^W-
ANNALES SCIENTIFIQUES DE L'^COLE NORMALE SUPERIEURE 68
524 s. GELBART AND H. JACQUET
Since fs is K-finite, we also see that

x = SK l(x)fs l l 1
4^ .
4"G l)] ^~ ~ {- x- ^M-^- ^)/^)
for | x | large enough. Thus
N
fJ^(1 ^Vj = I
E^.w^-^+eo^, s),
(=1
where the 6; are in y (F) and 9o (x, s) is defined by
00 for
eoOc^^N'T
[ 0 for !^;
|jc < 1.
But the Fourier transform of 60 (x, s) for Re (s) large enough is given by
eo(x,s)vl/^ l ^^^= f H~ S X~ l (x)v|/^2l ^1rfx,

J L2 J JMÎ L J
and since % is ramified it is easily found that this integral has the form ^"ws (p (y) % (y) \ y î
with (p in y (F). Thus:
w^ ?)==Ee,(3.)|^|l/2^ls+^-m^w|^|-^$(^
and
Fs*^)=E^-lsJe,M^+^wsf$^)x2M|y||2s-i dy
The sum on i is a polynomial in q~s, q5; the second term is the product of a polynomial
in q~\ q8 by L (2 ^-1, ^c2) c/. ([Ta], [Go Ja]). This concludes the proof of Lemma (7.3.2).
(7.4) Real case. In this case:
(7-4-1) F,(g)=/,(prg)(g.0)(0),
where 0 is an element of y (F) of a certain weight and/, is a function on G^ (F) such that
/'? A ^\ ^ ( a -^/ cos9sin9\~| , . i 0 i, i a 1/2
,o
ein9 <,
^•'•^ ^ô fcA-êcosejJ'^^l ! -b ^^2^
The integer ^ is the weight of/, and P is a polynomial. The possibilities for %, P,/,, and 0
are as follows (cf. Tables (6.5.11), (6.5.13) and the remarks made in establishing those
tables):
TABLE (7.4.3)
Case % wgt(/,) wgt(0) wgt (F.) P (s)
(0....... 1 0 1/2 1/2 1

(ii).. . . . . . sgn -1 1/2 - 1/2 1
(iii). . . . . . sgn -1 5/2 3/2 s
4° S^RIE — TOME 11 — 1978 — N° 4

Since P, transforms according to a character of MK, to prove Proposition (7.2) for F = R,

it will suffices to show there is a polynomial P* such that
(7.4.4) ^(e)=L(2s-l,'x2)P*(s),
up to an exponential factor.
Now, as in (7.2),
(7.4.5) TO= W,
\y1 o Wd'y,
where
(7.4.6) W^^J^/Ju^ ^g]^(-x)dx.
Again, for each s (with Re s > 1) W, is in the space denoted iy (xo^"1, 1; \h?) in [JL].
Therefore we proceed as in (7.2). We let $1 be an element of y (F2) such that the function
(7.4.5) /o,(g)=f ^[(O.OgjxOM^^detg^detgl^ 2

JR-
has the same weight as fs. Then the function W^ defined by
(7.4.6) W.*(g)=f/,Jw^ ^'k(-x)^
is va."y (xo*~1, 1; vj/p) and proportional to W,.

More precisely, we take
(7.4.7) d» 1(r
v
^ = J ^(-i^^+y2)) case(i),
' -" [ exp ( - it (x2 + y2)) (x - iy) cases (i) and (h).
Then O'i [cf. (6.6.6)] is given by
(7.4.8) d>'1 (Y v) = I exp( ~ "(x22+y2)) case (i)

'
'• ' y ' I exp (- it (x + y2)) (x + y) cases (i) and (ii).
In all cases
/o,(e)=L(s,x)
so that
L(2s,x 2 ) P(s)
(•+0
-jj,2 o
(7.4.9) F?(e)= w.^ OOOd^.
L(s, x) J-a
Since (in all cases) $ (—j0 = ^ (y), up to exponential factors we also have
(7.4.10) F;(e) = L(2S> ^(s) f ^ W ^ f f .u^ °)1$(^)d^.

L(s, x) Jo L\ ^/J
\ /»
(7.4.11) (
W,* ^ ^)=\^(yt,t-l)^t)\t\s-ldxn(y)\y\ll-i'2.
/ v
This time
(7.4.12) J^Y*[(7 i0)]!^^17^^
2 w>
'-TA^ ?)]H-" -lJ: • [(s ;)]M-"^
=jffol(x,^)x(x)|x| s + ^ l |^|^ x xd^
-^ffî^^xM^.IÎ^" 1 ^-!^!^^^^
2JJ H M
Thus this integral can be easily computed.

Similarly it is easy to compute
fô)^)!^--172^
JO
since:
^ (^\ = J exp (- TT x2) cases (i) and (ii),
w
[ exp ( -- n x2) (8 n x2 - 2) case (ii).
Using a Baines-Mellin lemma as before we can (for Ret large enough) compute the integral
/•+00 / Q\ ,___
(7.4.13) j^ W^ ^ Ô^jOH"-5^.
[From (7.4.11) we see this integral converges for Re (t—s)^0.^ Thus we may evaluate
(7.4.10) by setting s = t in the expression for (7.4.13). This will give the required
result [with P* ^ 1 only in case (iii)].
(7.5) Complex case. - If F = C recall that
F,(g) = L(s, x) fo[(0, Qg] | ^r^xO)^,
where 0 is a suitably chosen function of the form
0(x, y) = exp[-2n(xx+yy)]P(x, x, y, y).

Now:
Fs*te)=jF,[u^ ^g]^=L(5,x)Jf^[atx)g]|f| s + ^
=L(s,^SS^x)g]\t\s-l/2^t)dytdx.
4® S^RIE — TOME 11 — 1978 — N° 4

As in Section 6, ifgis in SU (2),
F,*(g)=L(5,x)L^-J,x)Q(s)(g)
where Q is a polynomial. So since
2
•(.-|,x)-
L(5,x)Lf^J,^=L(2s--l,x )
up to exponentional factors, F^ has the required analytic properties, and the proof of (7.2)
is complete.
8. Analytic continuation of the main integral
In this Section F is a number field. As before we let cr = ®v °v be an automorphic

cuspidal representation of G^ (A) and S the (finite) set of finite places v where CT(, is not
quasi-unramified. We let / be a character of Fj[/F^ and we assume /„ is highly ramified
for v e S \cf. (5.3.3)]. We assume in addition that there is at least one finite place w such
that %^ is ramified. Since this is so if ^ is of high enough ramification, we may take w
to be in S if S ^ 0.
Our purpose in this section is to prove the following:
THEOREM (8.1). — Suppose a and % satisfy the above conditions. Then L^ (s, CT, 30) is
entire and bounded in vertical strips of finite width.
To prove this theorem we need only consider the "main integral" introduced in Section 5.
Indeed for each v let F^ 5 be the function introduced in paragraph 6. (Recall that the
choice of F^ s depends on Oy and %„). Define F, (g) by
w=nF^) v
ifs-ts^ [if- (5.5.5)]. Then;
"•K'o ^)<; i'>]-xMi"r"Ffe).

If we set
Efe^)= S F,(y-F(Y)g)
yeB(F)\G(F)
then this converges for Re s large enough. Moreover, there is a ^¥ in y (A) and a cusp
form (p on G^ (A) belonging to q such that
L2(5, o, x) = f E(g, s)Q^W^(g)dg, (pr(/i) = g)

J G (F)\G (A)

528 $. GEL&ART AND H. JACQUET
for Re s large enough [c/. Prop. (6.2) and (5.6.14)]. Thus Theorem 8.1 reduces to :
PROPOSITION 8 . 2 . — Suppose F, and % are as above, and (p is a rapidly decreasing genuine
function on Mp (F)\Mp (A). Then the function
51-^ Ê(h,s)n>(h)dh,
foerF(G(F))T\Mp(A),
initially defined for Re (s) sufficiently large, extends to an entire function ofs which is bounded
in vertical strips of finite width.
This Proposition is tantamount to the analytic continuation of the Eisenstein series
E (g, s). Thus we need to recall a crucial fact from the theory of Eisenstein series.
LEMMA (8.2.1). - Let F, be a function on Mp (A) such that
^(S a 01 )^ ^=KH(k)\ar1'2^
where H is a fixed finite function on MK. Then the function F^, defined for Res large
enough by
F*(g)=fF,L(M;)^ ^g"L,
continues to a me romorphic function ofs without poles on the line Re (s) = 1/2. Moreover,
if Re s = 1/2, then:
f F,(fe)F,(fe)dfe=[ ¥^W(k)dk.
J MK J MK
We take this Lemma for granted; it follows from the more general results of [La 3].
(Recall F is a number field!)
Now our function F, is clearly of the form
(8.2.2) F,(g) = L(25, ZÊ^^F^g),
<w
where, for each i, F^ satisfies the conditions of the Lemma and A, (s) is an elementary function
of s, i. e., a sum of terms like asê^. If we set
(8.2.3) F,*(g)=fF,L(w)^ ^g^x,

then we have also
(8.2.4) F^g) = L(25, XÊM^F^g).

Thus it follows from Lemma (8.2.2) that
(8.2.5) ]F,(fe)F,(fe)dfe= S¥!(k)¥~!(k)dk if R e s = 1 .
4° SERIE — TOME 11 — 1978 — N° 4

RELATION BETWEEN AUTOMORPHIC REPRESENTATIONS OF GL (2) AND GL (3) 52$
Actually we have more information about the analytic behavior of F^. Namely (by
Proposition 7.2):
(8.2.6) F;(fe)==L(25-l, ÊITO^)
where Hy is some finite function on MK and h^ is elementary.

To continue, we need to use a familiar "truncation process", (see [DL] or [Ge Jl] for
more details.) For c > 0 we let ^ be the characteristic function of the "Siegel set":
{^^(S a01)^ ^M^^MK}.
We set
(8.2.7) F;(g)=E°(g,s)x,(g),
where E° is the constant term of E (g, s):
(8.2.8) E°(g,,)=JÊ[^ ^g.^.
Then by the Bruhat decomposition,
(8.2.9) E°fe,s)=F,(g)+F,*(g).
Thus E° has a known analytic behavior. But from reduction theory we hnow that if c
is large enough then the series
(8.2.10) E'(g,5)= E F;(rF(y)g)

B(F)\G(F)
has only finitely non-zero many terms when g is in a siegel set. Let us study then the
scalar product
(8.2.11) E'(fc,s)9(/Od/L
To evaluate this integral consider what happens when h is in a Siegel set S. There is a
compact subset 0 of S such that for h e S—Q,
E^sÊ0^);
c/. [Ge Jl]. On the other hand, for h in Q, the series defining E' has only finitely many
non zero-terms, so since (p is rapidly decreasing at infinity in 6, it is clear that (8.2.11)
converges for all s and defines an entire function of s bounded at infinity in vertical strips.
Now set
(8.2.12) E" (A, s) == E (A, s) - E' (A, s).

530 S. GELBART AND H. JACQUBT
Let also L2 denote the space of genuine functions on Mp (A) which are rp (G (F)) invariant
and such that
f(h)f(h)dh < +00, AerF(G(F))T\Mp(A).
Clearly (p is in L2. Then Proposition (8.2) will follow from:

PROPOSITION 8.3. — jy*Re s is large enough, EJ is in L2, and the map s ^-> EJ continues to
an entire function ofs with values in L2. Moreover, the map s i-> 11 E^' | j is bounded in vertical
strips of finite width.
Proof. — Again, from the known theory of Eisenstein series, one knows that EJ is
square integrable and s \-> E'y holomorphic, for Re s large enough. Moreover for
Re s^ > Re s^ > 0,
(8.3.1) (EÊ^)=f H^-^ff ¥^k)¥^k)dk)

JM<c \JMK /
-f X^H52"81^ F^k)Pf,(k)dk
JM>C J MK
-f \a\l-sl-sldxaS F^(k)^(k)dk
JM>c J MK
+f X'OOH51-52^ f F,*(fe)F^(fe)dfe.
J|o|<c J MK
In each integral above the integration over F^ is taken modulo F". But %2 is not a principal
character of F^F^. Therefore the second and fourth terms drop out, leaving
(8.3.2) (E;,, E:,)=A(^ s,)(s,+sîr\
with
(8.3.3) A(5l,5,)=c s l + s 2 - l f F^WF^Wdfe-c 1 - 51 - 5 ^ F,*(fe)F^(fe)^.

J MK J MK
Now A Cs-i, s^) is holomorphic in C2. Moreover, if s^ = (1/2) +iy and ^ = 0/2)-?>
with y e R, we JBind that A vanishes [c/. (8.2.5)]. Thus A (s^ s^) = 0 if ^ +^ -1 = 0,
and it follows that
(8-3.4) (EÊ^)
extends to a holomorphic function of (s^, s^) in C2. A standard argument (cf. [La 3] or
[Ge J 2]) then shows that s h-» E^ continues to an entire function (with values in L2), and
by analytic continuation, formula (8.3.2) is still valid when s^+s^—l ^ 0.
To complete the proof of Proposition 8.3 (and hence Theorem 8.1) we have to show
that (E^, E^) is bounded in vertical strips. For this we write *s- = (l/2)+x+iy with
4® SERIE — TOME 11 — 1978 — N° 4

-a ^ x ^ a and y e R. Then:
(8.3.5) (E:,E:)=AQ+^+^j+x-^(2x)- l =B(x,^)(x)- l ,
where
B(x, y) == ^ A / " +x+^, 1 + x - f A
Now B(0,j0 = 0. Thus:

r 1 /®
B(x,30=x —(^,^)A
Jo ox
and
(E;,E / ;)=f laB (^3.)A.

J o 5x
But
^B(.^)=l^fl4-x+^14-x-^l^
ax 2 5si\2 2 ) 2 &,\2 2 /
So since the partial derivatives of A can be expressed in terms of elementary functions

and L-functions and their derivatives, it easily follows that 8B/9x is bounded for
— a ^ x ^ a , and the same is true of (E^', E^').
Q.E.D.
9. The Main Theorem

Again F is a number field.
(9.1) We first review and complete the results of [JPSS]. We let T be a finite set of
places (possibly empty). For each v ^ T we let n^ be an admissible irreducible represen-
tation of Ga^ (or its Hecke algebra). In addition we make the following assumptions
(and fix notation as follows):
(9.1.1) For each v, the representation TCy has trivial central quasi-character;
(9.1.2) Suppose v is finite; if?^ is generic, i. e. admits a "Whittaker model" in the sense
of [JPSS] (§ 2), then we set ^ = n^; if Ky is not generic, then it is a quotient of a certain
induced representation ^ which is said to be attached to Ky (he. cit.) for instance, we may
have
(9.1.2.1) ^=Ind(G3^P,;l^l),
where P^ is the parabolic subgroup of type (2.1), 1^ == n (a^2, a^172) the trivial represen-
tation of G^, „, and
(9.1.2.2) ^=Ind(G3.,, B3.,; a,172, 1, a;172).
ANNALES SCIENTinQUES DE L'^COLE NORMALE SUP&UEURE 69

(9.1.3) Suppose v is archimedean; then either n^ is unitary generic, in which case we set
^v = ^y? or ^v ls one of the representations considered in paragraph 4, namely:
(9.1.3.1) ^^(Gs^Ba.,;^!,^,1^^,
in which case ^ = ^ or ^ is (9.1.2.1), in which case ^ is (9.1.2.2).

In all cases, Tiy is a quotient of^y, and the representation ^ is semi-simple only if i;y = Tiy.
(9.1.4) For almost all finite v, n^ is unramified, thus of the form
^=^1,^ H 2 , ^ ^ 3 , ^
tilv
where ^ y(x) = | x \ : this means that Tiy is the only unramified component of the
induced representation
Ind(G3,,, Bs^; H,^, ^,1,, Hs,*;);
we also assume there is a t > 0 (independent of v) such that
-^ ti,v ^ t (i = 1, 2, 3, almost all v).
Given the notation and conditions above, we can now form the representations
7^= ® ^ , ^= ® ^
t^T v^T
of the Hecke-algebra ^T of the restricted product group
G^rio. (^T)
and form for each character % of F^/F^ the infinite products
L(s,nT,t)c)= ft L(5,^,^),
u^T
and
L(5,y,x)= riL(5,t,7; 1 ).
^T
These products converge absolutely for Re (s) sufficiently large.

(9.2) Suppose that the above functions are entire, bounded in vertical strips, and satisfy
(9.2.1) L(5, T^, x) = Ft s(5, ^ ®X., ^) Ft e(5, X., ^LO-S, S^ x~ 1 )

v^T ueT
provided the ramification of % is sufficiently high at each place v in T. Then we conclude

as in [JPSS] (§ 13) that there is a space -T of C°° functions on G^ (F) Z^ (A)\Ga (A) which
is invariant on the right under convolution by Jf1' and realizes the representation ^T.
Moreover, the elements of ^ are "slowly increasing". The various possibilities for T
and f are now discussed case by case.
4° SERIE — TOME 11 — 1978 — N° 4

(i) Suppose first that T = 0. Then we set

71=^=®^, ^=^=®^,
so that
L(s, 7^, x) = L(s, 7i ® x)» L(s, y, /) = L(s, n ® x)»
and the functional equation reads
(9.2.2) L(s,7i®x)=6(s,7C®x)L(l-s,7t®5C~ 1 ).
In this case the elements of ^ are cuspidal, thus cusp-forms.

Thus each ^ is semi-simple and unitary generic; in particular 7iy = ^ for all u, Tiy is
unitary generic, and n^ cannot be a representation of the form (9.1.3.1) or (9.1.2.1).
(ii) Now suppose T + 0 but the elements of ^ are cuspidal. Then again they are cusp
forms. For v ^ T, we again find n^ = ^ is unitary generic. On the other hand, for v e T,
there is a unitary generic representation n^ with trivial central character such that n = ® Tty
(all v) is automorphic cuspidal. Moreover (9.2.2) is satisfied.
(iii) Next suppose the elements of ^ have zero constant terms along
r/i * *M
N3= 0 1 *
\0 0 I /
but not along
/I 0 .
U=^ 0 1 *
\0 0 1
(c/. [JPSS], § 13).
Then (c/. the proof of [13.8] in [JPSS]), there is a quasi-character [i of F^/F" and an
automorphic cuspidal representation T ofG^ (A) with central quasi-character n"1 satisfying
the following conditions: for any v set
^=Ind(G3.t,,P^;^,^);
then for all v^T the representations ^ and r^ have a common irreducible component;
moreover, for almost all finite v both ^y and r|y contain a (unique) component which is
unramified and that component is the same for both representations (and equal to n^).
Note that \i need not be a character and therefore T need not be unitary. (The precise
definition of the induced representation r|y-for v infinite—is irrelevant since allt hat
matters to us is the infinitesimal class of r|y and its components.) If v is finite then Ty is
generic and r|y has exactly one generic component. Therefore, for v e T, let n^ be that
component and form the representation
^=(00®^.
»eT

We will write also TC' = 00 n^ so that n^ ^ n^ for v ^ T. Then (again see the proof of
(13.8) in [JPSS]) L(s, n' ® 50) is meromorphic and satisfies (9.2.2) for all /.
(iv) Suppose now that the elements of i^ have zero constant terms along N but not along
r/i * *M
V= 0 1 0 .
l\° 0 1/J
Then a similar conclusion can be reached.

(v) Finally suppose that the elements of ^ have non zero-constant term along N. Then
there are 3 quasi-characters v; of F^/F" such that v^ v^ V3 = 1 and, if we set,
TL = Ind(G3.,, 83.,; v^, v^. Vs.),
then for all v ^ T the representations ^ and r|y have a common component. For almost
all v they contain also the same unramified component which is then n^ For v e T let
n^ be the unique generic component of r\y and set
7C' = ® TCy ® 7^.

veT
Then L (.y, 71' ® /) satisfies (9.2.2) for all % (/o^. c^.).

We can now formulate the main theorem of this paper.
(9.3) THEOREM. — Let a be a unitary irreducible representation ofG^ (A) which is auto-
morphic cuspidal. Assume that for any character % o/F^/F", % + 1, the representations a
and a ® % are inequivalent. Then:
(1) for any %, L^ (s, <r, /) is entire;
(2) /or any place v the representation a^ admits a lift n^ to G^^;
(3) set n = ® Tty (all v). Then n is automorphic cuspidal;
9
(4) no representation of the form n (^, K^) with (1/4) ^ t occurs as a component of a.
The proof will occupy the rest of paragraph 9. (The case when CT ® % w a for some 50
has already been discussed in Section 3.7; see also Remark 9.9).
(9.4) We prove first (9.3.2). Let So be the set of places where G^ is extraordinary.
There is nothing to prove if So = 0. So assume So ^ 0, and let S => So be the set of
places v where ^ is not quasi-unramified. For v ^ S, let T^ be the lift of or.,. By (8.1)
the assumptions of (9.2) are satisfied with T = S. From (9.2) we conclude that there is,
for each v e S, a generic representation Uy, trivial on the center, such that, if
^ = ® ^ ® ®^,
ves vis
hen L (j, n' (g) 50) is meromorphic and satisfies the functional equation
L(5, n' ®5c) = s(s, 7C' ®x)L(l-5, IT' ®5c~1)
4® SERIE — TOME 11 — 1978 — N° 4

RELATION BETWEEN AUTOMORPfflC REPRESENTATIONS OF GL (2) AND GL (3) 535
for all ^c. On the other hand, L^ (s, cr, 50) is meromorphic and
L2(5, a, %) = 82(5, a, %) L2(l~s, CT, 5C~1)
for all 50. Since the local L and e factors in both equations are in fact the same for all
v ^ S we conclude that
(9.4.1) ft L(l-s, H^K^MS, <®X,; ^)/L(5, <®x,)

yeS
= [I L2(l-s, a,, ^"^(s, cr.,, 7,; x^)/!^, a,, /,).

veS
Now fix w e So and let T| be a character of F^. Choose / in such a way that /„ = TI and
Xy is highly ramified for v e S, t? ^ w. Then for t^ e S, y =^ w,
f L(5, ^ ® 3C,,) = L(5, ^ ® xJ'1) = 1»

(9.4.2)
1 e(s, < ® X.; ^.) = s(s, /„; v|/,)3,
and
(9.4.3) ( L2(5, a,, /,)= L2(5, a,, x;1) = 1,
1 S2 (s, 0^, ^; \|/,,) = e(s, /„; ^)3;
[c/. [JPSS] Theorem (5.6) and [Ja] Theorem (16.1)]. We conclude that
(9.4.4) E(s,<®^^;v|/JL(l-s,^®^^- l )/L(5,<®1^)

= 82(5, CT«,, -H; ^)L2(1-5, CT^, iT1)/!^, CT^, -n).
But now <!„ is extraordinary. Therefore
L2(s, cr«,,il) = L2(s, CT«,, r|~1) = 1
and the left hand side of (9.4.4) is a nomomial in q^3. This, however, can happen only
if < is super cuspidal ([JPSS], § 7). Therefore
L(s, < ® 11) = L2 (s, <r^, 11) (= 1),

and
e(s, < ®il; ^) = £2(5, cr«,, T|; ^).
Note that the functional equation for n' also reads
L(5, ?c' ®x) = e(s, TC' ®x~ l )L(l-5, TC' ®x~ 1 ),
so since Ttp ^ TCy ^ TC,, ^ Hy for y ^ S, we find
PI L(l-5, n, ®x^ l )e(5, ^ ®7,; ^)/L(5, < ®z,)

veS
= n L(l-5, < ® ^ l ) £ ( 5 , < ®X, ; ^)/L(5, 7C, ®X,).

reS
ANNALES SCIENTIFIQUES DE L'^COLE NQRMALE SVPERIEURE

Threfore we conclude as before that
L(l-s, ÔOrr^s, ^®TI;\|/J/L(S, <®T|)
is equal to the same expression for n^. But as Tiy, is trivial on the center, this implies
n^ ^ TT^ (cf. (7.5.3) of [JPSS]). Thus < is indeed the lift of CT^.
(9.5) Now that (9.3.2) is proved, for each v we let Tiy denote the lift of (7y. For v e S
[S being as in (9.4)], Uy is unitary generic and we set E;y = Tiy. For v ^ S we define £;„
as in (9.2). Note that the only places where ^y 4^ ^v are those for which
(9.5.1) a, = 7i (^ a174, ^ a-174)
in which case T^ is given by (9.1.2.1) and £;„ by (9.1.2.2). Then ^ has another component
^ which is unitary generic, namely
(9.5.2) ^ = Ind(G3.,, P,; c^2, a;172), 1,),
where o-(ni, ji^) is defined as in (1.10.1).

Now set n = ® TC(, (all u). Then:
L(s,7C®7)==L2(s, CT, 50),

(9.5.3)
6 ( 5 , 7 1 ®7) = 8 2 (S,CT,X),
and L (s, n ® %) is meromorphic and satisfies the functional equation
(9.5.4) L(5,7r®x)=e(s,iT®x)L(l-s,7c®5c- 1 ).
Finally, set T = S if S ^ 0, and T = { w } (where w is any finite place) if S = 0.

If / is sufficiently ramified at each place ofT, then we know [c/1 (8.1)] that L (s, n ® %)
is entire and bounded in vertical strips; moreover (cf. (5.1) of [JPSS]),
(9.5.5) £(s,7r,®/,,vl/,)=[E(5,x,,\|/,)] 3 , and L(5,7i,®7,)=l for ceT.
Therefore the discussion in (9.2) applies. In particular, we are in one of the cases 9.2 (ii)
to 9.2 (v). What we are going to show is that 9.2 (ii) is the only possibility and this will
imply the theorem.
(9.6) Suppose we are in case (iii). With the notations of (iii) we know therefore that the
representations ^ and Tj,, have a common irreducible component for all v ^ T. Although
the representations r|y, ^y, T| = ® r|y, ^ = ® ^ may fail to be irreducible, the factors
L (5, ^), L (s, r^), L (s, ^), L (s, T|), c (s, ^; v|/^), s (s, ^; \|/^), 8 (^, ^), and s (^, T|) are
defined. Moreover, for all character % of F^/F^,
L(5, ^ ®x) = e(5, ^ ®X)L(1-5, i ®X~ 1 ),

with a similar statement for L (5', T| ® %).
4° SERIE - TOME 11 - 1978 - N° 4

Arguing as we did in (9.4) we conclude that if v is any place in T and X is a character

ofF^, then:
s(s, ^ ®?i; vK)L(l-5, i (g)^1)/!^, ^ ®X)
= e(s, 11,®^; v|/^)L(l-5, ii,®^"1)/!^, Tl,®?i).
But ^y and r|y have unique irreducible generic components i;y and r\y and the previous
identity is true for these pairs too. So once more we conclude that ^. ^ T|^. In particular,
/or all v, ^ and r|y have a common irreducible component. Recall also that for almost
all finite v, Ky is unramified and equal to the common unramified component of ^ and T|(,.
To prove that (iii) is impossible we first suppose that |A is a character.
(9.6.1) If (A is a character then T is unitary, each representation T,, unitary generic, and
it follows that r\y is irreducible (in fact unitary generic, cf. [JPSS], § 6). Thus x\y is
an irreducible component of £;„ for all v (at tê infinite places in the infinitesimal sense
only). If, in addition, ^ is irreducible then r|y = §y = Tiy and
(9.6.1.1) L(s,7^,®^l; l )=L(s,^^,®^l,~ l )=L(s,T,®^l; l )L(5, 1,).
On the other hand, if ^ is not irreducible, then Tty is given by (9.1.2.1), ^ by (9.1.2.2)
(v being finite or not), and i;y admits another irreducible component which is unitary
generic, namely ^ (9.5.2). Since r|y is unitary generic, we find
^v = ^
in particular, r|y is ramified if v is finite. Thus the set E of places where ^ is reducible
(and r|y = ^y) is finite, and for v e E we find
(9.6.1.2) L(., ^ ®n;1) = L(5, T, (g)^1)!^, 1,). L(5? 7l(a^ ^^ 2))

L(s, G ^ 7 , a, 1/2))
The factor on the extreme right is (l—^ 72 " 5 )" 1 if v is finite, and c^(^—(l/2))~ 1 ify is
infinite.
In either case, it does not vanish for s = 1. Thus we find from (9.6.1.1) and (9.6.1.2)
that
L(., n ®n- 1 ) = L(5, T (g)^1)!^, l)x fl L(5? 7l(0^ / 2^^l /))

peEL(s,CT(a; ,a, 2 ))
and
L(5, (a®^- l )xa)=L(5, T®^- 1 )L(5, l)L(s, n- 1 )^ H L ( 5 ^ ( a ^ >/2a ^l/2

l,6EL(s,CT(a; ,a„ ))
But if \i is note a principal character then L (1, î"1) 7^ 0. Also L (1, T ® \i~1) ^ 0 by the
main result of [J Sh]. So since the finite product above does not vanish at 1, we find
that L(s, (cr ® n'^xcr) has a pole at s = 1. This means cr ® H~ 1 ^ a, which is a
contradiction. Thus n cannot be a character.
ANNAI.ES SCIENTIFIQUES DE I/ECOLE NOHMALE SVPERIEURE

(9.6.2) Now suppose \i = a 2 ^ po where ô ^ a character and ^ is real and ^ 0. For

almost all finite v, po is unramified and the representation T(, is of the form Ty = TC (a^, a^)
The unramified component ofr^ is therefore the representation n (a^, a^, a 2 ^ ô.t») which
must (for almost all v) coincide with the unramified component of §y which is TCy. Now Ky
has the form n (a^, 1, a^", ) and so the triples (a^, a^, a^) and (a^, \^pv) must be the
same (up to order).
From the discussion of paragraph 3.3 and 3.4 it follows that Oy must be in the comple-
mentary series for almost all finite v and, after a change of notations,
a,=-2f, &,=0, Uo.t.=l» 0<|f|<1.
Thus T, = ^(a^ 2 ^ 1) for almost all v. This situation, however, is impossible. Indeed
it implies the existence of a unitary cuspidal automorphic representation p of G^, (A) with
Pv = ^ (o^, a;"1), 0 < t < (1/2), for all v not in some finite set of places E. This in turn
implies [by (9.7)] that
L(s, p x p ) = L ( s + f , l)L(s-^, 1)L(5, I)2
xv^L(s+t,
n L(s,P.xp.)
UL(s-f, 1,)L(5, I,) 2
But at ^ = t, L(s+t, 1) and L(.y, 1) do not vanish. Neither does the finite product, as
is easily checked. Thus L (s, p x p) has a pole at s = t ^ 1, an obvious contradiction.
In summary, possibility (iii) cannot occur; similar arguments show that (iv) too is
impossible.
(9.7) Suppose finally we are on case (v). Using the notations of (v) we see (just as
before) that T|(, and ^ have a common irreducible component for all v.
(9.7.1) Suppose first that the v, are characters. Then for each v the representation T^
is irreducible unitary generic ([JPSS], § 6 and 10). Thus it is a component of ^. If ^ is
irreducible we find ^ = n^ = r\y and
( 9 - 7 - 1 -!) L(s, TC, ®x.) = L(s, 11, ®/,) = ]~[L(5, v^x,).

i
If ^ is not irreducible then, since Ky is the lift of a,, TC, is given by (9.1.2.1) and ^ by
(9.1.2.2). Since n^ is not unitary generic, T|, must be the other component of^ [c/. (9.5.2)]
i. e. :
TI, = Ind(G3.,, P,; ^(^l\ a;172), 1,).
On the other hand,
11, = Ind(G3^, B3,,; Vi^, V2^, V3^)
and—except in the complex case—these two equalities are incompatible.
4° SERIE — TOME 11 — 1978 — N° 4

Thus we let E denote the set of (complex) places where this happens. For v e E,
(9.7.1.2) L(5,^®x,)=L(5,i;,®/,)
=
ni^ ^ L(5+l/2,x,)L(5,x,)L(^l/2,x.)
HL(S, Vi,V%v)'—————————————T=f—————————————-—————————————•
i-1 n^^X.)
Moreover we then have (up to order):

v^(z) = z(zz^l/2, v,., = 1, V3., = z(zz)-112
and
o(a,l/2,a;l/2)=7^(v^„V3.,).
From (9.7.1.1) and (9.7.1.2) we find that

3
L(s, (a ®/) x CT) = ]"[ L(s, V(/)L(S, x)
1=1
^ rr L(5+l/2, x.)L(s, X,)L(5~1/2, xJ
Ê- n^v^/,,)
Now set x = vF1- Then L (^, Vi 7) has a pole at s = 1 which cannot be compensated
for by a zero of the other L-factors (nor by a zero of the finite product, as is easily checked).
Thus we get a pole for L (s, (a ® v^1) x o) at s = 1. But if v^ ^ 1 this contradicts our
hypothesis (since it implies CT ^ <r ® v^'1). On the other hand, if v^ = 1, we get a pole
of order at least two, and this is also a contradiction.
(9.7.2) Now suppose the v» are not all characters. Using again the fact that ^ and r\y
have (for almost all v) the same unramified component (namely Uy) we conclude as before
that Oy is in the complementary series for almost all v and that (changing notation if
necessary):
Vâ^Vi, V3=a~2tV3, V2=V2,
where v;' is a character and t ^ 0 is real. In fact we even find v^v = 1 for almost all v
so that Vi = a2^ V3 = a"2^ v^ = 1. Moreover 0 < t < (1/2).
Suppose first that t ^ (1/4). Then:
l^,=Ind(G3.„B3.„a?,l,(x; 2( )
is irreducible for all v and is a component of ^, for all v. If ^ is also irreducible then
^ = TC^ = r|y. On the other hand, if ^ is not irreducible then ^ has (as seen before)
2 components; one is n^ the other
^ = Ind(G3.,, P,; â,172, a;172), Q.
But ^ does not contain the trivial representation ofK^y and r\y does. Thus Tty = r\v again.
ANNALES SCffiNTIFIQUES DE L'^COLE NORMALE SUP^REBURE 70

Summing up, if t 1=- 1/4, then for all v,
L(s, TC,) = L(5, 11,) = L(5+2r, 1,)L(5, l,)L(s-2r, !„)

and
L(s, CTXCT) = L(s, 7c)L(s, 1) = L(5+2r, l)L(s-2f, l)L(s, I)2.
Again the pole of L (5--2 t, 1) at 2 tis not cancelled by a zero and we conclude L 0, CT x 5)
has a pole at s = 2 r ^ 1, a contradiction.
Now suppose ^ = 1/4. Then T|(, has always two irreducible components:
^=Ind(G3.,,P,;l^,U
i1; = Ind(G3.,, P,; oCo^2, a;172), 1,).
One of these is always a component of ^. Since ^ is unramified if v is finite, we find

T}y = TC(, for almost all v. If ^ is irreducible then either r|y or 11; is a component of £;„. If
^ is reducible then ^ = r^ and ^ = r^. In conclusion, for almost all v, ^ = 7^ and
L(5, 71,) = L(5, 11^) = I/S+ 1 1^L(S, UL^S- 1 , lA
However, at a finite set of places (say E) we may have r|y' = n^ and
L(,, ..) - L^ 1 , l.)L(. l,Lf,- 1 , ,.)LC^.^»».

2
\ / \ 2 / L(s, 12^)
So
LQ. qxa) = I/.4.1, iVf.- 1
,l)Lfe 1)- [1 L(s
' CT(otcl/2> a;l/2))
.
\ 2 / \ 2 / i;eE L(S, l^^)
Again the pole of L(^-(l/2), 1) at s = 3/2 cannot be cancelled by a zero of the other
L-factors or the finite product and we again get a contradiction.
(9.8) We have now shown that the only possibility is (ii). Thus n^ = ^y is unitary
generic for any v t T. If S ^ 0, S = T and the same conclusion is true for v e S [c/. (3.41]).
On the other hand, i f ^ = 0 , then T = { w }. Since w is really arbitrary the same conclu-
sion is true for all w. Thus n^ = ^ is unitary generic/or all v.
We have seen in [9.2 (ii)] that there is a cuspidal automorphic representation K ' with
trivial central character and n^ ^ Uy for v ^ S. Now for any %:
(9.8.1) L(s, 7i' ®x) = 6(5, ^ ®x)L(l-s, T? ®5c-1).
Similarly L(s, n ® 7) satisfies (9.5.4). So using an argument used before, these two
equations imply that for any v e T and any character ^ of F^,
L(l-5, T^®^-1)^, 7T,®X;^)/L(5, 7C,®^)
= L(1-S, 7?, (SK-1)^, K, ®^;^)/L(S, 7<, ®^).
4® SERIE — TOME 11 — 1978 — N° 4

However, since both Ky and n'y are generic, this implies TCy ^ 7iy. Thus TT = ® 7iy is aufco-
morphic cupsidal and we have proved part (3) of Theorem (9.3). But part 1 follows
from the identity 1^ 0, a, 7) = L^ 0, TT ® 50), and since the lift T^ of ^ is always unitary
generic, the last assertion of the Theorem also follows.
Remark (9.9). - If cr ^ a ® T|, with T| ^ 1, then T^2 = 1. Let H be the corresponding
quadratic extension of F. Recall that there is a character 0 of H^/HP such that CT is the
automorphic representation TI (a^) of G^ (A) attached to Q. Let Q' be the conjugate of 0.
Then the lift 71 of CT is
TT = Ind (Gs (A), P (A); 7t (CT^ -1), 11)
and
L(s, TC ®x) = LÔ?, a, x) = L(s, OQ'-1. X°NH/F)L(S, ^).
Recall that n is automorphic. On the other hand, L(s,n ® %) has some pole (for % = T|
for instance), so 7C is ^ cuspidal.
Concluding Remark. — We wish to comment finally on the classical significance of
part (4) of Theorem 9.3. Suppose
/(z)= fU^""2
n=l
is a cusp form of weight k on TQ (N) (which is not a theta-series with grossencharakter a la

Hecke-Maass). By the Ramanujan-Petersson conjecture recently established by Deligne
we know that
(*) aÔ^ 2 - 1 / 24 - 6 ).
On the other hand, if F = Q and n corresponds to the form / above, then part (4) of
Theorem (9.3) implies that
(**) aÔ^2-174^).
This estimate is stronger than the one made by Rankin in his well-known work [Ra] but
obviously weaker than (*). According to [Se] (**) also results from Weil's work on
exponential sums (cf. [W 2]). For number fields, however, our result seems to be new.
Indeed the generalized Ramanujan-Petersson conjecture—though established for function
fields now by Drinfeld-has not yet been proven in general.
REFERENCES
[Bo] A. BOREL, Automorphic L-functions (Proc. of Symposia in Pure Math., Corvallis Conference,
Oregon 1977, American Math. Soc., Vol. 33, 1979).
[De] P. DELIGNE, Les constantes des equations fonctionnelles des fonctions L (Springer Lecture Notes,
Vol. 349, 1973, pp. 501-598).
[Ge] S. GELBART, WeU^s Representation and the Spectrum of the Metaplectic Group (Springer Lecture
Notes, Vol. 530, 1976).
ANNALES SCIENTIFIQUES DE I/ECOLE NORMALE SUP^RIEURE

[Ge Ja] S. GELBART and H. JACQUET, A Relation between Automorphic Forms on GL (2) and GL (3) (Proc.
Nat. Acad. Sc. U.S.A., Vol. 73, No. 10, October 1976, pp. 3348-3350).
[Ge J 1] S. GELBART and H. JACQUET, Forms ofGL (2) from the Analytic Point of View (Proc. of Symposia
in Pure Math., American Math. Soc., Vol. 33, 1979).
[Go Ja] R. GODEMENT and H. JACQUET, Zeta-functions of Simple Algebras (Springer Lecture Notes, Vol. 260>
1972).
[Ja] H. JACQUET, Automorphic Forms on GL (2): Part II (Springer Lecture Notes, Vol. 278, 1972).
[Jl] H. JACQUET, Principal L-functions of the linear group (Proc. of Symposia in Pure Math., American
Math. Soc., Vol. 33, 1979).
[JL] H. JACQUET and R. LANGLANDS, Automorphic Forms on GL (2) (Springer Lecture Notes, Vol. 114,
1970).
[JPSS] H. JACQUET, I. I. PIATETSKI-SHAPIRO and J. SHALIKA, Automorphic Forms on GL (3) (Annals of
Math., Vol. 109, 1979).
USh] H. JACQUET and J. SHALIKA, A non-vanishing Theorem for zeta-functions ofGLa (Inventiones Math.,
38, 1976, pp. 1-16).
[LL] J.-P. LABESSE and R. P. LANGLANDS, L-indistinguishability for SL (2), Institute for Advanced Study
Princeton, 1977, (preprint).
[La 1] R. P. LANGLANDS, Problems in the Theory of Automorphic Forms (Springer Lecture Notes in
Mathematics, Vol. 170, 1970).
[La 2] R. P. LANGLANDS, On the Functional Equation of the Artin L-Functions, mimeographed notes,
Yale University, 1968.
[La 3] R. P. LANGLANDS, On the Functional Equations satisfied by Eisenstein Series, mimeographed notes,
Yale University, 1967.
[La 4] R. P. LANGLANDS, Base Change for GL (2) : The Theory of Salto-Shintani with Applications,
Institute for Advanced Study, Princeton, 1976, (preprint).
[La 5] R. P. LANGLANDS, On the Notion of an Automorphic Rrepresentation [Proc. of Symposia in Pure
Mathematics, Corvallis Conference, Oregon 1977 (to appear)].
[Ra] R. RANKIN, Contributions to the Theory of Ramanujan's Function T (n) and Similar Arithmetical
Functions (Proc. Camb. Phil. Soc., Vol. 35, 1939, pp. 351-356).
Si] A. SILBERGER, The Langlands quotient theorem for p-adic groups (preprint).
[Se] A. SELBERG, On the Estimation of Fourier Coefficients of Modular Forms (Proc. of Symposia in
Pure Math., Vol. 8, A.M.S., Providence, R.I., 1965).
[Sh] G. SHIMURA, On the Holomorphy of Certain Dirichlet Series (Proc. London Math. Soc., (3) 31»
1975, pp. 79-95).
^ Ta] J. TATE, Fourier Analysis in Number Fields and Hecke's zeta-function, in Algebraic Number Theory
(Proc. of the Brighton Conference, Academic Press, New York, 1968).
[We 1] A. WEIL, Sur certains groupes d'operateurs unitaires (Acta Math., Vol. Ill, 1964, pp. 143-211).
[We 2] A. WEIL, On some Exponential Sums (Proc. Nat. Acad. sc., U.S.A., Vol. 34, 1948, pp. 204-207).
(Manuscrit recu Ie 4 janvier 1978,
revise Ie 21 mars 1978.)
S. GELBART,
Cornell University,
Ithaca, New York 14853,
U.S.A.
4° SERIE -- TOME 11 — 1978 - N° 4

Asens 1978 4 11 4 471 0

Uploaded by

Copyright:

Available Formats

Asens 1978 4 11 4 471 0

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Asens 1978 4 11 4 471 0

Uploaded by

Copyright:

Available Formats

A NNALES SCIENTIFIQUES DE L’É.N.S.

© Gauthier-Villars (Éditions scientifiques et médicales Elsevier), 1978, tous droits réservés.

Article numérisé dans le cadre du programme

The purpose of this paper is to establish a relation between automorphic forms on

* Both authors have been supported by grants from the N.S.F.

ANNALES SCEENTIFIQUES DE L'ECOLE NORMALE SUPERIEURE

should correspond to an irreducible admissible representation n^ = p* (a,,) of GL (3, Fy).

4° SfiRIE — TOMB 11 — 1978 — N° 4

L(s, TCiXTia) =^ L ( S » ^i^x71^)?

L(s, (a ® /) x a) = L^ (s, a, x) L(s, x),

ANNALES SCIENTDFIQUES DE L'ECOLE NORMALE SUPERIEURE

1. GL (2) x GL (2): local theory

^ ( S ^I^F^)^] for each ^eF and geG^F).

The existence and uniqueness of such a space is proved in Chapter I of [JL].

(1.1.3) y(5,Wi,W2,<D)= f Wi(g)W20lg)0[(0, l)g]|detg|^g,

4° SERIE — TOME 11 — 1978 — N° 4

(1.1.5) ^(l-s, Wi, ^2, 6)/L(1-5, ^ x^)

(1.1.6) ^(X, }Q == $(M, l;)\|/F(MX+^)rfM^

^W,W,$)=JJ^W,[(;; ?)^]W3[(7 ^fW^d-adk,

K = GL(2, R^) and f(g) = | detg| 5 fo[(0, Qg] | t^^^d^.

But 71:2 is supercuspidal. Therefore ([JL], Prop. (2.16)) the function

lim(s-so) p[(0, 0] |f| 2s (o)l0)2)- l (0d^= c0(0, 0),

with c a non-zero constant independent of 0. It follows that

lim(s-so)^(s, Wi, W2, 0) = c0(0, 0)b(Wi, W2),

ANNALES SCIENTIFIQUES DE L'ECOLE NORMALE SUPERIEURE 62

&(7Ci(g)W,, 7r,(g)W,) = Idetgl^^Wi, W,).

COROLLARY (1.3). — Suppose n is a supercuspidal representation and T| is the unramified

^ 4 ^ f L(S, ^l X7^) = L(5,-TCi ®|^)L(S, TCl 8)^

Suppose TT^ ^ ^^ special representation 0(^2, V2) m^ 1^2 ^'I1 :== a

4° StKEB, — TOME 11 — 1978 — N° 4

FWfy suppose TC( = CT(|A(, v<), H^v,"1 = ap, i = 1,2. 77^;

^,. 4. ^j; f L(s, ^i X7C2) = L(5, Hi U2)L(s, Vi ^2).

Here the L ^zrf e factors on the right are those of [JL].

L(S, 7t2 ®Pi)L(s, 7T2 ®Vi) = L(s, ^^^(s, V^^

and there is a similar assertion for the s-factor.

ANNALES SCDBNTIFIQUES DE L'^COLE NORMALE SUP&UEURE

should satisfy the following properties:

(1.5.1) n (oT =71(0), 7i:(CT®5c)=7r(a)®/,

(1.5.2) det (cr) = central quasi-character of n (cr),

(1.5.6) L(S, 7r(CT)X7T(T)) = L(5, CT ®T),

(1.5.7) E(S, 7I;(CT)X7C(T); V|/F)=£(S, <T®T;^F).

PROPOSITION (1.6). — Regard F as the completion at some place v of some global

4° S^RIE -~ TOME 11 -- 1978 — N° 4

Proof. - By Corollary 19.16 of [Ja],

= L(l—s, a®T)s(s, <J®T; \|/p)/L(5, a®T).

(1.9.1) L(5, 9oNK/F) = L(5, 9)L(5, 9Q,

1.9.2) X(K/F, v]/F)8(5, 9oNK/F ; vk) = s(5, 9; ^(s, 9^; v^).

ANNALES SCIENTIFIQUES DE L'ECOLE NORMALE SUPERIEURE

If % is any character of K^ let CT^ be the representation defined by (1.8.1). If a is

L(s, CT x 7c(CT^)) = L(s, n ® %)

L(5, cr x n(a^)) = L(s, n ® 50)

We note that if CT exists at all it is unique. Moreover, suppose CT = n (r) where T is a

E(S, CTXTl(CT^); \|/p)L(l-5, CTX7C(CT^))/L(5, CTXTC(CT^),

= e(5,7i ®%; \|/K)L(I-S, n ®3C~1)/L(5, n ®%).

Then n is a supercuspidal representation of G^ (F) and a lifting of CT.

L(S, CT X ^t(CT^)) = L(S, CT X CT(CT^) = 1.

L(5, TC ®;c) = L(5, ii ®x"'1) = 1,

and (1.9.4) is satisfied.

4° SERIE — TOME 11 — 1978 — N° 4