{2. Metaplectlc Groups and .Representations.

In [21] Kubota constructs a non-trivlal two-fold covering group

of GL2(g) over a totally imaginary number field. In this Section

I shall describe the basic properties of such a group over an

arbitrary number field and complete some details of Kubota's con-

struction at the same time. I shall also recall Well's construction

in a form suitable for our purposes.

I start by discussing the local theory and first collect some

elementary facts about topological group extensions.

Let G denote a group and T a subgroup of the torus regarded

as a trivial G-space. A tvp-cocyc.le (or multiplie.r, or factor set)

on G is a map from Ox G to T satisfying

(2.1) ~(glg2,g3~(gl, g2 ) = c(gl, g2g3)~(g2, g3)

and

(2.2) ~(g,e) =~(e,g) = 1

for all g'gl in G . in additlonl if G is locally compact, a

will be called Borel if it is Borel measurable.

Following Moore [28], let Z2(G,T) denote the group of Borel

2-cocycles on G and let B2(G,T) denote its subgroup of "trivial"

cocycles (cocycles of the form s(gl) s(g2)s(glg2 )-I with s a map

from G to T). Then the quotient group H2(G,T) (the two-

dimensional cohomology group of G with coefficients in T)

represents equivalence classes of topological coverings groups of

G by T which are central as group extensions.

To see this, let ~ be a representative of the cohomology

class ~ in H2(G,T). Form the Borel space GX T and define

multiplication in GX T by

(2.3) [gl'~;1][g2'~;2) = [glg2'~(gl'g2)r

 12

One can check that G• T is a standard Borel group and that the

product of H a a r m e a s u r e s on G and T is an invarlant measure

for G X T. Thus by [25] G• T =~ admits a unique locally compact

topology compatible with the g i v e n Borel structure.

Note that the n a t u r a l maps from T to ~ and ~ to G are

continuous. (They are homomorphisms, and o b v i o u s l y Borel, hence

they are a u t o m a t i c a l l y continuous.) The latter map, moreover,

induces a homeomorphism of ~/T with G. Thus we have an exact

sequence of locally compact groups

i ~ T -~ ~ - ~ G-~ i .

This sequence is c e n t r a l as a group extension and depends only on

the e q u i v a l e n c e class of e . Its natural Borel cross-sectlon is

~: g ~ (g, I)
2. I. Local T heor[.

Let F denote a local field oT zero characteristic. If F

is archimedean, F is ~q or C; if F is non-archimedean, F

is a finite algebraic extension of the p-adic field Qp

If F is non-archlmedean, let 0 denote the ring of integers

of F, U its group of units, P its m a x i m a l prime ideal, and

a generator of P . Let q = lwl-I denote the residual characteristic

of F .

The local m e t a p l e c t l c group is defined by a two-cocycle on

GL2(F) which involves the Hilbert or quadratic norm residue symbol

of F .
o

The Hilbert sy~ibol (',') is a symmetric b i l i n e a r map from

F xx Fx to Z2 which takes (x,y) to I iff x in Fx Is a norm

from F(J~). In particular, (x,y) is identically 1 if y is a

square. Thus (',') is trivial on (FX) 2 • (FX) 2 for every F and

trivial on Fx• Fx itself if F = ~ .

Some properties of the Hilbert symbol which we shall repeatedly

use throughout this paper are collected below.

Proposition 2.1. (i) For each F, (',.) is continuous,

(2.41 (a,bl = (~,-ab) = (a,(l-a)bi ,

and

(2.5 (a,b) = (-ab,a+b) ;

(ii) If q is odd, (u,v) is identically i on U x U;

(ill) If q is even, and v in U is such that vml(4),

then (u,v) is ident.lcally 1 on U .

The proof of this P r o p o s i t i o n can be gleaned from Section 63

of O'meara [31] and Chapter 12 of A r t i n - T a t e [i].

ab
Now suppose s = [cd] 6 SL2(F) and set x(s) equal to c

or d according if c is non-zero or not.

 14

Theorem 2.2. The map ~: SL2(F) • SL2(F) ~ Z 2, defined by

(2.6) ~(Sl, S2) :(X(Sl),X(S2))(-X(Sl)X(S2),X(SlS2) ) ,

is s Borel two-cocycle on SL2(F ). Moreover, this cocycle is

cohomologically trivial if and only if F = @ .

This Theorem is the main result of [20]. According to our

preliminary remarks it determines an exact sequence of topological

groups

1 ~ z2 ~ s~2(F) ~ SL2(F) ~ I

where SL2(F) is realized as the group of pairs [s,~] with multl-

plication given by (2.3). The topology for SL2(F), however, is

not the product topology of SL2(F) and Z2 unless F = @ . Indeed,

suppose N is a neighborhood basis for the identity in SL2(F).

Then a neighborhood basis for SL2(F) is provided by the sets (U, 1)

where UeN and ~(U,U) is identically one.

Proposition 2. 3 . If F/@, each non-trivial topological

extension of SL2(F) by Z2 is isomorphic to the group SL2(F )

just constructed.

Proof. If F =~, each such extension is automatically a con-

nected Lie group, hence isomorphic to "the" two-sheeted cover of

SL2(F) obtained by factoring its universal cover by 2~. Suppose,

on the other hand, that F is non-archlmedean. What has to be

shown is that H2(SL2(F),Z2) = Z 2. For this we appeal to a result

of C. Moore's.

Let EF denote the (finite cyclic) group of roots of unity -

in F . Consider the short exact sequence

l~Z 2 ~ ~ / z 2~ l

The corresponding long exact sequence of cohomology groups is

 15

'''~I(s~(F),~/Z2) ~2(SL2(F),Z 2) ~H2(SS2(F),S ;)

H2(SL2(F),EF/Z2) +H3(SL2(F),Z2) ~''' .

Recall that HI(sL2(F),T ) =Hom(SL2(F),T). Moreover SL2(F) equals

its commutator subgroup. Therefore Hl(sL2(F),EF/Z2) =[1] and
H2(SL2(F),Z 2) imbeds as a subgroup of H2(SL2(F),EF). But from
Theorem 10.3 of [29] it follows that H2(SL2(F),EF) = E F . Thus the
deaired conclusion follows from the fact that H2(SL2(F),Z 2) is
non-trlvial and each of its elements obviously has order at most
two. []

Remark 2.4. As already remarked, a non-trlvial two-fold cover

of SL2(F) for F a n on-archimedean field seems first to have

been constructed by Well in [47]. His construction, which we shall

recall in Subsection 2.3, is really an existence proof. His general
theory leads first to an extension

where T is the torus and Mp ( 2 ) i s a group o f u n i t a r y operators

on L2(F). Then it is shown that Mp(2) determines a non-trlvlal

element of order two in H2(SL2(F),T).

Remark 2. 5 . In [20] Kubota constructs n-fold covers of SL2(F ).

His idea is to replace Hilbert's symbol in (2.6) by the n-th power
norm residue symbol in F (assuming F contains the n-th roots
of unity). In [29] Moore treats similar questions for a wider range
of classical p-adic linear groups.

Now we must extend ~ to

a = as2(F)
In fact we shall describe a two-fold cover of G which is a trivial
non-central extension of SL2(F) by F x, i.e. a seml-direct product
 16

of these groups.
ab i 0
If g = [cd] belongs to G, write g = [0 det(g) ]p(g) where

(2.7) P(g) = [ ca ~cSL2(F)

For gl'g2 in G, define

(2.8) cz*(gl, g2) =ct(p(gl )det(g2) , p ( g 2 ) ) v (det(g2),p(gl))

where

I 0 -I I 0
(2.9) sY = [0 y] S[o y]
and
=f 1 i~ e/O
(2. lO) v ( y , s) \ (y,d) otherwise

if s = [c ]' Note t h a t the restriction of c~~ to SL2(F) X SL2(F)

coincides with a 9

Proposltion 2.6. If y c F x, and ~=[s,~] c~2(F) , put ~Y

equal to {sY,~v(y,s)]. Then s-~s y is an automorphism of SL2(F),

and the seml-direct product of SL2(F) and Fx it determines is
isomorphic to the covering group ~ of G determined by (2.10).

Proof. It suffices to prove that

~ ( s l, s 2) = ~ ( s l Y , s2Y) v ( y , s 1) v ( y , s 2) v(y, s 1 s 2)

and this is verified in Kubota [21] by direct computation.

Remark 2.7. We shall refer to ~ as "the" metaplectic group

even though there are several (cohomologically distinct) ways to

extend ~ to G 9 We shall also realize it as the set of pairs

[g,~] with g ~ G, ~ ~ Z 2, and multiplication described by

[gl,~l][g2,~2] = {gl, g2,~*(gl, g2)~l~2] 9

Now let B,A,N, and K denote the usual subgroups of G 9

Thus
 17

A= , ai~F ,
a2

a2

and K is the standard maximal compact subgroup of G (U(2) if

F =~, 0(2) if F =R, and GL(2,0) otherwise).

If H is an[ subgroup of G, H will denote its complete

inverse image in ~ . Moreover, if ~ splits over H, then

is the direct product of Z2 and some subgroup H' of

isomorphic to H . We shall denote H' by H even though H'

need not be uniquely determined by H .

In general, it is important to know whether or not ~ splits

over the subgroups listed above. In particular, the Proposition

below is useful in constructing the global metaplectic group. (Recall

that if x belongs to a non-archimedean field its order is defined

by the equation x=w~ u eU.)

Proposition 2.8. Suppose F/C or ~ and N (as usual)

is a positive integer divisible by h . Then ~ splits over the

compact group

= ab
[[c d ] 6K: a ~l, c ~0(mod ~)]

ab
More precisely, for g = [c d ] e G, set

a b f(c,d det(g)) if cd~0 and ord(c) is odd

(2.11) s([ c d]) -
L 1 otherwise

Then for all gl, g 2 e K N ,

(2.12) ~*(gl, g 2) = S(gl)s(g2)s(glg2) -1

Proof. Theorem e of [21] asserts that (2.12) is valid for all

ab ab
gl, g 2 in [[c d ] e K: [c d ] ~ 12(m~ N)]. But careful inspection
 18

of Kubota's proof reveals that the conditions b--0(mod N) and

d i l(mod N) are superfluous. Indeed Kubota's proof is computational

and the crucial observation which makes it ~ork is the Lemma below.

We include its proof since Kubota does not.

Lemma 2..9. Suppose a b d ] c KN 9

k = [c Then

s(k) =f(c,d(det k)) if c%0 and c~U

(2.13) \ 1 otherwise.

Proof. Throughout this proof assume [ac ~d]b KN . In particular,

dot(k) --ad-bc belongs to U .

Suppose first that c~0 and c ~U . Then clearly cd~0 .

Indeed d =0 implies that dot(k) =-bc, a contradiction since

c~U implies -bc~U ~ Thus if order (c) is odd,

s(k) = (c~d(det(k))

by definition.

On the other hand, if ord(c) = 2 n (where n~0 since c~U)

it remains to prove that

(c,d(det k)) = 1 .

But d(det k) = a d 2 - d b c , so kcK N implies that d(det k) -~l(mod 4).

Thus

(c,d(det(k))) = (w2nu, u ') = (u,u') = i

by (iii) of Proposition 2.1.

To complete the proof it suffices to verify that c ~0 and

c~U if cd~0 and ord(c) is odd. This is obvious, however,

since if c cU, then ord(c) =0, a contradiction, since ord(c)

must be odd. F~
Note that when the residual characteristic of F is odd,

~ - - K : GL(2,0F).
 19

Definitien 9.10. If F =~ or C, let s(g) denote the

function on G which is identically one. If F is non-archimedean,

let s(g) be as in (9.11), and in general, let ~(gl, gg) denote

the factor set ~*(gl,g2 ) s(gl) s(g2)s(glg 2)

Obviously ~ determines a covering group of G isomorphic

to ~ . But according to Proposition 2.8, its restriction
KNX KN is identically one. Thus ~N is isomorphic to K~• Z 2 ,
and KN lifts as a subgroup of ~ via the map k ~ {k,l]. For
this reason we shall henceforth deal exclusively with ~ .

Lemma 9. II. Suppose

F
~i xi]
gi = ~ B, i=1,2
o

Then ~(gl, g2) = (~i,~2)

Proof. Since

~i xi = 0 ~i xi

0 hi ~i~DL 0 ~[l

det(gi) P(gi) ,

it follows that

-i
= (~{I,~I)(-WIIw2 ,

But using (2.4) together with the symmetry and billnearlty of Hllbert's

symbol, this last expression is easily seen to equal (Wl,~2).

Thus the Lemma follows from the identity
 20

(2.1~) sIE~0 ~Jl = 1

valid for all [ ~] r B. []

Corollary 2.12. Suppose Y = [~ 0]. Then V = {Y,~ is

central in ~ iff y is a square in A.

Proof. Suppose V' = {Y',C'] = [[~0' O,],C, is a r b i t r a r y in

~. Then V V' = Y' Y iff {YY',~(Y,Y')CC] = [Y'Y,~(Y',Y)~'] iff

/3(y,y') = #3(y',y) iff

(2.z5) (~,~,) = (~,,x)

4

So suppose first that y is a square in A. Then both ~ and

are squares in Fx and (2.15) obtains by default (both sides equal

one for all ~',k').

On the other hand, if ~, say, is not a square in F x, then

by the n o n - t r i v i a l i t y of Hilbert's symbol,

(~,k') = -i for some h' ~ F x

This means that (2.15) fails for ~' = I, say. Thus V will not
I 0
commute with {[0 ~,],I} and the proof is complete. []

Corollary 2.12'. The subgroup N of G lifts as a subgroup

of ~. So does A2 with

A 2 = {y ~ A: y = 8 2 , 6 ~ A] 9

Proof. Obvious.

Corollary 2.13. Fix F J g. Then:

(a) The center of ~ is

z o (F x) 2]
z(g) = {[[o z]'g]: z ~

(b) Suppose

-'2
G = [{g,(:] ~'G: det(g) c (FX) 2]

Then the center of -J2

G is g = [[[~ 0z ] , ( ~ ] : z ~ F x] 9
 21

Proof. Since g=[g,~] eZ(~) only if geZ(G), i.e. only

if g = [~ oZ] with z eF x, it suffices to prove that {[0z o],r
z
commutes with ever[ g' =([ca b ],~') e G iff z
is a square in Fx
Clearly [[~ 0],~] commutes with g' iff ~([0z O z ]' g, ) =
~(g,, [~ oz]), But a simple computation shows that

a,(K >= <z,c/

and
0
~*(g',[~ Z]) = (z,c)(z, det(g')

So since

s(g,)-ls([~ o z o z o z o g,),
z])-Is(g'[o z])=s([ 0 z ] -Is(g' )-iS( [o z ]

0
{[~ z],(~} ~ Z(~) if and only if

(2.16) (z, det(g')) = i

for all ~' ~ G, or, since Hilbert's symbol is non-trivial,

[ [~ oz],~] c z(~) iff z is a square. On the other hand, if
det(g') ~ (FX) 2, (2.16) holds for all z, and hence (b) too is
established. []
2.2. Global Theory.

In this paragraph, F will denote an arbitrary number field,

v a place of F, F v the completion of F at v, and ~ the

adeles of F. Thus the G of Section 2.1 becomes G v = GL2(Fv),

its maximal compact subgroup is K v, A is A v, etc. Our interest

henceforth is in the global group

G~ = GL 2 ( ~ )

and its two fold cover %.

If g and g' are arbitrary in G~, put ~v(g,g') = ~(gv, g~),

if g = (gv) and g' = (g~). According to Proposition 2.8,

~v(g,g' ) = i for almost every v. Thus it makes sense to define

on %• by

(2.17) ~(g,g') = ~ ~v(g,g')

v

the product extending over all the primes of F. Since ~ is

obviously a Borel factor set on GA it determines an extension

of G~ which we shall denote by ~A and realize as the set of

pairs [g,~], g e G~, ~ e Z 2, with group multiplication given by

{ g l , ~ l ] { g 2 , C2] = [ g l g 2 , ~ ( g l , g2)cl~2 ].

Important subgroups of G~ include

N = H KN
KO v<~ V

and

GF = GL 2 ( F ) .

Proposition 2..!~.. The subgroup ~0 of G~ lifts to a subgroup

of ~ via the map k 0 ~ [ko, l]-

Proof. Proposition 2.8.

Proposition 2...15. For each u e GF, let

s~(u = ~
v
Sv(V)
 23

the p r o d u c t extending over all (finite) primes v of F. Then

the map

V ~ {V, s~(v) ]

provides an i s o m o r p h i s m b e t w e e n GF and a subgroup of %"

Proof. Note first that if Y = [c

a d
b ] e GF, the v-order of

c is 0 for almost every v. Thus Sv(Y ) : i for almost every

v and the product appearing in (2.18) is finite, i.e. sA(y ) is

well-defined. To prove the P r o p o s i t i o n it will suffice to prove that

(2.18) s~(y)sA(y')/~(y,V') = s~(YV')

for all y, y' in G F.

So fix y and y' in G F. For almost all v, all the entries

of y and y' will be units. Thus ~v(y,y') = i for almost all v.

(By P r o p o s i t i o n 2.1, Hilbert's symbol is trivial on units if v

is finite and the residual c h a r a c t e r i s t i c of Fv is odd.) On

the other hand, for all v,

~v(y,u = (rl, r2)v(rs,r~) v

with r I, r 2, r 3, r 4 e P. Consequently, by the p r o d u c t formula

for Hilbert's symbol (quadratic reciprocity for F),

~v(y,y,) = 1 .
v
That is,

~(~,y' ) : s~(x)s~(~' ) s~(yx' )

as was to be shown. []

Because of this P r o p o s i t i o n we can make sense now out of the

homogeneous space

where Z 0oo denotes the subgroup of the center of G oo ,consisting of

positive real matrices [~ 0]. (Cf. Corollaries 2.12' and 2.13. )

This space will be the focus of our a t t e n t i o n in Section 3.

 24

The P r o p o s i t i o n below makes it possible to relate functions

on ~ with classical forms defined in the upper half-plane. Before

stating it we collect some facts concerning the quadratic power

residue symbol.

Suppose a,b ~ F x with b relatively prime to a and 2 9

Let S denote the set of a r c h i m e d e a n places of F. Suppose that

Ordv( b )
(b) = I1 v
v
where (b) denotes the F-ideal generated by b and the product extends

over all prime ideals of OF (the ring of integers of F). 0nly

f i n i t e l y many v will be such that Ordv(b) / 0 and each such v will

be relatively prime to 2 and a .

The quadratic power residue symbol (a) is then 1 if x 2 =a

has a solution in 0v and -i otherwise. The quadratic power residue
symbol (b) is
ordv(b)

v/S
42
Now consider the congruence subgroup

(2,20) rl (~) = {[ ~ bd] { SL(2,0): a ~ d ~ 1, o ~ O(N)].

Clearly

(2.21) rz(N) : a F n a ~ I<oN

where GO
~ = ~ Gv,
0 the product of the connected components of
vES
9 ~ a b
the a r c h z m e d e a n completions of Go Note that for any [c d ] in

SL(2,0), d is relatively prime to c and 2o

Proposition 2.16. If d is relatively prime to c and 2, let

(2.22) (-~)s : (-~) n (c,d)

VE S V

If y = [a b] belongs to FI(N) , put

(2.23) x(y) =
((~) ife/0
s
otherwise.
 25

Then

(2,24) %(Y) = x(~)

for ~ rl(~)-
Proof. From the definition of everything in sight it follows
tha t
s~(~) = n Sv(~) = n sv(Y)
v v finite

Thus the Proposition is obvious if c = 0 (since both sides

of (2.24) are then one).

Assume now that c / 0 and recall that c and d are

relatively prime. By Lemma 2~

#c,d, det(y)) v if c / 0 and c ~ U

v
Sv(Y)
= t~ 1 otherwise

Therefore
= I(~ aet Y)v if vfc
sv(~)
otherwise

and consequently

s&(y) =v~c(C,d(dety ))v"

By assumption, det(Y) is a unit for each v. Moreover,

det(y) m I(4). Thus by Proposition 2.1, and the product fromula
for Hilbert's symbol

vINc(c'det Y)v = v~c (c'det(Y)) v~SH (c, d e t y ) = I.

Consequently v ~c (c,d(det y) =v~c (c,d)v, which means it suffices

to prove

(2.25) v~c(C'~)v = (~)s-

 26

To simplify matters, assume F has class number one.

(The Proposition is undoubtedly true without this assumption.)

To prove (2.25) recall the following basic formulas ([14],

Chapter 13): for each prime integer w in F dividing both

c and 2,

(2.26) (~) = ~ (w,d)v

YES
v]2
(Supplementary Reciprocity Law)~ for each prime w which divides

c but not 2,

(2.27) (~) = (w,d) w H (w,d)v

vcS
v12
(Power reciprocity formula). Note that if d is a v-unit for

odd v dividing c, (w,d)v = i. Consequently (2.26) can be

rewritten as

(2.28) (~) = H
vlc (w,d) v ~
ws (w,d) v

vZ2 vl 2 ~

ordw(C)
Now suppose c : ~ w By the m u l t i p l i c a t i v i t y
wLc
of the power residue symbol, (2.28) and (2.27) can be multiplied

for each w]c to obtain

= cIC,dlvl
(v V~Soo
(c,dlv.

Thus the proof is complete. (Units play no role since [ac ~] ~ FI(N)

implies (u,d)v = i.) []

Two special cases of Proposition 2.16 will be of particular

interest in SeCtion 3. The first assume F is a totally imaginary

number field, in which case

(2.29) X(Y) = I (~) if c ~ 0

0 otherwise
l
 27

since (c,d)~ is identically one.

Corollary 2.17. (Cf. Kubota [19], cf. Theorem 6.1 of Bass-

Mi!nor-Serre [2]). If F is totally imaginary, X(u defines a

Character of FI(N ), i.e. X(u165 : X(YI)X(Y2) for all u165

Proof. Since F is totally imaginary, ~ is identically

one on G0K
~ N0 in G~. So suppose u and Y2 are arbitrary in
FI(N ). Since YI and u belong to GF,

s~(~l~ 2) : ~(yl,u165

(by (2.18)). But ~A(yI,u = i. Thus the Corollary is immediate

from Proposition 2.16. []

The existence of such a character (with the assumption that

F is totally imaginary) provides a starting point for Kubota's
recent investiations ([21] through [23]). Its relation to the
congruence subgroup problem is discussed in [2].
The second example I have in mind assumes F = Q. In this case,
if c 0, c > 0 or d>0

X(') = I - ( i ~ if
0 c f=i 0.
0' c f( 0 and d (

Now X no longer defines a character of FI(N) (~ is not a

trivial extension of G ). However X(Y) is still a "multiplier
system" for ~I(N). More precisely, if u = [a ~], let J*(y,z) =
C

(ez+d) I/2. Here w I/2 is chosen so that -~/2 < arg(w I/2) ~ ~/2.

Then IX(Y)1 : I and

X(Ylu J~ (u J~(u z)
(2.31) ~I~l)~(Y2) j* (~1u ,z)

for all y ~FI(N). I.e., X(~) is a multiplier system for FI(N)

of dlmension 1/2.
2.3. Well's Metaplectic Representation.

In this Section I want to sketch Well's general theory of the

metaplectic representation and reformulate parts of it in a form

suitable for the c o n s t r u c t i o n of automorphic forms on the m e t a p l e c t i c

group.

Roughly speaking, to each abstract symplectic group, one

associates a projective representation (Weil's m e t a p l e c t i c represen-

tation). This representation operates in L2 of the group and its

associated multiplier is of order two. Thus Well's representationre-

produces a two fold covering group which repreduces ST2(F) when the

underlying group is SL(2).

We begin by recalling some basic properties of projective

representations and the group extensions they determine.

If H is a Hilbert space, the torus imbeds in the obvious way

as a central subgroup of U(H) equipped with the strong operator

topology. A projective (unitary) representation of G in H is

then a continuous homomorphism ~* from G to U(H)/T 9

Our interest will be in the cocycle (or multiplier) represen-

tations of G canonically associated to such projective representations

~*. These are obtained by choosing a Borel c r o s s - s e c t i o n f of

U(H)/T in U(H) and introducing the map m =fo~*. If f(T) = i,

this map from G to U(H) must a u t o m a t i c a l l y be Borel and satisfy

(2.32) r e) ::I

and

(2.33) ~(gl)~ g2 ) = ~(gl, g2)~(gl, g2)

for all gl, g2 c G. Here ~ is a Borel f u n c t i o n from G • G to

T which (by the associative law in G ) must be a two-cocycle

on G with values in T .

Note that ~ above obviously depends on the choice of

cross-section f 9 However, if f' is another such cross-section,

 29

the cocycle it determines will be c o h o m o l o g o u s to a . Thus each

projective r e p r e s e n t a t i o n u n i q u e l y determines an element of H2(G,T).

A n y Borel map f r o m G to U(H) satisfying (2.32) and (2.33)

is called a m u l t i p l i e r representation with multiplier a (o~

simply, an a - r e p r e s e n t a t i o n ) and all such r e p r e s e n t a t i o n arise f r o m

projective representations as above. More precisely, if ~ is

any a - r e p r e s e n t a t i o n , set ~* = po~, where p is the n a t u r a l

projection from U(H) to U(H)/T . Then ~ : f~*, and

lles in the c o h o m o l o g y class d e t e r m i n e d by ~* 9 In this sense,

each p r o j e c t i v e representation is e s s e n t i a l l y a f a m i l y of m u l t i p l i e r

representations with c o h o m o l o g o u s multipliers.

T h r o u g h o u t this paper we shall want to d i s t i n g u i s h b e t w e e n

representations of the m e t a p l e t i c group w h i c h factor through Z2

and those that do not. Moreover, we shall often want to confuse

those that don't w i t h cocycle r e p r e s e n t a t i o n s of GL(2).

In general, if ~ is an extension of the locally compact

group G by some subgroup T of the torus we shall call

~enuine if ~(t) = tol for all t e T ~ G 9 For the sake of

e x p o s i t i o n we assume below that m u l t i p l i c a t i o n in ~ is d e t e r m i n e d

by some fixed cocycle a .

P r o p o s i t i o n 2.18. Let g : G~ denote the (Borel) cross-

section g(g) = (g,l) and suppose ~ is a genuine

r e p r e s e n t a t i o n of ~. Then:

(a) The map ~og from G toU(H) is an a - r e p r e s e n t a t i o n

of G;

(b) The c o r r e s p o n d e n c e ~ ~ ~og is a b i j e c t i o n b e t w e e n

the c o l l e c t i o n of genuine r e p r e s e n t a t i o n s of ~ and ~ - r e p r e s e n t a -

tions of G;

(c) This c o r r e s p o n d e n c e p r e s e r v e s u n i t a r y e q u i v a l e n c e and

direct sums.

Proof. Part (a) follows i m m e d i a t e l y from the fact that

 3O

g(glg2 ) = e(gl, g2)t(gl)g(g2). To prove (b), suppose ~' is an

e-representation of G in H and define

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

Then ~ is a Borel homomorphism of ~ into U(H), whence

continuous, and obviously "genuine". The rest of the proposition

follows from the fact that a n operator A intertwines the

e-representations ~i and ~ if and only if it intertwines the

corresponding ordinary representations ~I and ~2"

We shall now describe Weil's construction in earnest.

The symplectic groups of Weil's general theory are attached to

locally compact abe!ian groups G so the notation <x,x*> will

denote the value of the character x* in G* (the dual group to G)

at x e G. Since Weil's theory is essentially empty unless G is

isomorphic to G* we shall assume throughout that this is the case.

Example 2.19. Let denote a local field of characteristic zero

and V a finite-dimensional vector space defined over F. Fix q to be a

non-degenerate quadratic form on V and T the canonical non-trivial

additive character of F described in [46]. The identity

<X,Y> = T(q(X,Y)), (X,Y c V)

where

q(X,Y) = q(X+Y)-q(X)-q(Y),

establishes a self-duality for the additive group of V

equipped with its obvious topology. Thus Well's theory will be

applicable in particular to G = V together with (q,T). This

example in fact will suffice for the applications we have in mind.

For each w = (u,u*) ~n G x G* let U'(w) denote

the unitary operator in L2(G) defin@d by

u'(w)~(x) = <x,u*>~x~u).
 31

Then U'(Wl)U'(w2) = F(Wl, W2)U'(Wl+W2) where

(Wl, W 2) = <ul,u >

if W i = (Ui, U[). That is, U' is a multiplier representation

of G • G* with multiplier I/F and the family of operators

U(w,t) = tU'(w) (w e G ~ G, t e T)

comprises a group with composition law given by

(2.38) (wl, tl)(w2, t 2) = (Wl+W2, F(Wl,W2)tlt2).

In particular, the multiplier representation U' determines an

extension of G x G* by T which is cal~ed the Heisenberg group

of G and denoted by A(G) o Yn case G = ~, A(G) is (the

exponentiation of) the familiar Heisenberg group

0 i x*
0 0 i

In general, T is the center of A(G); A(G) may be viewed

either as G • G* • T equipped with the group law (2.34) or as the

group of operators [U(w,t)] (in which case it is denoted by ](G)).

The crucial fact is that U(w,t) is an irreducible representation;

in fact, U(w,t) is the unique irreducible representation of A(G)

which leaves T pointwise fixed. Therefore at least the first

part of the result below is plausible.

Theorem 2.20 (Segal). Let B0(G ) denote the group of

automorphisms of A(G) leaving T pointwise fixed,

the normalizer of ~ in L2(G), and let

Bo(a)
be the natural projection. Then PO is onto with kernel T.

In other words, there exists a multiplier representation

 32

of B0(G) in L2(G) whose range (choosing the constant in all

possible ways) coincides with BO--~--

~. Indeed if s c B0(G), the
formula uS(w,t) = U((w,t))s) defines an irreducible unitary

representation of A(G) which fixed A(G) pointwise and hence is

equivalent to U. Therefore U((w,t)s) : r-l(s)U(w,t)r(s) for

some unitary operator r(s) in L2(G) (uniquely determined up to

a scalar in T) and

s ~ r(s)

determines a multlp]!er representation of B0(G ) (the Well

representation) satisfying P0(r(s)) : So

This "abstract" Weil representation is relevant to SL(2)

since B0(G) is the semi-direct product of (GX G*)* and an

abstract symplectic group which reduces to SL2(F) when G = F.

More precisely, s~nce s in B0(G ) fixes (l,t), it is

completely determined by its restriction to G• G* • [i] where it is

of the form (w,l) ~ (w~,f(w)). Thus on G • G* • T it is of the

form (~,f) where

(w,t)s = (w,t)(a,f) = (wa,f(w)t),

a is an automorphism of G • G*, f~ G • G* ~ T is continuous, and

F(Wl~,W2~) f(w1+w2)
(2.35) F(Wl,W21 : f(wl)f(w7

Conversely, each pair (o,f) satisfying (2.35) defines an

automorphism of A(G) fixing T pointwise, so B0(G):[(~,f)] ,

with group law

(~,f)(~,~,) : (~,,f")

if f"(w) : f(w)f'(w~).

Now let Sp(G) denote the abstract symplectic group of

automorphlsms of G ~ G* which leave invariant the bicharacter

 33

F(Wl,%) <Xl,X~>
F(~2,~l) = ~ (~i = (xi'x~))
Using (2.35) one checks that if (a,f) = s e Bo(G) then a r Sp(G).

Our claim was that the exact sequence

(2.36) 1 + (GX G*)* + B0(G) ~ Sp(G) + 1

actually splits.

To check this it is convenient to describe Sp(G) in

m a t r i x form. Since each ~ e A u t ( G x G*) is of the form

(x,x*) ~ (x,x~)(y ~)

where ~:G ~ G, t3:G ~ G*, u + G*, we s h a l l , following Well,

write

o= [u ,

and define
~I = [ _ ~ - ],

where ~* denotes the a p p r o p r i a t e map dual to ~. Then

~ A u t ( G • G*) is symplectic iff a~ I = I.

Given a bd ] = a ~ Sp(G),
[c define fq on G • G* by

f~ (u, u~) = <u, 2 - 1 u ~ > < 2 - 1 u ~ y 6 ~, u~><u*y, u~> .

Here we are assuming, as we do throughout, that x ~ 2x is an

automorphism of G. Then f~ and a satisfy (2.35) and the map

-, (q, f)

is a m o n o m o r p h i s m of sp(G) into B0(G ) which splits (2.36).

In short, Theorem 2.20 produces an exact sequence

(2.37) 1 ~ T-~ BO--(-C7, B0(G) + 1

and Bo(G ) contains a copy of Sp(G).

 Example 2.21. Suppose F is local and G = (V,q,T) is as

in Example 2.19. Then SL2(F) can be imbedded homomorphically in

Sp(G) by allowing each element of F to act on V via

scalar multiplication. (Since ~* = a for each ~ e F, ~ = [a

c ~]eSL(2,F)
obviously satisfies a~ i = i; note that Sp(G) = SL2(F) if G = V

but in all other cases properly contains it.) From this it follows that

one can associate to each quadratic form (q~V) a natural projective

representation of SL(F) in L2(V) which we shall call the Well

representation attached to (q~V) and denote by rq.

Before further analyzing this representation we need to adapt

Weil'sgeneraltheorytothespecialcontext of Examples 2.19 and 2.21.

Recall that V is a vector space over F and ~ is a non-trlvlal

additive character of F (fixed once and for all). If X* belongs

to the linear dual V* we shall denote its value at X in V

by [X,X*]. The natural isomorphism between V and V* is then

[X,Y] = q(X,Y).
Using [.,.] in place of <-,.> one can now "linearlze"

the theory Just sketched, introducing:

(a) the bilinear form B(Zl, Z2) = [XI,X~] in place of the

blcharacter F(WI,W2);
(b) the Heisenberg group A(V) in place of A(G);

(c) the symplectic group Sp(V) in place of Sp(G); and

(d) the pseudo-symplectic group Ps(V) in place of B0(G).

A typical element of Ps(V) is of the form (~,f) where

a e Aut(VXV*) belongs to Sp(V) and f : V • V* ~ F satisfies

(2.38) f ( z l + z 2 ) - f ( Z l ) - f ( z 2) = B(Zl~,Z2~)-B(zl, z2).

This relation, of course, is the linearlization of (2.35): taking

of both sides of it yields (2.35). Note that Ps(V) ~B0(G). In fact
Ps(V) is the subgroup of B0(G) consisting of pairs (o,f) with
f quadratic; cf. (2.39) below.
Note now that (2.38) associates to each o in Sp(V) one
and only one quadratic form on VxV*. ~We are assuming F
 35

has characteristic zero, in particular, characteristic not equal

to two.) Therefore the homomorphism

(a,f) § f

is an isomorphism between Ps(V) and Sp(V)

Since the same map from B0(G) to Sp(G) has kernel
isomorphic to (G• the analogy between the exponentiated

and unexponentiated ~heories breaks down here. However, the map

(2.39) (~,f) ~ > (~,~~

does imbed Ps(V) homomorphically in Bo(G) so we can still

obtain an extension of Ps(V) by pulling back (2.37) through ~:

1 > T > Mp(V) ~ > Sp(V) > 1

The group

Mp(v) ={(s,~) ~ P s ( V ) x ~ ) : p0(~) =~(s)]

is Well's general metaplectic group. It is a central extension

of sp(v) by T .

Weil's results concerning the non-triviality of Mp(V) include

the following:

(a) Mp(V) always reduces to an extension of Sp(V) by

Z2, i.e. the cohomology class it determines in H2(G,T) is of

order 2 ;
(b) the extension Mp(V) is in general non-trivial; in

particular, if F~@, and V is one-dimensional, Mp(V) is always

a non-trivial cover of Sp(V) by Z2 .

These results yield a (topological) central extension

(2.4o) l~ z2 ~ % 7 ( v ) ~ sp(v) + i
 36

which by Proposition 2.3 must coincide with

1 ~ Z 2 ~ ~r2(F ) ~ S ~ ( F ) ~ 1

when V=F. I.e., Well's metaplectic group generalizes the meta-

plectic cover of SL2(F ) constructed in 2.2.

Although Well's construction does not immediately yield an

explicit factor set for S-L2(F) it does a priori realize this group

as a group of operators. In fact, it provides a host of representations

for this group. These correspond to quadratic forms over F and in

explicit terms are realized as follows.

Fix a Haar measure dTy on F such that the Fourier transform

~(x) = ~f(y)~(2xy)d y

satisfies (f) (x) =f(-x). Recall that T is the canonical character

of F whose conductor is OF . Given (q,V), fix an orthogonal

basis XI,...,X n of V such that if X :~xiXi, then q(X) =q(xl,...,Xn)

= ~i x , with ~i=5 q(Xi'Xi)" Let Ti denote the character

Ti(Y) = T(giY) of F, i = l,...,n, and let diY denote the

corresponding Haar measure on F, normalized as above.

If F is non-archimedean~ the limit

Y(Ti) = me~lim~pm~i(y2)diY

is known to exist (by Well [AT] it is an eighth root of unity).

Consequently the invariant

v(q,~) = n ~(~i )
i=l

is well-defined. If F=R, and ~i(x) = exp(~ r set y(~i )

equal to exp(88 sgn(%i) ) and define Y(q,T) as above. Finally,

if F =C, set Y(q,T) identically equal to i . The Fourier transform

on L2(V) is defined to be
 37

$(X) = ~ %(Y)T(q(X,Y))dY
V
n
where dY= Z diY 9
i=l

Theorem 2.22. For each t eF x let Tt denote the character

of F defined by Tt(x) = ~(tx) and denote y(q,r ) by y(q,t).

Suppose we use (q,V,T) to imbed SL2(F) in Sp(V) (and B0(V)).

Then a cross-section of S-p(V) over SL2(F) (c.f. (2.40)) is

provided by the maps

(2.4l) 1 b
[0 1 ] +
r( 1 b ) ~ ( X )
[0 1 ]
=
Tt(bq
(X))~(X)

0 -i r( 0 -i
(2.42) [1 0] ~ i1 0 ])~(x) = ~(q't)-l~(-X)

(~ e L2(V)). More precisely, for each t e F x, these maps extend

to a multiplier representation of SL2(F) in L2(V) whose associated

cocyele is of order two.

Remark 2.23. In (2.42) the Fourier transform is taken with

respect to Tt and Haar measure dtY = Itln/2dy. Thus

A
(X) = ~ 9(Y)~t(q(X,Y))Itln/2dy 9
v

Proof
of Theorem 2.22. Without loss of generality we assume
n
t :I. Since d Y : H d.Y it is easy to check that the operators
t:l ~
in question are tensor products of the operators in L2(F)

corresponding to ~i' namely

r([ol ~])f(• = ~i(b~2)f~(xl,

and

r([ ~ -IO]If(x) =~(~)-l?(-x) ,

i = l,...,n. Thus the Theorem is reduced to the case V:F,

q(x) = x 2, precisely Theorem I.A.I of Sha!ika [35].

We note that SL2(F) is generated by elements of the form

 38

I b 0 -I
[0 1 ] and [I 0 ] subject to certain relations. Thus one could

follow Shalika directly and prove Theorem 2.22 by checking that these
i b 0 -I
relations are preserved by the operators r([0 1 ]) and r([l 0 ])"

In any case, the resulting m u l t i p l i e r representation of SL2(F)

will be denoted rq and called "the" Well representation attached

to the quadratic form q. (This is an explicit realization of the

representation r of Example 2.21).

q

Corollary 2.24. r is ordinary if and only if V is even

q
dimensional.

We conclude this Section with some remarks. Although the group

of operators B0(G) is irreducible (it contains T(G)), the

representations rq are never irreducible. In fact, their

decomposition is now known to be of great importance for the theory

of automorphic forms and group representations.

For the case when q is the norm form of a quadratic field,

see [35], [45], [36], and [18]; for the case when q is the norm

form of a division algebra in four variables see [37] and [18]. As

already remarked in the Section I, no such complete results have

yet been obtained when q is a quadratic form in an odd number of

variables. This is because until recently no one seriously attacked

the r e p r e s e n t a t i o n theory of the two-fold covering groups of SL2(F).

In these Notes we shall describe how a We~l representation in

one and three variables decomposes and how its decomposition relates

automorphic forms on GL(1) and GL(2) to automorphic forms on

the m e t a p l e c t i c group. The general p h i l o s o p h y is explained in

Subsection 2.4.
2.4. A p h i l o s o p h y for Well's representation.

The purpose of this Subsection is to describe a simple principle

which (though unproven) underlies most of the results of these Notes.

Roughly speaking, the idea is that quadratic forms index I-i

correspondences b e t w e e n automorphic forms on the m e t a p l e c t i c group

and automorphic forms on the orthogonal group.

Let F denote a local or global field of characteristic zero,

(q,V) a quadratic space over F, and r the corresponding

q
representation of SL2(F ). Let H denote the orthogonal group

of q. This group acts on L2(V) through its n a t u r a l action on

V. The resulting representation of H in L2(V) m a y be assumed

to be u n i t a r y and we denote it by A(h).

Now fix F to be local. From T h e o r e m 2.22 it follows that

for h c H and ~ ~ S-L2(F) the operator A(h) commutes with

rq(~). In fact it seems plausible that rq and A generate

each others commuting algebras. Suppose for the m o m e n t that this

is so. Then the p r i m a r y constituents of rq correspond i-I to

pr~ary constituents of A. In particular, the commuting diagram

L2 (V)

SL 2 H

leads to a c o r r e s p o n d e n c e b e t w e e n irreducible representations of

H which occur in A and irreducible representations of SL 2

which occur in rq. F o l l o w i n g R. Howe, we call this correspondence

D for "duality". Globally, D should pair together automorphic

forms on H which occur in A with automorphic forms on SL 2

which occur in rq. This is the c o r r e s p o n d e n c e alluded to above.

Since the existence of this correspondence D seems to

rest on the hypothesis that r and A generate each others

q
commuting algebras, some further remarks are in order. The precise
 40

formulation of this hypothesis, in the greater generality of "dual

reductive pairs", was communicated to me by Howe; as such, it is

but one facet of his inspiring theory of "oscillator representations"

for the metaplectic group (manuscript in preparation). In the

context of SL 2, at least over the reals, this fact had also been

suspected by S. Rallis and G. Schiffmann (cf. [52]). On the other

hand, important special examples of D had already appeared in the

literature. In [36] Shalika and Tanaka discovered that the Well

representation attached to the norm form of a quadratic extension of

F yielded a correspondence between forms on SL(2) and S0(2). This

seems to have been the first published example of D. (Actually the

correspondence of Shalika-Tanaka was not i-i since they dealt with

S0(2) instead of 0(2).) For quadratic forms given by the norm forms

of quaternion algebras over F the resulting corresponding between

automorphic forms on the quaternion algebra and GL(2) was developed

by Jacquet-Langlands ([18],Chapter III) following earlier work of

Shimizu. In both cases, the fact that r and A generate each others
q
commuting algebras was not established apriori; rather it appeared as

a consequence of the complete decomposition of rq.

One purpose of these Notes is to describe the duality correspon-

dences belonging to two forms in an odd number of variables, namely
2 2--T
gl(x) = x 2, and q3(xl, x2,x3) = X l - X l - X 3. The inspiration for our
discussion derives directly from general ideas of Howe's, an initial
suggestion of Langland's, and earlier works of Kubota [ ], Shintani
[ ], and Niwa [ ]. Our results, specifally Cor.4.18, Cor.~.20, and
T~eorem 6.3, lend further evidence to the general principle asserted
above. However, as in earlier works, the fact that r q and A generate
each others commuting algebras, appears as a consequence ofthe existence
of D.
2.5 Extending Well's representation to GL 2

Well's construction of the m e t a p l e c t i c group produces a multiplier

representation rq of SL 2 . There are several reasons now why

it is advantageous to extend this c o n s t r u c t i o n to GL 2 . One is

that Well's representation for SL 2 often depends not only on q

but also on the choice of additive character ~ . By contrast, the

natural analogue of rq for GL 2 depends only on q .

In this Subsection I want to explain how Weil's original

representation depends on T More precisely, i want to define

an analogue of rq for GL 2 which is independent of ~ The case

when q is the norm form of a quadratic or quaternionic space is

discussed in [18]. I shall treat the cases

ql(x) : x 2
and

q3.xl,
x2,x3)(
2
: xI -
x~- x~

following a suggestion of Carrier's.

Throughout this p a r a g r a p h the following conventions will be in

force:

l) F will be a local field of characteristic zero;

2) n=l or 3 according as q =ql or q3 ;

rn will denote rql or rq3 ;

a b
4) [[a b],l] will be abbreviated by [c d ] "

Note that the formulas for rn given in T h e o r e m 2.22 depend on Tt "

Therefore, for emphasis, I shall denote rn by rn(~ t) ; the

notation r will be reserved for the r e p r e s e n t a t i o n eventually

n

introduced for GL 2 .

Proposition 2.27. (Cf. [18] Lemma 1.%.). If a ~ F x, and

a
= [s,(] ~ST2(F), define ~a to be is ,~ v(a,s)] (cf. Proposition

2.6). Then
 42

rn(Ta~(Y) = rn(T)(Fa)
for all a cF x and ~ SL2(F ) .

Proof. We may assume without loss of generality that n =I

and ~ is a generator of -SL2(F).

- Suppose first that s=w=[[ Ol 0],I].
i

Then

0 a a o ~ {1,(a,~)]
2a : ~ a : [-a-lo] : [o a - z ]
and

rn(T)(~a)~(X) = (a,a) V(qn, T)(rn(T)([O a a-o l ])~)(x)

Some tedious computations with Hilbert symbols also show that

a 0 i a -I i a -I
-- 1 a
[0 a -1] = W[o I ] ~ [0 1 ] ~ [0 I ][l,(a,a)]

Consequently

(2.~3) rn(~)([a. O1])%(X) = (a,a) lal n/2 ~(qn'Ta)~ ~(aX)

0 a- Y~qn 'Tj
and

rn(T)(~a)~(X ) = [al n/2 y(qn, Ta)~(ax).

But laln/2y(qn, Ta)~(aX) : rn(Ta~)%(X). Therefore rn(T)(~a ) :

rn(Ta)(~), as was to be shown. The analogous identity for

~ [[0i bl ] ,I] is completely straightforward. []

Corollary 2.28. The representation rn(Ta) is independent

of a eF x iff rn(T ) extends to a representation of G-L2(F).

proof. Suppose rn(Ta) is equivalent to rn(T ) for all

a cF x. Then for each a eF x there is a unitary operator Ra
in L2(F n) such that

rn(T a)(~) = R a rn(T)Ral

for all s e SL2(F). Equivalently, by Proposition 2.27,

 43

rn(T)(~a) = R a rn(Ta)R[l

i 0
But G-L2(F) is the semi-direct product of ST2(F) and FX=[[ 0 a] ].
Thus rn(T) can be extended to a representation of G-L2(F) by
defining rn(T ) ( [0i 0a ] , i) to be Ra
I 0
Conversely, if rn extends to G-T2(F), rn(T)([O a],l) will
intertwine rn(T ) with rn(~a) []

Example 2.29. If a ~ (FX) 2, say a = 2, then Ra~(X)=I~Ii/2h(~x)

intertwines rl(Ta) and rl(T); in particular, rl(~) extends

to a representation of

G-L22(F) : Jig,c] ~ G-L2(F) : det(g) ~ (FX) 2]

In general, rq(Ta) will not be equivalent to rq(T). Indeed

rq(T) will extend only to a representation of the subgroup

[[g,~] ~ G-~2(F) : det(g) fixes rq(~)]

To get around this problem we "fatten up" rq(~) before attempting to

extend it. This way there is more room in the representation space
for intertwining operators to act.

Definition 2.30. Let dt denote the restriction of additive

Haar measure on F to Fx . Let r denote the direct integral
q
of the representations rq(T t) with respect to dt .
Note that the space of r
is isomorphic to L2(F n• FX).
q
Moreover, the methods of Corollary 2.28 imply that r q extends to
a representation of G---L2(F) satisfying

(2.44) rq([ol 0a])r = laI-i/2~ta_l(X)

Proposition 2.31 9 The action of rq in L2(F m• F x) is given

by the formulas

(2.45) rq([ 0I b1 ] )~(X,t) = rt(bq(X))~(X,t)

 44

A
r (w)@(X,t) = ~(q,t)9(X,t)
(2.46) q
: ~(q,t)~ ~(Y,t)Tt(q(X,Y)dtY
Fn

(2.87) rq( [a0 a-O 1 ] ) ~ ( X , t ) = a,a)laln/2 ~~(q'Tat) ~(aX, t)

and

(2. ~8) r q ( [ o1 0a] ) ~ ( X , t ) = la -1/2%(y~,ta-l)

Proof. Apply the definition of direct integral. []

Proposition 2.31 implies we could have defined rq directly

for GL 2 using formulas (2.45), (2.46), and (2.48). In fact thi{

construction of r was communicated to me by Cartier awhile ago

q
and my Definition 2.30 simply reformulates his ideas.

Note that

a o ~(q~,ta) -1/2~ ( 2)
(2.49) rl([ o a])r = y(ql, t ) ral aX, ta - .

Thus one e a s i l y checks t h a t r l ( [ 0 a a])

0 commutes w i t h rl(g) for
all g ~ GL2(F ) iff det(g) c (FX) 2 This is consistent with

Corollary 2. 13(a).

Now it is natural to ask what shape the philosophy of Subsection

2.Z~ takes in the present context? Roughly speaking, the answer

is that the orthogonal group of q is simply replaced by the group

of similitudes of q .

Suppose first that q :ql " The group of similitudes of ql

is FX=GLI(F) and its action in L 2 ( F x F x) is given by

(A(y)~) (x, t) = l al - 1 / 2 9 ( a x , ta -2)

Note that A(y) clearly commutes with rI .

Now suppose q =q3" If F3 is realized as the space of 2x2

symmetric matrices with coefficients in F then q(X) :det(X).
 45

The group of similitudes of q3 is essentially GL2(F ). More

precisely, define

go X = tg X g

when g e GL2(F) and X c F 3. Then q(g~X) = (det g)2 q (x). The

action of GL2(F ) in L2(F 3 • F x) is given by

(A(g)~)(X,t) = I det gl 2 ~ ( g , X , t ( d e t g)-2)

and A once again commutes with r3 .

The result now is that rI (resp. r3 ) should decompose

into irreducible representations of ~ indexed by irreducible

representations of GLI(F) (reap. irreducible representations of

GL2(F) which occur in A ). Globally this means automorphic forms

on the metaplectic group which occur In rI (resp. r3 ) should

be indexed by automorphic forms on GL I (reap. automorphic forms

on GL 2 which occur in A ). The global results that can be

obtained or expected are described in Subsections 6.1 and 6.2.

The local analysis is carried out in Subsections 4.3, 4.4, and

5.5.
2.6, Theta-functions

Classically, a.utomorphic forms are constructed from theta-

series attached to quadratic forms. This is the procedure reformu-

lated in representation theoretic terms by Well in [47]. For

SL 2 the idea is this.

Suppose F is a number field and q is an F-rational quad-

ratic form in n variables. Define a distributuon ~ on Fn by

e(~) : s ~(r

Piecing together local Well representations a representation rq

of ST2(~ ) is defined with the property that 0(rq(~)~) is

SL2(F)-invariant. In particular, ~(rq(S)~ = 0(~) for all s e SL2(F).

This is the theta function attached to q. For GL2, the point of

departure is Proposition 2.32 below.

For v a non-archimedean place of F let 0v denote the ring

of integers of Fv and Uv its group of units. Let rv(q) denote

the corresponding Weil representation of GL2(Fv) in L2(F n x Fx).

As in Subsection 2.1, K vN is the subgroup of GL2(0v) consisting

of matrices [~ bd] with a ~ I and c ~ 0 (mod N). Here N is

a positive integer divisible by 4. Thus K~ = GL2(O v) if Fv

has odd residual characteristic.

Proposition 2.32. Suppose q = ql or q3" Then for almost

every v, rv(q ) is class I, i.e., the restriction of rv(q) to

GL2(0v) has at least one fixed vector. More precisely, for each

place v, define ~v0 e L2(F n x Fx) by

n
(2.50) ~v(Xi,
O .... Xn, t ) = (~ 10 ) | 1U N
i=l v
v

Here i0 denotes the characteristic function of OF and IuN

v
v
denotes the characteristic function of Ur~v = { y e U v ' y i 1 (mod N ) ] .
 47

For all odd v,

(2.51) rv(q)(k)~o = go

for k e K N.
V

Proof. The group GL2(0v) is generated by the matrices

[oi b1 ] (b
e ~
W

and
i 0 (a
[o a ] ~ Uv).

Thus it suffices to check (2.51) for these generators.

Recall that our canonical additive character T has conductor

V

0v. In particular, ~v(tbq(X)) = i if q(X) e O v, b e 0v, and

t e U v. Therefore rv(q)(l b)~~ = ~v~176 Note also that
the Fourier transform of i0 is i0 . Thus rv(q)(w)~~ ) = ~v~
V V

The fact that rv(q)([ 01 a])~v(X,t)

o o = lal-i/2~~ ta -I) = oix ' t) is
obvious since a e U v.

To define rq globally, and to introduce theta-functions on

G~, we need first to define an appropriate space of Schwartz-functions

on ~ • ~. We shall say that ~ on ~ • ~ is "Sehwartz-Bruhat"

if ~(X,t) = N ~v(Xv, tv) and:

V

(i) ~v(Xv, tv) is infinitely differentiable and rapidly de-

x
creasing on • Fv for each archimedean v~

(ii) for each finite v, ~v is the restriction to ~v x ~v

of a locally constant compactly supported function on F~v+l;

(iii) for almost every finite v, ~v = ~o (the function defined

V

by (2.5O)).
Denote this space of functions by A ( ~ • F~). Since ~ ( ~ • FI)
is dense in L2(~ • F~), we can define a representation rq of
in L2(~ • F~) through the formula
 48

By virtue of Proposition 2.32 this definition is meaningful for at

least all % e ~(~ x F~). Indeed for almost all v, ~v e GL2(Ov)

O
and %v = @v" Thus rv(q)(~v)%v = %v and rq(~)% again belongs to

~(~ X F~). By continuity, rq(~) operates (unitarily) in L 2.

The role of ~v in the local theory is now played by a non-trivial
character T of F~.

The theta-functions on ~ corresponding to q (or rq) are

defined as follows. For each ~ e ~(~ • F~) define 8(~,g) on

a~ by

8(~'g) : Z (rq(~))@({,D) .

Proposition 2.33. For each y e GL2(F),

(2.52) 8(%,Yg) = @(~,~);

i. e. , e (~,~) is GL2(F ) invariant.

Proof. We may assume without loss of generality that ~ = [I,i]

i b
and y is a generator of GL2(F)o Suppose first that Y = [0 1 ]
with b s F. Then

rq(y)~(X,t) = T(btq(X)@(X,t)

with T = H~ trivial on F. But q is F-rational. Thus

V

rq(~)%({,~) = ~({,9) for ({,~) c F n • F x and (2.52) is immediate.

Now suppose y : wo From Well's theory (cf. Theorem 5 of [47])

it follows that y(q, Tt) -" 1. Moreover, Itl : i if t e F x.
Therefore,

r~ rq(W)~(g,n) = ~1~1n'/2 y(q, Tn)$(~g,~ )

: ~ $(~,~)
 49

= z ~(g,~).

The last step above follows from Poisson's summation formula.

Finally suppose i aO]

Y = [0 with a e F x. By formula (2oI$8)

rq(~)~(y,~) = lal - l / 2 ~(~,~a-1)o But lal -1/2 = 1. Therefore

rq(y)%({,9) = Z 9({,q) as d e s i r e d .

Observation 2.34. The f u n c t i o n 8(9,g) always defines an

automorphic form on ~, i.e., 8(~,g) is always GF invariant.

Nevertheless, it may often be the case that 8(~) is identically

zero. Suppose, for example, that ~(-X,t) = -@(X,t), i.e. ~ is an

odd function in Xo Then rq(~)~ is also odd for all ~ e ~.

For simplicity, suppose also that F = Q. From formula (2o~7) it

follows that rq([ -i0 - ~])~(X,t) = -~(X,t), i e.

-1 o
e(~, [ o ]7) : - e(~,~).

But by the GF-invariance of 8,

-i ~]~)

Therefore 8(~,~) i 0~
We close this paragraph by demonstrating how 8(~,~) generalizes

the basic theta function

2~in2z
n ~

Proposition 2~176 Fix n = i, F = ~, and rq = r I. Suppose

= n ~p where ~p = ~o for all finite p and ~ (x,t) =

P
Itl -l/q e- I t ! 2 ~ x 2 Then if g = [[yl/2 xy-l/2
-1/2 ]' 1 ],
0 y

8(%,g) = yl/As(x+iy)o
 50

Proof. Since
[yl/2
0
xy-1/2
y-I/2 ] =
I x
[0 1 ] 0
[yl/2 o
y -I/2]'

Z rq(~)%({,~).
- . - = lyll/4 e-2W7~ 2 e2~i~2x~.
({,~)

Here we choose T = ~ Tp with T (X) = e 2~ix and Tp trivial on

0p for p < ~. Thus ({,~) contributes to the summation above

only if { e ~ and 9 = i (recall ~2 = 102~176 l.e.,

e2~in ~ ( x + l y )

as was to be shown~

Note that for a c (0,~), and @ as above,

a o -I/2
r([ 0 a ] ) ~ ( ~ , ~ ) = ai ~(ax, ta -2)

= al-i/2 Ita-21-1/4 e-ltl2~x2

= tl-i/~ e-ltI2~x2

Thus 8(~,~) is actually defined on

7 = z~