This action might not be possible to undo. Are you sure you want to continue?
The trace fortmila
THE TRACE FORMULA FOR REDUCTIVE 'GROUPS*
James ARTHUR
This paper is a report on the present state of the trace formula for a general reductive group. The trace formula is not so much an end in itself as it is a key to deep results on automorphic representations. However, such applications have only been carried out for groups of low dimension
( 51,
171,
[8(c)],
[ 3 ] ) . We will not try to discuss them here.
For reports on progress towards applying the trace formula for general groups, see the papers of Langlands [8(d) I and Shelstad [ll] in these proceedings Our discussion will be brief and largely confined to a description of the main results. On occasion we will try to give some idea of the proofs, but more often we shall simply refer the reader to papers in the bibliography. Section 1 will be especially sparse, for it contains a review of results which were summarized in more detail in [l(e)]. This report contains no mention of the twisted trace formula. Such a formula is not available at the present, although I do not think that its proof will require any essentially new ideas. Twisted
~ e c t u r e sfor Journges Automprphes, Dijon, Feb. 1619, 1981. I Wish to thank the University of Dijon for its hospitality.
1
Janes ARTHUR
trace formulas for special groups were proved in r8(c) I and [ 3 1 . As for the untwisted trace formula, we draw attention to Selberg's original papers [10(a)l, [10(b)], and also point out, in addition to the papers cited in the text, the articles [ 5 1 ,
C9 (b)I .
[l(a) ], [ 4 ] ,
[12], [9 (a)l and
1, THE TRACE FORMULA  FIRST VERSION,
Le G be a reductive algebraic group defined over split component of the center of G, and set
2).
Let AG be the
where X(G)^ is the group of characters of G defined over 9. Then a r e vector space whose dimension equals that of AG. Let G ( A )
is be the
kernel of the map H
:
G (A) *
^,
which is defined by
Then G (Q) embeds diagonally as a discrete subgroup of G (A)
and the
coset space G (Q)\G (A) has finite invariant volume. We are interested in the regular representation R of G (A) on (G(Q)\G (A) 1)
. If
1)
f e c (G(A) 1) , R(f) is an integral operator on L2 (G(Q)\G
(A)
. The
sour
ce of the trace formula is the circumstance that there are two ways to express the integral kernel of R(f).
We shall state the trace formula in its roughest form, recalling
briefly how each side is obtained from an expression for the integral kernel. This version of the trace formula depends on a fixed minimal
The trace f o m ' i a
parabolic subgroup Po, with Levi component Mo and unipotent radical N o , and also on an appropriate maximal compact subgroup K = II K v In addition, it depends on a point T in of G ( A ) 1
.
which is suitably regular with respect to that
a(T)
is large for each root
a
of
Po, Po
in the sense
A0 =
On
\.
The trace formula is then an identity
We describe the left hand side first.
0 denotes the set
of equivalence classes in
G($),
in which two elements in
G ( Q ) are deemed equivalent if their semisimple components are
G(Q)
conjugate.
This relation is just G ( Q )
conjugacy if G
is anisotropic, but it is weaker than conjugacy for general G If
P
.
is a parabolic subgroup of Po, and
G
which is standard with
respect to
o
e
0,
set
where
N
is the unipotent radical of
P
P
and
Mp
is the unique
Levi component of
which contains M
.
Then
James ARTHUR
is the integral kernel of the operator convolving f on
Rp(f)
obtained by
L ( N @)M (Q) (A) \G I). If P = G
,
it is
just the integral kernel of T k(x,f) =
R(f).
Define
P3P0 The function
I
dim ( A / A 1 (1)
' 6eP (Q)\G (Q) p e
I
K
(fix,<5x)~p(~p(5x)~).
Tp ( H (  )  T) ,
1 if
whose definition we will recall in and vanishes on a large compact P s G
a moment, equals
P=G,
neighborhood of 1 in G ( A ) I if T k&(x, f) i$ obtained by modifying hood of infinity in G ( @ )\ G ( A )
.
In other words,
.
in some neighborT The function k(x,f) turns
( y) KG ,& x ,
out to be integrable, and the distributions on the left of the trace formula are defined by
If P
is a standard parabolic subgroup,
characteristic function of a certain chamber. and P I Ap  AMp If Q
P is the Write a = en.
, T
is a parabolic subgroup that contains
o<_
there is a natural map from
onto
a . Q'
We shall
write
for its kernel. (P , A ) .
Let
Ap
denote the set of simple roots of
It is naturally embedded in the dual space
The trace formula
of in
.
There is associated to each Let
a e Ap
a "coro~t" av
.
8
be the basis of t^p/Agwhich is dual to ,Then
*
*
{av : a e
A }
.
<
is the characteristic function of
Let by
Hp be the continuous function from G ( A )
to
%
defined
T In the formula for k@(x,f)
above, the point
T
belongs It
to a , but it projects naturally onto a point in A . is in this sense that
is defined. The set
X
which appears on the right hand side of (1.1) If
0
is best motivated by looking at 0.
0,
consider those
standard parabolic subgroups B which are minimal with respect to the property that union of M ( Q )
a meets M B . Then
e n
MB
is a finite
conjugacy classes which are elliptic, in the
sense that they meet no proper parabolic subgroup of MB which is defined over
Q.
Let
W
be the restricted Weyl group of
Jmea ARTHUR
(G
,A).
It is clear that
0
is in bijective correspondence (MB , cB), and cB where B is a
with the set of
Woorbits of pairs G,
standard parabolic subgroup of conjugacy class in MB(Q). representations of GA' () G(@),
is an elliptic
If we think of the automorphic as being dual in some sense to the we can imagine that the c u s p i d a l
conjugacy classes in
automorphic representations might correspond to elliptic conjugacy classes. of pairs of G and ( M ,r ) , r
X
is defined to be the set of Worbits where B is a standard parabolic subgroup
is an irreducible cuspidal automorphic represen
tation of MB@) of groups B
.
For a given
x
E
X ,
let
P
x
be the set
obtained in this way.
It is an associated class
of standard parabolic subgroups. associated class Suppose that
L 2 ( N (&)M~(Q)\G (A)'
Similarly, we have an
Pn.
for any
<ye P a
0.
P
0
x 6 )x
X
and
are given.
Let in For every
be the space of functions @ with the following property. with B
c
L~ [ N (A)% (Q)\G () ) A'
standard parabolic subgroup B ,
x e G @I)
P
, and almost all
,
the projection of the function
I
under MB(~)' (MB I rB)
1 (nmxldn ,
1 meMg(A) ,
% (fa)\%@I
L2 ( M (Q)\MB@I1 ~
onto the space of cusp forms in
)
transforms in which
as a sum of representations
rB ,
is a pair in
x.
IÂ there is no such pair in
x,
B , X will be orthogonal to the space of cusp forms on % (Q)\Mc () . It follows from a basic result in Eisenstein A'
The trace fornula
series that group in
L (Np@IMP (Q)\G (A)' )
will be zero unless there is a P. Moreover, there is an
P
x
which is contained in
orthogonal decomposition
Let R f
Kp
1
x (x,y) ^x
be the integral kernel of the restriction of
to
L (Np(AIMp(Q) (A) I ) \G
.
One can write down a We have
formula for
(x,y) in terms of Eisenstein series.
each side being equal to the integral kernel of define the modified functions dim ( A h G 1 kT (x,f) = [ (1) p3p0 6eP ( Q ) \G (Q)
Rp(f).
Xf we
I
K ~ l (fix , fix) (~~(5x1T ) x Tp 
we immediately obtain an identity
T It turns out that the functions kx(x,f) are also integrable.
In fact, the sums on each side of the identity are absolutely integrable. The distributions on the right of the trace formula are defined by
James ARTHUR
The trace formula (1.1) follows. Most of the work in proving (1.1) comes in establishing the kT (x, f) and kT (x, . Of these, the second f) integrability of
x
function is the harder one to handle.
To prove its integrability G(Q) \ ( ) GA'
it is necessary to introduce a truncation operator on Given on T
.
as above, the truncation of a continuous function h is the function
G(Q)\G@I1
1 2 truncating the function
If
x
X ,
let
ATAT K (x,y) be the function obtained by
x
in each variable separately. operator, one shows that
From properties of the truncation
is finite (see [l(d), 511 ) that
.
One can also show, with some effort,
([l(d), Â § 2 ] ) from which one immediately concludes that
The trace forrnula
is also finite
( [l(d), Theorem 2.1 ] ) . This was the result that
In the process, one shows that $he integral of
was required for formula (1.1). for any
x
e
XI
is zero for sufficiently regular T other words,
( l (d), Lemma 2.41). [
In
T This second formula for Jx(f)
is an important bonus.
We shall
see that it is the starting point for obtaining a more explicit T formula for Jx ( f )
.
2,
SOME REMARKS
~t is natural to ask how the terms in the trace formula T.
It is shown in Proposition 2.3 T of [1(Â£] that the distributions JT (Â£ and J(f) are poly& nomial functions of
T
(1.1) depend on the point
T,
and so can be defined for all points To in
in
<x.
There turns out to be a natural point
%
such that the distributions
James ARTHUR
and
are independent of the minimal parabolic subgroup Po. For a better version of the trace formula, we can set T = To to obtain
(incidentally, To is strongly dependent on the maximal compact subgroup K For example, if G = G L and Ho is the group of diagonal matrices, . T0 will equal zero if K is the standard maximal compact subgroup of GI,@). However, if K is a conjugate of this group by Mo(A.), To might not be zero). The distributions J
Y
and J
x will
still depend on No and
K. Moreover, they are not invariant. There are in fact simple formulas to measure how much they fail to be invariant.
Let L(M~)be the set of subgroups of G, defined over Q, which contain Mo and are Levi components of parabolic subgroups of G. Suppose that M e L(Mo). Let L(M) be the set of groups in L(M~)which contain M. Let FI) be the set of parabolic subgroups of G, now no longer standard, (1 which are defined over Q and contain M. Then if Q also 0 contains M. Let P(M) be the set of groups Q in F(M) such that M equals
e
M
F(M) ,
M.
Suppose that L e KM)
. We write
Q
L~ (M) , F~ (M) and F^ (PI)
n L(Q) is a disjoint
for the analogues of the sets 1 (M), F(M) and P(M) when G is replaced by L. Now suppos'e that 0 is a class in 0. Then '
(?
The twice formula
union
of equivalence classes in the set, OL
,
associated to L . on LA' ()
We
can certainly define the distributions JBi
.
set
By definition, J
(9
is zero unless
0
meets L(Q).
If
x
X
,
we can define a distribution J Formula (2.1), applied to L
x
on L) @' yields
in a similar way.
,
where now
f
is any function in C ~ ( (A)' ) L
.
The formulas which measure the noninvariance of our distributions depend on a certain family of smooth functions
indexed by the groups L e I ( M ) and define these functions here. map
Q e F (M)
.
We will not
They are used to define a continuous
James ARTHUR
f r o c~(L(A)~) to m
: ~feft)'
.
C(@' ;M))
is given by
for each
y
E
L) &'
.
If
fQY(m)
where
6Q
is the modular function
Q(A).
Then the formulas
alluded to are
and
where [l(f)
tfc 0 .
y c X I ,
.
Theorem 3.21
.)
f c cL&' "()) M Here 1 wOQ1
and
y c L(A).
(see
stands for the number of
elements in the Weyl group of function
( M ,A )
.
As usual,
fY
is the
f will equal f by definition. We therefore QrY obtain a formula for the value of each distribution at fy  f as a sum of terms indexed by the groups Q c FL (Mo), with Q * L . If Q = L r The distribution will be invariant if and only if for each f L and y . the sum vanishes. For example, J^, will be invariant if and only if
O'n M(Q)
is empty for each group M e F ~ M )
The trace f o d a
with
M
*
L.
3, T H E T R A C E F O R M U L A IK I N V A R I A N T F O R M
There is a natural way to modify the distributions J, and
J
x
so that they are invariant.
This was done in the paper
[I(Â£)] under some natural hypotheses on the harmonic analysis of the local groups G(Q). of this construction. If H is a locally compact group, let
H(H)
We shall give a brief discussion
denote the
set of equivalence classes of irreducible unitary representations of
H.
Suppose that M
is any group in
L(M).
We shall agree
to embed of
M(A)'
II (M (A) ) I
and
AM@?)
in
0
II (M (A) ; for )
M (A) is the direct product
,
so there is a bijection between
H (M (A) )
II (M (A) ) I
on
AMOR)
0
and the representations in
which are trivial
.
Let
IItemp
(M(/A)
1
)
be set of tempered representa
tions in
II ( M ( < A ) )
.
From HarishChandra's work we know that
there is a natural definition for the Schwartz space, C ( M l & ) ) , of functions on
C (M
1 (a))
M(A)'
.
There is also a linear map
7 "
from
to the space of complex valued functions on
%emp ( M @ ) ~ ) , ' given by
In [l ( f) of
T
,
Â5
I we proposed a candidate, l (M ( A ) ) , for the image
l (M (A)I )
.
Roughly speaking,
is defined to be the space which are ~chwartz
of complex valued functions on
I H t e (M (A) )
James ARTHUE
functions in all possible parameters. T maps
It is easy to show that 1 (M (A) ) maps
C (M (&) I)
continuously into
.
~t is also easy
to see that the transpose space of butions on I(M(A)), M @I )
T ;
of
T
1 (M (A) ) ' , the dual
'
into the space of tempered invariant distri
.
For each M
e
HYPOTHESIS 3.1:
? (M (&)
(Mo), TM maps
C (()) MA'
onto
') .
Moreover, the image of the transpose,
is the space of aH tempered invariant distributions on M) @'
.
I
This hypothesis will be in force for the rest of 53. is any tempered invariant distribution on M @ )
If
,
we will let TM (I) = I
I
A
I
be the unique element in
7 (M ( A ))
such that
.
Important examples of tempered invariant distributions are the orbital integrals. valuations on
Q,
Suppose that
S v
is a finite set of in Set
1
and that for each M defined over Q .
S,
T
is a
maximal torils of
T = ( TT T ( Q ) ) n M W ;
ves
,
in T ; whose
and let T ' be the set of elements y St reg centralizer in
The trace formula
equals
T ;
.
Given
f e C (M~A)')
and
1 y e TSrreg,
the orbital
integral can be defined by
ID(^) 1 %
on
is the function on
T ;
which is usually put in as a
normalizing factor.)
C (M (!A)) .
on
2 (M @ ) A'
I is a tempered invariant distribution Y By Hypothesis 3.1 it corresponds to a distribution
?Y
) . Now suppose that
Â
has compact support.
Then the map
has bounded support; that is, the support in closure n
T
T1
S ,reg
has compact
.
Let
I (@' M)
)
be the set of functions Ts
4
2 (M (a))
such that for every group
,
the function
has bounded support. such that. T
There is a natural topology on
C :
I (~($4))
).
(M @A) )

maps
(M () ) &I
continuously into
1 (M @I1
HYPOTHESIS 3.2 : For each group H
onto
c
L (, , M)
maps
C :
1 (M(A) I . )
Moreover, the image of the transpose,
James ARTHUR
is the space of all invariant distributions on'M(A)
1
.
be
If I is an arbitrary invariant distribution on M(A) l , let 1 the unique element in I (M(A) ) such that TA(1) = I.
A
Much of the paper Cl(f)l is devoted to proving the following theorem
THEOREME 3.3
:
There is a continuous map
for every pair of groups M
c
L in L(MO), such that
and
We will not discuss the proof of this theorem, which is quite difficult. Given the theorem, however, it is easy to see how to put the trace formula into invariant form. PROPOSITION 3.4
:
Suppose that
J~ : C(() :LA
1
)
+ (E
,
L
6
L(M~),
is a family of distributions such that
The trace formula
for a
h L
f
1 c~(L(A)) and y e LA' ()
. Then there is a unique
family
of invariant distributions such that for every L and f,
PROOF
:
Assume inductively that I
M e
has been defined and satisfies (3.1) L. Define
for all groups
L(Mo) with M
5
for any f
c_(L(A) 1) . Then
is certainly a distribution on L(A) 1
.
The only thing to prove is its invariance. We must show for any
y e L (A)
that IL (fy) equals IL (f). This follows from the fornula for
~ ~ ( f the) formula for 1(l(fy), and our induction assumption. ~ ,
n
According to (2.3) and (2.3) , we can apply the proposition to L the families { J } and {JL1. We obtain invariant
x
James ARTHUR
distributions
] : I {
and
{; I}
which s a t i s f y t h e analogues o f
(3.1).
The t r a c e formula i n i n v a r i a n t form i s t h e c a s e t h a t of t h e f o l l o w i n g theorem.
L=G
T E R M 3.5: HOE
For any group
L
in
i(Mo),
PROOF:
Assume i n d u c t i v e l y t h a t t h e theorem h o l d s f o r a l l groups with
M e I ( M )
M
;L .
For any such
M
we a l s o have
f o r any
$ e I ( M @ ) ) . Then
by ( 2 . 2 )
and ( 3 . 1 ) .
The trace formula
4, INNER
I@, Jx
PRODUCT OF TRUNCATED E I S E N S T E I N S E R I E S ,
An obvious problem is to evaluate the distributions J _ , and I
x
explicitly.
How explicitly is not clear, but
we would at least like to be able to decompose the distributions as sums of products of distributions on the local groups G(Q).
0.
In [l(c)] w e d e fined the notion of an u n r k i f i e d class in If 6  is unramified, JT f ) can be expressed as a weighted ( f ([1 (c), (8.7)1 )
orbital integral of express I(f)
.
It is possible to then
as a certain invariant distribution associated (see [l(f), 5141 for the case
to a weighted orbital integral. of GL.) If
o
is not unramified, we would expect to express
J(f) Then
as some kind of limit of weighted orbital integrals. I*(Â£ would be a limit of the corresponding invariant In any case, the lack of explicit formulas for with
o
distributions. Je(f) and
I&(f),
ramified, should not be an
insurmountable impediment to applying the trace formula. One can also define an unramified class in
X .
For any
such class, it is also not hard to give an explicit formula for ~ ( 5 ) . (see [l(d), p. 1191 .)
x
Unlike with the classes
@
,
however, it seems to be essential to have a formula for all in order to apply the trace formula.
x
We shall devote the rest
of this paper to a description of such a formula. Suppose that
A (P)
P
e
F(M)
is a parabolic subgroup. Let
be the space of squareintegrable automorphic forms on whose restriction to Mp (A)' is square integrable.
N (AIMp( Q ) \G iA)
There is an Eisenstein series for each
6 e A (P) given by
James ARTHUR
~t converges for
Re(X)
in a certain chamber, and continues
analytically to a meromorphic function of and in
IT
6
X eÂ¥^p,a If
*
x
e X
(P) be the space of vectors XI A (P) which have the following two properties.
2
n ( l > ~ ( A ) ) , let
A2
+
(i) The restriction of
6
to
G(A)
belongs to
L~(N~Ãˆ)M~(Q)\G(A)~ (ii) For every m
>
.
x
in
I
G (A) the function , m
6
o> (mx)
fL> (A),
IT.
transforms under Let
Mp(<A.) according to
xrn inner product
i2 be the completion of A2 (P)
(PI with respect to the XI'"
For each of G(A) on
*
&P,C
there is an induced representation (PI, defined by
x2 XI
Ox,
IT
The representation is unitary if
X
is purely imaginary.
Now, suppose again that a minimal parabolic'subgroup
The trace formula
Po e P ( M )
h a s been f i x e d .
Let
T
be a p o i n t i n
f f L
which
W s h a l l begin by e Po. d e s c r i b i n g t h e r i g h t hand s i d e of (1.2) more p r e c i s e l y . The kernel Kx(xly) can be e x p r e s s e d i n terms of E i s e n s t e i n s e r i e s
is s u i t a b l y regular with respect t o
where
@
i s summed o v e r a s u i t a b l e orthonormal b a s i s of
A?
X^
(P).
To o b t a i n
A
A K (x,y),
x
w e j u s t t r u n c a t e each o f t h e two
Then JT(f)
E i s e n s t e i n s e r i e s i n t h e formula. x=y
x
i s given by s e t t i n g
i n t h e r e s u l t i n g e x p r e s s i o n , and i n t e g r a t i n g o v e r
G ( Q ) \G @)I
.
x
I t t u r n s o u t t h a t t h e i n t e g r a l over
G ( Q ) \G(A)'
may be t a k e n i n s i d e a l l t h e sums and i n t e g r a l s i n t h e formula T T f o r A,A K ( x , x ) . T h i s p r o v i d e s a s l i g h t l y more convenient expression f o r
P
J (Â£
T
( [ l ( d ) , Theorem 3.21)
.
Given
Po
,
T
II ( ~ $A)) , and p
on
A?
A
c
ioii
,
d e f i n e an o p e r a t o r
nT ( P I ^ ) XI^
XI^
(P)
by s e t t i n g
f o r any p a i r of v e c t o r s becomes t h e formula
and
@
in
A
X,v
(PI.
Then (1.2)
James ARTHUR
where
We hasten to point out that (4.2) does not represent an T explicit formula for Jx(Â£) It does not allow us to see how to decompose J
x
into distributions on the local groups G(Q). J(Â£
T
Moreover, we know that
is a polynomial function of
T.
However, this is certainly not clear from the right hand side of (4.2). The most immediate weakness of (4.2) is that the definition of the operator (P,A) is not very explicit. However, there XlT A) is a more concrete expression for QT (PI , due to Langlands. X^ (see [8(a), 591 .) It is valid in the special case that P
Q
belongs to the associated class
Px ;
that is, when the
Eisenstein series on the right hand side of (4.1) are cuspidal. To describe it, we first recall that if in P and
P
are groups
F ( M ) and s belongs to W (CT. a z ,
Ofp onto
isomorphisms from elements in W
M
) , the set of 1 obtained by restricting
to .<7~,,
"PI then there is an important function
pllp
(s,A).
Forany
$ed2(p),
is defined to be
The trace formula
The integral converges only for the real part of chamber, but Mpl,;(s,A)

X
in a certain
can be analytically continued to a
meromorphic function of a maps from
A e uc
A (P) to
I ( M (A) I )
with values in the space of PIC A (PI). Suppose that T is a
2
*
representation in
.
Then M
l p
(s, A)
maps the subspace XI* (P,A) be the
2 (P) to AxrsTT(Pl). If XI* value at A = A of '
A*
A e i<^ , let uT
where
Here,
Z(A ) pl
G is the lattice in .<rtp
generated by 1
{a":
Then
OJ
a e A } .
1
T ( P I A) is an operator on A (P) X Ã ˆ XI* Lanqlands' formula amounts to the assertion that if T belongs to P the operators Q~ (PI ) and u A (PI A)
.
P are
xr
XI*
x IT
equal. This makes the right hand side of (4.3) considerably more explicit. However, J (f) is given in (4.2) by a sum over all
T
James ARTHUR
s t a n d a r d p a r a b o l i c subgroups belong t o
P
x'
n o t be e q u a l .
U n f o r t u n a t e l y , i f P does n o t T t h e o p e r a t o r s CIT ( P , l ) and w (PIX) may XI^ x,Tf The b e s t w e can s a l v a g e i s a formula which i s
P
.
asymptotic w i t h r e s p e c t t o Set
T
.
W s h a l l say t h a t e
to
Po
T
approaches i n f i n i t y s t r o n g l y u i t h r e s p e c t infinity, but
T
if
1 1 ~approaches ~ ~
remains w i t h i n a
region
f o r some
6 > 0.
T E R M 4.1: HOE difference
If
d>
and
4'
a r e vectors i n
A^
x,v
(P) ,
the
approaches z e r o a s to
Po.
T
approaches i n f i n i t y s t r o n g l y w i t h r e s p e c t
X
The convergence i s uniform f o r
i n compact s u b s e t s
T h i s theorem i s t h e main r e s u l t of [ l ( h ) l . t h e formula of Langlands a s a s t a r t i n g p o i n t .
The proof u s e s
11
The trace formula
5, CONSEQUENCES
A
OF A PALEYWIENER THEOREM,
The asymptotic formula of Theorem 4.1 is only uniform for in compact sets.
\
However, the formula (4.2) entails i/f/ict. ,
integrating if P
. '
over the space
*
*
which is noncompact
G.
Therefore Theorem 4.1 apparently cannot be exploited.
Our rescue is provided by a multiplier theorem, which was proved in [l(g)] as a consequence of the PaleyWiener theorem for real groups. The multiplier theorem concerns C ( G OR) ,
~,r,) ,
the algebra of smooth, compactly supported functions on
GQR)
which are left and right finite under the maximal compact subgroup of G(R). Set
where
M (R)
/kand
is the Lie algebra of some maximal real split torus in
Then of
/kn
JhrK
and Let
is the Lie algebra of a maximal torus in
nM
(El.
is a Cartan subalgebra of
G(C),
,kE(
q Ã £ the Lie algebra
,
of W
is invariant under the Weyl group, W
(qÃ, $C) .
distributions on
4
$)
be the algebra of compactly supported
which are invariant under
y
t
.
The and
multiplier theorem states that for any
f_
C :
E) l' $
%,y
c
(G OR)
, %) ,
there is a unique function
(G fJR) , \ , )
with the following property.
If IIÃ
in is any
representation in
Il (G OR) )
,
then
James ARTHUR
where
{v%}
is the
Worbit in and
associated to the
A
infinitesimal character of transform of support of
y
y
is the FourierLaplace
.
The theorem also provides a bound for the
in terms of the support of y and of f . & %,Y We apply the theorem to C ~ ( ) , K) , the algebra of K AG '
~ () AG '
finite functions in C
)
.
For each group
is a natural surjective map kernel of on h .
h :
y e
4
1
P
3
Po
 + Ã ˆ
Suppose that
E ( Vis~actually supported
is the restriction to
.
Let
.bl
,
there
be the
fyl .
Any function
f e c ~ G @ ) , K)
G @)
of a finite sum
where each
f
is a
G) @'
K
finite function in C (G ( Q ))
m
.
The
restriction to
of the function
depends only on and
TT
f
.
We denote it by
f
pxl
.
TT
Suppose that
(P
P
3
Po
e
II (M (A)). Then the operator
p X r T (P
,A
f ) Y
will be
a scalar multiple of
A
f) . For if
there is a Weyl orbit
{vn}
in
associated to the Then
infinitesimal character of
ii_.
The trace formula
We shall now try to sketch how the multiplier theorem can be applied to the study of J~
x.
The key is the formula (4.2) ,
and in particular, the fact that the left hand side of (4.2) is a polynomial function of for points on in f
C :
T.
Now this formula is only valid
T
which are suitably regular in a sense that depends
.
If
N > 0
,
let G ( : C
1 (A)
, K)
be the space of functions
(G (A)
, K)
which are supported on
is the usual kind of function used to describe estimates on Gift). See [l(c),
Â
11
.I
Then it turns out that there is a N
constant C
such that for any
,
and any
f e c~(G@)
1
, K),
formula (4.2) holds whenever
(see [l(i), Proposition 2.21 .) ~f
f
belongsto
C;(G(A)~,K)
and
lc&($)'
is
supported on
James ARTHUR
then
f Y
will belong to
A' CN+Ny (G ()
, K) .
We substitute f
Y
into the right hand side of (4.2).
We obtain
This equals
where
for
H
and
IT e
II ( M (A) ~
H
1
1.
The function
T ifi^ (H)
depends
only on the projection of finitely many function of
H
IT.
on A .
It vanishes for all but
It can also be shown to be a smooth bounded
.
T whenever
The expression (5.1) is a polynomial in
F
H c
f^ .
~ e ty
be the Dirac measure on
.$
at the
The trace formula
point
H ,
and set
Then
N
Y
=
[IH[~ .
The expression (5.1) equals
This function is a polynomial in
T whenever
Its value at H = 0 is just the right hand side of (4.21, which equals J T (f) as long as dp (TI is greater than C (1+ N)
.
0
Suppose we could integrate (5.2) against an arbitrary Schwartz function of
H e
4. '
By the Plancherel theorem on
inner $ the resulting* inner product could be replaced by an were * product on i i . If the original Schwartz function
/ikG
taken from the usual PaleyWiener space on
.bl we would be ,
able to replace (4.2) by a formula in which all the integrals were over compact sets. Unfortunately this step cannot be taken
immediately, because (5.2) may not be a tempered function of For each
1 e
H
.
I1 ( M ( A ))
, let
James ARTHUR
be the decomposition of any nonzero point
X
vT
into real and imaginary parts.
Then
will cause (5.2) to be nontempered.
Nevertheless, it is still possible to treat (5.2) as if all the points
X
were zero.
This can be justified by an elementary We shall forgo
but rather complicated lemma on polynomials.
the details, and be content to state only the final result. Let i
S (ib*/ici.)w
be the space of Schwartz functions on W
*
/iaG
i r
*
<:
which are invariant under
.
If
B r S (i,b /iaG)
*
*
and
II ( M (A) l),
set
B (A) = B ( i Y + A) ,
A
r ifip/i^t ' G
*
*
It is a Schwartz function on
i<x/i^ G "
*
*
THEOREM 5.1:
(i) For every function B r pT(B) in
T
s (i.b*/i<)'
there
is a unique polynomial
such that
approaches zero as to
Po.
T
approaches infinity strongly with respect
(ii) Suppose that
B(0) = 1
.
Then
where
The trace f o m d a
See [l(i), Theorem 6.31.
0
If the function B
happens to be compactly supported, the
The first statesame will be true'of all the functions B . ment of Theorem 5.1 can be combined with Theorem 4.1 to give
THEOREM 5.2:
Suppose that
B c cW(i.b*/i~Q)
c
.
Then
P (B)
is the unique polynomial which differs from
by an expression which approaches zero as strongly with respect to
Po.
T
approaches infinity
See [1(i) Theorem 7 1 .1
.
5,
AN EXPLICIT FORMULA,
Theorems 5.1 and 5.2 provide a two step procedure for T if f is any function in C ~ (A) I) which G calculating Jx f ) is
K
finite.
One first calculates
as the polynomial which is asymptotic to
James ARTHUR
I I( pI PaPo~eII ~ (A) )
One then chooses any T J (f) by
P M ~ )
rlJ * * ~r(~x,n(P,A)~x,~(~,A,f))~T(~)d . i^p/imG
T
B
such that
B(0) = 1 ,
and calculates
The second step will follow immediately from the first. first step, however, is more difficult.
The
It gives rise to some
combinatorial problems which are best handled with the notion of a (G , M) family, introduced in [I(Â£ I. Suppose that M e L (No) .
A fG
, M) family is a set of smooth functions
indexed by the groups
Q
in
P(M),
which satisfy a certain Q and
Q'
compatibility condition. groups in P(M) and
Namely, if
are adjacent
A
lies in the hyperplane spanned by the
Q
common wall of the chambers of c (A) = c
Q
Q'
(A).
iffl then M' A basic result (Lemma 6.2 of [l(f 1 ) asserts that
Q'
(G ,M)
and
in
*
if
{c (A) is a }
Q
family, then
extends to a smooth function on
iNM.
*
A second result, which
is what is used to deal with the combinatorial problems we
The tvaee f o d a
mentioned, concerns products of
( G IM)
families.
Suppose that
{d (A) } is another (G ,M) family. Q associated to the (G ,M) family
Then the function (6.1)
is given by
([l(f), Lemma 6 . 3 1 ) .
For any
S e F(M),
s cM(A)
is the function
(6.2) associated to the
(Ms ,M)
family
and
c(A)
l
is a certain smooth function on
iffl
only on the projection of For any
(G , M)
A
onto
iaM
*
family
{ c (A)}
and any
s
.
M
which depends
,
L e L (M) ,
there
is associated a natural to lie in
i<xL
( G IL)
family. Let
A
be constrained
*
and choose
Ql e P(L).
The compatibility
condition implies that the function
is independent of
Q.
We denote it by
c
1
(A).
Then
James ARTHUR
is a
(G , L)family.
We write
for the corresponding function (6 2). value at cL A typical example of a (G ,M)
A = 0
We sometimes denote its
simply by

family is given by
Q c P(M), A
t
ia,
where
M compatibility condition requires that for adjacent Q
{ yQ
: Q c
P (M)}
is a family of points in
4 %

The and Q'
,
where
a
is the root in
A
Q
Q
which is orthogonal to the common and Q'
wall of the chambers of
.
If each
c
is actually
positive, the function ct4(A) admits a geometric interpretation. It is the Fourier transform of the characteristic function in
OLM
of the convex hull of
= cM(0)
{yQ : Q e P
(M)1 .
The number (G , M)
c
is just the volume of this convex hull.
families of this sort are needed to describe the distributions
Jn, and
IT
in the cases where explicit formulas exist.
(see
11(c), 571, [l(f), 5141 and also [ K b )I .) For another example, fix also a point
P
e
F(M)
and let M = M put
X
in
i m .
*
.
Fix
For any
Q e P(M),
The trace formula
Then
M P
Q
A = M
Q p
I
(A)'MQiP(A+R) ,
A
6
ioi
*
MI
i s a f u n c t i o n on
on
imM
*
with values i n t h e space of operators
A* ( P ) .
I t c a n b e shown t h a t
is a
(G , M )
family of
(vector valued) functions.
I n o r d e r t o d e a l w i t h ( 6 . 1 ) we must l o o k back a t t h e T ( P , i ) i n 54. The e x p r e s s i o n (4'.4) c a n be XI^ w r i t t e n a s t h e sum o v e r s e W (mP ,K P ) o f d e f i n i t i o n of
ii)
Given
P
a n d t W P , f t . w t of
),
set
Q = w q l pIwt ' t
fox
any r e p r e s e n t a t i v e
t
in
G(Q).
Then
i s a b i j e c t i o n between p a i r s which o c c u r i n t h e sum above and
groups
Q e P(M).
Notice t h a t
James ASTHUR
I t can a l s o be shown t h a t
equals
MQJp(A)
 IM Q l P ( s , W
e
(SP\}
(Y(T))
where
YQ(T)
is t h e projection onto
%
of t h e p o i n t
l ( T  T o ) Theref o r e ,
+
To.
t h e f u n c t i o n which must be s u b s t i t u t e d i n t o ( 6 . 1 ) , can b e o b t a i n e d by s e t t i n g
A' = A
i n t h e sum o v e r
s
W(ffl.
I&P)
of
Formula ( 6 . 3 ) s u g g e s t s a way t o h a n d l e (6.4)
.
W set e
and d e f i n e
A(Y(T))
c (A) = e
Q
The trace fommla
and
d (A) = t r (Q ~ P ( X ) l ~ Q j p ( s ~ r *')P x r 71 IP,X,~)) Q for any { d (A))
Q
are
P(M),. (G ,M )
It is not hard to show that families.
{cQ(A) }
and
The function (6.4) equals
an expression to which we can apply (6.3). of terms indexed by groups
S
e
The result is a sum
F(M).
The contribution to
(6.1) of each such term can be shown to be asymptotic to a polynomial in T
.
The sum of all these polynomials will be the pT ( B )
required polynomial
.
Once again , we will skip the
details and state only the final result. In the notation above, set
This is a product of two
(G
( G , M ) families, so it is itself a is any group in
L (Mp),
,M
)
family.
If L
ML (PIX) = lim 1 M~ (PfX,A)8 ( ) ' A" A+O QcF(L) Q~ I
T
is defined.
It is a polynomial in
(P)
T with values in the space
of operators on
.
James ARTHUR
If L
M
are any two groups in
L ( M ), let
U C
W1'
reg
be the set of elements in W (mM aM) for which , space of fixed vectors.
is the
THEOREM 6.1:
P
2
The polynomial
pT (B)
equals the sum over
P,
,
n ( M Ãˆ '), ~
L e L ( M ~ )and
s
e $(@Ã£p)re
of
the product of
with
See [ l ( j ) , Theorem 4.11.
n
The theorem provides an explicit formula for pT (B) From T this we can obtain a formula for J (f) and, in particular, for
.
x
It i s easy to show that
Then
Jx(Â£ can be obtained from the formula of the theorem by
simply suppressing T .
The trace formula
The formula for function B . the function
J (Â£ will still depend on the test
x
It would be better if we could remove it. f is still required to be
K
Moreover,
finite.
I. )
Our formula The
ought to apply to an arbitrary function in C(@ ;G)
dominated convergence theorem will permit these improvements provided that a certain multiple integral can be shown to converge absolutely. The proof of such absolute convergence turns out
to rest on the ability to normalize the intertwining operators between induced representations on the local groups G(Q). At
first this may seem like a tall order, but it is not necessary to have the precise normalizations proposed in [8(b), Appendix 111. We require only a general kind of normalization of the sort established in [ 6 ] for real groups. The analogue for padic In any case, we
groups should not be too difficult to prove.
assume the existence of such normalizations for the following theorem.
THEOREM 6.2.
Suppose that
f e
C ~ ( (A) G l)
.
Then
J (f)
equals
the sum over M e L ( M I , L
L (M),
e
n (M(A) l)
and
of the product of
with
J a m e s ARTHUR
This is Theorem 8.2 of C l ( j ) I. Implicit in the statement if the absolute convergence of the expression for J (f).
x
Let D (f) be the sum of the terms in the expression for J (i) for
x
x
which and s
In particular, the distribution D of J
XI
x
is invariant. As the "discrete part"
it will play a special role in the applications of the trace
formula.
BIBLIOGRAPHY
1. ARTHUR,
J. ,
( a ) The S e l b e r g t r a c e f o r m u l a f o r g r o u p s o f F  r a n k one, Ann. o f M a t h .  ( 1 9 7 4 ) , 236385. 100 ( b ) The c h a r a c t e r s o f d i s c r e t e s e r i e s a s o r b i t a l i n t e g r a l s , I n v . M a t h . 32 ( 1 9 7 6 ) , 2 0 5  2 6 1 . ( c ) A t r a c e formula f o r r e d u c t i v e groups I : terms a s s o c i a t e d 45 t o c l a s s e s i n G(Q), Duke M a t h . J .  ( 19781, 9 1 1  9 5 2 . ( d ) A t r a c e formula f o r r e d u c t i v e groups I 1 : a p p l i c a t i o n s o f a t r u n c a t i o n o p e r a t o r , Comp. M a t h . 40 ( 1 9 8 0 ) , 8 7  1 2 1 . ( e ) E i s e n s t e i n s e r i e s and t h e t r a c e f o r m u l a , P r o c . Sympos. P u r e M a t h . , v o l . 3 3 , P a r t I, Amer. M a t h . Soc., P r o v i d e n c e R.I. ( 1 9 7 9 ) , 253274. ( f ) The t r a c e f o r m u l a i n i n v a r i a n t f o r m , Ann. o f M a t h . 113 1 9 8 1 ) 3 114. ( g ) A PaleyWiener theorem f o r r e a l r e d u c t i v e groups, p r e p r i n t . ( h ) On t h e i n n e r p r o d u c t o f t r u n c a t e d E i s e n s t e i n s e r i e s , t o a p p e a r i n Duke Math. J.

(i) a f a m i l y o f d i s t r i b u t i o n s o b t a i n e d f r o m E i s e n s t e i n seOn r i e s I : A p p l i c a t i o n o f t h e PaleyWiener theorem, p r e p r i n t .
The trace formula
(j) On a f a m i l y o f d i s t r i b u t i o n s o b t a i n e d f r o m E i s e n s t e i n series
I1 : E x p l i c i t f o r m u l a s , p r e p r i n t . 2 . FLATH, D . , E o m p o s i t i o n o f r e p r e s e n t a t i o n s i n t o t e n s o r p r o d u c t s , P r o c . Sympos. P u r e M a t h . , v o l . 3 3 , P a r t I , Amer. Math. S o c . , P r o v i d e n c e , R. I. ( 1 9 7 9 ) , 179184.
3. FLICKER, Y . ,
The a d j o i n t l i f t i n g f r o m S L ( 2 ) t o P G L ( 3 ) , p r e p r i n t .
4 . GELBART, S . , a n d JACQUET, H . , Forms o f GL(2) f r o m t h e a n a l y t i c p o i n t o f viewg, P r o c . Sympos. P u r e Math., v o l . 3 3 , P a r t I , A m e r . Math. S o c . , P r o v i d e n c e , R . I . ( 1 9 7 9 ) , 213252.
5 . JACQUET, H., a n d LANGLANDS, R.P., A u t o m o r p h i c Forms o n G L ( 2 ) , Lect u r e N o t e s i n M a t h . , 114 ( 1 9 7 0 ) . 6 . KNAPP, A.W., and STEIN, E.M., I n t e r t w i n i n g o p e r a t o r s f o r s e m i s i m p l e g r o u p s 11, I n v . Math. 60 ( 1 9 8 0 ) , 984. 7. LABESSE, J . P . , a n d LANGLANDS, R.P., Can. J . Math. 31 ( 1 9 7 9 1 , 726785. Lindistinguishability f o r SL(2),
8 . LANGLANDS, R.P., (a) E i s e n s t e i n series, P r o c . Sympos. P u r e M a t h . , v o l . Math., S o c . , P r o v i d e n c e , R . I . ( 1 9 6 6 ) , 235252.
9, Amer.
( b ) On t h e F u n c t i o n a l E q u a t i o n s S a t i s f i e d b y E i s e n s t e i n S e r i e s , (1976). L e c t u r e Notes i n Math.,
(c) B a s e Change f o r G L ( 2 ) , A n n a l s o f Math. S t u d i e s , 1980. ( d ) L e s d e b u t s d ' u n e f o r m u l e des t r a c e s s t a b l e s , P u b l i c a t i o n d e
l t U n i v e r s i t 6 P a r i s V I I , volume 1 3 , 9. OSBORNE, M.S., a n d WARNER, G., ( a ) The S e l b e r g t r a c e f o r m u l a I : r  r a n k J . , v o l . 324 ( 1 9 8 1 ) , 1113. (1982). one l a t t i c e s , C r e l l e ' s
( b ) The S e l b e r g t r a c e f o r m u l a I1 : P a r t i t i o n , r e d u c t i o n , t r u n c a tion, preprint.

1 0 . SELBERG, A., ( a ) Harmonic a n a l y s i s a n d d i s c o n t i n u o u s g r o u p s i n w e a k l y s y m m e t r i c R e i m a n n i a n s p a c e s w i t h a p p l i c a t i o n s t o D i r i c h l e t series, J . I n d i a n Math. S o c . 20 ( 1 9 5 6 1 , 4787. ( b ) D i s c o n t i n u o u s g r o u p s a n d h a r m o n i c a n a l y s i s , P r o c . I n t . Cong. Math. 1 9 6 2 , 177189.
11. SHELSTAD, D . ,
t o appear i n t h e s e proceedings.
1 2 . WARNER, G . , S e l b e r g ' s t r a c e f o r m u l a f o r n o n  u n i f o r m l a t t i c e s : I R  r a n k o n e c a s e , A d v a n c e s i n Math. S t u d i e s 5, 1142.