The Selberg Trace Formula For Groups of F-Rank One
The Selberg Trace Formula For Groups of F-Rank One
The Selberg Trace Formula For Groups of F-Rank One
Introduction
An important tool for the study of automorphic forms is a nonabelian analogue of the Poisson summation formula, generally known as the Selberg trace formula. There have been a number of publications on the subject following Selberg's original paper [lo], the most recent being [2] and [7, 3 161. With t h e exception of Selberg's brief account [ I l l , however, most authors have restricted themselves to the groups SL(2) and GL(2). In this paper we develop the formula for a wider class of groups. We shall work in an ad& framework so our group G will be a reductive algebraic group defined over a number field F . We require t h a t t h e F-rank of the semisimple component of G be one. To simplify our introduction, let us assume t h a t G itself is semisimple. If A is the adele ring of F , let \ be t h e regular representation of G on L^G^/Gd. It is important to A t r y to decompose \ into irreducible representations. To begin with, \ splits into a sum of two representions \, and \ such t h a t kOis a direct sum of irreducible representations while hi decomposes continuously. The theory of Eisenstein series provides us with a fairly good understanding of the decomposition of \. However, virtually nothing is known about how to pick out t h e irreducible components of \. I t is a t this problem t h a t the Selberg trace formula is aimed. \ may be regarded as a representation of L1(G.^). Our first aim will be to prove t h a t the operator \(f ) is of trace class when f is a suitably regular function on GA. Besides imposing the usual conditions on j* we shall be forced to make an additional assumption. If GL is a product of groups of real rank one this assumption is harmless, b u t in general i t is unsatisfactory. After two preliminary sections, we s t a t e our assumptions in 3 3, where we also establish the desired properties of \(f ). Once we have proved t h a t the operator Lo(/) is of trace class, we can go about calculating its trace. To do this we must take the kernel, KO, \(f), of and integrate i t over the diagonal. KOcan be expressed as the difference
327
of the kernels K and Ki of h(f) and hi(f) respectively. To study Ki we shall need to quote a number of results from the theory of Eisenstein series. The basic references a r e [5], [8], and [9] where the results a r e proved for discrete subgroups of real Lie groups. This covers our situation because any automorphic form on G\IGF can be regarded as a finite sum of functions, each of which is an automorphic form on the quotient of Go by some arithmetic subgroup. In Sections 4 through 8 we analyze the functions K(x, x) and Ki(x, x), breaking each one up into a number of components. Although neither of these functions is integrable, we find in 5 6 t h a t the non-integrable components of each function cancel. All the remaining terms t u r n out to be integrable, although in 5 8 we need to appeal to the integrability of Ko(x, x) itself to verify this. We integrate each term as we go along, leaving the results to be collected in 5 9 in our final formula. The treatment of these last five sections is strongly motivated by [7, pp. 526-5461. Most of the methods used in this paper originate with Selberg (see [Ill), including the ideas behind t h e proof of Theorem 3.2 and the convergence of the integrals in 8 8. These were described to me by Robert Langlands whom I would like to thank for his encouragement. I am also grateful for the comments of Stephen Gelbart, who read through the original manuscript.
Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .,327 . . The Spectral Decomposition. . . . . . . . . . . . . . . . . . . . . . . . . . ,336 ... . The Operator /."(f) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .346 An Arrangement of the Terms in the Kernel.. . . . . . . . ,354 The Elliptic and Singular Terms . . . . . . . . . . . . . . . . . . . . . ,362 . The First Parabolic Term . . . . . . . . . . . . . . . . . . . . . . . . . . ,366 The Second Parabolic Term. . . . . . . . . . . . . . . . . . . . . . . . . . .,372 The Third Parabolic Term . . . . . . . . . . . . . . . . . . . . . . . . . . .,378 . Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ,384
1. Preliminaries
Let G be a connected reductive algebraic matrix group defined over a number field F. For any place v of F we shall write Gfor GFv,the group of ^-rational points of G. We shall denote the adeles of F by A, and we write G for the corresponding adelized group. If A and f stand for the s e t of infinite and finite places of F respectively, we can write
328
JAMES ARTHUR
GA. If H is any F-subgroup of G, we write Cr(HA) for the space of linear combinations of functions
f=ILf.
t h a t satisfy the following conditions: (i) If v is infinite, f,e C:(Hv); (ii) If v is finite, f,, is locally constant and has compact support; (iii) For almost all finite places v, f is the characteristic function of Gov. We shall also sometimes write C(HAfor the space of linear combinations of functions of the form
=f - f f
where f a differentiable function Has fr is a locally constant function is and on Hf. The radical of G is a torus which is defined over F. Let Z be its Fsplit component. Let X(G) be the group of rational characters on G, and let X(G)= be those characters in X(G) which are defined over F. Define
by
e ( % v H ~ (=)1 X(x) 1 '' ,
X(G)=, x E GA .
Let G be the kernel of HG. A We shall define a subgroup Z2 of 2-. Fix a basis xi, . -, xr of -^(GIF. The restriction of these characters defines an F-homomorphism $ from Z to GL(1)'. This in turn defines a homomorphism $- from the identity component of Zy,onto the identity component of GL (1, m)'. For any positive = 1 for every finite real number X we let [(k) be the idele such t h a t place v of F and ?(k)m= k for every infinite place w of F. The collection {((k): k > O} defines a subgroup, GL+ (1, m), of the identity component of under $-. GL (1, m). Let Z2 be the inverse image of GL+ (1, A The restriction of Htimaps Z2 bijectively onto 3. Therefore G is the direct product of Z2 and G\. The group Z2 is independent of the basis XI, Xr.
[a (
+
" ' 7
329
We would like to specify Haar measures on certain subgroups of GA. On any discrete group we will use the Haar measure that assigns to any point the measure 1. On any quotient of unimodular groups to which we have assigned Haar measures we will use the corresponding quotient measure. Let dx be the Tamagawa measure on GA. We recall the definition. The representation a of the Galois group g(F/F)on the vector space X(G) @F defines an Euler product
u s , a}
J-J u s , a ) .
The order of the pole of L(s, a) a t 1 equals the multiplicity of the identity representation in a, which is dim (X(G)p (x) 8 ) r . '= Let + be a nontrivial character on A which is trivial on F. At each place v of F, + defines a nontrivial character +, of Let d& be the measure dSi. Let a> be a left on F,,, self-dual with respect to h,and let dS = invariant form of highest degree on G, defined over F. For each v define the measure dx, on G, to be
c.
IJ
where 1 ft) It, is the measure defined by the form a> and d:,,. If Ar is the discriminant of F, dx is the Haar measure on Gs. which equals
This measure is independent of the choice of and u>. Any basis zi, - ., xr of X(G)F defines an isomorphism between 2: and (RE)". We take as measure on Za that which corresponds to the Euclidean measure on (RT)'". This measure is independent of the choice of - ., Our measures on Gs. and Za define a measure on G which we also denote s A by dx. It is well known that the number
zi, x,.
is finite. r(G) is the Tamagawa number of G. Let P be a parabolic subgroup of G defined over F. Let N be the unipotent radical of P. N is connected and defined over F. Fix a Levi component, M, of P. M is connected. It is known that M is defined over F and that P is the semi-direct product of M and N. In particular the maps
M x N-P, NxM-P,
330
JAMES ARTHUR
regarded as morphisms of algebraic sets, are isomorphisms defined over F. It follows that Pr = MrNr = NFMF Let A be the F-split component of the radical of M. A contains 2. If we replace (G, 2 ) by (M, A) we can define the vector space a, the map H.y, the groups M and A;, and the measures on MA,A2 and M l as above. A There is an isomorphism of affine varieties exp: n N
defined over F, from n, the Lie algebra of N, onto N. On nA choose the Haar measure d X that makes the measure of nA/nFequal to 1. Let d n be the Haar measure on NA which is the image of d X under the above map. We define left and right Haar measures on PAby
P A
R such that : ,
A A We shall write P for the group M m . P is unimodular and a Haar measure d p on Pi is defined by our choices of Haar measures on M i and NA. Pi, is a discrete subgroup of Pji, and the volume of P/ p ^ P is T(M), the is Tamagawa number of M. Finally, it is obvious that the group PA a semidirect product of A t and P . A Suppose that O P is a fixed minimal parabolic subgroup defined over F. It is known ([I, Theorem 4.131) that O P is unique up to conjugation under F. Now for any finite place v, G o is an open compact subgroup of GI t is known ([3, p. 101) that the double coset space
(II, GOV)\Gf/OPf
f
is finite. It follows that for almost all finite v, G, = Go'Pv A recent unpublished theorem of Bruhat and Tits states that for any finite place v there is an open compact subgroup K, of Gwhich, among other things, has the property that For a statement of this theorem see [6, Theorem 51. At any finite place v we shall define K, to be G o if G,
Go-OPPvFor
S E L B E R G TRACE FORMULA
331
the other finite places we take K, t o be any open compact subgroup of G, that satisfies (1.1). A t each infinite place v, we take Ky to be any maximal compact subgroup of Gv such t h a t the Lie algebras of Ky and A , are orthogonal under the Killing form. If we define K = K K v then
GA
K-OPp. .
Let d k be the normalized Haar measure on K. There is a positive constant cG such t h a t for all f e Cr(GA)
We shall assume from now on t h a t the F-rank of G / Z is 1, and we shall write P for the minimal parabolic subgroup OP. The dimension of A / Z is 1, and W I M Pis compact. We fix for once and for all an isomorphism
t-ht
teR
from R onto a subgroup T of A: such t h a t A: is the direct product of Z t and T and for any p e C;(A:)
Any element of x e GA has the decomposition for M,m e MA, t e R, n e NA, and z e Z. The number t is uniquely determined. We shall denote i t by H(x). There is a real number p such that
ap(p) =
2,oHlvl
pepA.
Let us agree upon some additional notation t h a t we shall later need. Suppose t h a t Y E GF and t h a t H i s a connected F-subgroup of G. We shall write H+(Y)for the centralizer of 7 in H. It is clear t h a t H ^ ( Y )is defined over F. It is also obvious t h a t H^(Y)* is the centralizer of Y in HF. We reserve the notation H(7) for the identity component of H+('"/). H(7) is
332
JAMES ARTHUR
also defined over F and i t is a normal subgroup of finite index in H ( 7 ) . I n particular, the group of rational points, H ( Y ) ~has finite index in t h e group , for H + ( Y ) ~We shall write n T a this latter index. . We will be interested primarily in t h e case where t h e group H is reductive, and Y is a semisimple element in HF. Then i t is known ([I, p. 701) t h a t H(Y) is reductive. An important tool in analyzing the elements of GF is t h e Bruhat decomposition. N(A), the normalizer of A in G, is defined over F. N(A)/M is a group of order 2, and the nontrivial coset of N(A)/M has a representative w which is rational over F. w normalizes M, and for a e A*,
Then according to the Bruhat decomposition, Gp is the disjoint union of Pp and N F . PF. We shall need to appeal to some results from reduction theory. We refer to [3] where the results are proved for the case F = Q . The results can be applied to our situation by restriction of the ground field F to Q . For any number c > 0, let
w e
S(c)
{x e GL H ( x ) 5 log c} .
LEMMA 1.1. There i s a constant c For a proof see [3, p. 161. From [3, Theorem 91 we also have LEMMA 1.2. For a n y c
= S(c).GF.
i s finite modulo P p .
We write Az(c) for S(c) n A. : PA the set If
a~
G(c) = K-A;(c)
is called a Siege1 d o m a i n for GA. By Lemma 1.2 we may choose a Siege1 domain (c such t h a t G A
=
S(C).GF
In order to prove the above results one employs a strongly F-rational representation p of G. For a general account of such representations see [I, 8 121. We shall also need to use a strongly F-rational representation, which we now describe.
333
M. Let B be a
em& ,
em> 0, en Z
such t h a t A lies in x(A) is the highest weight of an irreducible representation of G. This representation is strongly rational if and only if -V is the restriction to A of a character in X(M)i. We shall now take A to be any fixed character t h a t satisfies this property. Let p be the strongly F-rational representation of G whose highest weight is A and let p act on the vector space V defined over F. If /3 e X(A) is t h e simple F-root of (G, A), then the restriction of t h e character 2A. to A equals %./3 for a positive integer n,. We have a decomposition
I I
334
JAMES ARTHUR
The function 1 1 1 1 is called the height function associated to the basis {en, , ed}. I t is easy to verify t h a t if v e VAis primitive then p(x)v is also primitive for any x e G A . The map is continuous with respect to the topology defined by ) [ A that for any x e G and p e PA,
\\o(xp)eo
[I.
I t is also clear
I1
ebzrfpl p(x)ea\l \\
.
11 p(k)eo 11.
ncNA. For n e NA
> go -
Then
\\P(nw)eo I 1
=
so that
ebHfnwl
^>
T1 p(nw)eo I 1 I
=
\\ p(nw)eo I 1
\\ p(n)ed I 1
We can write
x
=
Then
k - n l w a l m w -h p z , w
where
n, = n -m l ~ p h l m - ~
eHixuw)
_ -
e H ( n i w ) -t
Now
e-t
= e-H(x'
> ;
> =
.
< e < c0 such that
Therefore e"
> 1. >0
there i s a n s with 0
i Y e GF and f
S ( E ) S(C)Y # 0 n
then Y lies in P F . Proof. The corollary follows from Lemmas 1.2 and 1.4.
Suppose that 8 is a Siege1 domain. A function f on Gn is said to be slowly increasing on 8 if there are constants C and N such that
1f
(a;)
1 5 C e c ~ ~, ( ' ) ~
xe8.
\ f (x)1 5 C.ye-vHIx) ,
is called the constant t e r m of h. Let 3 be the center of the universal enveloping algebra of the Lie algebra of Gm. I t is clear how to define the action of S on C m ( G J . A function h in Cm(GJGPZ2) called an automorphic form if is
J A M E S ARTHUR
(i) 11 is left K-finitel (ii) h is Sfinitel (iii) h is slowly increasing on any Siegel domain. The following we11-known principle is basic to the theory of automorphic forms.
1.6. S u p p o s e t h a t h i s a n automorplzic f o ~ m n Gi/GrZz. T h e n o LEMMA tile f u n c t i o n 11 - lzp i s y a p i d l y decreasing o n a n y Siegel d o m a i n .
The left regular representation, A, of GA on t h e Hilbert space L2(GA/GFZi) unitary. A fundamental problem in t h e theory of autois morphic forms is to decompose this representation into a direct integral of irreducible representation of GA. L e t L2({G})be t h e space of functions h in L2(Gi/GFZz) such t h a t for almost all x
It is clear t h a t L2({G})is a closed A-invariant subspace of L2(Gi/GFZ:). It is called t h e space of c u s p f o r m s . Analysis of this space is t h e deepest part
of t h e above problem. The theory of Eisenstein series provides an intertwining operator between t h e restriction of to the orthogonal complement of L2({G})in L2(Gi/GFZ;) and a direct integral of certain induced representations. I n this section we describe this intertwining operator. Let Z.* be t h e center of t h e universal enveloping algebra of t h e Lie algebra of M-. Suppose t h a t T is an irreducible representation of K and is a homomorphism of SAlf into C. L e t L be t h e vector space of that functions qi in Cm(GA/MFA2Ni)such t h a t f o r any x E GAl (i) qi(x; z) = ~ ( z ) # ( xZ E SJI1 )~ and (ii) t h e function p(k) = # ( k l x ) is contained in a subspace of L 2 ( K )on which t h e right regular representation of K is equivalent t o T. I t is known t h a t L is finite-dimensional. We define an inner product on L by
337
write V(P) for the collection of all simple (K, S.,J types. For any complex homomorphism x of ZAtflet V(P, x) be the collection of simple (K, S.,J types associated to x. I t is easy to see t h a t V(P, x) is not empty for only countably many We fix for once and for all an indexing of these homomorphisms by the natural numbers. Then
x.
=
V(P)
u:=lv(P,
x"J -
For any
2E
) for 4 e X7 and x, y E G.k/ZL. ~ ( z is just a n induced representation. Each space K(n) is invariant under ~ ( z ) . For f a measurable function of compact support on GA/Z:, define as usual
~ ( zf:) is a bounded operator on -3C. LEMMA 2.1. F o r z E C a n d y E GA, the adjoint operator ~ ( zy)* equals : T(- 2: y-I). NA. Our earlier normalization of Proof. Let Q be the group M F . A z e Haar measures defines a left Haar measure d l q on Q. Since GA/Qis compact, we can find a real valued function p E C:(GA) such t h a t xG Then i t is clear t h a t for 4,
+E X ,
=
e-2nH
(G,-,))
1 6 ~
'~(x)4(x)rndx
*
338
J A M E S ARTHUR
equals
0%
z(-z: f *)
where f * ( y ) = f F ) , y
For any f
E
GA/Z:.
is continuous on the compact space (GA/MFA?NA) x (G.k/MFA?NA). For fixed x, y and f i t is a Schwartz function of z in R.
339
equals ( ~ ( zf)$)(x) for almost all x. We will be able to conclude from the : last lemma that R(z: f : x, y) and P(z: f : x, y) are equal for almost all x and y. If R(z: f : x, y) happens to be continuous separately in x and y the two functions will be equal for all x and y, Fix L in V(P). For 6 E L i t is known (15, Lemma 23, Cor. 31) t h a t the series $(xa)e(z-Piff(z8i eGF,pF E($: " x, =
z6
converges uniformly for x in compact subsets of GA/Z: and z in compact = subsets of D,, {z: Re z < - p}. In fact i t can be shown ([5, Lemma 241) t h a t for any Siege1 domain 3 there is a locally bounded function c on (- X , - p) and an integer N such t h a t for x E 3 and z in Do, e G F , p p 1 $(xa)e(z-o)ff 1 2 (Re ~ ) e ( - ~ y + R e z - ~ ~ H ( z )
ca
E($: z: x) is called the E i s e ~ s t e series associated to $. Let us review its i~ basic properties. First of all, E($: z: x) is a left K-finite eigenfunction of Z on GA/GpZ2, so i t is an automorphic form. The constant term, Ep($: z : x), equals ,$(x)e(2-p)~(~1 ( ~ ) $ ) ( x ) e ( - ~ -,~ ~ ~ ( ' ) + (M where M(z) is a uniquely defined analytic function which maps Dp into the space of linear operators on L ([5, Theorem 51). M(z)*, the adjoint of M(z), equals M(Z) ([5, Lemma 481). For any x E GAand $ E L, E($: z: x) and M(z)$ can be continued to meromorphic functions on C which are regular on the imaginary axis. Any poles which lie to the left of the imaginary axis are simple, and must occur on the interval [ - p , 0). All poles must occur simultaneousIy for E($: z: x) and M(z)$. If D is the set of points in C where M(z) is holomorphic, E($: z: x) is continuous on D x Gk. In addition, for any Siege1 domain g, any compact subset u of D,and any $ in L, there a r e numbers C and N such t h a t (2.1)
Z E W ,X E G .
,
QEL.
E(M(z)$: - Z: X) ,
340
JAMES ARTHUR
(See [5! Theorem 71.) There is a useful formula for M(z). For m E Mal let mw = U Re z LEMMA 2.5. F i x L C V(P) a n d $ E L. Then j%r k < -pl the integral
E
~ - ~ ~ Z U .
K, m c Mtl a n d j % v
is absolutely convergent.
U is By the Bruhat decomposition, {e} {NF0w} a set of representatives for GF/PF. Therefore the above integral equals
kmhtnz ,
e -'-o'~\
\ A
$(knwmU)e '
Since t
must equal (M(z)$)(km). The absolute convergence of this integral is an immediate consequence of the absolute convergence of the Eisenstein series. 0 For any L E V(P) let :T(L) be the space of entire functions with values in L which are Fourier-Laplace tranf orms of functions in C:(iR) 8L. Then on any vertical stripl functions in X ( L ) decrease a t infinity faster than any polynomial. For any a E X ( L ) define
S E L B E R G TRACE FORMULA
where zr is any real number smaller than -p. By an estimate quoted earlier t h e integral is absolutely convergent, and i t is independent of zn. It is known ([5, Lemma 26, Cor. 1 and Lemma 401) t h a t for a and b in X ( L ) , Ea and Eb are in L2(G,/GFZa)and t h a t
If L2({P})is t h e orthogonal complement of L2({G})in L\G.JGrZa,), known t h a t L2({P}) t h e orthogonal direct sum is
@.elP
i t is
L2({P {Ll) . },
The integrand is clearly meromorphic in z. Now i t is known ([5, Lemma 1011) t h a t t h e norm of M(z) is bounded a t infinity in t h e strip {z: z, 5 Re z 5 O}. Therefore we can use the residue theorem t o shift our contour of integration to t h e imaginary axis. For z e [ - p , 0) let p(z) be t h e residue of -27rM(Q a t S = z. p(z) vanishes for all but a finite number of z. Our integral becomes
The expression defines a positive semi-definite inner product on X ( L ) . It follows from a simple approximation argument, which we leave t o t h e reader, t h a t t h e linear operators /i(z), z e [ - p, 0) , a r e all positive semi-definite. L e t V be t h e vector space of functions from [-p, 0) to L . Define a positive semi-definite inner product on V t o be
JAMES ARTHUR
Factoring out by the space of null vectors, we obtain a finite-dimensional Hilbert space which we denote by Q ( { P } ,{ L } ) . Let ?({P} } ) the space of square integrable functions a^ from i R { L be to L such that
ai(- 2 )
M(z)al(z) ,
zeiR.
{ be {L ) Let L2({P},L } ) the orthogonal direct sum of G({P}, } and L ( { P } ,{ L } ) . Any function a in X ( L ) obviously defines a vector a,, in L ~ { P {L}). }, We define a function a1in Q{P},} ) { L by al(z)= -(a(z)
2
+ M(- z)a(- z ) ) ,
zeiR.
(a,,, 1 ) 0
E a
Let us denote the restrictions of E to G({P}, } ) { L and ;({P}L } by E,, { ) and El respectively and we will write L I ( { P } , and L ~ ( { P { L } 'for the {L}) }, ) { we corresponding ranges in L m , {L}).For i = 0 or 1 and a; e L:({P}, L } ) need a formula for the function
343
This latter function is square-integrable over any Siege1 domain so i t follows from Lemma 1.6 t h a t En($:z: x) is square-integrable on GA/GpZ2. Another poperty of E0($:z: x) is t h a t i t is orthogonal t o LVG}). To see this, choose any automorphic form h in L2({G}). By Lemma 1.6 the function E(C: 6: x)h(x) is integrable for C e C. If C
< -,Q,
By analytic continuation,
Our assertion then follows from the well-known fact t h a t there is an orthonormal basis of L2({G}) consisting of automorphic forms. LEMMA 2.6. Suppose that we a r e given a p a i r (an,a,) such that (9 a G({PI, {A}), n f {L}). (ii) a, i s a function o compact support i n Then for almost a l l x the function
m,
Proof. Let f(x) be the function defined by (2.3). I t s constant term, fp(x), equals
This formula, together with a strengthened version of Lemma 1.6 ([8, Lemma 3.4]), insures t h a t f is square-integrable. It is clear t h a t f is orthogonal to L2({G}). Choose a sequence {b'} of functions in X ( L ) such t h a t as n approaches , (b?, by) approaches (a0,a J in L2({p},{L}). The constant term of EP equals
JAMES ARTHUR
which equals
1 -Ez 27T
r-p,o
(/Wbn(z))(x)e
Ret=o
-'-O
"'
27T
+ (M(z)bn(z))(x)e
Rez=o
-^-O
}d 1 z
$E~~! (p(z)b;(z))(x)e~+-'-( 7~
P,O
(b;(z))(x)e d
By choosing t h e sequence {bn} suitably we can force this last expression t o approach fp(x) for all x, as n approaches infinity. On t h e other hand, since Ebn converges t o EoaO E,a, in t h e mean, t h e constant terms of Enao Ela, and f a r e equal almost everywhere. Now t h e function
lies in L m G p Z ^ } and is orthogonal t o L\{G}). Since its constant t e r m equals 0 almost everywhere, t h e function itself must vanish almost everywhere. Fix i = 0, 1. Define
Q({P}) Qiw Ll({P}, {L}) = G({P}) @LeT.:P!L:({P}, {L}) = By taking t h e direct sums over V(P) of t h e maps
LyGJGpZ:) ,
V(P)
we obtain a map from q { ~ } into LyGJGFZ,;) which we denote again by ) Ei. The image of this map is L:({P}). Consider t h e adjoint map
Then t h e map E^Ef is t h e orthogonal projection of L2(G,4/GFZ:) onto L;({P}). The space L\{P}), which is t h e orthogonal complement of L\[G}) in L\GJGi,Za), equals For any L, and z e [-,o, O), define an operator r(z) on L t o equal zero
345
on the kernel of p(z) and to equal p(z)-' on the orthogonal complement in L of the kernel of ~ ( z ) .
that
L e V ( P ) and
m.
T h e n for
any
for almost all z in iR. Proof. These formulas follow easily from the last lemma. Let us prove only (ii). Let ai be any continuous function of compact support in ~ Q P {L}). }, Form the inner product of E T h and a,. On the one hand this equals
which is t h e same as
by Lemma 2.6. Since a , is arbitrary, we g e t t h e required formula. Our final task f o r this section is to verify the intertwining property of the operators E and Er LEMMA 2.8. Suppose that f is a left a n d right K-finite f u n c t i o n in C:(G JZ)F i x L e V ( P ) a n d $ e L. T h e n
a s meromorphic functions of z. Proof. For Re z < - p t h e formula is a direct consequence of t h e definitions of n ( z :f ) and E(p:z: x ) . The lemma follows in general by analytic continuation.
COROLLARY F o r f a s above, 2.9.
a s meromorphic functions in z.
346
Proof. Fix L and $. we observe t h a t
JAMES ARTHUR
+ + 0,
i>e L , E(+:
2:
subspaces o f
We shall sometimes write L:(GJGpZ^,) for L~({P}). We shall denote the direct sum
by L w G F Z a ) . (This notation is different from [5] and [9] where t h e symbol L2, is used for the space of cusp forms.) For any x in GA/Z+we denote the LKGJGFZ2), and LKG.JGpZ^) by restrictions of \(x) to L2({G}),LI({P}), \,({G}: x), \{{P}: x), \(x), and \(x) respectively. For f e C:(G.4/Z+) we can f Lo(/) is the define the operators \({G}: f ) , \({P}: ) , U f ) , and \(f). operator we shall be most interested in. I t is t h e sum of \({G}: f ) and
\({PI: f ): I t should be noted t h a t the results of this section a r e but special cases of 18, 3 71, where the spectral decomposition is carried out for groups of arbitrary rank.
3. The operator \(f)
Suppose t h a t f is a complex-valued function on GA/Z; which is t h e convolution over GA/Z; of two left and right K-finite functions f and f " I t seems likely in C3GA/Z:). \{f) is a bounded operator on LYGJGpZ:). () t h a t \ f is of trace class. Our objective will be to prove this fact under an additional assumption on f and then to find a formula for t h e trace. In calculating the trace of Lo(/) we shall integrate its kernel over t h e diagonal. This necessitates finding the integral kernels of L(f) and hi(/). While we are a t it, we may as well find a formula f o r t h e kernel of \({P}: f). This will perhaps have to be studied t o prove t h a t \(f) is of trace class
347
without using t h e additional condition (Assumption 3.5) t h a t we shall impose. For any function h in C3GA/GpZ^)
CeGF l ) f(w-
is finite for x and y lying in fixed compact subsets of GJZy,. Therefore \(f) is a n integral operator with kernel K. I t is clear t h a t K is a smooth function. In addition K is slowly increasing. In fact, given any Siege1 domain and a K-finite function g in Cr(GJZ^), it is known ([5, Lemma 91) t h a t there a r e constants C and M such t h a t (3.1)
1 EyeGF ) 1 ^ Ce-.lrH(zl g ( W ,
x, y e
In studying the operators \i({P}: ) and \(f) we must examine certain f X-invariant subspaces of L'({P}) and L;({P}). For any positive integer N define LXN)
=
@zeUP,,nl~ ( { P }{L}), I ,
i=0,1.
For T > 0 let C(N, T) be the subspace of elements alin U(N) such t h a t the projection of a,onto any of t h e direct summands {H({P}, {L}):L uV, _, ( :_P
L)}
is a function supported on the interval [-iT, zT\. Let Li(N) and L'[(N, T ) be t h e images of &N) and L[(N, T ) under t h e maps Eo and El. Li(N) and LXN, T) a r e binvariant subspaces of LyGJGpZ;). Let \(N: f ) and \(N, T: f ) be the compositions of \(f) with t h e projections of L2(GA/GFZa) onto these subspaces. In order to g e t a formula for t h e integral kernels of these operators we shall fix for once and for all an orthonormal basis
of X. We may assume t h a t each basis vector lies in some space L e V(P). For each positive integer n let I., be t h e set of indices a for which there is an L e V(P, yn) such t h a t pa lies in L . Then For a , 8 e I define
JAMES ARTHUR
and
Since f is left and right K-finite, ~ , ( z f: ) vanishes for all but finitely many a and ,L? in In. It follows t h a t the above sums are finite. K o ( n :z: f : x, y ) vanishes for all but finitely many z in [-p, 0). LEMMA 3.1. \(N: f ) a n d \(N, T: f ) a r e i n t e g r a l operators whose kernels a r e and
Proof. The result follows from a routine use of Lemmas 2.6, 2.7, and 2.8. We remark that the kernels defined by this lemma are both continuous functions on (GA/GpZ^) x (GJGpZ-n) . THEOREM G i v e n a n y Siege1 d o m a i n w e c a n choose constants C 3.2. a n d M such t h a t for a l l x a n d y in the expressions and
a r e both bounded by C e A r H ^ ) e - ' f H ( y l . Proof. Recall that f = f l * f " where f' and f" are both K-finite. For any n we fix a finite-dimensional subspace !^Cf(n) of X ( n ) which contains the ranges, and the orthogonal complements of the kernels, of the restric) : tions of both ~ r ( z : f ' and ~ ( zf " ) to X ( n ) . For x e GA/ZL and z e [-p, O ) , define a vector E { ( z : x ) in X f ( n )by
349 6 e X f ( n ).
( 6 , E { ( z : x ) ) = E ( # : z: a;) ,
Then
1 K,(n: z: f
:x, y) \ =
= ^- \ ( r ( z ) E { ( z : ~ ( zf :)*E{(z: x ) ) y),
2r
\ \
1 -1 2Tc
( r ( z :f " ) r ( z ) E { ( z :y),
T(Z:
f ' ) * E { ( z :x)}
f')*
^-Zf") n ( z : ( f ' r y , =
by Corollary 2.2. Theref ore 1 K,(n: z: f : x , y ) \ equals
(3.4)
-1 21 71
( r ( z ) ~ c ( ( f:" ) x ) * E l ( z : ) , ~ ( zf ':) x E { ( z : x ) ) 1. z y
P,
X f (n),
1 -(Y(z)T~(z:
VVL
and
( f ")*)*E{(z:y ) ,
1 -(r(z)z(z:
v2^
Define
lf
= f'*(f')*,
2f
( f " ) * *f"
Then the above product equals Ko(n: 2/: y, y)1/2Ko(n: lf a;, x)'I2 . z: z:
JAMES ARTHUR
Schwarz' inequality t o t h e sum over n and z in (3.2), we observe only necessary t o establish t h e bound for (3.2) in t h e case t h a t y = x. we have
1 K,(n: z: \fx, x) 1
By Lemma 3.1
Ko(n: z: \fx, x)
is t h e value on the diagonal of the kernel of \(N, '/). But \(N, '/) is t h e restriction of the positive, semi-definite operator \('/) to t h e invariant subspace Ll(N). Therefore its kernel, which is continuous, can be bounded on t h e diagonal by K(x, x). However, according to (3.1) this last function can in t u r n be bounded on a Siege1 domain by a function Ce^'-". But this function is independent of N. Consequently i t majorizes for all x ~ , ( n z: x, x) : We deal with (3.3) in exactly t h e same manner. First of all we show t h a t for z imaginary,
e [-(,,,
C2C z
is bounded by Consequently, i t is enough to establish t h e theorem when f is replaced by Yand y = x. We then resort to Lemma 3.1, verifying t h e required result as above. COROLLARY The functions 3.3.
and
a r e the integral kernels of \,({P}: f ) a n d \(f ) respectively. Proof. This follows from Lemma 3.1 and Theorem 3.2. I t follows from the corollary t h a t t h e kernel of Lo(/")equals
Now we t u r n to t h e task of proving t h a t \(f) is of trace class. L e t G be the set equivalence classes of irreducible unitary representations of
351
MJA;. Let \J,i be the regular representation of M J A ; on L w M p A f ) . I t is a well known that there is a \,.-decomposition
where for each a in &.=, the representation of M a obtained by restricting \v to V ( a )belongs to a nonnegative integral multiple of the class a . Let X ( a ) be the space of functions in X such that for any x e Gn the function
$dm)
lies in V ( a ) . I t is clear that
$(xm) , @oee,r X ( 0 )
m e Mi/MFA2 ,
We assume from now on that the basis {$a}aei, chosen earlier, is compatible with this decomposition. The element w, representing the nontrivial Weyl group element, defines a coset in the group of automorphisms of MA modulo the group of inner automorphisms. In this way w defines an involution on Sy.We say that a class a e & is unramified if a w + a , and ramified if a" = 0 .
is LEMMA 3.4. Suppose that 6 unramified. For a n y L e V ( P ) let L ( a ) = L fl X ( a ) . Then for a n y z e C the space L ( a ) L ( a w ) s i n v a r i a n t i under M(z). Furthermore the restriction of M(z) to this space i s regular for z e [-p, 0).
Proof. It is clear from the formula in Lemma 2.5 that M(z) maps L ( a ) into L(aw). Of course the formula is only true for Re z < -p, but our assertion follows by analytic continuation. Similarly M(z) maps L ( a w )into L ( a ) , so that L ( a ) + L(@") an invariant subspace of M(z). is For z e [ - p , 0 ) let ^{z) be the residue of -2rM(C) a t = z. Let I, be is an orthonormal basis for the subset of indices in I such that {$^sic L ( a ) L(aw). The trace of the restriction of p(z) to L ( a ) + L(a") equals
<
E J e I c ( ~ ( z ) $3), . $j
We have assumed that each p j is either in L ( a ) or in LC@'),so since L ( a ) is orthogonal to L(a"), this expression equals 0. On the other hand, we saw in 2 that p(z) was positive semi-definite. Therefore p(z) = 0, so M is regular a t z. If a e Sw and g is any function in C^GJZ^), we shall denote the restricz: tion of r ( z : g ) to X ( a ) by ~ ( 0 :g). At this point it is necessary to place an additional restriction on our function f . We summarize all the requirements in the following:
352
JAMES ARTHUR
ASSUMPTION f i s the convolution of two f u n c t i o n s f a n d f" in 3.5. C Z G J Z ; ) . f a n d f " are left a n d right K-finite a n d in a d d i t i o n
*: f ) = x(a: z: f " ) = 0 , z:
zec,
iSx.
We remark that this last condition is always true if the Lie group MJA^ is compact. THEOREM Suppose that f satisfies A s s u m p t i o n 3.5. T h e n \ ( f ) i s 3.6. of trace class. Proof. L o ( / ) is the sum of \{{G}: f ) and \{{P}: f ). \({G}: f ) equals \({G}: f ')ho({G}: ") f
I t is known ([5, p. 141) that the operators \({G}: f ') and \({G}: f ") are of Hilbert-Schmidt class. Therefore \)({G}:f ) is of trace class. : Since f' and f" are left and right K-finite, the operators ~ ( az: f ) and ~ ( az:: f " ) are of finite rank for any z e C and a e &. Therefore, by Assumption 3.5 and Lemma 3.4, both & ( { P } f ') and \)({P}: f ") are of finite : rank. I t follows that \)({P}: f ) is of trace class. Now that we have proved that \(f ) is of trace class, we would like to be able to say that KO(%,) is integrable over the diagonal, and that its y integral yields the trace of h o ( f ) . However, we have not shown t h a t the y kernel KO(%,) is continuous, so we must proceed cautiously. y LEMMA 3.7. The f u n c t i o n KO(%,) i s continuous in each variable separately. Proof. Following the notation of the proof of Theorem 3.2, we know that
1 K l ( n : z: f : x, y ) 1
is bounded by
is no greater than
For y lying in any Siege1 domain this last expression can be bounded by
by Theorem 3.2. I t follows t h a t for any fixed x t h e integral defining Kl(x, y) converges uniformly for y in compact subsets of GJGpZt. Therefore Kl(x, y) is continuous in y. Since K(x, y) is continuous, KO(%, = K(x, Y) - K1(x, Y) Y) is also continuous in y. Similarly, Ko(x,y) is continuous in x. Our operator \(f) is the product of the two Hilbert-Schmidt operators \(f)and V f " ) . Recall that for any Siege1 domain 8 we may choose constants C and M such t h a t for x and v in 8,
GF
f '(mv-l)
1 5 C~-L'"'('~.
If follows that for any x there is a unique function hk(v) in L K G J G F Z ~ ) , such t h a t for any <^> e LXG.JGFZ2),
Combining the dominated convergence theorem with the above inequality we see that for any <^> e LKGJGFZ2) the function
hxv)
Then H1(x, v) is a Hilbert-Schmidt kernel for \(f). Suppose that H"(v, y) is a fixed Hilbert-Schmidt kernel for our second operator \(f"). We may assume t h a t for every y, H1'(v, y) is squareintegrable in v. Then
is well defined for each x and y. I t is a Hilbert-Schmidt kernel for Lo(/"), and for any fixed y it is continuous in x.
LEMMA 3.8. The kernel H i s integrable over the diagonal a n d its integral equals the trace o \r(f). f
Proof. The integrability follows from Schwarz' inequality. suppose t h a t {*a} is an orthonormal basis for LxGJGFZ;). Define h'ab and
=
Now
(Mf ')A,
JAMES ARTHUR
Then
equals
where the convergence of the infinite sums is in the mean. We may interpret this integral as as the inner product on L y G A / G F Z Lx GA/GFZ:). Since the inner product is continuous on any Hilbert space we may interchange the integral and summation signs. The result is
Em,,; ,
^m ^
THEOREM 3.9. The kernel KO(%,) is integrable over the diagonal a n d y its integral equals the trace of L o ( f ) .
y Proof. The functions H ( x , y ) and KO(%, ) a r e both Hilbert-Schmidt kernels for h 0 ( f )so they must be equal almost everywhere on ( G J G Z ; ) x (GJGFZL). For any positive integer n let S n be the set of points y such that the measure of the set
T y = { x : KO(%,) y
H ( x , y)}
s = U"
S%
is also zero. y For any y not in S , H ( x , y ) equals KO(%,) for almost all x. But these two functions a r e continuous in x, so they must be equal for all x. In particular, the set of points y such that
m y , Y ) # KdY, Y )
has measure 0. Our theorem now follows from Lemma 3.8.
4. An arrangement of the terms in the kernel
From now on we require that f satisfy Assumption 3.5. We have just shown that
KO(%, ) = K ( x , x ) x
is integrable and that
K&, w)
Before we calculate this integral, we must group the terms in the integrand in a suitable manner. Recall that K(x, x) equals
(W) .
An element in Gp is said to be elliptic if it is not GF-conjugate to any element in Pp. Any such element is semisimple. Let G, be the collection of elliptic elements in Gr. Before classifying the remaining elements of GFwe shall first prove a few simple lemmas. Suppose f-t E Mp. Then p is semisimple. Recall that G+(p),P T ( / 4 , and W ( p )are the centralizers of /-<in G, P, M, and N respectively. MT(p),
4.1. For any LEMMA
[/ E
MF
P+(p)= M+(u)N+(/-<) .
is in P+(p). Then
m n = ppp-I
. .
m and pn/' = n
For any such /!, G(p) is reductive, and P ( p ) is a minimal parabolic subgroup defined over F. From the lemma,
P ( ^ ) = ML'w(!-i)
is a Levi decomposition for P(p). In the discussion of 1 we may replace (G, P, M, N ) by (G(p),P(p), M(p), N(p)) and make all the corresponding definitions. We use them without further comment.
356
JAMES ARTHUR
our ordering on X(A) (g) Q. Fix j = 1, 2. Let n(p)j be the set of all X in n(p),the Lie algebra of N(p),such that
n(p)-'is a subspace of n(p),defined over F. Let us write n-' for n(e)-'. Then Ad ( p l )is a semisimple linear operator on n which is defined over F. Let 3 . n ( ^ ) be a complementary subspace of n(p)-'in n-' which is invariant under j . Ad (p-I). n(p)-'is also defined over F. Notice that the linear operator
Ad ( p ) - id is invertible on n(,V I t is clear that
N ( P )= exp ~
are F-closed subsets of N . Define
357
This corollary is just a restatement of the lemma. While we are a t it, we may as well state an ad& lemma. version of the last
Proof. Fix j = 1, 2. Our Haar measures on n{ and n(p)i define a :Haar . measure d p on n ( , ~ ) iI. t is an easy consequence of the product formula for F that
{ <)' ( "7. (exp (Ad (p")Xj
)
(-
for any compactly supported measurable function oj on N(p){. To prove our lemma we just repeat the argument of Lemma 4.2, replacing each sum over a set of F-rational points with an integral over the corresponding adele space. LEMMA 4.5. Any element in Pp is Pi-conjugate to an element pv with p e My and v e N(p)p.
j Proof. Any element in PF can be written as p for p e M,, and 11 e NF. By Corollary 4.3 there are elements 8 e N p and v e N ( P )such that ~
7
In other words
p-'S^tv8-1
prj = Qpv8-I
Let Mr be the set of elements p in MF such that N(p) is trivial. Let Ms be the complement of Mr in Mp. Recall that M is a subgroup of index 2 in N(A), the normalizer of A. Mr and M, are both stable under conjugation by elements in N(A)F. LEMMA 4.6. Suppose that
&pl"l
"p2"'
for two elements pl and p2i n MT. Then 8' i s in the same N(A)F-coset as Proof. Let e = Val.Then e e ~ =, e. Either e is in Pi, or it is in Np.w-Pp. In the first case
&=m,
and
w M F ,v e N F l
JAMES ARTHUR
pv
prlpvpi = ^ w W ^ P l
=
I t follows that v e N(pJF, so that v = e. Therefore 8, On the other hand, suppose that Then vw7T
=
8#.
p;lvw~pl = p;'vp,w
~-lp','~w^:U
By the Bruhat decomposition, and Since N(ps)p {e},v must equal el so that
6,
=
a2wr .
TT
LEMMA 4.7. Suppose that 8, a n d 8, are elements i n GF a n d that 8,p1v18y1 O,^,V,; = for pl, pa e M,, and v, a n d v, nontrivial elements i n N(fi)? a n d N(pJF respectively. Then 8, a n d 8, are i n the same PF-conjugacy class. Proof. Let e If e is not in PF,
z = VW~T ,
=
O;'O1.
Then
V^?&/W
= &.
veNF,T T E P ~
Then vwx
=
This implies that v, = @uy-'. By Corollary 4.3, v, equals the identity element. This is a contradiction, since we assumed that v, was nontrivial. Therefore S^S,e PF. We shall write {G,} for a fixed set of representatives of Gp-conjugacy classes in G,. Let {Mr}and {ME} denote fixed sets of representatives of Mr-conjugacy classes in Mr and Ms respectively. Finally, fix a set {{M,}}of representatives in {ME} those GF-conjugacy classes in GF which intersect of
S E L B E R G T R A C E FORMULA
359
Ms. It is clear t h a t the contribution to K(w, x ) from elements which are GF-conjugate to an element in M. is
I~"(f:
=
Z;<E
f (x'~'-'x-l)
In this formula we have had to include the integer n,u,g which, as we recall, is the index of G(& in G + ( P ) ~The contribution from the elliptic elements is .
I e ( f : 3 ) = Ere { G e l ( n r , ~ ) - ' E a G ~ / G ( (r x ~ ^ x - l ). f
Lemmas 4.6 and 4.7 account for the contribution from the remaing elements. K ( x , x ) becomes the sum of
I.(f: x)
+ I - ^ f : x)
together with
(4.1)
and
(4.2)
u#e
Suppose that co is the positive number defined by Lemma 1.3. Fix a number e between 0 and c0. Let y, be the characteristic function of the set S(e). Since M y is stable under conjugation by w,the term (4.1) equals
and
and
360
JAMES ARTHUR
We have used the equality of the integers n,,,v and npp , which follows from Lemma 4.1 and the fact that N L ( p ) is connected. Next we shall break up the function K l ( x , x). Suppose that 0 belongs to some L e V ( P ) . Recall that Ep(p: x ) , the constant term of E(p: z: x ) , z: equals o(x.ei~-,o)~~~) -t ( M ( z ) p ) ( ~ ) e ( - ^ - p . ~ ( ' ) ) For any number e between 0 and q we define E,(o: z: x) to be ,
where, as before, yt is the characteristic function of the set S(e). For any x the sum is finite by Lemma. 1.2. We set E m : z: x ) equal to E(6: z'. x ) Ei((>: x). z: For convenience, set H p ( n : z: f : x ) equal to
n e NA
over the compact set NJNr.. Therefore by Torelli's theorem and Theorem 3.2 we can associate constants C and N to any Siegel domain 6 such that the inequality
4.8. G i v e n a n y Siegel d o m a i n G there are constants C a n d N LEMMA such that for all x e 6 the expression
is bounded by Ce--vH!x'.
By Lemma 1.2 the number of terms in the sum over GF/PF is no greater than some integer which is independent of x e 2. The lemma then follows from (4.3) and Corollary 1.5. Let us define K"( f : x: e) to be Kl(x, x) - K \ f : x: e). Then we may write K(x, x) terms:
-
and
We shall refer to these terms respectively as the elliptic, singular, and first, second, and third parabolic terms. We would like to evaluate the integrals over GJGFZ2 of each of these five terms. However, integrals arise in the third parabolic term whose convergence is not a t all obvious. One way to surmount this difficulty is to prove that each of the first four terms is integrable over GA/GFZ2. This would verify the integrability of the fifth term. I t will be sufficient to prove a weaker result. Let us say that a function h is weakly integrable over GA/GFZ2if (i) it is locally integrable, and (ii) for some c > 0 the integral
is finite. When we come to the first parabolic term it will be easier to prove only that it is weakly integrable. Of course this will weaken our conclusion on the integrability of the third parabolic term, but that will not matter.
JAMES ARTHUR
Our first concern will be to prove t h a t the elliptic term is integrable over GA/GFZ2. The integral
is also a relatively compact subset of Pi. Suppose that S 2 GA is the support off. S is compact modulo 2;. Let C be the closure in G of the set
w;'K-S-kw0.
C is compact modulo Z t .
LEMMA 5.1. Suppose that C is a compact subset of G modulo 22. Then there is a number e > 0 such that if 7 e GF and hi7h2 lies in C for some t < log e, then Y is in Pp.
Proof. Let p be the strongly F-rational representation of G on the vector space V defined in 5 1. We use the basis { g o , ., e A and t h e : . height function on V introduced in 5 1. Now p is trivial 2 I t follows that
is finite. We set this supremum equal to e 2 ' , where b is the positive rational number defined in 5 1. If Y is not in PF,
S E L B E R G T R A C E FORMULA
I I ,o(ht7h;l)eo [I = ecwr ]I p ^ v ) e , \\
If ht7hr1lies in C,
e-2bt
2 e- 2 b H ( h O 1 [ e,
11
ec2"
<
- E-2b -
It follows from this lemma that the function on G J G Z defined by (5.1) has compact support. Therefore the elliptic term is integrable over GA/GFZ+.Its integral equals
Now for y e { G J , Z is the split component of the radical of G(7). We have agreed to use the Tamagawa measure on G(7)^. However, we cannot immediately insert the Tamagawa number of G ( 7 ) in the above formula because our measure on Z t does not define the appropriate quotient measure. We must correct by a factor which we define to be the index in X(G(7))F of the group obtained by restricting the characters in X(G)v to G(7). If we write ?(Y, G ) for the number
(n^)-l(r^)-l^))
the integral of the elliptic term becomes
Before discussing the singular term we shall prove two more lemmas.
LEMMA 5.2. Let C be a subset of G A which i s compact modulo Z t . T h e n there i s o n l y a finite n u m b e r of elements p in { M y }U { M s }such t h a t there is a n x in G a n d a n n in NAfor which x p m l lies in C.
Proof. Let Cl = {kck-l: c e C , ke K } . Since P A is closed in GAPCl n P A is compact modulo Zt. We can choose a subset Cv of M A which is compact modulo Z: such that C1 n PA != C,rNA
.
h K , P~PA?
ppnp-I lies in C&. Let o> be a relatively compact set of representatives in P A for the compact double coset space A+\PA/PF. If
364
JAMES ARTHUR
then the element v7c. p n n ^ v l lies in the set a-'.CjrN*.a = CvN* . Choose a subset C.^ of MA, compact modulo Zt, such that w 1 o C M N A 0 ~ is contained in C^fNA. Then np.nn-' lies in C'vN,.. In particular p is MFconjugate to an element in C.{v. However, the projection of MF onto MJZt is a discrete subgroup of MJZt so that MF C',, is finite. Certainly only finitely many MF-conjugacy classes in MF meet C.{v. The lemma is proved. 0
LEMMA 5.3. F i x p e {M,.}U {Ms}. Suppose that C is a compact subset of P . Then there i s a compact subset Ci o PA/P(p)A such that if p e P^P{^)\ A f and
(P *pN(p)A'P-')
n ^0
then p lies i n Cl. Proof. Let ft) be a relatively compact fundamental set in Pi for PI/Pp. Denote the closure in P of w l C w by C'. Let 7 be the collection of cosets A 8 in PF/P(p)Fsuch that ( 8 0 p N ( p ) ~ -n C' V
+ 0.
The main point of the lemma is to show that 7 is finite. Assuming this fact for the moment, we let C D e the closure in PA/P(P)~ the set of
\Js^w.
Then if
(P-PMPL-P-~) + 0 nC
for some p e PA/P(,u)~,p must lie in CL If C, is the projection of C b n t o P p ( p ) A , then Ci is the required set. w It remains to show that 7 is finite. For j = 1, 2, define N(p)j, N(p)j, and N3 as in the proof of Lemma 4.2. If {MLis a set of representatives of MF/M(p)Fin My, it is clear that {ML* M & - W % is a set of representatives of P ~ / P ( P in~PFe ) Now there is a compact subset Calfof MS,, such that C' CvNA.
/u w
be the set of all elements 8 in {M}ysuch that @ - lies in CM.8' Let Keeping in mind that M(& is of finite index in M + ( P ) ~and using the fact , that MF n Cu is finite, we conclude that & is finite. It follows that the
365
union over all 8 E Y Mof the sets P C ' O is a compact subset of PA and is certainly contained in
ML~~-N(^-N;
for some compact subset of N(p)A. Let Y'v be the set of all elements
w
p-lv1,q
C;.
A^ .
3:-of the sets v:lO-lC'h
M:. N;
for some compact subset v2e N ( p ) i such that Then the finite set
C^ ~ ( p ) ;
Let 7 - be the finite set of elements \
6of N(p):.
p-^pT^
C^ N(/t)l .
is bounded by
The function f is compactly supported on GA/ZLso by Lemma 5.2 the sum over p is finite. For any p e {M.}, G ( p ) is a reductive group defined over F. G ( p ) contains A, but since p lies in M,, A is not contained in the center of G(p). Therefore the 8'-split component of the radical of G ( p ) is Z . In particular, with respect to the Tamagawa measure on the volume of G(p)A/G(p)FZ2 G ( p ) is the quotient of r ( G ( p ) )and FftG, correction factor introduced in the our discussion of the elliptic term. The integral (5.1) equals
I t follows from Lemma 5.3 that for a fixed p the function on PA/P(v)^ defined by
366
J A M E S ARTHUR
is of compact support. We conclude that the singular term is integrable over GA/GFZa. Its integral equals
In this section we shall prove that this term is weakly integrable over G J G f Z a and that its integral approaches 0 as e approaches 0.
Now
Therefore J ^ ( f : x: e) equals
(xO) . For , {M?},the group N ( p ) is trivial. I t follows from Lemma 4.2 that JF( f : x: e) equals
E p e
lvTl(n~ TI)-'
EaeGF,lf[/i)F
/& ',)( ~ d
= e
111~t("P
~ ~ ( ~
(n!'l f ) l ~ S e G F / P I P ) F Z ^ i . M / i f ( d @ - ' x ^ Y / A x 8 )
and
(6.2)
S E L B E R G TRACE FORMULA
with
I t follows directly from Lemmas 5.2 and 5.3 that this last expression is finite. The integral
is obviously finite and approaches 0 as e approaches 0. By Lemmas 4.1 and 4.2 the term (6.1) equals
Since {M,.} {MJ is a set of representatives of the conjugacy classes in Mp, U we have shown that
J ^ f : x: e)
+ W:x: e)
and an expression whose integral over G4/GFZ: approaches 0 as e approaches 0. The space it4 is a locally compact abelian group under addition which contains v.r as a discrete subgroup. Let X. be the unitary dual group of n4 and let X p be the subgroup of characters in X4 which are trivial on nF. Let 11 - 11 be the height function on X A associated to some fixed basis of X F . I t is easy to verify that there is an N such that
y Y = E(Ad (ht) ) , () Y
y \ \ 2 edtllEll . For fixed y e GA and p e MF the function
Yen..
a,
J A M E S ARTHUR
f (Y- ,u exp
is of Schwartz-Bruhat type on
IT.&.
Y-')
Yen.
For f e XA,define
If we sum the absolute value of (6.5) over p e Mr and 8 e GF/PF,and then integrate over G.JGFZ2, the result is bounded by
If a) is a relatively compact fundamental domain for P^/Pr in PA, this integral equals
a)
for every t
<: 0.
Then the
Keep in mind that T ( - ,/t, kv) is the Fourier transform of a Schwartz-Bruhat function and is continuous in kv. We observe by a slight restatement of Lemma 5.2 that there are only finitely many p e M such that T(f, p, kv)
#
for some f e XA and some kve K x a). Therefore, for any N there is a constant rlv such that for any primitive f e XA, P, ^) < : I t follows that for every N, (6.6) is bounded by
~ c . , f F
9
kveKxa).
For sufficiently large N this last expression is finite and approaches 0 as e approaches 0. If we sum (6.4) over /a and 8 we arrive a t the expression (6.7)
^fie s r F E 8 e
G ~ / P ~
TO P,&)x~(x
For fixed x there are only finitely many 8 E G& such that xc(x8)+ 0. Therefore the inner sum is finite. From this fact it is easily seen that the outer sum is also finite. To summarize what we have shown so far, the expression
equals the sum of (6.7) and a function whose integral over G^/GpZL approaches 0 as e approaches 0. The function (6.7) is not integrable. We leave it for the moment. In the first parabolic term we still have to consider the contribution from - K'(f: x: e). The function K'( f : x: e) equals
We may formally write this expression as the sum of the following four terms:
JAMES ARTHUR
and
and
a r e a l l finite. Proof. Fix a positive integer n and a n imaginary number z. It is clear that t h e function
R(n:z: f : Y , 4 =
(*
f)i,ft)(~)^o ,
which is continuous in y and v , is the kernel of t h e restriction of x ( z : f ) t o 3C(n). Therefore, if we define ' f and y a s in t h e proof of Theorem 3.2, t h e absolute value
I R(n:2: f :
is bounded by
Y, Y )I
I R(n:z: \fy,
By Schwartz' inequality we need only show t h a t (6.12) is finite when f is replaced by Y . Now
I ; I.( n : 2: 'f: Y , Y ) I ; =R
is bounded by the function P ( z : 'f: y, y ) defined in 3 2. It follows t h a t t h e series
371
: R ( n : z: Y y, v )
is absolutely convergent and defines a function R ( z : lf: y, v ) which is the : kernel of ~ ( zl f ) . By an argument similar to that used in the proof of Lemma 3.7, R ( z : 1/: y, v ) is continuous in y and v separately. Therefore, as we remarked in 3 2, R ( z : 1/: y, v ) equals P ( z : 1/: y, v ) for all y and v . The formula for P ( z : 'fy, y) is given in 5 2. It is clear that the integral
is finite. Therefore the expression (6.12) is finite. For any z and n, the set
{M(z)pf}^ 1% is an orthonormal basis for X ( n ) . It follows that the function
is the kernel of the restriction of n ( - z : f ) to X ( n ) . The finiteness of (6.13) follows by the above argument. The function
(6.16)
E .(M(z)@: f ) ^ ) { ~ ) ^ ( v ) .
M{/-)z(z:f')(M(z)n(z: f t ) ) *
=
is the kernel of the restriction of M(z)n(z: f ) to X ( n ) . We recall that ~ ( zf : = z ( z : f ' ) ~ ( zf :"). We have )
M ( z ) ~ ( zf :) n ( z : (f')*)M(z)-I ' = M(z)n(z:f l M ( z ) - ' = K ( - Z : l f ) .
=
y is bounded by
I R ( n : - z: lf y, y) \'I2. 1 R ( n : z: 2/: y,
y)
\112
.
0
I t follows that (6.14) is finite. Similarly (6.15) is also finite. The term (6.8) equals
for all
a".
372
JAMES ARTHUR
for all x . This expression also equals (6.17). Therefore, the contribution to - K ' ( f : x: e) from the terms (6.8) and (6.9) is the product of (6.17) with ( - 2 ) . The result exactly cancels out the term (6.7).
equals
Our use of Fubini's theorem is justified by the compactness of K and P. * By Assumption 3.5 and the proof of Lemma 3.4 the function is zero for all but finitely many B E I. Therefore, in the above integral over z, we can change the contour to a line {z: Re z = a}, for 8 < 0. The assertions of the lemma for the function (6.10) follow immediately. The result for (6.11) is proved the same way. 0 This lemma accounts for the last of the components in the first parabolic term We have completed the proof promised a t the beginning of this section.
7. The second parabolic term
The second parabolic term equals the sum of I F ( f : x: e) and I s p ( f :x: e). In this section we shall prove that both these functions are integrable over GJGFZ^,. We shall then calculate their integrals. The function I ^ f : x : e) equals
The integral
'E?,V (n?3..f)-11K 2
.(I
-
,Tf(p)FAL
r--
If
(kpup-'k-^
X O ^ - 7^(phtw))dtdrp dl? For any p e Mp let rf,.II the index in X ( M ( P ) )of the group obtained be ~ by restricting the characters in X ( M ) p to M ( p ) . We define
?(^,M ) = ( n ? , . T r ) - l - (I\.Tr)-'. ^(M(,u))-
has compact support on PA/Z2, the integral over P/M(v) can be taken over a compact set, by Lemma 5.3. Finally, it is clear that for any p the function
xs(pht) - xs(phtw)
tec-w,
a ) ,
has compact support. Therefore I f ( f : x: e) is integrable over GA/GpZ^,. Its integral equals
-(
-w
H ( m ) , H ( n w ) - loge - H ( m ) ]
374
JAMES ARTHUR
and
After changing the variable of integration on N Awe may appeal to Lemma 4.4, rewriting (7.2) as
We now consider the function I:(f: x: 6). Our discussion will include integrals over the groups P(f-t\ and P(p)A, for elements p in { M s } . We note that according to our understanding on the choice of Haar measures, the product measure on P(u}\ x At is a multiple of our right Haar measure on PWA r,,.y.Define by ~pw(p) e2P(P)H(p) = P e P(,U)A , to be the modular function of P(p)A.
where d p * is the invariant measure on PA/P(p)A defined by our Haar measures on Pi and P(p)\.
S E L B E R G TRACE FORMULA
375
Proof.
\
is bounded by
IIp(f:~:e)\dx
~,A/GFzL
*(I X ~ ( P ) ) ~ P ( P ) ~ ~ P dk ,
By Lemma 5.3 the integral over PA/P(/i}\ can be taken over a compact subset Cl of PA/P(p)A or equivalently, over a fixed compact set C(p) of representatives of Cl in Pi. For p e { M a } and n e N(p)A, define Op( : n) to be f
The support, U(p), of this function is a compact subset of N(&. pression (7.4) equals
The ex-
3 76
JAMES ARTHUR
P ( p ) i . We can choose a positive number to large enough so that the intersection of the set
{v-'h;'. n. h t v : v
E
~ ( p )t ? to?n E U ( p ) } 2
LEMMA 7.2. F o r a n y z~ C the i n t e g r a l d e f i n i n g Ie(z) i s absolutely convergent. It(z) i s a n e n t i r e f u n c t i o n whose v a l u e a t z = 0 equals the integral
Proof. All statements of the lemma are obvious conseqences of the above discussion.
It remains to calculate 16(0). Specifically, we shall express It(0) as a sum of an expression which is independent of c, and a term whose dependence on c is quite transparent. The idea? which we take from [7]?is to replace the integral over {t 2 log c} which appears in the definition of I&(z) by the difference of an integral over R and an integral over {t 5 log E } . Fix p~ { M a } . Let X(p)A be the unitary dual group of n(plA and let X(p)F be those characters in X(p)A which are trivial on ~ ( p ) For any ~ . p E P ( p l Aand t E X(p)A define
By choosing a height function on X(p)A we can repeat the argument of $ 6 to show that for any z E C the expression
> -1.
E
I t equals
At z
approaches 0.
> 0.
I t equals
For Re z
to be
LEMMA 7.3. T h e i n t e g r a l d e f i n i n g g(p: z: f ) i s absolutely convergent f o Re z > 0. It c a n be a n a l y t i c a l l y continued to a rnerornorphic f u n c t i o n ~ o n C whose o n l y s i n g u l a ~ i t < e sa r e s i m p l e poles a t z = 0 uwd z = -1. F i n a l l y , modulo a t e r m w h i c h approaches 0 a s E approaches 0, I t ( 0 ) equals the l i m i t a s z approaches 0 of
Proof. This lemma follows from the above discussion and Lemma
0 of the function
we write simply as
J A M E S ARTHUR
It is clear that
"
equals
If we write the integral over P i / P ( p ) i as an iterated integral over P;/P(p);N* and P ( p ) i N * / P ( p ) iwe arrive a t the expression
by virtue of Lemmas 4.1 and 4.4. It follows that (7.7) may be written as
Our discussion of the second parabolic term is now complete. We have shown that the integral over G A / G F Z zof this term equals the sum of the expressions (7.1) and (7.6), the term
obtained by combining (7.8) with (7.3), and an expression which approaches 0 as & approaches 0.
8. The third parabolic term
We have proved that the first four terms in the kernel of h o ( f )are weakly integrable. It follows that the final term, - K 1 ' ( f : x: &), is also weakly integrable over GA/GFZ2. In this section we shall calculate its integral. For convenience we set H ( n : 2 : f:x ) equal to
S E L B E R G TRACE FORMULA
379
We also define Hi(%:z: f : x ) and H:'(n: z: fi x ) by replacing all the functions E ( 4 : z: x ) in this definition by Es1(*: x ) and E!'($ z: x ) respectively. Then z: K t ' (f : x: c) equals
By Theorem 3.2 and Lemma 4.8 we can associate to any Siege1 domain g constants C and N such that for a11 x E i3 the inequality
is valid. For 0 < t < co let S ( t ) be the projection of S ( t ) onto GJGJ.2. Let G ( t ) be the closure of the complement of S ( t ) in GA/GFZL. G ( t ) is a compact subset of G A / G F Z zand our integral
equals
and
Ep(+$: x ) 2:
whenever x
380
JAMES ARTHUR
rapidly decreasing on any Siegel domainl by Lemma 1.6. But E:(oe:z: x ) is slowly increasing on any Siegel domainl so that the function
E:(oe: z: x)Es'r($p: x ) z:
is integrable over GA/GFZ:. Its integral equals
The expression in the brackets equals zero identically in x. On the other hand, since t < E ~ , integral the
equals
EP(oe: z:
z: x)xt(x)dx ,
.S[t)
an expression which equals 0. Since G ( t ) is the complement of S ( t ) in GA/GFZzl lemma follows. our It follows from the lemma that (8.2) equals
is finite.
The proof folIows the same idea as that of Theorem 3.2. For P~oof. and as in the proof of Theorem 3.Z1 we can easily see that when z is imaginary
1 H:'(n: z: f : x) 1
is bounded by
1 H:'(n: z: l j ?
x ) 11/2-
1 Hi1(%:z:
x ) 1/'
381
By Schwartz' inequality we need only prove the lemma when f is replaced : by Y. From the fact that the operator ~ ( zY ) is positive semi-definite it follows that
H ' / ( n : z:
x)^0
lf
which completes the proof. This lemma enables us to conclude that the integral over G J G Z ; of - K 1 ' ( f :x: e) equals
and
Proof. First of all, suppose that \ and ,G are distinct complex numbers whose real parts are both less than - p . Then it is known that
382 (8.11)
(X -
JAMES ARTHUR
This formula is stated in [8] and proved in [9]. It follows from a straightforward argument, which we will not reproduce, based on the formula (2.2). The functions defined by (8.9), (8.10), and (8.11) are all meromorphic in approach z. The limit of (8.11) X and ,E. We set p equal to z and let equals (8.8). The limit of (8.10) is the limit as t approaches 0 of
which is the sum of (8.6) and (8.7). On the other hand, the limit of (8.9) is the required inner product, so the lemma follows from analytic continuation. LEMMA 8.4. F o r z i m a g i n a r y the operator
is self-ad joint.
Proof. The adjoint of this operator equals
by virtue of the fact that M(-z)M(z) equals the identity. Let us write n(n:z: f ) for the restriction of the operator ~ ( zf : to )
yC(n).
.
is finite.
Proof. For any E between 0 and the sum of (8.4) and the expression
E~
where Y and Y a r e defined in the proof of Theorem 3.2. A glance a t Lemma 2.3 and the formula preceding it confirms that this expression is also finite. Finally, by Assumption 3.5 and the proof of Lemma 3.4, the sum over 6 in (8.13) is finite. For any 6 the function
is regular a t z = 0, and is in fact integrable over the imaginary axis. This concludes the proof. To complete our calculation we substitute each of the three terms of Lemma 8.3 into the expression (8.5). The first one yields
logs;-imt r rk:f )d I
which equals
i-
since our kernel P is continuous. After inserting the formula for P given in $ 2 and applying the Fourier inversion formula we obtain
This expression cancels (7.9). The contribution of (8.7) to the formula (8.5) is
and
&
384
JAMES ARTHUR
expression approaches 0 by the Riemann-Lebesgue lemma, while the second term approaches
9. Concluding remarks
Our computation is now complete. We have shown that the trace of xo(f) is the sum of a certain number of terms, each of which is independent of c, and an expression which approaches 0 as c approaches 0. Since we started off by letting e be any number between 0 and cothis latter expression must vanish. The remaining terms are scattered throughout the earlier sections. They a r e (i) The elliptic term,
En
;G,
and
- t r {M(O)z(O:f)}
4
We need hardly remark that our formula is not yet in a reasonable form. Considerably more work is required in several directions before we might hope to obtain information about the space of cusp forms. In the first place we have left the term (9.2) in an unsatisfactory state. What is needed is some sort of analysis on the orbit structure of
385
P(p) in N ( p ) , for elements p in Ms. Once (9.2) has been put into a more tractable form we can s t a r t analyzing the various terms of the trace formula as distributions on G^. In view of Harish-Chandra's work on the Schwartz space, it makes sense to ask whether a distribution on any of the local groups Gv is tempered. Every term in the trace formula is, in all probability, a linear combination of products of tempered distributions on the groups Gv. The problem would be to calculate the Fourier transforms of these distributions. If v is a place of F, a distribution T, on G , is said to be invariant if for any pair of functions f and g in C:(G,) The distributions defined by the elliptic and singular terms are all invariant, as are those defined by (9.4). On the other hand, the distributions defined by the remaining terms are not invariant. This complicates the problem of calculating the Fourier transforms. Of course the sum of the terms (9.1), (9.2), and (9.3) defines an invariant linear functional on f. However, it will not be possible to see how the noninvariant components of these terms cancel without calculating the appropriate Fourier transforms.
[ I ] A. BOREL and J. TITS, Groupes rkductifs, Pub. Math., I.H.E.S., 27 (1965), 55-151. [ 2 ] M. DUFLO and J. P. Labesse, Sur la formule des traces de Selberg, Ann. Scient. EC. Norm. Sup., (4) 4 (1971), 193-284. [ 3 ] R. GODEMENT, Domains fondamentaux des groupes arithmktiques, Skminaire Bourbaki, 257 (1963). [4] , Introduction a la theorie de Langlands, Skminaire Bourbaki, 321 (1966). [ 5 ] HARISH-CHANDRA, o m o r p h i c f o r m s o n s e m i s i m p l e L i e g r o u p s , Springer-Verlag, 1968. Aut , 161 -H a r m o n i c a n a l y s i s o n reductive p-adic g r o u p s , Springer-Verlag, 1970. A 171 H. JACQUET and R. P. LANGLANDS,u t o m o r p h i c f o r m s o n GL(2), Springer-Verlag, 1970. O [ 8 1 R P. LANGLANDS,n t h e f u n c t i o n a l e q u a t i o n s satisfied b y E i s e n s t e i n series, Mimeographed notes. , 191 -Eisenstein series, in Algebraic G r o u p s a n d D i s c o n t i n u o u s S u b g r o u p s , Amer. Math. Soc. (1966), 235-252. [lo] A. SELBERG, Harmonic analysis and discontinuous groups in weakly symmetric Riemannian spaces with applications to Dirichlet series, J. Indian Math. Soc. 20 (1956), 47-87. D [ l l j A. SELBERG, 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 , Proc. Int. Cong. Math. 1962, 177-189. (Received December 4, 1972) (Revised September, 17, 1973)