(Note that then Z/rr;i(.s) = &>(*)•) Furthermore, for any g e GLr(A), note that H°(AF, g) is discrete, so from the definition, the associated measure is simply the standard counting measure. Thus by writing down precisely, H°sa(F,g) = 1+
J]
exp(~"J]\g
aeH°(SpecOF,£(g))\{0}
=:l+H's&(F,g). In this expression, the first term is simply the constant function 1 on the moduli space, while each term in the second decays exponentially. Hence, by a standard argument about convergence of an integration of theta series for higher rank lattices, see. e.g, [Ne], and the fact that 1 in the first term H^a(F, E)A cancels with 1 appeared as the second term in the combination H^a(F, E)A - 1, we conclude that II(s) and II(-s - A) are all holomorphic functions, provided A > 0. • 1.5
First Example
1.5.1 Epstein Zeta and High Rank Zeta Accordingly, the rank r zeta function ^q,r(s) of Q is given by
&,,(*) = f
(eft°«-A) - l) • ( e -) deg(A) dtfA),
Re(,) > 1,
where h°(Q, A) := log I £ exp ( - n\x\2) I, And we have
deg(A) = - log Vol(Rr/A).
250
L. Weng
Theorem', (i) £Q,I(S) coincides with the (completed) Riemann-zeta; (ii) ^q,r(s) can be meromorphically extended to the whole plane; (iii) £Q,,-(S) satisfies the functional equation £JQ,r(s) = fQ,r(l - s); (iv) ^Q,r(*) has two singularities at s = 0, 1 only, all simple poles with the same residues Vol (A1Q,,-[1]J. Decompose according to their volumes, Mq>r = there is a natural morphism MQ^[T] —> AtQ,r[l], A quently,
and r ? • A. Conse-
U^OAIQ^IY], H
(e"0«-A>-i)-(Odeg(A)^(A)
€oA*)= f J<JT>oMQ,r[T]
Jo
l JAVm \
'
But h°(Q, Tlr • A) = log (-ZxeA exp ( - n\x\2 • 7"?)) and
r
T
B
\Bl
we have ^r(s)=rx-n-isT(r-s)-
f
2
2
\x\~rs
V
JMQAU UAMO)
J
Accordingly, introduce the completed Epstein zeta function for A by E(A;S):=n-sns)-
\x\~2s.
£ *eA\{0)
We then arrive at Proposition. (Eisenstein Zeta and High Rank Zeta) r f ^Q,rW = 2
JMQAD
r E(A,-s)d/ii(A). l
251
1 High Rank Zetas for Number Fields 1.5.2
Rankin-Selberg & Zagier Method
Consider the action of SL(2, Z) on the upper half plane (H. Then a standard 'fundamental domain' is given by D = {z = x + iy 6 *H : \x\ < \,y > 0, x2 + y2 > 1}. Recall also the completed standard Eisenstein series E(z;s):=7r-Sr(s)-
YJ _
y> \mz + n\2s
(m,n)eZ2\{(0,0)) '
Naturally, we are led to consider the integral fD E{z, s)-^. However, this integration diverges. Indeed, near the only cusp y = oo, by the ChowlaSelberg formula, E{z, s) has the Fourier expansion YJ^n(y,s)eZninx
E(z;s)=
n=—oo
with
an{y S)
'
(£(2s)ys + t(2-2s)yl-s, \2MH
ifn = 0; if n * 0,
where £,(s) is the completed Riemann zeta function, crs(n) := Y^d\n ds, and Ks(y) •= \ £° e-y{t+^l2ts% is the K-Bessel function. Moreover, \Ks(y)\ < e-y/2KKe(s)(2), if y > 4,
and
Ks = K-s.
So an±o(y, s) decay exponentially, and the problematic term comes from ooCy. s), which is of slow growth. Therefore, to make the original integration meaningful, we need to cutoff the slow growth part. There are two ways to do so: one is geometric and hence rather direct and simple; while the other is analytic, and hence rather technical and traditional, dated back to Rankin-Selberg. (a) Geometric Truncation Draw a horizontal line y = T > 1 and set DT = {z = x + iy € D : y < T),
DT = {z = x + iy e D : y >T).
252
Then D = DjUD7.
L. Weng Introduce a well-defined integration
y2
JDT
(b) Analytic Truncation Define a truncated Eisenstein series Ej(z; s) by
\E(z,s)-a0(y;s),
ify>T.
Introduce a well-defined integration
4™(s):=
dxdy [ET(Z;S)-
f
JD
With this, from the Rankin-Selberg method, one checks that we have the following: Proposition. (Analytic Truncation = Geometric Truncation in Rank 2)
S-
1
5
J
For a proof, please see Ch. 7. Each of the above two integrations has its own merit: for the geometric one, we keep the Eisenstein series unchanged, while for the analytic one, we keep the original fundamental domain of "K under SL(2, Z) as it is. Note that the nice point about the fundamental domain is that it admits a modular interpretation. Thus it would be very nice if we could keep the Eisenstein series unchanged, while offer some integration domains which appear naturally in certain moduli problems. Guided by this, in Ch. 2, we will introduce non-abelian L-functions using integrations of Eisenstein series over generalized moduli spaces. (c) Arithmetic Truncation Now we explain why the above discussion and Rankin-Selberg method have anything to do with our high rank zeta functions. For this, we introduce yet another truncation, the stability one.
253
1 High Rank Zetas for Number Fields
So back to the moduli space of rank 2 lattices of volume 1 over Q. Then classical reduction theory gives a natural map from this moduli space to the fundamental domain D above: For any lattice A, fix xi 6 A such that its length gives the first Minkowski minimum A\ of A. Then via rotation, we may assume that xi = (A\, 0). Further, from the reduction theory j-A may be viewed as the lattice of the volume A^2 = yo which is generated by (1,0) and u) = xo + iyo £ D. That is to say, the points in DT constructed above are in one-to-one corresponding to the rank two lattices of volume one whose first Minkowski minimum A~[2 < T, i.e, A\ > T~i. Set AC 5 2 ° S [1] be the moduli space of rank 2 lattices A of volume 1 over Q all of whose sublattices Ai of rank 1 have degrees < \ log T. As a direct consequence, we have the following Proposition. (Geometric Truncation = Arithmetic Truncation) There is a natural one-to-one, onto morphism
In particular, M|° 2 [1] = M Q>2 [1]-£>1. Consequently, we have the following Example in Rank 2.
£Q,2(S)
=
f(2s) 5-1
f(2*-l)
2 Non-Abelian L-Functions 2.1
Automorphic Forms and Eisenstein Series
To facilitate our ensuing discussion, we make the following preparation. For details, see e.g., [MW] and [We7]. Fix a connected reductive group G defined over F, denote by ZQ its center. Fix a minimal parabolic subgroup PQ of G. Then Po = MQUQ, where as usual we fix once and for all the Levi MQ and the unipotent radical UQ. A parabolic subgroup P is G is called standard if P D PQ- For such groups write P = MU with Mo c M the standard Levi and U the unipotent radical. Denote by Rat(Af) the group of rational characters of M, i.e, the morphism M —» Gm where Gm denotes the multiplicative group. Set a*M : = Rat(M) ® z C,
oM : = Homz(Rat(M), C),
and Rea^ := Rat(M) ® z R,
ReaM := Homz(Rat(M), R).
For any x e Rat(M), we obtain a (real) character \%\ : M(A) —> R* defined by m = (mv) i-> m^ := fives \mv$v with | • |v the v-absolute values. Set then M(A)1 := r\eRat(M)Kerl^|, which is a normal subgroup of M(A). Set XM be the group of complex characters which are trivial on M(A)1. Denote by HM '•= logM : M(A) -» OLM the map such that SX € Rat(M) c o^, <x, logM(m)> := log(mM). Clearly, M(A)1 = Ker(logM);
logM(M(A)/M(A)1) - ReoM.
Hence in particular there is a natural isomorphism K : a*M =* XM- Set ReXM := K(Rea*M),
IHIXM
:= K(i • Rea*M).
Moreover define our working space XM to be the subgroup of XM consisting of complex characters of M(A)/Af(A)1 which are trivial on ZG(A)Fix a maximal compact subgroup K such that for all standard parabolic subgroups P = MU as above, P(A) n l = (M(A) n K ) • (t/(A) nK). Hence we get the Langlands decomposition G(A) = Af (A) • f/(A) • K. Denote by mP : G(A) -» M(A)/M(A)1 the map g = m • n • k \-> M(A)1 • m where g € G(A), m e M(A), n € t/(A) and A: € K. 254
255
2 Non-Abelian L-Functions
Fix Haar measures on Afo(A), Uo(A), K respectively such that (1) the induced measure on M(F) is the counting measure and the volume of the induced measure on M(F)\M(A) 1 is 1. (Recall that it is a fundamental fact that M(F)\M(A) 1 is of finite volume.) (2) the induced measure on Uo(F) is the counting measure and the volume of U(F)\U0(A) is 1. (Recall that being unipotent radical, U(F)\U0(A) is compact.) (3) the volume of K is 1. Such measures then also induce Haar measures via logM to the spaces CLM0, a*M , etc. Furthermore, if we denote by po the half of the sum of the positive roots of the maximal split torus To of the central ZMQ of Mo, then ft-*
I Jilfo(A)-£/0(A)-K
f(mnk)dkdnm~2podm
defined for continuous functions with compact supports on G(A) defines a Haar measure dg on G(A). This in turn gives measures on M(A), U(A) and hence on CLM, &*M, P(A), etc, for all parabolic subgroups P. In particular, one checks that the following compatibility condition holds I
f{mnk) dk dn m~
Po
dm
JM0(A)U0(A)K
= I
f(mnk) dk dn m~
pp
dm
JM(A)U(A)K
for all continuous functions / with compact supports on G(A), where pp denotes one half of the sum of all positive roots of the maximal split torus Tp of the central ZM of M. For later use, denote also by Ap the set of positive roots determined by (P, Tp) and Ao = A/>0. Fix an isomorphism To - G^. Embed R+ by the map t \-* (l;f). Then we obtain a natural injection (R*+)R ^-> To(A) which splits. Denote by AM0(A) the unique connected subgroup of To(A) which projects onto (R* ) R . More generally, for a standard parabolic subgroup P = MU, set n AM(A) '•= -AMO(A) %M(A) where as used above Z» denotes the center of the group *. Clearly, M(A) = AM(A) • M(A)1. For later use, set also ^ L A ) := [a e AM(A) • logG a = 0}. Then AMW = AGm ©^M(A)Note that K and U{F)\U{A) are all compact, and M(F)\M(A) 1 is of finite volume. With the Langlands decomposition G(A) = C/(A)M(A)K in
256
L. Weng
mind, the reduction theory for G(F)\G(A) or more generally P(F)\G(A) is reduced to that for AM(A) since ZG(F) n ZG(A)\ZG(A) n G(A)1 is compact as well. (The reader should know that one of the chief purposes to have a nice reduction theory is to make sure that the integrations of moderate growth functions are controlled. Thus with M(F)\M(A)1 of finite volume, a good control is not a big problem. This then leads us to the affine part A.) As such, for to € Mo(A) set AM0(A)('O)
:= [a e AMo(A) • a" > ffla e A0}.
Then, for a fixed compact subset a> c Po(A), we have the corresponding Siegel set S(CJ; t0) := {p • a • k : p e OJ, a e AMo(A)(to), k e K}. In particular, for big enough a> and small enough to, i.e, t^ is very close to 0 for all or 6 Ao, the classical reduction theory may be restated as G(A) = G(F) • S(a>; to). More generally set A
M0(A)('o) := {a 6 AMo(A) : aa > igVor e A£},
and Sp{io; to) := {p • a • k : p € CJ,a € ApMo{K){to),k e K}. Then similarly as above for big enough a> and small enough to, G(A) = P(F)-Sp(a>; to). (Here Ap denotes the set of positive roots for (PoPiM, To).) Fix an embedding ig '• G «-» SLn sending g to (gif). Introducing a height function on G(A) by setting \\g\\ := fives sup{|g,yiv : V/,;'}. It is well known that up to 0(1), height functions are unique. This implies that the following growth conditions do not depend on the height function we choose. A function / : G(A) —» C is said to have moderate growth if there exist c, r € R such that |/(g)| < c • \\g\\r for all g e G(A). Similarly, for a standard parabolic subgroup P = MU, a function / : U(A)M(F)\G(A) -> C is said to have moderate growth if there exist c, r 6 R, A e ReX^o such that for any a e A^(A)> k e K, m € M(A)1 n Sp(co; to), \f(amk)\ < c • \\a\\r • mPo{m)\
257
2 Non-Abelian L-Functions
Also a function / : G(A) —» C is said to be smooth if for any g = gf goo € G(Af) x G(Aoo), there exist open neighborhoods V* of g* in G(A) and a C°°-function / ' : V*. -» C such that f(g'f • g^) = / ' ( g ^ ) for all gj € V, and g'n, € Vco. By contrast, a function / : S (o»; fo) -> C is said to be rapidly decreasing if there exists r > 0 and for all A e RCXMQ there exists c > 0 such that for a e AM(A),g € G(A)1 nS(w; f0), |0(ag)| < c • ||a|| • mPo(g)x. And a function / : G(F)\G(A) —» C is said to be rapidly decreasing if f\s(w,t0) is so By definition, a function 0 : U(A)M(F)\G(A) -» C is called automorphic if (i) 0 has moderate growth; (ii) (k\ • * -^2) parametrized by (k\, Z^) € K x K is finite dimensional; and (iv) is 3-finite, i.e, the C-span of all S(X)(f> parametrized by all X e 3 is finite dimensional. Here 3 denotes the center of the universal enveloping algebra u := U(LieG(Aco)) of the Lie algebra of G(ATO) and 6(X) denotes the derivative of 0 along X. For such a function 0, set fa '• M(F)\M(A) —> C by m i-> m~pp(t>{mk) for all k € K. Then one checks that fa is an automorphic form in the usual sense. Set A(U(A)M(F)\G(A)) be the space of automorphic forms on U(A)M(F)\G(A). For a measurable locally L1 -function / : U(F)\G(A) —» C, define its constant term along with the standard parabolic subgroup P = UM to be fP : C/(A)\G(A) -> C given by g H-> fU(F)\G{A) f(ng)dn. Then an automorphic form e A(U(A)M(F)\G(A)) is called a cusp form if for any standard parabolic subgroup P' properly contained in P, (pp> = 0. Denote by A0([/(A)M(F)\G(A)) the space of cusp forms on U(A)M(F)\G(A). One checks easily that (i) all cusp forms are rapidly decreasing; and hence (ii) there is a natural pairing (•,•) : Ao(U(A)M(F)\G(A))xA(U(A)M(F)\G(A)) defined by ( 0 , 0 ) := / z „ ( A ) t / ( A ) M ( F n G ( A ) Hg)4>(g) dg.
-» C
258
L. Weng Moreover, for a (complex) character £ : Zu{K) —* C* of ZM(A) set A(U(A)M(F)\G(A))€
:= {0 € A(U(A)M(F)\G(A))
:
:=
J]
A(U(A)M(F)\G(A))f
feHom(ZM(A),C*)
and Ao(U(A)M(F)\G(A))z
:=
£
A0(U(A)M(F)\G(A))f.
feHom(ZM(A).C*)
One checks that the natural morphism C[ReaM] ® A(U(A)M(F)\G(A))Z
-> A([/(A)M(F)\G(A))
defined by (Q, ) i-» (g H> 2(logM(m/>(g))) • 0(g) is an isomorphism, using the special structure of A^(A)-finite functions and the Fourier analysis over the compact space AM(A)\ZM(A>- Consequently, we also obtain a natural isomorphism C[ReaM] ® Ao(U(A)M(F)\G(A))z
-+
A0(U(A)M(F)\G(A))f.
(See 4.2 below for a more concrete and useful form.) Set also rio(M(A))f be isomorphism classes of irreducible representations of M(A) occurring in the space Ao(M(F)\M(A))^, and n 0 (M(A) := U^Hom(ZA,(A),c«)n0(M(A))f. (More precisely, we should use Af(A/) x (M(A)nK, Lie(M(Aoo))®RC)) instead of M(A).) For any n e n 0 (M(A)) f set Ao(M(F)\M(A))„ to be the isotypic component of type n of AQ(M(F)\M(A)\, i.e, the set of cusp forms of
259
2 Non-AbeJian L-Functions
M(A) generating a semi-simple isotypic M(A/)x(Af(A)nK, Lie(M(Aoo))®R C))-module of type TT. Set A0(f/(A)M(F)\G(A))^ := {> 6 A0(£/(A)M(F)\G(A)) : ^ € A0(M(F)\M(A))n,
Vk e K}.
Clearly A 0 (t/(A)M(F)\G(A)) f = e^no(M(A))^o(f/(A)M(F)\G(A))^. More generally, let V c A(Af(F)\M(A)) be an irreducible M(A/) x(Af(A) n K, Lie(Af(Aoo)) ®R C))-module with no the induced representation of M(A/) x (M(A) n K,Lie(M(Aoo)) ®R C)). Then we call n0 an automorphic representation of M(A). Denote A(M(F)\M(A))^0 the isotypic subquotient module of type no of A(M(F)\M(A)). One checks that V ® HomM(A/)x(M(A)nK,Lie(M(A„))®RC))(V; A ( M ( F ) \ M ( A ) ) )
-A(M(F)\M(A))„ 0 . Set A(t/(A)M(F)\G(A))„0 := {> € A(C/(A)M(F)\G(A)): <j>k e A(M(F)\M(A))„0, Vk e K}. Moreover if A(M(F)\M(A))no c A0(M(F)\M(A)), we call n0 a cuspidal representation. Two automorphic representations n and no of M(A) are said to be equivalent if there exists A e X^ such that n ^ no ® A. This, in practice, means that A(M(F)\Af(A))^ = ,1 • A(M(F)\M(A))no. That is, for any „0(m). Consequently, A(£/(A)M(F)\G(A))^ = (A ° mP) • A(f/(A)M(F)\G(A)V0. Denote by ^P := [no] the equivalence class of no. Then ty is an X^ -principal homogeneous space, hence admits a natural complex structure. Usually we call (M, ty) a cuspidal datum of G if no is cuspidal. Also for n e ty set ReTT := Re;^ = \x„\ e ReXjtf, where Xn is the central character of n, and Inur := n <8> (-Re?r).
260
L. Weng
Now fix an irreducible automorphic representation n of M(A) and an automorphic form (p e A{U{A)M(F)\G{A))n, define the associated Eisenstein series E( C by
E(4>,7r)(g) :=
J]
0(t%).
<5eP(F)\G(F)
Then one checks that there is an open cone C c ReX^ such that if Re7r e C, E(A •
0, Va e A^}. As a direct consequence, then E(<j),n) e A(G(F)\G(A)). That is, it is an automorphic form. 2.2
Non-Abelian L-Functions
2.2.1
Definition
Being automorphic forms, Eisenstein series are of moderate growth. Consequently, they are not integrable over G(F)\G(A) 1 . On the other hand, Eisenstein series are also smooth and hence integrable over compact subsets of G(F)\G(A) 1 . So it is very natural for us to search for compact domains which are intrinsically defined. As such, let us now return to the group G = GLr. Then form 1.3.8, we obtain compact moduli spaces M g [ A | ] := {g € GLr(F)\GLr(A):
degg = 0, p* < p)
for a fixed convex polygon p : [0, r] -* R. For example, A ^ t l ] = AiQ,r[l], (the adelic inverse image of) the moduli space of rank r semistable Z-lattices of volume 1. More generally, for a standard parabolic subgroup P of GLr, we introduce moduli spaces MP£p[A}r] := {g € P(F)\GLr(A) :tegg = 0,pgp
p-<
r
-p).
By 1.3.8, these moduli spaces MF'~p[Aj.] are all compact. As usual, take the minimal parabolic subgroup Fo to be the one consisting of upper triangular matrices corresponding to the partition ( ! , . . . , ! )
261
2 Non-Abelian L-Functions
with Mo consisting of diagonal matrices. Then P = Pi = UJMI corresponds to a certain partition / = (n,...,r\p\) of r with Mi the standard Levi and Ui the unipotent radical. (See e.g., 3.2.5.) Now for a fixed irreducible automorphic representation n of M/(A), choose
r
E{(p,n){g)dg,
ReTreC.
JM%[A1]
More generally, for a standard parabolic subgroup Pj = UJMJ D Pi (so that the partition 7 is a refinement of /), we obtain a relative Eisenstein series Ej(cf>,n)(g) :=
YJ 6eP,(F)\Pj(F)
WS>>
Vg e
PJ(F)\G(A).
There is an open cone C\ in ReXfy such that if Re;r 6 C\, then EJ(
Definition. Define the non-abelian L-function by LpfW, n) : - f Pp < , F/(0, n)(g) dg, JM Ff[tf\
Rene
C\.
Remark. Here when defining non-abelian L-functions we assume that
262
L. Weng
only serves the purpose of giving the constructions and results in a very neat form. We end this subsection by pointing out that the discussion for rank r Lfunctions Lppr holds for general non-abelian L-functions LFfp as well. So from now on, we will leave the discussion on general LF^~P to the reader, while concentrate ourselves to the special case LFpr. 2.2.2
Meromorphic Extension and Functional Equations
With the same notation as above, set ^P = [n]. For w 6 W the Weyl group of G = GLr, fix once and for all representative w e G(F) of w. Set M' :wMw~x and denote the associated parabolic subgroup by P' = U'M'. W acts naturally on automorphic representations, from which we obtain an equivalence classes w^3 of automorphic representations of M'(A). As usual, define the associated intertwining operator M(yv, n) by (M(w,n)4>)(g) := f
1
l
(p(w-
n'g)dn', Vg e G(A).
J U'(.F)nwU(F)w- \U'(A)
One checks that if (Re7r, a v ) » 0, Vor e A£, (i) for a fixed automorphic depends only on the double coset M'(F)wM(F). So M(W,TT)> is well-defined for weW; (ii) the above integral converges absolutely and uniformly for g varying in a compact subset of G(A); (iii) M{w,n)
,ri)forRen eC is well-defined and admits a unique meromorphic continuation to the whole space ^3; (II) (Functional Equation) As meromorphic Junctions on ^8, LFPr(
, wn),
VweW.
Proof. This is a direct consequence of the fundamental results of Langlands on Eisenstein series and spectrum decompositions and explains why only L2-automorphic forms are used in the definition of our Ls. (See e.g, [Arl], [Lai], [MW] and/or [We7]. ) •
2 Non-Abelian L-Functions 2.2.3
263
Holomorphicity and Singularities
Let n e ty and a e AM. Define the function h:ty-*Cbyn®Ai-> (A, av), VA 6 X^j =* a^. Here as usual, av denotes the coroot associated to or. Set H :- {n' e ^B : h(n') = 0} and call it a root hyperplane. Clearly the function h is determined by H, hence we also denote h by hn- Note also that root hyperplanes depend on the base point n we choose. Let D be a set of root hyperplanes. Then (i) the singularities of a meromorphic function / on ^8 is said to be carried out by D if for all n e ^8, there exist nn : D —» Z>o zero almost everywhere such that n' i-» (Il//e£)/*ff(7r')">r(//)) • /(^') is holomorphic at TT'; (ii) the singularities of / are said to be without multiplicity at n if n„ e {0,1}; (iii) D is said to be locally finite, if for any compact subset C c $ , {H 6 D : H n C * 0} is finite. Basic Facts of Non-Abelian L-Functions. (III) (Holomorphicity) (i) When Ren eC, L^fty,n) is holomorphic; (ii) Lj?r(, n) is holomorphic at n where Ken = 0; (IV) (Singularities) Assume further that <\> is a cusp form. Then (i) There is a locally finite set of root hyperplanes D such that the singularities ofL~fr(
0,VaeA^; (iii) There are only finitely many of singular hyperplanes ofLj.p((p, n) which intersect [n e ^ : 0, Va e AM}Proof As above, this is a direct consequence of the fundamental results of Langlands on Eisenstein series and spectrum decompositions. (See e.g, [Arl], [Lai], [MW], and/or [We7]). a
3
Geometric and Analytic Truncations
3.1 3.1.1
Geometric Truncation: Revised Slopes, Canonical Filtrations
Following Lafforgue [Laf], we call an abelian category ft together with two additive morphisms rk : ft -> N,
deg : ft -» R
a category with slope structure. In particular, for non-zero A e ft, (1) define the slope of A by ^(A) := ^ g ? ; (2) If 0 = A0 c A\ c • • • c A/ = A is a filtration of A in ft with rk(A0) < rk(Ai) < • • • < rk(A/), define the associated polygon to be the function [0, rkA] -» R such that (i) its values at 0 and rk(A) are 0; (ii) it is affine on the intervals [rk(A;_i),rk(A,-)] with slope ^(A,/A,_i) -
tfA); (3) If a is a collection of subobjects of A in ft, then o is said to be nice if (i) a is stable under intersection and finite summation; (ii) o is Noetherian, i.e., every increasing chain of elements in a has a maximal element in a; (iii) if Aj e a then A\ * 0 if and only if rk(Ai) + 0; and (iv) for A\, A2 e a with rk(Ai) = rk(A2). Then A\ c A2 is proper implies thatdeg(Ai)<deg(A 2 ); (4) For any nice a, set fi+(A) :=sap[n(Ai): Aj e a,rk(A0 > l}, p~(A) :=inf{/i(AMi): A\ e Q,rk(Ai) < rk(A)). Then we say (A, 0) is semi-stable if fi+(A) = p{A) = pT{A). Moreover if rk(A) = 0, set also p.+(A) = -00 and n~(A) = +00. Proposition 1. ([Laf]) Let ft be a category with slope structure, A an object in ft and 0 a nice family of subobjects of A in ft. Then 264
3 Geometric and Analytic Truncations
265
(1) (Canonical Filtration) A admits a unique filtration 0 = Ao C A\ C • • • c A/ = A with elements in a such that (i) Ai, 0 < i < k are maximal in a; (ii) Ai/Ai-i are semi-stable; and (Hi) p(M/A0) > /i(A2/Xi > > n(Ak/Ak-iy, (2) (Boundness) All polygons of filiations of A with elements in a are bounded from above by ~p, where ~p :=~p is the associated polygon for the canonical filtration in (1); _ (3) For any A\ e a, rk(AO > 1 implies p(Ax) < p(A) + Pifft1'))); (4) The polygon ~p~ is convex with maximal slope p+(A) - n(A) and minimal slope n~(A) - n(A); (5) If (A', a') is another pair, and u: A —» A' is a homomorphism such that Ker(w) € o and \m(u) e a'. Then pT(A) > p.+(A') implies that u = 0. This results from a Harder-Narasimhan type filtration consideration. A detailed proof may be found at pp. 87-88 in [Laf]. As an example, we have the following Proposition 2. Let F be a number field. Then (1) the abelian category of hermitian vector sheaves on Spec OF together with the natural rank and the Arakelov degree is a category with slopes; (2) For any hermitian vector sheaf (E,p), a consisting ofpairs (E\,pi) with Ei sub vector sheaves ofE andp\ the restrictions ofp, forms a nice family. Proof. This is a simple reformulation of Prop. 1.3.6. Indeed, (1) is obvious, while (2) is a direct consequence of the following standard facts: (i) For a fixed (E,p), ideg(Ei,pi) : (E\,p\) e a} is discrete subset of R; and (ii) for any two sublattices Ai, A2 of A, Vol(Ai/(Ai n A2)) > Vol((Ai + A 2 )/A 2 ).
• Thus we get canonical filtrations of Harder-Narasimhan type for hermitian vector sheaves over Spec OF- Recall that hermitian vector sheaves over Spec OF are Of-lattices in (R n xC 2 ) r = r k ( £ ) : Say, corresponding (9/r-lattices are induced from their H° via the natural embedding Fr <^-> (R n x C 2 ) r where r\ (resp. r 2 ) denotes the real (resp. complex) embeddings of F.
266 3.1.2
L. Weng A Micro-Global Relation
In algebraic geometry, or better in Geometric Invariant Theory, a fundamental principle, the Micro-Global Principle, claims that if a point is not GIT stable then there exists a parabolic subgroup which destroys the corresponding stability. Here even we do not have a proper definition of GIT stability for lattices, in terms of intersection stability, an analogue of the Micro-Global Principle holds. Let A = Ag be a rank r lattice associated to g € GLr(A) and P a parabolic subgroup. Denote the sublattices filtration associated to P by 0 = Ao c Ai c A2 c • • • c A|p| = A. Assume that P corresponds to the partition / = {d\, di,..., dn=-\p\). Consequently, we have rk(A,) = rt := d\ + di + • • • + di,
for i= 1,2,..., |P|.
Let p, q : [0, r] -» R be two polygons such that p(0) = q(0) = p(r) q(r) = 0. Then following Lafforgue, we say q is bigger than p with respect to P and denote it by q >/> p, if q{ri) - p(n) > 0 for all i = 1,..., |P| - 1. Introduce also the characteristic function l(p* < p) by
lo,
otherwise.
Here as before, JP denotes the canonical polygon for the lattice corresponding to g. Recall that for a parabolic subgroup P, pp denotes the polygon induced by P for (the lattice corresponding to) the element g e G(A). Fundamental Relation. For a fixed convex polygon p : [0, r] —* R such that p(0) = p(r) = 0, we have
KF>)=
S
(-i)17,1-1
P: standard parabolic
2
I(P?>PP).
6eP(F)\G(F)
Remarks. (1) This result and its proof below are motivated by a similar result of Lafforgue for vector bundles over function fields.
267
3 Geometric and Analytic Truncations
(2) The right hand side may be naturally decomposite into two parts according to whether P = G or not. In such away, the right hand side becomes
1G-
(-l) 1 ^ 1 -.
£ P: proper standard parabolic
This then exposes two aspects of our geometric truncation: First of all, if a lattice is not stable, then there will be parabolic subgroups which take the responsibility; Secondly, each parabolic subgroup has its fix role - Essentially, they should be counted only once each time. In other words, if more are substracted, then we need to add one fewer back to make sure the whole process is not overdone. Proof. (Following Lafforgue) Note that all parabolic subgroups of G may be obtained from taking 5-conjugates of standard parabolic subgroups with S € P(F)\G(F), and that ~p* = ~pgsps-X- Therefore, it suffices to prove the following Fundamental Relation'.
KP8
{-\f^\{pp>PP).
P: parabolic
To establish this equality, we consider two cases depending on whether (a) p^ < p; or (b) J?8 is not bounded from above by p. Assume that (a) holds. Then the LHS is simply 1. On the other hand, for a proper parabolic subgroup Q 4- G, by definition, PQ < p. Hence 1(/A >Q p) = 0. Consequently, the right hand side degenerates to the one consisting of a single term involving G only, which, by definition, it is simply 1 as well. We are done. Next, consider (b), for which p g is not bounded from above by p. This —g,p is a bit complicated. To start with, we set A% to be the unique filtration of A8 such that the associated polygon ~pPp is maximal among p8Q where Q runs over all parabolic subgroups contained in P. That is ~pp := max {/A : Q c P, parabolic subgroup].
268
L. Weng —8
Denote the associated parabolic subgroup by QP and call it P-canonical. Clearly, G-canonical is simply canonical. Moreover, since JP £ p, the LHS = 0. So we should prove that the RHS = 0 as well. For this, regroup the parabolic subgroups P appeared in the RHS according to the refined canonical parabolic subgroup Q8P. Then clearly, it suffices to establish the following Proposition. With the same notation as above, if ~pg ^ p, then for any fixed parabolic subgroup Q,
2 (-iri(ppP>PP)=o. P&P=Q
Proof. We break this into the following 3 steps. Step A. Fix a non-negative real number p.. For the lattice A = A8 and a fixed parabolic subgroup P of G, set (1) MQ to be the unique parabolic subgroup of G such that if^A, is the induced filtration of A8, then
fA!} = JA5:/,(A5/A5_1)>MA?+1/A3+4 where A» is the canonical filtration associated with Ag; —g
—g,P
(2) ^Qp to be the unique parabolic subgroup of G such that if ^A„ is the induced filtration of A8, then
i"A5'P] - « ) eP
where A • is the filtration of Ag induced from the parabolic sub—g,p group P, and A, is the P-canonical filtration associated with A8. In particular, from this construction, we have: (i) For —gP A, , so-called P-canonical filtration, the Harder-Narasimhan con—g,P i—g,P
ditions hold, i.e., A ; j^j-\ are semi-stable, and -rg,P tTS
rrsf rr8
.(Af/A-,)^;,/^).
3 Geometric and Analytic Truncations
269
except for those j appeared as that associated to P itself; (ii) While the canonical filtration A, of A8 satisfying ^ ( A J / A J . , ) > /i(X5+1/A*),
V; = 1,..., Kf I - 1,
A,} is the one obtained from this canonical one by selecting only the part where much stronger conditions
are satisfied. Consequently,
Q8 c "£f; (iii) The filtration PAf' } contains not only that part of the P-canonical filtration where stronger conditions
are satisfied, it contains also the full filtration of Ag associated to P. So, Q8P c "g£,
while
»~Q% c P.
Step B. For g e G(A), set Gl /(GO =jrkAf ,fl2 KQi) ={ikAj
MQi)
=(rkAf'e'
WQi)
=(rkAfe' G2
KP(Q2) =(rkA^
i=
l,2,...,\Qi\-l};
; = l,2,...,ie2|-l);
M[^i
/A/_i j > fi[AM /Az
)+H)\
(4 2 -p)(rkA^)>0).
Lemma. With the same notation as above, the following correspondence [PeP:Q8P
= Qu "Qp = Q2, PgP>p P]
- * ( l : I c I(Q2), J^Qx) c 7(J22), (/(fii) - •*>(&)) U (/(&) - MQi)) c S c ATp(j22)}
270
L. Weng
defined by P H+ /(/>) := (rkAf P : i = 1 , 2 , . . . , | P | - 1 } is a well-defined bijection. Proof of the Lemma. (I) Well-defined. For this, let us look at the conditions for P in {PeP: QP = QU tlQp = Q2,PgP>Pp}. —8
(i) Qp - Q\ gives the filtration (01 = Ag'Ql
c A8'Ql c Ag'Ql
c • • • c Ag'Ql
c A8,Ql
= A8
with p8Q be the maximal among all [p8Q : Q c p). Hence, (a) (2i c P; and (b) Except possibly at indices corresponding to the partion coming from P,
/4A*
/ V l j > A*(At+1 /A t j ,
by the maximal property, or more clearly, the Harder-Narasimhan property. Therefore, we have /(Gi) - MQi) c 1(F). (ii) ^Qp = Qi gives the filtration (0) = A8,Ql
c Ag'G2 c A*'22 c • • • c A8,Ql
c Ag'G2 = A8
which by definition coincides with {A8/
:j=l,2,...,|P|-l)
U {Af'^1 : ft[Ak
JAk_x ) > n[Ak+l JAk
) + n j.
By taking the rank for each lattices, we get the relation KQi) - I(P) U Jfi(Q\)- Therefore, (a) Q2cPor equivalently I(P) c 7(<22); and (b) KQi) - / „ ( & ) c /(/>). (iii) psp >p p is, by definition, equivalent to the following group of inequalities
(/ p -p)(rkAf)>0.
3 Geometric and Analytic Truncations
271
Consequently, I(P) c KP(Q2) since I(P) c I(Q2). Therefore, P i-> /(f) is well-defined. (II) Injectivity. This is clear since with a fixed partition /, there exists only one unique standard parabolic subgroup Pj such that I(Pi) = A-1. (III) Surjectivity. For a subset S of [l, 2 , . . . , A such that
I c/(fi2), ^(Gi) c /(&), (/(Gi) - /o(Gi)) U (/(fiz) - /„(&)) c S c KP(Q2), clearly, tiiere exists a parabolic subgroup P = P%. So we need to check whether QP = Qu * ~QP = Qi and p£ >/» p. _ 8 It is clear that p p >P p since E c KP(Q2). On the other hand, QP = Gi» ^Qp = Qi are direct consequences of the definition of P-canonical filtration and the conditions that (/(Gi) - MQi)) c I and (/(G2) - ^ ( G i ) ) c S as explained in the proof above of the well-defined statement. This completes the proof of the Lemma. Step C. With the Lemma, we have
(-lfll(pp>Pp)
£
=\
( 1)#(
2
-
s:(/(eiWo(ei))u(/(G2)-^(Gi))csc/rp(e2) =
jo, if (/(GO - /o(Gi)) u (/(G2) - ^(Gi)) * ^(G2); \ i , if (/(Gi) - MQi)) u (/(G2) - //i(Gi)) = ^ ( 6 2 ) .
Surely, we want to show that the value 1 cannot be taken for g such that ~pp is not bounded by p. To this end, as the final push to establish our Fundamental Relation, set// = 0 and Q\ = Q2 = Q in the above discussion so that
2
(-l)ml(pP >P p)
P<2P=QU»QP=QI
=
J]
(-lf\\{pp>pp)
P{QP=Q=°QP
= J] P-X>SP=Q
(-lfll(pp>Pp)
272
L. Weng
is the summation in the Proposition. Then, by definition and the convexity property of p, (/(0i) - MQi)) U (/(&) - Jp{Qi)) = KP{Q2) implies that KQi) = JoiQi), or better Ksp{Q2) = 0. Consequently, this then shows that JP < p, a contradiction. This completes the proof of the Proposition and hence the Fundamental Relation. •
3.2 3.2.1
Arthur's Analytic Truncation Parabolic Subgroups
Let F be a number field with A = Ap the ring of adeles. Let G be a connected reductive group defined over F. Recall that a subgroup P of G is called parabolic if G/P is a complete algebraic variety. Fix a minimal F-parabolic subgroup Fo of G with its unipotent radical No = Np0 and fix an F-Levi subgroup Mo = Mp0 of Fo so as to have a Levi decomposition PQ = MoNo- An F-parabolic subgroup P is called standard if it contains Fo. For such parabolic subgroups F, there exists a unique Levi subgroup M = Mp containing Mo which we call the standard Levi subgroup of F. Let N = Np be the unipotent radical. Let us agree to use the term parabolic subgroups and Levi subgroups to denote standard F-parabolic subgroups and standard Levi subgroups respectively, unless otherwise is stated. Let F be a parabolic subgroup of G. Write Tp for the maximal split torus in the center of Mp and T'p for the maximal quotient split torus of Mp. Set a> := X»(7» ® R and denote its real dimension by d(P), where X»(T) is the lattice of 1-parameter subgroups in the torus T. Then it is known that 5/> = X*(T'p) ® R as well. The two descriptions of 5/> show that if Q c F is a parabolic subgroup, then there is a canonical injection ap ^-» 5g and a natural surjection Q.Q -» ap. We thus obtain a canonical decomposition O.Q = 5^ © ap for a certain subspace 5^ of 5g. In particular, ac is a summand of 5 = ap for all F. Set ap := ap/aG and a^ := O.Q/OG. Then we have ag = apQ © ap and ap is canonically identified as a subspace of ag. Set % := ap0 and a£ = a£ then we also have ao = a£ © ap for all F.
273
3 Geometric and Analytic Truncations 3.2.2 Logarithmic Map
For a real vector space V, write V* its dual space over R. Then dually we have the spaces aj, a*p, (OQ J and hence the decompositions a
5 = ( a $T®( a Gf ea P-
So o|, = X(MP) <S> R with X(MP) the group HomF(MP,GL(lj) i.e., collection of characters on Mp. It is known that a*p - X(Ap) <8> R where Ap denotes the split component of the center of Mp. Clearly, if Q c P, then MQ C Mp while Ap c AQ. Thus via restriction, the above two expressions of a*p also naturally induce an injection a*p ^-> a*Q and a surjection aQ -» ap, compactible with the decomposition a*Q = (a^J © ap. Every x = Z -Wi in a/>c : = af> ® ^ determines a morphism P(A) -» C* by /? i-> p* := n Wi(p)\s'- Consequently, we have a natural logarithmic map Hp : P(A) -» ap defined by <#/>(/>),*> = ^ .
YY 6 a*p.
The kernel of ///> is denoted by P(A)1 and we set MP(A)1 := P(A)1 n Mp(A). Let also A+ be the set of a e Ap(A) such that (1) av = 1 for all finite places v of F; and (2) x(ao-) is a positive number independent of infinite places cr of F for all AT e X ( M P ) .
ThenM(A)=A + -M(A) 1 . 3.2.3
Roots, Coroots, Weights and Coweights
We now introduce standard bases for above spaces and their duals. Let Ao and Ao be the subsets of simple roots and simple weights in OQ respectively. (Recall that elements of Ao are non-negative linear combinations of elements in Ao.) Write AQ (resp. AQ) for the basis of ao dual to Ao (resp. Ao). Being the dual of the collection of simple weights (resp. of simple roots), AQ (resp. AQ ) is the set of coroots (resp. coweights). For every P, let A^> c aj be the set of non-trivial restrictions of elements of AQ to ap. Denote the dual basis of Ap by A„. For each a e Ap, let
274
L. Weng
a v be the projection of By to ap, where B is the root in Ao whose restriction to ap is a. Set A^ := |arv : a e Ap}, and define the dual basis of A^ by Ap. More generally, if Q c P, write APQ to denote the subset a e AQ appearing in the action of TQ in the unipotent radical of Q n MP, (Indeed, MP n Q is a parabolic subgroup of Mp with nilpotent radical NQ := NQ n Mp. Thus AQ is simply the set of roots of the parabolic subgroup (Mp n Q,AQ). And one checks that the map P i-» A^ gives a natural bijection between parabolic subgroups P containing Q and subsets of AQ.) Then ap is the subspace of OQ annihilated by A£. Denote by (A V )Q the dual of AQ. Let (AJJ)V := ja v : a e AJ) and denote by AJ the dual of (A£) v . 3.2.4
Positive Cone and Positive Chamber
Let Q c P be two parabolic subgroups of G. We extend the linear functiona l in AQ and A~ to elements of the dual space o* by means of the canonical projection from Oo to a^ given by the decomposition ao = a~ © a^ © ap. Let TQ be the characteristic function of the positive chamber {Heao:(a,H)>OVaeApQ} = a® © {# e apQ : (a,H) > 0 for all a e AQ] © aP and let TQ be the characteristic function of the positive cone [H 6 0o:(tiT,//>>OVtzTeAj} = a® © {// e apQ : (m,H) > 0 for all m e A^} © ap. Note that elements in AQ are non-negative linear combinations of elements in AQ, SO we have I*
3.2.5
>T
P
Example with 5 L r
Set G = SLr(R). Let P 0 = B be the Borel subgroup of G consisting of upper triangular matrices, and Mo be the Levi component consisting of diagonal matrices. One checks easily that the unipotent radical No consists
275
3 Geometric and Analytic Truncations
of upper triangular matrices whose diagonal entries are all equal to 1 and that (the positive part Ao+ of) the split component AQ consists of diagonal matrices diag(ai,a2,...,ar) with a\,a2,...,ar € R>o and II)=i a ' = *• Hence Oo = [H := (HUH2, ...,Hr) eW : £- = i Hi = OJ and the natural logarithmic map Hp0 : PQ -» ao is given by g i-» Hp0(a(g)J = loga\ + log«2 + • • • + loga r , where we have used the Iwasawa decomposition g = n-m- a(g) • k with n € No, m e MQ, a(g) = diag(<2i, a2, ...,ar)e A0+ and keK = SO(r). Set ei e aj such that e,(/T) = Hi, then Ao = [ei - e 2 , e 2 - e j , . . . , e r - \ - e r y We identify Ao with the set A := {1,2,..., r-1} by the map a, := e,-e,+i *-> i,i = l , 2 , . . . , r - 1. It is well known that there is a natural bijection between partitions of r and parabolic subgroups of G. More precisely, the partition I := (d\,d2,...,dn)of r(sothatr = d\+d2 + --- + dn andd,- € Z>o)corresponds to the parabolic subgroup Pj consisting of blocked upper triangular matrices whose diagonal entries are blocked matrices of sizes d\,d2,...,dn. Then clearly, the unipotent radical Nj consists of blocked upper triangular matrices whose blocked diagonal entries are identity matrices of sizes d\, d2,..., d„, the Levi component Mj consists of blocked diagonal matrices whose blocked diagonal entries are of sizes d\,d2,...,dr, and the corresponding A/+ consits of matrices of the form diagfai /, ,a2Id2,..., anIdn) where 7^, denotes the identity matrix of size d{, a,; > 0 for i = 1,2,..., n, and n"=i ai = 1. As such, a/ is a subspace of do defined by a7 := [(HuHz,...
,Hr) e do : Hi = H2 = • • • = Hn,
Hn+i = Hn+2 = • • • = Hn, . . . , Hrn_1 + i = Hrn_l+2
= ••• =
Hrn\
where we set r,- = d\ + d2 + • • • + d{ for i = 1,2,..., n. Moreover, with the help of Iwasawa decomposition, the corresponding logarithmic map may be simply seen to be the map Hi : A/+ -> a/ denned by diag{a\Idl, a2Idl,...,
anIdn)
.-» ( ( l o g f l l ) W l ) , 0 o g f l 2 ) ( * ) , • • • , (logfln)***).
276
L. Weng
Obviously, the natural bijection between parabolic subgroups P of G and subsets of Ao or better of A is given explicitly as follows: The parabolic subgroup P = Pj corresponds to the subset I{P) of A = | l , 2 , . . . , r - 1 j such that A - I{P) = {ru r2,..., rn}. That is, 7(7>) = { l , 2 , . . . , r 1 - l ) | J ( r 1 + l , r 1 + 2 , . . . , r 2 - l }
U " • U I'"-1 + lf r"~i+2' • • •'r" ~ AFor simplicity, from now on, we will use / to indicate both the partition and its corresponding subset I(Pi). For example, the subset 7 = 0 corresponds to the partition = ( 1 , 1 , . . . , 1) of r = 1 + 1 + • • • + 1 (since A - 0 = {1,2,..., r-1}) whose corresponding parabolic subgroup is simply Po the minimal parabolic subgroup. Consider now 0 c 7 c / . From A - / c A - 7 c A, we get 7Jo = Pq, c Pi c Pj. Set A- J = {s\,s2,..., sm} corresponding to the partition r = fi+h + --- + fm- Then f\ = d\ + d2 + • • • + dkl, f2 = 4,+i + dki+2 + ••• + dk2,...,fm = dkm_l+i + dkm_l+2 + --- + dkm and aj is a subspace of a0 defined by aj : = { < > , 7 7 2 ^ , . . . , / # " > ) 6 «o). Clearly aj is in a/ since a/ is a subspace of ao denned by a/:=((A?'U?\...,A^)eoo} and we have the map aj
->
0/
Moreover, with the help of Iwasawa decomposition, the corresponding logarithmic map may be simply viewed as the map Hj : Aj+ -» a.j denned by di&g(biIfl,b2If2,...,bmIfm) »
((log bi)&\
(log b2)^\
...,
(log
bm)^).
To go further, for i e A, set _ .
mi := wf =
i
^k-ak r
k=\
.
r-\
+r
^(r-k)-ak k=i+l
277
3 Geometric and Analytic Truncations
be the corresponding fundamental weight, mi is the Q-linear form on oq, = ao given by l
mi(H) =Hi+H2+-+Hi-
-(H\
+ H2 + --- + Hr)
= Hi+H2 + --- + Hi for all H = \H\,H2,...,
Hrj e oo, so that A/ = {trr,- : / e A - /} since A/ = (or,1
: i e A - /}.
la;
I
Note that aj = (A/)R and elements of A/ are non-degenerate positive linear combination of elements in A/, so both A/ and A/ generate c^. Consequently, we have a* = (mt : i e A - / ) R => ( ^ : j e A - j ) R = a}. Moreover, by definition, the space aj is the subspace of aj annihilated by a}. Therefore, aj = [H 6 a7 : a(tf) = 0 Var e a}) = JH e a/: GT,(#) = 0 V; e A - /}
= {(,...,^>)ea0: rfi#i + d2H2 + ••• + dklHkl = 0, dki+iHki+\ + dkl+2Hk1+2 + ••• + dk2Hk2 = 0,... j . Note that clearly dim aj + dim ay = dim aj and a^ n ay = 0, so we have the natural decomposition aj = aj® ay. Furthermore, note that n
I = {J{rj-\
+
hrj-i+2,...,rj-l}
i=i
={l, 2 , . . . , r - l} - {n, r2,..., r„_i}.
278
L. Weng
Set, for k = 1,2,-•• ,dj- 1, d-k
k
k
dj l
~
motivated by the construction of fundamental weights above. Then if, as usual, we define the coroot aY e oo by 1 (<*,v)* = ' - 1 0
if k=i, ifk=i+l, otherwise,
then jaY : i e /j is a basis of the subspace a^ of am and easily one checks that TiT,-(ap = Sij for all i,j e I. Consequently, jroj
;
: i € /} is its dual
0
basis in (a^)*. In particular, we also get A/ = [VJ{ : k e / - / = (A - /) - (A - /)). We end this preperation with the following remarks about the notation. With respect to the decomposition
we have that the space (ajH is generated by A^ = (or/: / € / - 0 = /} c A0 the root system, similarly (a'A is generated by A£ = [ak : k e I - 0 = /) c A0 while (aj) is generated by A^ which is not a subset of A0 but we have A, c A/ consists of the restriction of a* to aj with k e J -1 - (A -1) - (A - J). That is Aj = {ak\aJ :k€j-I = (A~I)-(AJ)}. In particular, if / = A we have <4 = ao, af = a/,
and
A^ = A0, Af = A/.
279
3 Geometric and Analytic Truncations 3.2.6
Partial Truncation and First Estimations
Denote T£ and T^ simply by Tp and T>. Basic Estimation. ([Ar2]) Suppose that we are given a parabolic subgroup P, and a Euclidean norm \\ • \\ on ap. Then there are constants c and N such that for all x € G(A)1 and X € ap,
Y;
-TP(H(SX) -
X) < c(\\x\\emf.
6eP(F)\G(F)
Moreover, the sum is finite. As a direct consequence, we have the following Corollary. ([Ar2]) Suppose that T e OQ and N > 0. Then there exist constants c' and N' such that for any function (f> on P(F)\G(A)\ and x, y e G(A)1,
£
\tf6x)\-Tp(H(6x)-H(y)-X)
6eP(F)\G(F)
is bounded from above by c'\\xf'-\\yf
• sup (|0(M)| • \\u\rN). ueG(A)1
3.2.7
Langlands' Combinatorial Lemma
In this subsection, following Arthur, we recall Langlands' combinatorial lemma. If Pi c P 2 , following Arthur [Ar2], set
oi(H):=o-Pp]:= £
{-\f^A^r\(H)
• f,{H),
for H € do. Then we have Lemma 1. If Pi c Pz, cr* is a characteristic function of the subset of H e ai such that (i)a(#)>0foralla€A2; (ii) cr{H) < 0 for all cr € Ai\A*; and
280
L. Weng
(iii) m(H) > 0 for all w e A2. As a special case, with Pi = P2, we get the following important consequence: Langlands' Combinatorial Lemma. IfQcP then for all H e %,
are parabolic subgroups,
(-l)dim(A*M,,)T£(tf)f£(tf) =6QP;
2 R-.QcRcP
YJ (-lf^^iHtfiH)
=SQP.
R-.QcRcP
We next give an application of this combinatorial lemma, which will be used in 3.2.8. Suppose that Q c P are parabolic subgroups. Fix a vector A 6 <1Q. Let 4(A)
:= (_1)#(-A-A(^)
and let
He QO,
be the characteristic function of the set r„ l"
£
w{H)>0, °° wm * 0, :
ifA(av)<0u .Pl v VQf£A ifA(a )>0' G|-
Lemma 2. With the same notation as above,
Z
B
p
p
R-.QcRcP 3.2.8
fo,
if A(or v )<0, 3 a e A£
I1'
Ulnerwl!
>e
Langlands-Arthur's Partition: Reduction Theory
Our aim is to derive Langlands-Arthur's partition of G(F)\G(A) into disjoint subsets, one for each (standard) parabolic subgroup. To start with, suppose that to is a compact subset of Aro(A)Mo(A)1 and that To 6 - o j . For any parabolic subgroup P\, introduce the associated Siegel set s''1 (To, co) as the collection of pak,
p e o), a
281
3 Geometric and Analytic Truncations
where a(H0(a) - To) is positive for each a e AQ. Then from classical reduction theory, we conclude that for sufficiently big co and sufficiently small T0, G(A) = Pi(F)sPl(T0, to). Suppose now that Pi is given. Let 5Pl(To,T,co) be the set of x in s Pl (To, co) such that VT(HO(X) - T) < 0 for each w e KP'. Let FPx (x, T) := Fx(x, T) be the characteristic function of the set of x e G(A) such that 5x belongs to sPl(Jo, T, co) for some 6 e P\(F). As such, Fl(x,T) is left Ai(R)°M(A)Mi(F)-invariant, and can be regarded as the characteristic function of the projection of the space sPi(T0,T,io) onto the space Ai(R)°Afi(A)Mi(F)\G(A), a compact subset of A i (R)°JVi (A)Ml (F)\G(A). For example, F(x, T) := FG(x, T) admits the following more direct description which will play a key role in our study of Arthur's periods: If Pi c P2 are (standard) parabolic subgroups, we write A~ :- Ap for Ap^A) 0 , the identity component of Ap,(R), and AZ2:=A™itP2:=APinMP2(A)\ Then the logarithmic map Hpl maps A™2 isomorphically onto cij, the orthogonal complement of Q2 in ai. If T0 and T are points in ao, set A™2(T0, T) to be the set [a € A~2 : a(H\{a) -T)>0,
ae A2X; m(H\(a) -T)
A?j,
where A^ := Ap,nM2 and Aj := AP]nM2. In particular, for T0 such that -T0 is suitably regular, F(x, T) is the characteristic function of the compact subset ofG(F)\G(A)1 obtained by projecting N0(A) •
MQ(A) 1
• Ap°oG(To,T) • K
ontoG(F)\G(A) 1 . All in all, by Lemma 2 of 3.2.7, we arrive at the following Arthur's Partition. ([Ar2]) Fix P and let T be any suitably point in T0+a+. Then
J]
J]
Pi:P0cPicPSePi(F)\G(F)
F\6X)-TP(H0(SX)-T)=1
VxeG(A).
282
L. Weng
3.2.9
Arthur's Analytic Truncation
Recall that an element T e aj is called sufficiently regular, if for any a e A0, a(T) » 0. Definition. Fix a suitably regular point T 6 a^. If )(X) to be the function (AT4>)(X)
:= £(-i)dim(A/Z) />
£
M&c) •
TP(H(.6X)
- r),
6eP(F)\G(F)
where
0H*) := I
(f>(nx)dn
JN(F)\N(A)
denotes the constant term of
)(x) = ^(-l) d i m ( A / Z ) <M*) • TP(H(X) - T), where the sum is over p
all, both standard and non-standard, parabolic subgroups; (b) If
283
3 Geometric and Analytic Truncations
This is because by definition, all constant terms along proper P : P ±G are zero. Moreover, it is a direct consequence of the Basic Estimation in 3.2.6 for partial truncation, (see e.g. the Corollary there), we have (c) If(f> is of moderate growth in the sense that there exist some constants C, N such that |0(x)| < c\\x\\N for all x e G(A), then so is A r 0 . Fundamental properties of Arthur's analytic truncation may be summarized as follows: Theorem. ([Ar2,3]) For sufficiently regular T in ao, (l)Let(p: G(F)\G(A) -^Cbea locally L1 function. Then ATAT (g) for almost all g.If is also locally bounded, then the above is true for all (2) Let 0i, 02 be two locally Lx functions on G(F)\G(A). Suppose that 0i is of moderate growth and 02 is rapidly decreasing. Then f
ATl(g)-4>2(g)dg= f
JZG(A)G(F)\G(A)
Mg)-ATcf>2(g)dg;
JZO(A)G(F)\G(A)
(3) Let Kf be an open compact subgroup ofG(Af), and r, r' are two positive real numbers. Then there exists a finite subset lX( : i = 1,2,..., N\ C < U, the universal enveloping algebra of g^, such that the following is satisfied: Let be a smooth function on G(F)\G(A), right invariant under Kf and let a € AG(A), g 6 G(A)1 D S. Then N T
\*
where S is a Siegel domain with respect to G(F)\G(A); and (4) The function A T 1 is a characteristic function of a compact subset of G(F)\G(A) J . The key to the proof is that nilpotent groups have simple structure. Say, for (1-3), while a bit complicated, it is based on the discussion in 3.2.6-8. As for (4), which has been considered to be only secondary so far, please see 3.3.2 below.
284
3.3 3.3.1
L. Weng Arthur's Periods Arthur's Periods
Fix a sufficiently regular T e Oo and let 0 be an automorphic form of G. Then by Theorem 3.2.9, A r 0 is rapidly decreasing, and hence integrable. In particular, the integration AT^g)dg
A(cf>;T):= f JG(F)\G(A)
makes sense. We claim that A{
AT
f JZc(A)G(F)\G(A)
A r (A r 0)(g) dn{g).
= f JZo(A)G(F)\G(A)
Moreover, by the self-adjoint property, for the constant function 1 on G(A),
I
ug)-AT(ATyg)dti(g)
Zc(A)G(F)\G(A)
(ATiyg) • (Ar0)(g) dub)
= r JZG( A ) G(F)\G(A)
A r ( A T l W ) •
= f JZG( A ) G(F)\G(A) r
T
since A 0 and A 1 are rapidly decreasing. Therefore, using A T o A r = A r again, we arrive at A(0;T)= f
A7" 1(g) • 0(g) dn(g).
(*)
JZG( A ) G(F)\G(A)
To go further, let us give a much more detailed study of Authur's analytic truncation for the constant function 1. Introduce the truncated subset 1(7) := (ZG(A)G(F)\G(A))r of the space G(F)\G(A) ! by S(T) := (ZG(A)G(F)\G(A))7, := [g e Z G ( A ) G ( F ) \ G ( A ) : ATl(g) = l).
285
3 Geometric and Analytic Truncations
We claim that E(T) or the same (z G( A)G(F)\G(A)) r , is compact. In fact, much stronger result is correct. Namely, we have the following Basic Fact. ([Ar4]) For sufficiently regular T e a+, A r l(*) = F(x, T). That is to say, A r l is the characteristic function of the compact subset S(T) of G(F)\G(A) 1 obtained by projecting N0(A) • Mo(A)1 • A~ G(T0, T) • K onto G(F)\G(A)\ Thus, by the equation (*) and the Basic Fact above, AT(f>(g)dp(g)
Jz, Zc(A)G(F)\G(A)
ATl(g)-4>(g)dfi(g)
= f JzG(A)G(F)\G(A)
= r
JJ.(T)
That is to say, we have obtained the following very beautiful relation: Arthur's Periods. For a sufficiently regular l e a o , and an automorphic form (p on G(F)\G(A),
f JUT)
AT(g)dn(g). 1
The importance of this fundamental relation can hardly be over-estimated. Say, when taking (f> to be Eisenstein series associated with an L2-automorphic forms, (i) (Analytic Evaluation for RHS) The right hand side can be evaluated using certain analytic methods. For example, in the case for Eisenstein series associated to cusp forms, as done in the next chapter, the RHS can be evaluated in terms of integrations of cusp forms twisted by intertwining operators, which normally are close related with L-functions; (ii) (Geometric Evaluation for LHS) The left hand, by taking the residue (for suitable Eisenstein series), can be evaluated using geometric methods totally independently. A good example is given in [KW], as to be explained in 4.6. See also 4.7-8, where this is used to obtain certain new yet fundamental relations among special values of zeta functions.
286
L. Weng
3.3.2
A r l Is a Characteristic Function
Proof of the Basic Fact is based on Arthur's partition for G(F)\G(A) and an inversion formula, which itself is a direct consequence of Langlands' Combinatorial lemma. Recall that for T e do, level P-Arthur's analytic truncation of
^P4>(g) := Z
2
(-D«-^>
fl:/fc/>
**<*> • T£(*<*> - 4
SeR(F)\P(F)
As such A r stands simply for A r , G . Thus, together with Langlands' combinatorial Lemma, we have the following: (a)Ar^):=^-1^(/eW(G) R
Z
M8g)TR(H(6g)-T);
and
6€R(F)\G(F)
(b) (Inversion Formula) For a G(F)-invariant function
#*> = Z
Ar,/
Z
P
V^) • TP(H(<5S) - T).
SeP(F)\G(F)
(c) Assume that T 6 Cp and (m) is rapidly decreasing on M(F)\M(A) 1 . In particular, the integration JM(F)\M(A\i AT,p
Y, P
FP Sx T
Z
( ' )"
T
P{HP(6X)
-T) = l
SeP(F)\G(F)
where r/> is the characteristic function of [H € do : a(H) > 0, a e AP] and Fp[nmk, T) = FMp{m, T),
n e NP(A), m e MP(A), k e K.
On the other hand, by the inversion formula, applying to the constant function 1, we have £ P
£ 6eP(F)\G(F)
(AT'Pl)(6x) • rp(HP(8x) - T) = 1,
287
3 Geometric and Analytic Truncations
where AT'P is the partial analytic truncation operator defined above. With this, the desired result is immediately obtained by induction. As a direct consequence of this proof, we conclude also the following Proposition 1. ([Ar4]) For sufficiently regular T, the characteristic Junction F@(g, T) coincides with the partial truncation along Pfor the constant Junction 1. That is to say, Fp(g,T) = AP>T1 =
sp
£ R-.PocRcP
£
?R(H(6x)-T).
6eR(F)\P(F)
Note that the right hand side makes sense even when T is not sufficiently regular. This then leads to the definition of Fp(g, T) for general
Fp(g,T) = AP-TI=
2
sp
R-.PncRcP
2
rR(Hm-T).
6eR(F)\P(F)
When preparing our Arthur period paper, we note that such a function is compactly supported. (See the lemma below.) As such, a natural question is whether Fp(g, T) is a characteristic function for a natural compact subset. As it appears, the answer is well-known among those working with trace formula: When I addressed it to Arthur in 2005, he immediately gave me his paper [Ar4], which leads the discussion above for sufficiently regular T. For general T, we have the following Proposition 2. ([Ar5], see also [Ho]) As a function of g e G(A)1, Fp(g, T) is bounded and compactly supported modulo U(A)M(F), uniformly for T varying in a compcat subset. Proof The proof is standard. Following Arthur, let
TQp(H,X):= YJ
44(Hrf;(H-X).
R-.PcRcQ
Basic properties of Tp(H,X) are given in Lemma 2.1 of [Ar5], from which, in particular, we know that Tj,{H,X) is compactly supported in if € aj,, uniformly for X varying in a compact subset. On the other hand, by the Langlands Combinatorial Lemma, easily, Fp(g,T + X)=
J]
£
FR(Sg,T)rp(H(6g)-T,X).
R:P0cR
As such, the final result comes from Prop. 1: for sufficiently regular T, Fp(g, T) is a characteristic function of a compact subset. •
288 3.4 3.4.1
L. Weng Geometric and Analytic Truncations A Micro Bridge
For simplicity, we in this subsection work only with the field of rationals Q and use mixed languages of adeles and lattices. Also, without loss of generality, we assume that Z-lattices are of volume one. Accordingly, set G = SLr. For a rank r lattice A of volume one, denote its sublattices filtration associated to a parabolic subgroup P by 0 = A 0 c Ai c A2 c • • • c A|/.| = A. Assume that P corresponds to the partition / = (d\, d2,..., d\p\). Recall that a polygon p : [0,r] -* R is called normalized if p(0) = p{r) = 0. For a (normalized) polygon p : [0, r] —» R, define the associated (real) character T = T{p) of MQ by the condition that
am = [p(i) - p(i - 1)] - [pii + 1) - pit)] for all i = 1,2,... ,r - 1. Lemma. With the same notation as above, T(p) = (p(l) - p(0),p(2) - p ( l ) , . . . , p(i) - p{i - 1), ...,p(r-l)-p(r-2),p(r)-p(r-l)). Proof. Let T = T(J>) = (hip), t2ip),...,
Up)) = (h,t2,...,
tr) 6 do.
Note thato/CT) = f,- - f,+i. Hence U - tM =[pii) - pii - 1)] - [pii + 1) - pii)] =Apii) - Apii + 1),
i = 1,2
r- 1
where Apii) := pii) - pii - 1). Consequently, t\ - ti+\ = Apii) - Apii +1),
i = 1,2,..., r - 1.
289
3 Geometric and Analytic Truncations In particular, h-tr
= Ap(l) - Ap(r) = p(\) - p(r - 1),
and tM =h+ Ap(i + 1) - p{\),
i = 1,2,..., r - 1,
since Ap(l) = p(\) - p(0) = p(\) and A(r) = p(r) - p(r - 1) = -/?(r - 1). But r
r-\
0=YJti = YJ{h-p(l) + &P(i+D) j=l
i=0
=(r-D(h-p(l))
+ {p(0)-p(rj)
=(r-l)(h-p(D). Therefore t\ = p{\) and hence t{+\ = Ap(i +1).
D
Now take g = g(A) e G(A). Denote its lattice by A8, and its induced filtration from P by 0 = A f c A f c - . - c A J p f = A*. Clearly, the polygon p^ = p^* : [0, r] —> R is characterized by (1) /7p(0) = /4(r) = 0; (2) /?p is affine on [r,-, r,-+i], i = 1,2,..., |P| - 1; and (3) Pp(n) = deg(Af) - n • *&?!, i = 1,2,..., |P| - 1. Note that the volume of A is assumed to be one, therefore (3) is equivalent to (3)' pgP(n) = deg(Af p ), i = 1,2,..., \P\ - 1. The advantage of partially using adelic language is that the values of Pp may be written down precisely. Indeed, using Langlands decomposition g = n • m • a(g) • k with n e Np(A),m 6 Mp(A)l,a € A+ and k e K := Up SL(Oqp) xSO(r). Write a = a(g) - di&g{a\Idi,a2Id2
a\P\Idm)
290
L. Weng
where r = d\ + d2 + • • • + d\p\ is the partition corresponding to P. Then it is a standard fact that
deg(Af') = " log ( f ] fl?)
=
- Z ^ l o ^ aJ>
' = 1 • • • • • 1^1-
Set now l(/?p >p /?) to be the characteristic function of the subset of g's such that p8p >p p. Micro Bridge. For a fixed convex normalized polygon p : [0, r] —* R, and g € S Lr(A), with respect to any parabolic subgroup P, we have tp(-Ho(g)-T(p))
=
l(pP>Pp).
Proof. First let us consider the right hand side. By definition, Upp >PP)
= 1
if and only if
ppin) - pin) > o,
pp(r2)
-p(r2)>o,
Pp(r\p\-\) - p(r\p\-i) > 0. That is, deg(AfP) - p{n) > 0,
deg(Af) - p(r2) > 0,
deg(A^_j)-/7(r| P |_ 1 )>0. In this way, we have established the following Subiemma 1. With the same notation as above, Upp >PP) = 1 if and only if
(-rfilogai)-p(n) >0, ( - d\ log ax - d2 log a2) - p(r2) > 0,
( - d\ logai -d2loga2
d\P\-i loga|/>|_i) -p(r\ P \-i) > 0.
291
3 Geometric and Analytic Truncations
Next let us consider the left hand side, that is, the function Tp(-Ho(g) T(p)). When does it take the value one? Recall that for P = P/, 77. is the characteristic function of the following subset of % = O.Q © a/ defined to be a j ® j f f e a , : m&H) > 0 Vi € A - /}. That is to say, if H e % belongs to a£0[H e a7 : m&H) > 0 i =
rur2,...,r\P\-i),
then T > ( # ) = 1. But H e a7 means H = (Hfx),H{*2\...,H^) Therefore, the conditions that H € a/
such that
and JJQ dM = 0.
vJi(H) > 0 for all i = r\, r2,..., r\p\-\
is equivalent to the conditions that \P\
H = (H^\ Ef\...,
flgW)
where
j
dtHt = 0
;=i
and d\ H\ > 0, diHi + d2H2 > 0 , . . . , d\H\ + d2H2 + ••• + d\p\-iH\p\-i > 0. On the other hand, for H = (H\ ,H2,..., Hr\ e ao, the projections of H to a/ (resp. to a!Q) is given by /,Hi+H2 + --- + Hn^dl) fHn+l+Hn+2 + --- + Hr2s ' r#/i f l -i+i
'1
+
^nfi-i+ 2 + • • • + Hw \(d\p<)\
^
) J e a/
292
L. Weng
(resp. Hx + + tiH22 + + •• •• •• + +H H\ tirn, d\ '
I
Hi+H2+-+Hri Hi+H2
d\ + --- + Hn
#r|/>|_i+l + -^n/1-1+2 + • • • + Hm • • • • >
H
r \ P \ - i + \ -
,
\p\ H
+2
m-\
d
^/•i-i =\H\,H2,..
.,HrJ
_ (,Hl+H2
lv ••'[
\P\ ^npi-i+2 + • • • + Hm >
+1 +
+ --- + Hn,(dl)
*
'
(Hn+1
+ Hn+2 + • • • + Hr2^(d2)
'^
J2 )eao->
)
d{p]
^ '
Therefore, Trp(-H0(g)-T(p))
=l
if and only if the coordinates H of -Ho(g) - T(p) =: H = (Hn satisfy the conditions that Hi + H2 + • • • + Hri > 0, (Hi+H2
+ --- + Hn) + (Hn+i + Hri+2 + --- +
(Hl+H2
+ --- + Hn) + (Hn+l + Hn+2 + --- + Hn) + ••• + (^n/i-i+i
+
Hr2)>0,
^n/i-i+2 + • • • + # w - i ) > °-
Recall now that, by definition, -H0(g) - T{p) = (-t\-\ogbx,
-t2 -\ogb2,...,-tr-
log br)
where aig) = diag(fei, b2, ...,br)
= diag( fl < dl) , af\...,
a$f)
293
3 Geometric and Analytic Truncations and Tip)=(tut2,...,tr) = (pil), pil) - pi\), ...,pir-\)-pir-
1), -pir - 1)).
As such, this latest group of inequalities, after replacing Hi's with the precise coordinates of -H0ig) - Tip), are then changed to the following group conditions: ( - log bx - pil)) + ( - log b2 - pil) + pi\)) + ••• + ( - logb n - pin) + pin - l)) > 0 (( - logfcj - pil)) + (-logb2-
pil) + Pil))
+ ••• + ( - log bn - pin) + Pin - 1)))
+ ( ( - iogbn+\ - pin + l) + Pin)) + ( - logfcn+2 - Pin + l) + pin + l)) + ••• + ( - log bn - pin) + pin - l))) > 0
( ( - \ogbx - pil)) + ( - iogb2 - pil) + pil)) + ••• + ( - log bn - pin) + pin - i))) + ( ( - iogfcri+i - pin + l) + pin)) + ( - logbn+2 - pin + 2) + pin + i)) + ••• + ( - log bn - pin) + pin - l)))
+ ((- l o g ^ j + i - /?(>|p|_2 + 1) + pir\p\-i)) + ( - l o g ^PI-2+2 - Pir\p\-2 + 2) + pir\p\-2 + 1)) + ••• + ( - log fcn/>H - pir\P\-{) + pir\Pl-i - 1)) > 0. This, after the obvious simplification with Bx := -logbi,B2
= -logb2
Br - -\ogbr,
rQ := 0,
294
L. Weng
becomes the group of conditions Bi + B2 + --- + Bri -p(n)
+ p(0)>0,
(B1+B2 + --- + Bn) + (fl n+1 + Bn+2 + --- + Br2)- p(r2) > 0,
(Bi+B2 + --- + Bri) + (Bn+1 + Bn+2 + --- + Bn) + BnPl_2+2
+ ••• + Bm_^
- p(r\P\-i)
> 0.
Thus note that (bub2, • • • A ) = a(g) = (a-«\ a2r^\...,
aj™0),
we arrive at the following Sublemma 2. With the same notation as above, TP(-H0(g)-T(p))
=l
if and only if - J i l o g a i -pin)
>0,
( - d\ log ai - d2 log a2) - pin) > 0
( - dx loga! - d2 loga 2
d\p\-i loga|/>|_i) - p(r|p|_i) > 0.
By comparing Sublemmas 1 and 2, we see that fp(-Hoig)-Tip))
=
l(pgp>Pp). D
3.4.2
Global Bridge Between Geometric and Analytic Truncations
With the micro bridge above, now we are ready to expose a beautiful intrinsic relation between algebraic and analytic truncations.
295
3 Geometric and Analytic Truncations
Global Bridge. For a fixed normalized convex polygon p : [0, r] —> R, let T{p) := (Kl), P(2) - p{\),...,
p{i) - p(i - 1),.. •, p(r - 1) - p(r - 2), -p(r - 1))
be the associated vector in do. IfT(p) is sufficiently positive, then Kp8
(AT^l)(g).
Proof. This is a direct consequence of the Fundamental Relation in 3.1.2, the definition of Arthur's analytic truncation and the Micro Bridge in 3.4.1.
•
4 Rankin-Selberg & Zagier Method Global bridge between geometric and analytic truncations in 3.4.2 shows that at least for polygons p whose associated T(p) e ao are sufficiently positive, the corresponding non-abelian L-functions may be viewed as a special kinds of Arthur's periods. Therefore, to understand our Ls, it is quite helpful to understand the structure of Arthur's periods. However, to write down them precisely, Arthur's periods for automorphic forms seems to be too general. It is for this purpose that we then narrow our class by considering Eisenstein periods: By definition, Eisenstein periods are integrations of Eisenstein series over truncated compact domains. Simply put, the aim of this chapter is to show that a special kind of Eisenstein periods and hence also certain classes of Ls can be evaluated via an advanced version of Rankin-Selberg & Zagier method. 4.1
Regularized Integration over Cones
To begin with, we here give a review of Jacquet-Lapid-Rogawski's beautiful yet elementary treatment of integrations over cones. Let V be a real r dimensional vector space. Let V* be the space of complex linear forms on V. Denote by S(V*) the symmetric algebra of V*, which may be regarded as the space of polynomial functions on V as well. By definition, an exponential polynomial function on V is a function of the form r ;=i
where the At are distinct elements of V* and the P,(JC) are non-zero elements of S(V*). One checks easily that (1) The Xi are uniquely determined and called the exponents of f. By a cone in V we shall mean a closed subset of the form C:= [xeV:
(pi,x)>0Vi]
where {p,} is a basis of Re(V*), the space of real linear forms on V. Let {ej\ be the dual basis of V. We shall say that A e V* is negative (resp. 296
297
4 Rankin-Selberg & Zagier Method
non-degenerate) with respect to C if Re {A, ej) < 0 (resp. (A, ej) * 0) for each;'= l , . . . , r . (2) The function f(s) = Ej_i e^^P&x) is integrable over C if and only if A{ are negative with respect to Cfor all i. To define regularized integrals over C, we study the integral
7c(/U):=
ff(x)eMdx.
It converges absolutely for A in the open set [A e V* : Re(At - A,ej) < 0 VI < i < n, 1 < j < r), and can be analytically continued as follows: In the case when Re(A) > 0 and P is a polynomial of one variable,
f Jo
m>0
We then can use the right hand side to obtain the continuation. To deal with more general case, fix an index k and let Ck be the intersection of C with the hyperplane Vjt := \x : {^k,x) = 0j. Then, we can write x = ink, x)ek + y with y e Vjt. Thus, for a suitable choice of Haar measures,
i=im>o[(A-Ai,ek))
Jc
*
This formula then gives the analytic continuation of Icif', ty to the tube domain defined by Re(/l, - A, e*) > 0 for j' ± k, 1 < i < n with singularities on hyperplanes //*,,- := i,A : (A,ek) = *,-,«*>}, 1 < « < «. Moreover, since this is true for all k = l,2,...,r, the function iAf;A) has analytic continuation to V* by Hartogs' lemma with (only) hyperplane singularities along Hit, 1 < k, < r, 1 < i < r. In particular, from such an induction, one checks (3) Suppose that f is absolutely integrable over C. ThenIc(f\A) morphic at 0 and Icif', Oj = fc f(x) dx; and
is holo-
298
L. Weng
(4) The function lAf; AJ is holomorphic at 0 if and only if for all i, A{ are non-degenerate with respect to C, i.e., (At, e^) =£ Ofor all pairs (i, k). Denote the characteristic function of a set Y by TY. For A e V* such that At - A are negative with respect to C for all i = 1,2,..., n, set F(A; C, T) := f f(x) • rc(x - T) • e~{x'x)dx Pi(x)TC(x = J]e-<^T)Ic{Pi(*
T)e~(x-Xi'x)dx + T)e<M;A).
The integrals are absolutely convergent and FyA; C, TJ extends to a meromorphic function on V*. With this, following [JLR], we call the function f(x) • ^(x - T) #-integrable if FLA; C, Tj is holomorphic at A = 0. In this case, set
r-
f(x) • TC(X - T)dx := F(0; C, T).
By (4), the #-integral exists if and only if each exponent Aj is non-degenerate with respect to C. To go further, suppose that V = W\ © W2 is a decomposition of V as a direct sum, and let C, be a cone in Wj. Write T = T\ + T2 and x = x\ + X2 relative to this decomposition. If the exponents Aj of / are non-degenerate with respect to C = C\ +C2, then the function
f
r A , 1(W\ - Ti)d\Vl W2 >-» I f(w\ + W2)T 1 JWi
is defined and is an exponential polynomial. And it follows by analytic continuation that (5) J
f(x)TC(x-T)dx
= \ [ I Jw2 \ JWi
f(wi+W2)TCl,(wi-Tl)dwi)-T^2(w2-T2)dw2; J
299
4 Rankin-Selberg & Zagier Method
and as a function ofT, it is an exponential polynomial with the same exponents as f. More generally, let g(x) be a compactly supported function on Wi and consider functions of the form g{w\ - h)Tc(w2 - T2), which we call functions of type (C). It is clear that the integral e~Mdx
F(A) := f f(x) • g(Wl - h)TC(w2 - T2) •
converges absolutely for an open set of A whose restriction to W2 is negative with respect to C2- Furthermore, F(A) has a meromorphic continuation to V*. Following JLR, the function f(x)g(wi - h)Tc(w2 - T2) is called #integrable if F(A) is holomorphic at A = 0. If so we define J
/(*) • g(wx - h)TC(.w2 - T2)dx := F(0).
Note that as a function of W2, L f(w\ +W2)g{w\ -T\)dw\ polynomial on W2, hence I f(x) g(wi -1\ )TC(W2
Jw Jw, )W22 \ \JWi
is an exponential
-T2)dx
f{w\ + W2)g{w\ - T\)dw\ )T£(vi>2 - T2)dw2.
We end this discussion by the following explicit formulas: (6) J e^\c(x-
Jv
T)dx = {-\)rNo\(e v ue2,...
,er) •
* '
. , where
' 115=1 <^.«y>
C := I £y=i a ;' e ; : aj > 0} and Vol(ei,e2,-•• ,er) is the volume of the associated parallelepiped I 2;=i ajej '• aj
r
cone C c R , J
eM(i
_ T c ( ; c _Tj)dx
=
e
[0,1]). M
e dt =
_ r
eAT
A
and hence for any
x x) C
e<
' T (x - T) dx
since 1 - 1 ° is the characteristic function of the cone - C .
300 4.2
L. Weng JLR Periods of Automorphic Forms
Now let us follow [JLR] to apply the above discussion on integration over cones to periods of automorphic forms. Let Tk(X) be a function of type (C) on ap that depends continuously on k € K, i.e., we assume that there is a decomposition ap = W\® W2 such that T/t has the form gk(\Vl - T\)TC2k{W2 ~ T2)
where the compactly supported function gk varies continuously in the L1norm and linear inequalities defining the cone Czk vary continuously. Also with the application to automorphic forms in mind, let / be a function on the space M(F)N(A)\G(A) of the form s
aj{Hp{a),k)e{^pp'Hp(a))
f{namk) = J ^ / ™ . * ) • ;=i
for n 6 N(A), a e Ap, m € M(A)1 and k e K, where, for all j , (a) (pj(m, k) is absolutely integrable on M(F)\M(A) 1 x K; and (b) Aj e a*p and aj(X, k) is a continuous family of polynomials on ap such that, for all keK, aj{X, k)e(AJ-X)Tk(X) is #-integrable. Remark. This is really nothing but the micro structure for automorphic forms stated more formally at 2.1. The point is also related to the reduction theory: with M(F)\M(A) 1 of finite volume, (pj behaves very well. So the problem when say taking integration is about the A-part, which is affine the exponential polynomial function can be controlled nicely provided the exponents are negative. As such, following [JLR], define the ^-integral
f
f{g)-rk(Hp{g))dg
P(F)\G(A)
Y
f(
f
J-}JK\JM(F)\M(A)1
j(mk)dm)[ f I
\Jap
aj(X,k)Tk(X)dx)dk. J
Remark. The reader should notice that in this definition there is no integration involving JN(F)XN(A) dn : In practice, / is supposed to be the constant term along P, so is iV(A)-invariant.
301
4 Rankin-Selberg & Zagier Method
Clearly, the set £/>(/) of distinct exponents [Aj] of / is uniquely determined by / , but the functions 4>j and a, are not. So we need to show that the #-integral is independent of the choices of these functions. This goes as follows: The function f(exmk)dm
Xv-> I 1 JM(F)\M(A) /M(F)\M(A)>
is an exponential polynomial on dp and the above #-integral is equal to f
M
f(exmk)dm)e-
f
JK Jap \ JM(F)\M(A)1
I
This shows in particular that the original #-integral is independent of the decomposition of / as used in the definition. Moreover, if each of the exponents Aj is negative with respect to Cik for all k e K, then the ordinary integral
f
m-Tk(HP(g))dg
JP(F)\G(A)1
is absolutely convergent and equals to the #-integral by Basic Property (3) of integration over cones listed in 4.1. That is to say, we have the following Lemma. (1) The above ^-integral fp{F)KGWf(g) • Tk(HP(gj)dg is welldefined; and (2) If each of the exponents Aj of f is negative with respect to C2k far all k € K, then the ordinary integral
f
)P{F)\G(A)11 JP(F)\G(A)
f(g)-Tk(Hp(g))dg v
'
is absolutely convergent and equals to the ^-integral. Fix a sufficiently regular element T € at. Then the above construction applies to Arthur's analytic truncation A^'PxP(g) and the characteristic function TP(HP(g) - T) where Y € JHP(G) := A((/(A)M(F)\G(A)), the space of level P = MU automorphic forms. Indeed, with ^(namk) = ^T Qj(HP(a)) • if/j{amk) 7=1
302
L. Weng
for n e N(A),a e Ap,m e M(A)1 and k e K, where £?/ are polynomials and \f/j € 3ip(G) satisfies
Mag) =
e^VW^jig)
TtP for some Aj e aj, and all a e Ap, fM p. lxK A ij/j(mk)dm dk is wellT,p defined, since the function m \-> A if/j(mk) is rapidly decreasing on the space M(F)\M(A) 1 x K. Now Tp is the characteristic function of the cone spanned by the coweights A^, so
1
AT'Py(g)-Tp(Hp(g)-T)dg
JPU P(F)\G(A)
exists if and only if (AJ,VJV)*0
VweAji
and
Aj e SpQV).
The same is true for Ar-/'xI'(g)-T/>(#/>(,?-x)-F) for x e G(A). (For x e G(A), let K(g) € K be an element such that gK(g)'1 e P0(A), then HP{gx) = HP(g) + HP(K(g)x) and hence TP(X - T - HP(K(g)xj) is the characteristic function of a cone depending continuously on g.) Set then IP/W):=
AT'PV(g)-TP(HP(g)-T)dg.
f
For any automorphic form 0 6 !h{G), write £/>(0) for the set of exponents &p{ * 0
VGTV
6 A^, A e &P{
G).
If e J?l(G)*, then I^T((f>p) exists for all P; and define JLR's regularized period by
f
p):=ITG(ct>).
In the follows, let us expose some of the basic properties of JLR's regularized periods. For this, we make the following preparation. As above, for x € G(Af), let p(x) denote right translation by x: p(x)(f>(g) - (gx).
303
4 Rankin-Selberg & Zagier Method
The space 31(G) is stable under right translation by G(A/). Furthermore, p(x)P(namk) = J^ Qj(HP(a)) • ej(mk). j
Since amkx = n*aa'mm'K(kx) for some n* e N(A), we have (ppinamk • x) e^+ppMa)+^a'))
= J ] Qj(HP(a) + HP(a')) •
This shows that Gp(p(x)(p) - &p(
f
JG(F)\G(A) 1
4.3
(g)dg=
f
JG(F)\G(A)1
then $ e 3\(G\ and
<[>(g)dg.
Arthur's Periods and JLR's Periods
For / e 3lP(G) with
f(namk) = ^tjimk)
•
aj{H(a))e{x^pp'H(a))
;=i
where n e N(A), a e AP, m e M(A)1, k e K, and for all j , aj(X) is a polynomial, and (f>j(g) is an automorphic form in 3\p(G) such that
304
L. Weng
(j)j{g) for aeAp, following [JLR], define
f
P(F)\G(A)1 I
m-TP(H(g)-T)dg (pj(mk) dm dk
^ri
JKJM K JM(F)\M(A)! X:
( |
aj(X)e{xi'X)TP{X - T) dx\.
This is well-defined provided that the following two conditions are satisfied: _ v V (i*) (n, tz7 ) * 0 V£ c P, SJ e (A v )£, /i € S e (0); and (ii*) <^, orv) * 0 Vor e A/., A € 6 P (0). Let ${{G)** be the space of e ^(G)**,
X
G(F)\G(A)
= Y(-1)^ G > f 4.4
M*) • rp(//(g) - r)
Bernstein Principle
With above, to understand our L-functions at least for T » 0, i.e., for T € QJ sufficiently positive, we need to study Eisenstein periods. For this, the following result plays a key role. Bernstein's Principle. ([JLR]) Let P = MU be a proper parabolic subgroup and let o- be an irreducible cuspidal representation in the space L 2 (M(F)\M(A) ! ). Let E(g,(f>,A) be an Eisenstein series associated to a cusp form 4> e ^/>(G), A)) are defined.
305
4 Rankin-Selberg & Zagier Method
We believe that this principle is the main reason behind special yet very important relations among multiple and ordinary zeta functions in literature. 4.5
Eisenstein Periods for Cusp Forms
As a direct consequence of Bernstein principle, with the explicit formula for constant terms of Eisenstein series associated to cusp forms, we have the following closed formula for Eisenstein periods associated with cusp forms: Theorem. ([JLR]) Fix a sufficiently positive T e aj. Let P = MU be a parabolic subgroup and cr an irreducible cuspidal representation in L 2 (M(/ r )\M(A) 1 ). Let E(g,(f>,A) be an Eisenstein series associated to , A) dg is equal to (1) 0 if P ^ Po is not minimal; and (2) Vol(( Z a e A o aaa- : aa e [0,1)}) • £ n ^ Z - P ^ w<=W
X I
(M{w,A)
JM0(F)\MO(A)1XK
ifp = pQ = MQUQ is minimal. Consequently, from the very definition, we have the following Theorem'. Fix a convex polygon p such that T(p) e CLQ is sufficiently positive. Then for a P-level cusp form , Ljfitp, X) is equal to (1)0 if P is not minimal; (2) Vol({ Z« £ A„ <*? : aa e [0,1)}) • £ n ^ g U weW X
I
)M0l(F)\M0n(AyxK JMr (A)]xK
(M(w,A)(f>)(mk)dmdk,
ifP-PoMoNo is minimal. We remind the reader that this result is merely the beginning of a quantitive study about our L-functions. Say, (1) When the polygon p satisfies T(p) » 0, we still need to understand LFf{(p, A) where P' D Pisa, parabolic subgroup of G;
306
L. Weng
(2) We have to weak the restriction that T(p) » 0; and (3) We need also to study our non-abelian L-functions for principal lattices, i.e., that for general reductive groups. In fact all this proves to be very interesting and quite fruitful: In (1), Artin L-functions naturally appears; to (2), the above discussion is valid as long as T{p) e aj U {0}; and for (3), no essentially difficulties appear despite its generality.
4.6
Volumes of Truncated Fundamental Domains
As an application, we here recall a recent result of H. Kim and the author on the volumes of Arthur's truncated domains [KW].
4.6.1
The Results
Let then G be a split, semi-simple group of rank r. Recall that for sufficiently regular r e a j , Arthur's analytic truncation A r l coincides with the characteristic function of a compact subset I(T) := (G(F)\G(Ay J c G(F)\G(A) 1 . (See e.g., 3.3.1.) Our result calculates the volume of E(r). In particular, letting T —> oo, we recover Langlands' famous formula on the volumes of fundamental domains G(F)\G(A) 1 . To state the result, for any w e W, let Jw := {a e AQ : wa e AQ] and Pw be the standard parabolic subgroup corresponding to Jw. Let rw := {a e Ao : wa < 0}, and, for i>\, set m :=#{<* > 0 : = /} - #{a > 0,:
= / - 1}, n,> :=#[a > 0, wa < 0 : (p, orv> = i] - #{or > 0, wa < 0 : (p, av> = i - 1}. As usual, denote by A? the discriminant of F, r-i the number of complex embeddings of F, and £F(1) the residue of the completed Dedekind zeta function £F(S) at 5 = 1. Theorem. ([KW]) For a sufficiently positive T e a+, the volume of the
4 Rankin-Selberg & Zagier Method
307
truncated fundamental domain (G(F) \G( A)1J is given by Vol({ J ] aa<*v : aa e [0,1)}) • W b f
["[frOr*
FlaeAo-w^ (1 - <Wp, Or v »
w^Vo
Note that Z(oo) = G(F)\G(A) 1 , and in the above formula, as T -* oo, only the term corresponding to w = 1 survives. In this way, we then recover Langlands' formula on volumes of fundamental domains; Corollary. ([La2]) IfG is split and semi-stable, then the volume of a fundamental domain %for G(F)\G(A) 1 is given by Vol(5) =Vol(j YJ «*<*V : o« 6 [0> D}) oreAo
i>i
4.6.2
The Proof
Apply Theorem 4.5 to the special case where G is a split, semi-simple group and
where, by Gindikin-Karpelevich formula [La4],
Consequently, we have f
ATE(g,l,A)dg
JG(F)\G(A)'
Z w€n
e(W\-PJ)
— — —M(w,A) FLeAo( w * ~P>a >
n
(*) dm.
JM(F)\M(A)i
308
L. Weng
since
{wX+
P'H(-m)) = 1 for m e M(A)1. (Here v := Vol(( £ a e A o a a a v :
e
a a e [0, l)Jj appears because the measure on ao is defined using co-weights Izja : a e Ao}. Indeed, v is equal to the order of the fundamental group of 1 if a $. Ao, a > 0. Hence Res,i=p£(g, I, A) = Jim [~[ (U,a v > - l) • M(w0,A) aeAo
where wo is the longest Weyl group element. Since
""" * " Q & w . «»> + i:
we have
ReS/l=p£(g, l.A) - fr(l)r • ["JfrOT'. i>i
where r denotes the rank of G. Now by 3.3.1, ATE(g,hA)dg=
f
f
JG(F)\G(A)'
E(g,\,A)dg. Jw)
Hence Res^ p ( f JG(F)\G(A)'
Ar£(g,l,A)^)= f
ResA=pE(g,\,A)dg
Js(r)
^ 0
= Vol(Z(T)) • ResA=pE(g, I,A) = fra/ f l ^ " ' • Vol(Z(r)). «>i
On the other hand, by Theorem 4.5, the residue at A = p of the left hand side of (*) is
Vo,( W \r(A)').jta(v£,«-«M M n < £ $ p r >
309
4 Rankin-Selberg & Zagier Method
Here following Tate [T], Vol(r(F)\T(A) 1 ) = (Vol(F*\lJr))" = frW with £>(D := Res,=ifr(j). (See e.g., [T].) Note also that (wA, arv> = (A, (w -1 ar) v ), and for a given w € W, the sets {or € Ao : war < 0} and {a e Ao : war € Ao} are disjoint. Thus, if we let Wo be the set of all w e W such that {a € Ao : war < 0} U {a e Ao : war e Ao} = Ao, and rw := #{a e Ao : war < 0}, then we have lim( V e^-^M(w,A)
Rav>-1 U / T * v\ i )
•2
«wp,ar v >-l)
vfeWo
Therefore, Vol(l(D) = v • (27r)'2rA7 [~[ ^(i)-" 1 ' i>i
To go further, let us recall that: there is a one-to-one and onto correspondence between the elements of Wo and the standard parabolic subgroups ofG. Indeed, given w € Wo, let Jw = {a e A : wa e A] and denote by Pw the standard parabolic subgroup corresponding to Jw. Then P := Pw is the standard parabolic subgroup of G associated to w. Conversely, any standard parabolic subgroup P of G is associated to a unique subset / c A. Set then Dj := [w e W/Wj : wJ > 0}. It is known that the set Dj gives distinguished coset representatives of W/Wj. Let wj be the element of the maximal length in Dj. Then wj e W0 and in fact Wo = {wj : J c A}. We claim that if w e Wo, <wp,arv) < 0 for all a e A - wJw: if ar € A - wJw, then w/(ar) = -J3 for some/? 6 A. But by [Ca], if w/ is the element of the maximal length in W, and wij is the element of the maximal length in Wj, then wj = wiwij. So w-1(ar) = -wijw(J3) since w _1 = wijwwi. If now/? € Jw, then -wijw(fi) € Jw and w~l(a) e /„,. But w~l(a) £ Jw, since
310
L. Weng
a <£ wJw. Contradiction. Consequently, /? g Jw and wijw > 0. Therefore w~x(a) < 0. As such, (<wp,av)-l) = (-l)rk^
Y\
Y\
(l-(wp,av>).
This then completes the proof of the Theorem. The result may also be written as follows, oriented by standard parabolic subgroups; Theorem'. ([KW]) For a sufficiently regular T, Vol(SXr)) = Vol({ £
a
^
• a<* e [0,1))) • {2n)nrdF"5
aeAo
Ua€Ao-WjA^ ~
/cAo
(Wjp,av))
Remark. The formula, as it stands, suggests that the volumes of cuspidal regions corresponding to parabolic subgroups can be written in terms of special zeta values as well. Proof. By the Theorem, this is a direct consequence of the fact that for w e Wo, rw + 2 «,> = 0, and - £ "i = r, the rank of G. Indeed, - 2 «,• is the number of positive roots such that (p, av) = 1. It is exactly r. Similarly, - Z ni,w is the number of positive roots such that wa < 0, and (p, a v ) = 1. It is exactly rw = rankiV So rw + £](-«,• + n,>) = r for each w e Wo. o 4.7
When Parameters Are Small
By definition,
A^CS) := £(-l)<W-*G>
J]
P
SeP(F)\G(F)
^8) MH^
~ T)-
And, as indicated in 3.3.1, to have the formula
r JX(.T)
AT
we need the following basic properties of the truncation operator A r .
4 Rankin-Selberg & Zagier Method
311
Theorem. (Arthur) For T sufficiently large, (1) A r o A r = AT; (2) AT is self-adjoint; (3) A r 0 is rapidly decreasing, if is automorphic on G(F)\G(A); (4) A r l is a characteristic function of a compact subset Z(T) ofG(F)\G(A)] However for geometric truncation using stability, parameters T may not be very regular. So we need to check whether properties (1) ~ (4) still hold for such Ts. Let us look at (1) first. For T sufficiently large, this is proved by Author based on the following Lemma. ([Ar2]) Fix Pi = M\U\. Thenforcpe C(G(F)\G(A)1), I
AT (f>(u\g) du\ = 0
JUi(F)\Ui(A)
unless vj(H{g) - T) < Ofor each m € A/>,. By examining the proof of Arthur, we see that for S Ln, if T is taken to be in the positive chamber or T = 0, the whole proof still works. Indeed, only the part at p. 91 of Arthur's proof [Ar2] should be checked. But then this is quite obvious. (For general reductive groups, the extra element 0 should be replaced by a certain naturally defined element. For details see [Ar5].) Next, let us turn to (2) and (3). Clearly, formal calculations in Arthur's proof work for any T. So the point here is really about the convergence of all integrals involved. Put this in a more general term, convergence for the integrals appeared are really the central part of Arthur's truncation theory. Sure, one more key ingredient needed here is the combinatorial technique used. Luckily, this elementary yet highly non-trivial combinatorial part can be formally applied even when T is not that large. All in all, the up-shot is the so-called Arthur-Langlands partition for the fundamental domain listed at 3.2.8. It is here that we use the condition T being sufficiently regular to make sure that the function Fp(g, T) is a characteristic function of a certain compact subset. While it is a bit involved, however, by checking the proof on convergence very carefully, we can conclude that what one really needs in proving (2) and (3) for smaller T is that, as a function of g, Fp(g, T) is bounded and compactly supported modulo U(A)M(F) uniformly for T
312
L. Weng
varying in a compact subset. But this is then exactly what Prop. 2 of 3.3.2 provides. Finally, Let us turn our attention to (4). This is a very hard one: In our treatment, this is proved using the Global Bridge between geometric truncation and analytic truncation established in 3.4.2 and a micro-global principle called Fundamental Relation established in 3.1.2. As such, we have proved the following Theorem. Let P = MN be a parabolic subgroup ofG = SL„ and let (M, er) be a cuspidal datum. Let E(
JG(F)\G(A)1
ATE(g,
(i) is equal to zero ifP is not a minimal parabolic; (ii) is equal to
Vol({ YJ «<*<*V : 0 < a a < l}) aeAo e<,wA-p,i)
x
n
/ , Ti 1~> U^ ^ , YlaeAoiwA -p,av>
JM(F)\M(A)lxK
M(w,A)
if P is minimal. Consequently, we also have the following Theorem. For any T in the positive chamber a^orT = 0, Vol(Z(T)) = Vol({ J ] aaay : aa e [0,1))) •
(Wd^
aeAo
./cAo
4.8
YlaeAo-wjj(l-(wjp,(XV))
Fundamental Relations Between Special Zeta Values
By Corollary 4.7, with a detailed calculation, we then obtain the following result on the volumes of moduli space AIF^OAFI^), which reveals certain fundamental relations among special values of classical Dedekind and our zeta functions.
4 Rankin-Selberg & Zagier Method
313
Theorem. Let£F,r(l) := ResJ=i£F,rO) = Vol(Mf>(|A/r|5)). Level 1.
fr,i(D
= fr(l);
Level 2.
fr>2(l)
= fr(2)fr(D -
Level 3.
fed)
=
fr(l)fr(l);
fr(3)fr(2)fr(l) -^F(2)^(1)2-^F(2)^(1)2+^(1)3;
Level 4.
fr,4(l)
=
fr(4)fr(3)fr(2)fr(l)
- ^F(3)^(2)^F(1)2 - ^H2)2fr(D2 - ^ F ( 3 ) ^ F ( 2 ) ^ ( D 2 + ^(2)2fr(l)2 + ^F(2)^F(D2 + ^ F ( 2 ) ^ ( D 3 0
9
6
Level r. The r-level relation can be described as follows. (1) %F,r(\) is a degree r homogenous polynomial in %p(f)s, for i = 1,2,... , r; (2) there is a one-to-one and onto correspondence between the terms of this polynomial and partitions of r, or better, standard parabolic subgroups. More precisely, in the case r = r\ + r2 + • • • + r^, the corresponding term consists of three parts, that is, sign • coefficient • product of zeta values; (i) the sign is given by
(-l)N~l;
(ii) The coefficient is
; and (n + r2)(/-2 + r 3 ) • • • (rN-\ + rN) (iii) the product ofzeta values is given by: Ifri > 2, the product ^f(r,)^f(r,— 1) • • -£F(2) does appear. Moreover, these subfactors together with certain powers of %F{\) then gives all the product of zeta values (under the constrain that the final product is of degree r). It is well known that following a result of Siegel, in order to read the special value £/K«), we have to go to rank n lattices. The above relations may be viewed as a refined version of this basic result. Roughly put, our result says that
314
L. Weng
(i) among all rank ^-lattices, only stable lattices are essential; (ii) other rank n lattices can be built up from stable lattices of lower ranks; (iii) corresponding modular points of lattices in (ii) are sitting in cuspidal regions; and (iv) volumes of cupidal regions can be expressed in terms of £Hm)'s, m < n. We end this discussion by pointing out that our result above is closely related with the result of Harder and Narasimhan ([HN]) on counting stable bundles over curves on finite fields: Their method, an arithmetic geometry one, is algebraic while ours, a geometric arithmetic one, is analytic.
5 High Rank Zetas and Eisenstein Series 5.1
High Rank Zetas for Number Fields
Let K be an algebraic number field (of finite degree n) with A# the absolute value of its discriminant. Denote by OK the ring of integers. For a fixed positive integer r e N, denote by Atf.r the moduli space of semi-stable OKlattices of rank r with dp the associated (Tamagawa type) measure (induced from that on GL). Recall that for each A e Af/r,r, the associated 0-th geoarithmetical cohomology h°(K, A) is h°(K, A) := log ( YJ exp ( - * £ xeA
ll^ll^ - 2* £ \\xa\\2Pcr))
a-.R
cr:C
where x = (x0-)cr€sij<) and (pa-)cresm denote the cr-component of the metric p = PA determined by the lattice A with 5«, a collection Archimedean places of K; and that the rank r zeta function %KAS) °fK1S t n e integration
f^™- 1 ) • M"l08V°1(A) <WA), Re W > 1.
€*A* •= f or the same, Ms)
:= (A* )' • f
(e"°^)
• (e-fSW
dp(A), Re (,) > 1.
JAeMK,r
Theorem. (1) (Meromorphic Continuation) £KAS) is well-defined when Re(s) > 1 and admits a meromorphic continuation, denoted also by %KAS)> to the whole complex s-plane; (2) (Functional Equation) £*,,.( 1 - s) = &>(.$); (3) (Singularities & Residues) £KAS) has two singularities, all simple poles, at s = 0,1, with the same residues Vol(Aljt>r([A|])J. r
Here, as before, Atjf>r([A|]) denotes the moduli space of rank r semistable Ojf-lattices whose volumes are fixed to be A£. The proof of this theorem is fairly standard now since we have a good arithmetic cohomology which satisfies the Serre duality and the Riemann-Roch formula. For details, see e.g., 1.4.2. 315
316 5.2
L. Weng Space of 0/r-Lattices
Recall that an OK -lattice A consists of a underlying projective (^-module P and a metric structure on the space V = A ®z R = rio-eSeo Va-- Moreover, in assuming that the 0#-rank of P is r, we can identify P with one of the (r— 1 ^
A" := ^r;o, := 0K © o„ where a,, i = 1,... ,h, are chosen integral OKideals so that |[ai], [a2], . . . , [a/,]} = CL(^T) the class group of K. With this said, following Minkowski, we obtain a natural embedding for/5: P := tf£_1) © A M A*r) *-> (Rri x C r2 ) r 3 (R'")''1 x (C")1"2, which is simply the space V = A ®z R above. As a direct consequence, our lattice A then is determined by a metric structure on V = Yla-es^ V or the same, on (R r ) n x {C)n. It is well known that all these metrices are parametrized by the space (GL(r,R)/<9(r))n X (GL(r, C)/t/(r)) r \ Next let us shift our discussion from GL to SL. For this, we start with a local discussion on OK -lattice structures: For complex places T, by fixing a branch of the n-th root, we get GL(r, C)/f/(r) = (SL(r, C)/S U(r)) x R*+; while for real places cr, we have GL(r, R)/0(r) s (sUjr, R)/50(r)) X R^. Consequently, metric structures on V = Uo-es^ V
1
x
(CT 2 are
((5L(r, R)/50(r)) n X (SL(r, C)/s t/(r)) r2 ) X (R;) ri+r2 Furthermore, when we work with (^-lattice strucures A = A(P) on P, from above parametrized space of metric structures on V = Yl
X (S L{r, C)/S U(r)f} x (R;) ri+r2
317
5 High Rank Zetas and Eisenstein Series
Hence we need (a) to study the structure of the group AutoK(0^~l) © a) in terms of SL and units; and (b) to see how this group acts on the space of metric structures ((SL(r,R)/SO(r))ri
X (SL(r,C)/SU(r))n)
X (R*+1+/-2 )ri<
View Auto*(^ _ 1 ) ©a) as a subgroup of GUr, K). Then, A\itoK(0^~l)® a) = {A<=GL(r,K)tl
0K \a * •••
detA € UK a
a 0K)
+
Introduce groups GL (r,R) := {g e GL(r,R) : detg > o} and 0 + (r) := {A e 0 0 , R) : detg > o}, and also the subgroup A u t ^ ( 0 ^ _ 1 ) © a) of AutoK(P^~ } © a) consisting of these elements whose local determinants at real places are all positive. Then, by checking directly, we obtain a natural identification of quotient spaces between AutoK(0(K-l) © a)\((GL(r,R)/0(r)) ri x (GL(r,C)/U(r))r2) and Aut
OK(°K~l) © a)\((GL + (r,R)/0 + (r)) ri X (GL(r,C)/U(r))n).
Recall that for a unit e € UK, (a) diag(e,..., e) € AutoK(0%.~l) © a); and (b) detdiag(e,..., e) = sr e UrF := {er: e e t/jdSimilar to the subgroup GL+, introduce a subgroup UK of UK by setting £/£ := {E<EUK:s(r> 0, V
Aut^O^
© a). Thus, choose elements Ml,. . . , M ^ K ) eUK
318
L. Weng
such that | [ « i ] , . . . , [«^(r,A:)]} gives a complete representatives of the finite quotient group UK/(uK n C/^), where fi(r, K) denotes the cardinality of the group UK/(UK n UrK). Set also SL(0K~l) © a) := SL(r, /Q n GL(0K~l) © a). Then we have the following Lemma. ([We6,8]) There exist Ai,... ,A^r>K) in GL + ((9^ _1) © a) such that (i) det Ai = ut for i = 1,... ,ju(r, A"); and (ii) \A\,... tA^icu is a completed representative of the quotient resulting from A u t + ^ O j r 0 © a) modulo SL{0^~X) © a) x (ur£ • diag(l,..., 1)). That is to say, for automorphism groups, (a) AutJ (<9^ © a) is naturally identified with the disjoint union U ^ f }A( • (sL(0K~l)
© a) x (urK+ • diag(l
1)));
and, consequently, (b) The (^-lattice structures A(P) on the projective OK -module P = 0£~ ©a are parametrized by the disjoint union
VJ^Ai^SUd^
© a)\((SL(r,R)/SO(r)f
x (5L(r,C)/5f/(r)) r2 )) x (\UrK n £/+|\(R;) n+r2 )).
Therefore, to understand the space of OK -lattice structures, beyond the spaces SL(r, R)/S 0(r) and SL{r, C)/S U(r), we further need to study (i) the quotient space \UrK n ^|\(R.+) , which is more or less standard (see e.g. [Ne] or [We6,8]); and more importantly, (ii) the quotient space SL(0K~l) © a)\((SL(r,R)/SO(r)f
X
(SL(r,C)/SU(r)f2\
319
5 High Rank Zetas and Eisenstein Series 5.3
High Rank Zetas and Epstein Zetas
Choose integral Ox-ideals <x\ = OK, 02, • • •»°-h such that the ideal class group CL{K) is given by {[ai],..., [a/,]}, and that any rank r projective Otf-module P is isomorphic to Pai for a certain i, 1 < i < h. Here, Pa := Pr,a '•= 0^ - 1 ) ©o for a fractional (9^-ideal a. Introduce the partial high rank zeta function gK,r,a(s) associated to a by setting ^ ; « ( . ) := f
(
^
- 1) • (*"')" l0SV ° 1(A) ^(A),Re(,) > 1
where A1/f,r(a) denotes the part of the moduli space of semi-stable OKlattices whose corresponding modular points lie in (SL(0{K~1) © a)\((SL(r,R)/SO(r))ri
x
(SL(r,Q/SU(r)f2)"j
x(\UrKnUi\\