Local Theta Correspondence of Depth and Theta Dichotomy: Representations

J. Math. Soc.
Japan
Vol. 54, No. 4, 2002
Local theta correspondence of depth zero representations and

theta dichotomy
By Shu-Yen PAN
(Received Feb. 25, 2000)

(Revised Mar. 5, 2001)
Abstract. In this paper, we prove that depth zero representations are preserved by
local theta correspondence for any type I reductive dual pairs over a -adic field. $p$
Moreover, the minimal $K$-types of the paired depth zero irreducible admissible rep-
resentations are paired by the theta correspondence for finite reductive dual pairs. As a
consequence, we prove that the Iwahori-spherical representations are preserved by the
local theta correspondence. Then we obtain some partial result of theta dichotomy for
finite reductive dual pairs and -adic reductive dual pairs of symplectic and orthogonal $p$
group, which is analogous to S. Kudla and S. Rallis’ result for -adic unitary groups. $p$
0. Introduction.
Let $F$ be a -adic field with odd residual characteristic. Let $D$ be a central $p$
division algebra over $F$ with an involution, fl be the ring of integers, be the $\mathfrak{p}$
prime ideal, be the (finite) residue field and be a prime element. Let
$f_{D}$ $\varpi$
$\mathscr{V}$
(resp. ) be a finite-dimensional nondegenerate -Hermitian (resp. -Hermitian)

$\mathscr{V}^{\prime}$
$\epsilon$
$\epsilon^{\prime}$
space over $D$ where are 1 or -1 and . Let (resp. ) $\epsilon,$

$\epsilon^{\prime}$
$\epsilon\epsilon^{\prime}=-1$
$U(\mathscr{V})$ $U(\mathscr{V}^{\prime})$
denote the group of isometries of (resp. ). We can define a skew- $\mathscr{V}$

symmetric $F$-bilinear form on the space . Then the pair $\mathscr{W}:=\mathscr{V}\otimes_{D}\mathscr{V}^{\prime}$
forms a reductive dual pair (over

$(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$ $F$ ) in the symplectic
group (cf. [Hw2], [MVW]). In particular, we have embeddings
$Sp(\mathscr{W})$
:
$l_{\mathscr{V}^{\prime}}$
and :
$U(\mathscr{V})$
) . Let denote the met-
$\rightarrow Sp(\mathscr{W})$ $l_{\mathscr{V}}$ $U\underline{(\mathscr{V}}’$
$\rightarrow Sp(\mathscr{W})$ $Sp(\mathscr{W})$
aplectic cover of . Let ( esp. ) denote the inverse of $Sp(\mathscr{W})$ $U(\mathscr{V}\underline{)}$$r$

$U(\mathscr{V}^{\prime})$
$ima\underline{ge}$
(resp. )
$l_{\mathscr{V}^{\prime}}(U(\mathscr{V}))$
. It is ) and $l_{\mathscr{V}}(U(\mathscr{V}^{\prime}))$ $\underline{in}Sp(\mathscr{W})$ $k\underline{no}wn$
$t\underline{hat}U\overline{(\mathscr{V}}$
with each other in

$c\underline{omm}ute$ . Therefore is a subgroup of $Sp(\mathscr{W})$ $U(\mathscr{V})\cdot\underline{U(}\mathscr{V}^{\prime})$
. restricting the Weil $representation\underline{o}f$

$S\underline{p(}\mathscr{W})$ $\underline{By}$
to the subgroup $Sp(\mathscr{W})$
’
( and p lling
$U\underline{(\mathscr{V}})\cdot U$ to ) by the homomorphisms
$\underline{\mathscr{V}^{\prime})}$
$u$ $\underline{ba}ck$
$U\overline{(\mathscr{V}}$
$\times U(\mathscr{V}$ $)$
’
$U(\mathscr{V})$ . ’
, there exists a one-to-one correspondence (called
$\times U(\mathscr{V}$ $)$ $\rightarrow U(\mathscr{V})$ $U(\mathscr{V}$ $)$
the local theta correspondence) between irreducible admissible representations of
2000 Mathematics Subject Primary 1lF27; Secondary $22E50,20C33$ .

Classification.
Key Words and Phrases. local theta correspondence, depth of a representation, theta dichotomy,
minimal K-types.
794 S.-Y. PAN
and irreducible admissible representations of

$U\overline{(\mathscr{V}})$
) (cf. [MVW], [Wp]). $U\overline{(\mathscr{V}}^{\prime}$
To determine the explicit correspondence is extremely difficult. In fact very little

is known except for few special cases.
Let be a reductive dual pair. Let be a
$(G, G^{\prime}):=(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$ $L$
good lattice in i.e., a lattice such that where

$\mathscr{V}$ $L^{*}\varpi\subseteq L\subseteq L^{*}$ $L^{*}$
$:=$
{ for any $l\in L$ } and is an integer which will be specified in

$v\in \mathscr{V}|h(v,$ $l)\in \mathfrak{p}^{\kappa}$ $\kappa$
subsection 2.1. Let be the stabilizer of in and define $G_{L}$ $L$ $G$
$G_{L,0}+:=\{g\in G|(g-1).L^{*}\subseteq L, (g-1).L\subseteq L^{*}\varpi\}$ .

It is easy to see that is a normal subgroup of . A depth zero minimal K- $G_{L,0}+$ $G_{L}$
type for is a pair $G$

where is an irreducible representation of the finite $(G_{L}, \zeta)$ $\zeta$
reductive quotient $G_{L}/G_{L,0}+$ (the definition here is slightly different from the
original definition in [MP1] . We say that an irreducible admissible represen- $)$
tation $(\pi, V)$ of contains a minimal $K$-type if $G$

, the vectors fixed $(G_{L}, \zeta)$
$V^{G_{L,0+}}$
by , is nonzero and
$G_{L,0}+$ , as a representations of $G_{L}/G_{L,0}+$ , contains . $V^{G_{L,0+}}$ $\zeta$
An irreducible admissible representation containing a depth zero minimal K-type

is said to be of depth zero (cf. [MP1], [MP2]).
Fix an Iwahori subgroup I of . Let denote the inverse image of in $G$
$\tilde{I}$
$l_{\mathscr{V}^{\prime}}(I)$
. By fixing a splitting
$\tilde{G}$
: , the concept of minimal -types can be

$K$ $\beta_{I}$
$I\rightarrow\tilde{I}$
extended to the metaplectic covers (this was pointed out to me by Jiu-Kang $\tilde{G}$
Yu). In particular, an irreducible admissible representation $(\pi, V)$ of is said to $\tilde{G}$
be of depth zero if is not trivial for some good lattice such that
$V^{\beta_{l}(G_{L,0+})}$
$L$
. The following is our first main result.

$G_{L},0+\subseteq I$
THEOREM A. $\underline{Sup}pose$ and are irreducible admissible

$th\underline{at}(\pi, V)$ $(\pi^{\prime}, V^{\prime})$
representations of $U(\mathscr{V})$
and respectively and they are paired by the local
theta correspondence. Then the depth of is zero if and only if the depth of is $\pi$
$\pi^{\prime}$
zero.
The concept of reductive dual pair can also be defined over a finite field. In
particular, the theta correspondence for a finite reductive dual pair is also defined
although this correspondence is not one-to-one in general. It is not difficult to
see that if is a reductive dual pair over and are
$(G, G^{\prime}):=(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$ $F$ $L,$
$L^{\prime}$
good lattices in respectively, then is a reductive

$\mathscr{V},$
$(G_{L}/G_{L,0}+ , G_{L}^{\prime},/G_{L,0^{+}}^{\prime}, )$
dual pair over the residue field of . The following is our second main result. $F$
THEOREM B. Let be a reductive dual pair. Let $(G, G^{\prime}):=(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$
(resp.
$(\pi\underline{V},)$) be an irreducible admissible representation of
$(\pi^{\prime},$
(resp.
$V^{\prime})$ $U(\mathscr{V})$
such that two representations are paired by the local theta correspondence.
$U(\mathscr{V}^{\prime}))$
Suppose that has a minimal $K$-type for some good lattice in such
$\pi$ $(G_{L}, \zeta)$ $L$ $\mathscr{V}$
that Then contains a minimal $K$-type

$G_{L,0}+\subseteq I.$ such that and are $\pi^{\prime}$
$(G_{L}^{\prime},, \zeta^{\prime})$ $\zeta$
$\zeta^{\prime}$
paired by the theta correspondence for the dual pair . $(G_{L}/G_{L,0}+ , G_{L}^{\prime},/G_{L,0^{+}}^{\prime}, )$
Local theta correspondence 795
The relation between the local theta correspondence of depth zero repre-
sentations and the theta correspondence for finite reductive dual pairs has many
nice applications. For example, it implies that an irreducible Iwahori-spherical
representation (i.e., an irreducible admissible representation admitting nontrivial
vectors fixed by an Iwahori subgroup) corresponds to an irreducible Iwahori-
spherical representation. This result has been proved by A.-M. Aubert in [Ab]
for those special cases called unramifted reductive dual pairs. Another im-
plications is that an irreducible unipotent representation (cf. [Lt]) corresponds to
an irreducible unipotent representation under some restriction of residue char-
acteristic of . $F$
Let be a split dual pair i.e., a dual pair

$(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$ $reduc\underline{tive}$ $re\underline{duc}tive$
such that the two splittings and exist. An $U(\mathscr{V})\rightarrow U(\mathscr{V})$ $U(\mathscr{V}^{\prime})$ $\rightarrow U(\mathscr{V}^{\prime})$
explicit formula of the splittings with respect to a Schr\"odinger model of the

Weil representation is given in [K1]. If $D$ is also commutative, the splittings
:
$\tilde{\beta}^{L}$
depending on a good lattice in

$U(\mathscr{V})$
is given in [Pnl]. The
$\rightarrow U(\mathscr{V})$ $L$ $\mathscr{V}$
advantage of the splitting is that the depth of is zero if and only if the depth $\tilde{\beta}^{L}$
$\pi$
of is zero where is any irreducible admissible representation of

$\pi\circ\tilde{\beta}^{L}$
. $\pi$ $U(\mathscr{V})$
Under the splitting we can describe the correspondence of depth zero su- $\tilde{\beta}^{L}$
percuspidal representations for -adic reductive dual pairs completely in terms of $p$
the correspondence of cuspidal representations for finite reductive dual pairs

as follows. It is known that if is an irreducible cuspidal representation of $\zeta$
the finite classical group $G_{L}/G_{L,0}+$ , then the compactly induced representation
-Ind
$c$
is an irreducible supercuspidal representation of $G:=U(\mathscr{V})$ (cf. [MP2]).
$ G_{L}G\zeta$
THEOREM C. Suppose that is a split reductive dual $(G, G^{\prime}):=(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$
pair such that $D$ is commutative.

(i) Let $(\pi, V)$ an irreducible admissible representation of $(\underline{resp}. (\pi^{\prime}, V^{\prime}))\underline{be}$
the group (resp. ). Assume that has a minimal K-type $U(\mathscr{V})$ $U(\mathscr{V}^{\prime})$ $\pi$
for some good lattice in . Suppose that

$(G_{L}, \zeta)$
is a s per- $L$ $\mathscr{V}$ $\pi\circ\tilde{\beta}$
$u$
cuspidal representation of for some splitting : $U(\mathscr{V})$

$\tilde{\beta}$
$U(\mathscr{V})\rightarrow U(\mathscr{V})$
and OX is a first occurrence in the theta correspondence. Then has

$\pi$
$\pi^{\prime}$ $\pi^{\prime}$
a minimal $K$-type such that OX is a first occurrence in the $(G_{L}^{\prime},, \zeta^{\prime})$ $\zeta$
$\zeta^{\prime}$
correspondence for the reductive dual pair and $(G_{L}/G_{L,0}+ , G_{L}^{\prime},/G_{L,0^{+}}^{\prime}, )$
$bo$
th are cuspidal representations. $\zeta,$
$\zeta^{\prime}$
(ii) Suppose that OX is a first occurrence of theta correspondence for the $\zeta$
$\zeta^{\prime}$
reductive dual pair where both are cuspidal. $(G_{L}/G_{L,0}+ , G_{L}^{\prime},/G_{L,0^{+}}^{\prime}, )$ $\zeta,$
$\zeta^{\prime}$
Then OX is a first occurrence for the reductive dual pair $\pi$

where $\pi^{\prime}$
$(G, G^{\prime})$
is the irreducible admissible representation of

$\pi$
such that $\tilde{G}$
$\pi\circ\tilde{\beta}^{L}\simeq$
( OX ), is the irreducible admissible representation of

$c$ - $Ind_{G_{L}}^{G}$
such $\xi_{L}$ $\zeta$
$\pi^{\prime}$
$\tilde{G}^{\prime}$
that , and (resp. ) is a character of

$\pi^{\prime}\circ\tilde{\beta}^{L^{\prime}}\simeq c- Ind_{G_{L}^{\prime}}^{G^{\prime}},$
$(\xi_{L^{\prime}}\otimes\zeta^{\prime})$ $\xi_{L}$ $\xi_{L^{\prime}}$ $G_{L}$
(resp. ,) defined in subsection 8.2. $G_{L}^{\prime}$

796 S.-Y. PAN
A nice application of Theorem is that we can extend some of S. Kudla $C$
and S. Rallis’ result on theta dichotomy to other -adic and finite reductive dual $p$
pairs. First we recall their result on theta dichotomy for -adic unitary groups $p$
as follows. Let be an -Hermitian space over a quadratic extension $E$ of . $\mathscr{V}$

$\epsilon$ $F$
Let and be two -Hermitian spaces over $E$ defined in [HKS]. Let
$\mathscr{V}^{\prime+}$ $\mathscr{V}^{\prime-}$ $\epsilon^{\prime}$
$(\pi, V)$ be an irreducible supercuspidal representation of and (resp. ) $U(\mathscr{V})$

$\ell+$ $\ell^{-}$
be the smallest dimension of (resp. ) such that occurs in the theta $\mathscr{V}^{\prime+}$ $\mathscr{V}^{\prime-}$
$\pi$
correspondence for the pair (resp. ) with re- $(U(\mathscr{V}), U(\mathscr{V}^{\prime+}))$ $(U(\mathscr{V}),$ $U(\mathscr{V}^{\prime-}))$
spect to the splittings given in [K1]. Suppose that the dimensions of and $\mathscr{V}$
$\mathscr{V}^{\prime\pm}$
are all in the same parity. Then $\ell++\ell^{-}=2n+2$ where is the dimension of $n$
. It is interesting that the sum $\ell++\ell-does$ not depend on although each of

$\mathscr{V}$
$\pi$
$\ell+,$
does. $\ell^{-}$
Let $Z$ be an -Hermitian space over a -adic field or a finite field. Define $\epsilon^{\prime}$
$p$
$n_{0}(Z)$ as follows.
0if $Z$ is symplectic;

1, if $Z$ is finite Hermitian;
$n_{0}(Z):=\{$
2, if $Z$ is finite orthogonal or -adic Hermitian; $p$
4, if $Z$ is $p$ -adic orthogonal.
Define the sgn character of a classical group as follows. If is the trivial $G$ $G$
group or a symplectic group, let sgn be the trivial character of . If is a $G$ $G$
(nontrivial) orthogonal group (resp. unitary group), let sgn be the character of
order two whose restriction to the special orthogonal group (resp. special unitary
group) is trivial. The following two theorems are our results on theta di-
chotomy.
THEOREM D. Suppose that $(U(v), U(v^{\prime}))$ and are two related $(U(v), U(v^{\prime\prime}))$
finite reductive dual pairs deftned in subsection 12.1. Let be an irreducible $\zeta$
cuspidal representation of $U(v)$ . Let (resp. ) denote the smallest dimension of $\ell_{0}^{\prime}$ $\ell_{0}^{\prime\prime}$
(resp.
$v^{\prime}$
) such that (resp. OX $sgn$ ) occurs in the theta correspondence for the
$v^{\prime\prime}$
$\zeta$ $\zeta$
dual pair $(U(v), U(v^{\prime}))$ (resp. $(U(v),$ ). Then $U(v^{\prime\prime}))$
’
$\ell_{0}^{\prime}+\ell_{0}^{\prime\prime}=2n+n_{0}(v^{\prime})=2n+n_{0}(v$ $)$
where $n$
is the dimension of $v$
.
THEOREM E. Let be two related p-adic $(U(\mathscr{V}), U(\mathscr{V}^{\prime})),$ $(U(\mathscr{V}), U(\mathscr{V}^{\prime\prime}))$
reductive dual pair given in subsection 13.1 and 13.2. Suppose that is an $\pi^{\prime}$
irreducible depth zero supercuspidal representations of with a minimal K-type $U(\mathscr{V})$
. Let be the irreducible depth zero supercuspidal representations of

$(\underline{G_{L},}\zeta)$
$\pi^{\prime\prime}$
such that
$U(\mathscr{V})$
sgn OX Then $\pi^{\prime}\circ\tilde{\beta}^{L}=$ $(\pi^{\prime\prime}\circ\tilde{\beta}^{L})$
’ ’
$\ell_{0}^{\prime}+\ell_{0}^{\prime\prime}=2n+n_{0}(\mathscr{V}$ $)$
$=2n+n_{0}(\mathscr{V}$ $)$
where is the dimension of

$n$
and (resp. ) is the smallest dimension of
$\mathscr{V}$ $\ell_{0}^{\prime}$ $\ell_{0}^{\prime\prime}$
(resp. ) such that (resp.

$\mathscr{V}^{\prime\prime}$
) occurs in the theta correspondence for the

$\pi^{\prime}$ $\pi^{\prime\prime}$
reductive dual pair (resp. ).

$(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$ $(U(\mathscr{V}),$ $U(\mathscr{V}^{\prime\prime}))$
Theorems $D$ and should be true for more general class of irreducible
$E$
admissible representations but the author does not know how to tackle this
generalization. An interesting consequence of Theorems $D$ and is that there $E$
exist a chain of irreducible (super)cuspidal representations such that any two

successive representations are paired by the theta correspondence. A special case
of the chains is the chain of unipotent cuspidal representations of finite classical
groups. This special case is already studied by J. Adams and A. Moy in [AM].
Theorem $D$ provides another proof of their result.
The content of this paper is as follows. In section 1, we introduce basic
notation used in this paper and the concept of good lattices in an z-Hermitian
space. In this section, we also study the Schr\"odinger model and the generalized
lattice model of the Weil representation of a -adic symplectic group. In section $p$
2, we introduce basic definitions of reductive dual pairs and theta correspondence.

Depth zero minimal $K$-types for -adic classical groups and their metaplectic $p$
covers are defined in section 3. In section 4 we prove our first main result,
Theorem 4.2. From this theorem we conclude that depth zero representations
are preserved by the local theta correspondence. We note here that the proof
relies heavily on a deep result of J.-L. Waldspurger in [Wp]. In section 5 we
prove that the depth zero minimal $K$-types of paired representations are paired by
the theta correspondence for finite reductive dual pairs. Theorem 5.6 is our
second main result. In sections 67 we provide several consequences of Theorem
5.6. Iwahori-spherical representations and unipotent representations are studied
in these two sections respectively. In section 8 we recall the splitting with
respect to a generalized lattice model from [Pnl]. This splitting gives us the full
advantage of studying theta correspondence for split reductive dual pairs by using
minimal $K$-types. In section 9 we study the theta correspondence of irreducible
depth zero supercuspidal representations for split reductive dual pairs. We also
provide a few examples to illustrate the nice connection of theta correspondence
of depth zero supercuspidal representations of -adic groups and theta corre- $p$
spondence of cuspidal representation of finite groups. Section 10 is a remark on

Shalika’s and Tanaka’s work [Tn] on constructing representation of modulo
congruence group $SL_{2}(Z/p^{k}Z)$ from the point of view of theta correspondence.
In section 11 we recall Kudla and Rallis’ work on theta dichotomy for p-adic
unitary groups. Theorem 11.4 is the third main result in this paper, which re-
formulated Kudla and Rallis’ result from our point of view. Sections 12 and 13
798 S.-Y. PAN
are consequence of Theorem 11.4. In section 12, we study the theta dichotomy
for finite reductive dual pairs. In section 13 the theta dichotomy for p-adic
reductive dual pairs is discussed.
Part of this work is a chapter of the author’s Ph.D. dissertation at Cornell
University under the supervision of Prof. Dan Barbasch. The author would like
to thank Prof. Barbasch for his suggesting this research direction and many helps.
The author also would like to thank Prof. Jeffrey Adams and Prof. Stephen
Kudla for their interest in this work. Some of the work was written when the
author visited National Center for Theoretical Sciences at Hsinchu, Taiwan in the
summer of 1998. The author would like to thank Prof. Jing Yu and the institute
for their hospitality. This work was partially supported by a Hutchinson fel-
lowship of Department of Mathematics, Cornell University. Finally, the author
thanks the referee for several useful suggestions to improve the presentation of
this paper.
1. Preliminaries.
In this section, we provide the general setting of this work. Subsection 1.1
concerns the notation used throughout the paper. Materials in subsections 1.2
and 1.5 are from [Wp].
1.1. Notation.
Let be a nonarchimedean local field,$F$
be the ring of integers of be $\mathscr{O}_{F}$
$F,$ $\mathfrak{p}_{F}$
the prime ideal, be a uniformizer of be the (finite) residue

$\varpi_{F}$
$\mathscr{O}_{F},$ $f_{F}:=\mathscr{O}_{F}/\mathfrak{p}_{F}$
field, the identity automorphism of . We assume that the characteristic of

$\tau_{F}$
$F$
is odd. Let denote the cardinality of the finite field and ord :
$f_{F}$ $q$ be $f_{F}$
$F^{\times}\rightarrow Z$
the discrete valuation such that $ord(\varpi_{F})=1$ . We will fix a nontrivial (additive)
character of . $\psi$ $F$
Let be a quadratic extension of $E$

the ring of integers of a $F,$ $\mathscr{O}_{E}$
$E,$ $\varpi_{E}$
uniformizer of the residue field of the nontrivial automorphism of

$\mathscr{O}_{E},$
$f_{E}$ $E,$ $\tau_{E}$
$E$
over . We make the choice such that $F$
if is unramified, and $\varpi_{E}=\varpi_{F}$ $E$
if is ramified. Let
$\tau_{E}(\varpi_{E})=-\varpi_{E}$
be a central quaternion algebra over , $E$ $D^{\prime}$
$F$
be a uniformizer,
$\varpi_{D^{\prime}}$
the canonical involution of . We assume that $\tau_{D^{\prime}}$
$D^{\prime}$
. We fix to be one of the triples

$\tau_{D^{\prime}}(\varpi_{D^{\prime}})=-\varpi_{D^{\prime}}$
, $(D, \varpi, \tau)$ $(F, \varpi_{F}, \tau_{F})$
or
$(E, \varpi_{E}, \tau_{E})$
. Let fl be the ring of integers, be the maximal
$(D^{\prime}, \varpi_{D^{\prime}} , \tau_{D^{\prime}})$ $\mathfrak{p}$
ideal, be the residues field of $D$ . Let

$f_{D}$
: be the usual quotient map. $\Pi_{\mathscr{O}}$ $(!)\rightarrow f_{D}$
We shall assume that $D$ is commutative from section 8.

1.2. Good lattices in an -Hermitian space. $\epsilon$
Let be a (finite-dimensional) nondegenerate right -Hermitian space

$(\mathscr{V}, \langle , \rangle)$ $\epsilon$
over $D$ where is 1 or -1. Let be the group of isometries of

$\epsilon$
. $U(\mathscr{V})$ $(\mathscr{V}, \langle, \rangle)$
Let be a lattice in i.e., a (free) right -module whose rank is equal to the
$L$ $\mathscr{V}$ $\mathscr{O}$
dimension of . Fix an integer $\mathscr{V}$

$\kappa$
(more specific information about $\kappa$
will be
given in 2. 1. ). Define$($
$b)$
$L^{*}=$
{ $v\in \mathscr{V}|\langle v,$ $l\rangle\in \mathfrak{p}^{\kappa}$
for all $l\in L$
}. (1.2.a)
It is clear that is also a lattice in . The lattice
$L^{*}$
is called the dual lattice $\mathscr{V}$ $L^{*}$
(with respect to the integer ) of . The lattice is said to be self-dual if $\kappa$ $L$ $L$
$L^{*}=L$ . The lattice is called a good lattice if . Let be a

$L$ $L^{*}\varpi\subseteq L\subseteq L^{*}$ $L$
lattice in . A decomposition
$\mathscr{V}$
of subspaces is called L-admissible $\mathscr{V}=\oplus_{i=1}^{n}\mathscr{V}_{i}$
if . Define a valuation
$L=\sum_{i=1}^{n}(L\cap \mathscr{V}_{i})$
: given by $ord_{L}$ $\mathscr{V}\rightarrow Z$
$ord_{L}(v):=m$ if $v\in L\varpi^{m}-L\varpi^{m+1}$

. (1.2.b)
be a good lattice in . Then
Let $L$ $\mathscr{V}$
$l^{*}:=L^{*}/L$ and $ l:=L/L^{*}\varpi$ are vector
spaces over . Let : $f_{D}$ $\Pi_{L}*:L^{*}\rightarrow l^{*},$ $\Pi_{L}$ $L\rightarrow l$
be the quotient maps. We can
define sesquilinear forms , , , on $\langle$ $\rangle_{l}*$ $\langle$ $\rangle_{l}$
$l^{*}$
and respectively by$l$
$\langle\Pi_{L^{*}}(w), \Pi_{L^{*}}(w^{\prime})\rangle_{l^{*}}:=\Pi_{\mathscr{O}}(\langle w, w^{\prime}\rangle\varpi^{1-\kappa})$

,
(1.2.c)
$\langle\Pi_{L}(v), \Pi_{L}(v^{\prime})\rangle_{l}:=\Pi_{\mathscr{O}}(\langle v, v^{\prime}\rangle\varpi^{-\kappa})$
where ,$w,$ . Note that the forms ,

$w^{\prime}\in L^{*}$
$v,$
$v^{\prime}\in L$ $\langle$ $\rangle_{l^{*}}$
and , are non- $\langle$ $\rangle_{l}$
degenerate and depend on the choice of a prime element $\varpi$

. The following table
is from [Wp] lemme I.2.
A good lattice in is called maximal (resp. minimal) if it is a maximal

$L$ $\mathscr{V}$
element (resp. a minimal element) in the set of all good lattices in with the $\mathscr{V}$
partial order defined by inclusion. It is easy to see that is maximal (resp. $L$
minimal) if and only if the space (resp. ( , , )) is anisotropic. $(l^{*}, \langle , \rangle_{l^{*}})$ $l$ $\langle$ $\rangle_{l}$
Two lattices in are said to be equivalent if there is an element

$L_{1},$ $L_{2}$ $\mathscr{V}$
such that $g.L_{1}=L_{2}$ . The space

$g\in U(\mathscr{V})$ can be decomposed as an or- $\mathscr{V}$
thogonal direct sum where is anisotropic and is a direct sum of

$\mathscr{V}^{O}\oplus \mathscr{V}^{1}$ $\mathscr{V}^{O}$
$\mathscr{V}^{1}$
hyperbolic planes. There is a unique good lattice in . Choose a complete $A^{O}$ $\mathscr{V}^{O}$
polarization $\mathscr{V}^{1}=X+Y$
, a basis , . . . , of , and the dual basis
$X$ of $x_{1}$ $x_{r}$ $y_{1},$ $\ldots,y_{r}$
$Y$
where is the Witt index of . Then
$r$
$\mathscr{V}$
800 S.-Y. PAN
$B_{i}:=x_{1}\mathscr{O}\varpi+\cdots+x_{i}\mathscr{O}\varpi+x_{i+1}\mathscr{O}+\cdots+x_{r}\mathscr{O}+y_{r}\mathscr{O}+\cdots+y_{1}$ (!) (1.2. d)
is a good lattice in for each $0\leq i\leq r$ . Therefore $N_{i}:=A^{O}+B_{i}$ is a good $\mathscr{V}^{1}$
lattice in . It is not difficult to check that every good lattice in

$\mathscr{V}$
is equivalent $\mathscr{V}$
to one of for $i=0,$ . $N_{i}$

$\ldots,$
$r$
1.3. The “sgn” character.

Let be a (finite or -adic) symplectic, orthogonal or unitary group. We
$G$ $p$
will use the notation $sgn$ ”to denote the character of defined as follows. If $G$ $G$
is the trivial group or a symplectic group, let sgn be the trivial character of . If $G$
$G$
is a (nontrivial) orthogonal group (resp. unitary group), let sgn be the character
of order two whose restriction to the special orthogonal group (resp. special
unitary group) is trivial.
1.4. Weil representations and the metaplectic covers.
Let be a symplectic space over . Let be the Heisenberg
$(\mathscr{W}, \ll , \gg)$ $F$ $H(\mathscr{W})$
group associated to . Let be the irreducible representation of $(\mathscr{W}, \ll, \gg)$ $(\rho_{\psi}, \mathscr{S})$
with nontrivial central character by the Stone-Von Neumann theorem.

$H(\mathscr{W})$ $\psi$
symplectic group
$T\underline{he}$
acts on . Define the metaplectic cover $Sp(\mathscr{W})$ $H(\mathscr{W})$
of
$Sp(\mathscr{W})$
to be the topological subgroup of Aut(7) consisting
$Sp(\mathscr{W})$ $Sp(\mathscr{W})$ $\times$
of the pairs $(g, M[g])$ where $M[g]$ satisfies the condition

$M[g]\circ\rho_{\psi}(h)=\rho_{\psi}(g.h)\circ M[g]$ (1.4.a)
$fo\underline{rg}\in Sp(\mathscr{W})$ and any , $M[g]\in Aut(\mathscr{S})$ $h\in H(\mathscr{W})$ . The metaplectic group
$Sp(\mathscr{W})$
comes equipped with a representation $\omega\psi$ on given by
$\mathscr{S}$
$\omega\psi(g, M[g]):=M[g]$ . (1.4.b)
The representation of ) or the projective representation

$(\omega_{\psi}(g), \mathscr{S})$
$Sp\overline{(\mathscr{W}}$
of
$(M[g], \mathscr{S})$ $Sp(\mathscr{W})$
is called the Weil representation or the oscillator repre-
sentation.
1.5. Generalized lattice model of the Weil representation.

Let be the conductor of a character of i.e., is the smallest integer
$\mathfrak{p}_{F}^{\lambda_{F}}$
$\psi$ $F$ $\lambda_{F}$
such that is trivial. Let be a good lattice in (with respect to

$\psi|\lambda_{F}$
).
$\mathfrak{p}_{F}$
$B$ $\mathscr{W}$ $\lambda_{F}$
Let be the quotient $B^{*}/B$ . We know that

$b^{*}$
is a vector space over with a $b^{*}$
$f_{F}$
nondegenerate skew-symmetric form on by (1.2.c). Let $H(b^{*})$ be the $\ll,$ $\gg_{b}*$

$b^{*}$
Heisenberg group associated to the finite symplectic space . Let $(b^{*} , \ll , \gg_{b}* )$ $\overline{\psi}$
denote the character of defined by where . $f_{F}$ $\overline{\psi}(\Pi_{\mathscr{O}_{F}}(t)):=\psi(t\varpi_{F}^{\lambda_{F}-1})$ $t\in \mathscr{O}_{F}$
Let be the Schr\"odinger model of the Weil representation of the finite

$(\overline{\omega}_{\overline{\psi}}, S)$
symplectic groups $Sp(b^{*})$ associated to the data (cf. [Hwl]). $(b^{*}, \ll, \gg_{b}*,\overline{\psi})$
Although the skew-symmetric form and character depend on the choice $\ll,$ $\gg_{b^{*}}$
$\overline{\psi}$
of a prime element , the Weil representation does not. This can be seen $\varpi_{F}$ $\overline{\omega}_{\overline{\psi}}$
easily from the Schr\"odinger model. Let denote the representation of $H(b^{*})$ $\overline{\rho}_{\overline{\psi}}$
corresponding to the character on the space . Let . It $\overline{\psi}$

$S$ $H(B^{*}):=B^{*}\times \mathfrak{p}_{F}^{\lambda_{F}-1}$
is easy to check that $H(B^{*})$ is a subgroup of the Heisenberg group . We $H(\mathscr{W})$
have a reduction homomorphism

$\Pi_{H(B^{*})}$ : $H(B^{*})\rightarrow H(b^{*})$ by $(b, t)\mapsto(\Pi_{B}*(b), \Pi_{\mathscr{O}_{F}}(t\varpi_{F}^{1-\lambda_{F}}))$
,
where :
$\Pi_{B^{*}}$ $B^{*}\rightarrow b^{*}$
and $\Pi_{\mathscr{O}_{F}}$
: $\mathscr{O}_{F}\rightarrow f_{F}$
. Let $K_{B}$
denote the stabilizer of $B$
in
$Sp(\mathscr{W})$
, and
$K_{B}^{\prime}:=\{g\in K_{B}|(g-1).B^{*}\subseteq B\}$ . (1.5.a)
It is clear that is a normal subgroup of and is isomorphic to

$K_{B}^{\prime}$ $K_{B}$ $K_{B}/K_{B}^{\prime}$
$Sp(b^{*})$ . Let denote the representation of $H(B^{*})$ inflated from

$\tilde{\rho}_{\psi}$
by the $\overline{\rho}_{\overline{\psi}}$
projection , and be the representation of

$\Pi_{H(B^{*})}$ inflated from by the $\tilde{\omega}_{\psi}$ $K_{B}$ $\overline{\omega}_{\overline{\psi}}$
projection and the isomorphism . Let

$K_{B}\rightarrow K_{B}/K_{B}^{\prime}$ $K_{B}/K_{B}^{\prime}\simeq Sp(b^{*})$ $\mathscr{S}(B)$
denote the space of locally constant, compactly supported maps : such $f$ $\mathscr{W}\rightarrow S$
that
$f(b+w)=\psi(\frac{1}{2}\ll w,$ $b\gg)\tilde{\rho}_{\psi}(b).(f(w))$ , (1.5.b)
for any $w\in \mathscr{W},$ $b\in B^{*}$

. Define the action $\rho_{\psi}^{B}$
of $H(\mathscr{W})$
on $\mathscr{S}(B)$
by
$(\rho_{\psi}^{B}(w, t).f)(w^{\prime})=\psi(t)\psi(\frac{1}{2}\ll w^{\prime},$ $w\gg)f(w^{\prime}+w)$
for $w^{\prime}\in \mathscr{W}$

and . We can define $M_{B}[g]$ in Aut(7 (B)) such that
$(w, t)\in H(\mathscr{W})$
$\rho_{\psi}^{B}(h)$
and $M_{B}[g]$ satisfy (1.4.a). Moreover we know that
$(M_{B}[k] .f)(w)=\tilde{\omega}_{\psi}(k).(f(k^{-1}.w))$ (1.5.c)
for and$k\in K_{B}$. This realization of the Weil representation is known

$f\in \mathscr{S}(B)$
as a generalized lattice model. For any union of -cosets , define $B^{*}$ $R$
$\mathscr{S}(B)_{R}:=$
{ $f\in \mathscr{S}(B)|f$ has support in $R$
},
(1.5.d)
$\mathscr{S}(B)_{w}:=\mathscr{S}(B)_{B^{*}+w}$ for $w\in \mathscr{W}$
.
If $B=A$ happens to be self-dual, the is the well known lattice $(M_{A}[g] , \mathscr{S}(A))$
model of the Weil representation. In this case, it is clear that is trivial and $\tilde{\omega}_{\psi}$
is one-dimensional.
$\mathscr{S}(A)_{w}$
2. Reductive dual pairs and local theta correspondence.

In this section, we recall some basic definitions for reductive dual pairs and
local theta correspondence. Of course, the material in this section is well known.
Basic references are [Hw2], [MVW], or [Rb].
802 S.-Y. PAN
2.1. Reductive dual pairs.

Let $D$ be as defined in subsection 1.1. Let (resp. ( , , )) be $(\mathscr{V}, \langle , \rangle)$
$\langle$
$\rangle^{\prime}$
an -Hermitian (resp. ’-Hermitian) space over $D$ such that

$\epsilon$
. Define $\epsilon$
, which will be denoted by latter for simplicity. Define

$\mathscr{W}:=\mathscr{V}\otimes_{D}\mathscr{V}^{\prime}$ $\mathscr{V}\otimes \mathscr{V}^{\prime}$
a skew-symmetric -bilinear from on by $F$ $\ll,$ $\gg$ $\mathscr{W}$
$\ll,$ $\gg:=Trd_{D/F}(\langle, \rangle\otimes\tau\circ \langle, \rangle^{\prime})$

(2.1.a)
where $Trd_{D/F}$ denotes the reduced $trace$ from $D$ to . Recall that is a $F$ $\psi$
character of with conductoral exponent . Define if $D$ is or an

$F$ $\lambda_{F}$ $\lambda:=\lambda_{F}$ $F$
unramified quadratic extension of otherwise. Let (resp. ) be $F,$ $\lambda:=2\lambda_{F}-1$ $\kappa$

$\kappa^{\prime}$
the integer used to define the dual lattices in (resp. ) as in (1.2.a). We $\mathscr{V}$
make the following assumption

$\kappa+\kappa^{\prime}=\lambda$
(2.1.b)
throughout the paper. We also assume that the duality of lattices in is $\mathscr{W}$
defined with respect to the integer . $\lambda_{F}$
The pair is called a (type ) reductive dual pair in .

$(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$ $I$ $Sp(\mathscr{W})$
The reductive dual pair is called unramifted if it satisfies the $(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$
following two conditions: (1) $D$ is itself or a unramified quadratic extension of

$F$
$F,$(2) there exist self-dual lattices in both spaces and . $(\mathscr{V}, \langle , \rangle)$ $(\mathscr{V}^{\prime}, \langle, \rangle^{\prime})$
2.2. Local theta correspondence.

From the definition of form , we know that there is an $th\underline{e}$ $\ll,$ $\gg$
$embe\underline{ddi}ng$
:
$l_{\mathscr{V}^{\prime}}$
. Let be the inverse image of
$U(\mathscr{V})$
in .
$\underline{\rightarrow S}p(\mathscr{W})$ $U(\mathscr{V})$ $l_{\mathscr{V}^{\prime}}(U\underline{(\mathscr{V})})$ $Sp(\mathscr{W})$
The group is called the of . Let

$U(\mathscr{V})$
be defined $m\underline{etap}lectic$ $co\underline{ver}$
$U(\mathscr{V})$ $U(\mathscr{V}^{\prime})$
similarly. One can check that and ) commute with each other. Let $U(\mathscr{V})$ $\underline{U(}\mathscr{V}^{\prime}$
be the two-fold cover of

$U\overline{(\mathscr{V}})$
in . We know that ) is a totally $U(\mathscr{V})$ $U(\mathscr{V})$

$U\overline{(\mathscr{V}}$
disconnected group. A representations of is called admissible if $(\pi, V)$ $U(\mathscr{V})$
is multiplication by and
$\pi|_{C^{\times}}(z)$
is an admissible representation of a $z$
$\pi|\overline{U(\mathscr{V})}$
totally disconnected group.

Let be the Weil representation of ) with respect to the
$(\omega_{\psi}, \mathscr{S})$
of . It is known that
$ch\underline{ara}cter$ is an admissible
$\psi$ $F$
of $(\omega\psi, \mathscr{S})$ $rep\underline{rese}ntati\underline{on}$
’
. Then
$Sp(\mathscr{W})$
be as a representation $(\omega\psi, \mathscr{S})\underline{ca}n$ $\underline{re}garded$ $0\underline{fU}(\mathscr{V})$ $\times\underline{U}(\mathscr{V}$ $)$
’
the
$v\underline{ia}$
to and the homomorphism
$r\underline{estr}iction$ $U(\mathscr{V})\cdot U(\mathscr{V}^{\prime})$ $U(\mathscr{V})$ $\times U(\mathscr{V}$ $)$ $\rightarrow$
. Let (resp.
$U(\mathscr{V})\cdot U(\mathscr{V}^{\prime})$
be an admissible repre- $(\pi, V)$ $(\pi^{\prime},\underline{V^{\prime})})$ $ir\underline{red}ucible$
sentation of the metaplectic group (resp. ). The representation $U(\mathscr{V})$ $U(\mathscr{V}^{\prime})$
is to correspond to the representation

$(\underline{\pi,V})$ if there is a nontrivial
$s\underline{aid}$
$(\pi^{\prime}, V^{\prime})$
$U(\mathscr{V})$
-map $\times U(\mathscr{V}^{\prime})$
$\Pi$
: $\mathscr{S}\rightarrow V\otimes_{C}V^{\prime}$
. (2.2.a)
This establishes a correspondence, called the local theta correspondence or Howe
duality, between some irreducible admissible representations of and some $U(\mathscr{V})$
irreducible admissible representations of ). It is proved by R. Howe (cf. $U\overline{(\mathscr{V}}^{\prime}$
[MVW], chapitre 5) and J.-L. Waldspurger (cf. [Wp]) that the local theta cor-
respondence is one-to-one when the residue characteristic of is odd. $F$
2.3. Finite reductive dual pairs.

Reductive dual pairs can also be defined over a finite field. Let be a finite $f$
field and be or a quadratic extension of . Let (resp. ) be a z-Hermitian

$d$ $f$ $f$ $v$
$v^{\prime}$
(resp. -Hermitian) spaces over such that

$\epsilon^{\prime}$
. Then forms $d$ $\epsilon\epsilon^{\prime}=-1$

$(U(v), U(v^{\prime}))$
a (finite) reductive dual pair in the finite symplectic group . It is $Sp(v\otimes_{d}v^{\prime})$
known that the theta correspondence for a finite reductive dual pair is in general
not one-to-one. For convenience, we shall allow the dimension of (or ) to $v$
$v^{\prime}$
be zero. Therefore we make the following conventions. We define the Weil

representation of the symplectic group on a zero-dimensional space to be the
trivial representation (of the trivial group). If one of the spaces is trivial, $v,$
$v^{\prime}$
then the theta correspondence for the dual pair is just the trivial $(U(v), U(v^{\prime}))$
representation corresponds to the trivial representation.
3. Depth zero minimal $K$ -types.

In this section, we recall the definition of depth zero minimal $K$ -types for p-
adic classical groups modified from [MP1] and [MP2].
3.1. Result of Moy and Prasad.
Let be a point in the Bruhat-Tits building of a -adic classical group
$x$ $p$
$G:=U(\mathscr{V})$ . Let denote the stabilizer of , and $G_{X}$

be the pro-nilradical $x$ $G_{x,0}+$
of . It is known that
$G_{X}$
is a normal subgroup of . The quotient
$G_{x,0}+$ $G_{X}$
$G_{x,0}/G_{x,0}+$ is a finite classical group. A depth zero minimal -type of an ir-

$K$
reducible admissible representation of is a pair where is an $(\pi, V)$ $G$ $(G_{X}, \zeta)$ $\zeta$
irreducible representation of the group such that is nontrivial $G_{X}/G_{x,0}+$

$V^{G_{x,0+}}$
and, as a representation of , contains . An irreducible admissible $G_{X}/G_{x,0}+$ $\zeta$
representations containing a depth zero minimal $K$-type is said to be of

$(\pi, V)$
depth zero. We should notice that the definition of a minimal $K$-type here
is different from the original definition in [MP1] and [MP2]. Here we do not
require to be cuspidal. However, the notion of depth of an irreducible ad-
$\zeta$
missible representation is still the same. $\pi$
It is known that if is a point in the building of , then there exists a vertex

$x$ $G$
$v$
of a chamber such that . Therefore an irreducible admissible $G_{v,0}+\subseteq G_{x,0}+$
representation of is of depth zero if and only if it has nontrivial vectors fixed

$G$
by $G_{v,0}+$for some vertex . $v$
3.2. Good lattices and vertices of the building.

Let be the classical group and be a good lattice (cf. subsection
$G$ $U(\mathscr{V})$ $L$
1.2) in . Define $\mathscr{V}$

804 S.-Y. PAN
$G_{L}:=\{g\in U(\mathscr{V})|g.L=L\}$ ,
$G_{L,0}+:=\{g\in U(\mathscr{V})|(g-1).L^{*}\subseteq L, (g-1).L\subseteq L^{*}\varpi\}$ , (3.2.a)
$G_{L,1}:=\{g\in U(\mathscr{V})|(g-1).L\subseteq L\varpi\}$ .
It is clear that and are open compact subgroups of . In $G_{L},$ $G_{L,0}+$ $G_{L,1}$ $U(\mathscr{V})$
fact, is a maximal open compact subgroup of

$G_{L}$
. It is proved by H. $U(\mathscr{V})$
Hijikata [Hj] that any maximal open compact subgroup of a classical group
is conjugate to one of
$U(\mathscr{V})$
where is a good lattice defined in subsection $G_{N_{i}}$ $N_{i}$
1.2.
From [BT], it is not difficult to see that for a vertex in the Bruhat-Tits $v$
building of , there exists a good lattice in such that $G_{v}=G_{L}$ and

$G$ $L$ $\mathscr{V}$
$G_{v,0}+=G_{L,0}+$ . And it is not difficult to figure that an irreducible admissible

representation of is depth zero if and only if it has nontrivial vectors fixed by $G$
for some good lattice

$G_{L,0}+$ in . $L$ $\mathscr{V}$
3.3. Depth zero minimal $K$-types for metaplectic covers.

To extend the concept of depth zero minimal $K$-types to the metaplectic
cover , we need to fix a splitting :
$\tilde{G}$
where I is a (fixed) Iwahori subgroup $\beta$

of (this was pointed out to me by Jiu-Kang Yu). In this subsection, we want

$G$
to discuss this issue.

Let be the irreducible representation of the Heisenberg group
$(\rho_{\psi}, \mathscr{S})$ $H(\mathscr{W})$
with respect to a nontrivial central character . For , let $M[g]$ $\psi$ $g\in Sp(\mathscr{W})$
be an element in Aut satisfying (1.4.a). Define a cocycle : $(\mathscr{S})$ $c$ $ Sp(\mathscr{W})\times$
C’ with respect to the map $M$ :

$ Sp(\mathscr{W})\rightarrow$
Aut(7 ) by $Sp(\mathscr{W})$ $\rightarrow$
$M[g]\circ M[g^{\prime}]=c(g, g^{\prime})M[gg^{\prime}]$ . (3.3.a)
Let be a reductive dual pair in . Recall that the met-

$(U(\mathscr{V}), U(\underline{\mathscr{V}^{\prime})})$ $Sp(\mathscr{W})$
aplectic cover is the of elements of the form for

$U(\mathscr{V})$
$g\underline{rou}p$
$(l_{\mathscr{V}^{\prime}}(g) , tM[l_{\mathscr{V}^{\prime}}(g)])$
$t\in C’$ , and the extension is given by . $U(\mathscr{V})\rightarrow U(\mathscr{V})$ $(l_{\mathscr{V}^{\prime}}(g), tM[l_{\mathscr{V}^{\prime}}(g)])\mapsto g$
’
Let $H$ be a subgroup of . A function : is called a splitting of $U(\mathscr{V})$ $\beta$ $H\rightarrow C$
the cocycle , if it satisfies $c|_{l_{\mathscr{V}}’(H)\times l_{\mathscr{V}(H)}}$
$c(l_{\mathscr{V}^{\prime}}(g), l_{\mathscr{V}^{\prime}}(g^{\prime}))=\beta(gg^{\prime})\beta(g)^{-1}\beta(g^{\prime})^{-1}$
(3.3.b)
for any $g,$ $g^{\prime}\in H$ . If is a splitting, the map

$\beta$
: defined by $\tilde{\beta}$
$H\rightarrow\tilde{H}$
$g\mapsto(l_{\mathscr{V}^{\prime}}(g),\beta(g)M[l_{\mathscr{V}^{\prime}}\underline{(g})])$ is a group homomorphism where is the inverse $\tilde{H}$
image of $(H)$ in $l_{\mathscr{V}^{\prime}}$

$U(\mathscr{V})$
. We will also called the splitting of the metaplectic $\tilde{\beta}$
cover . $\tilde{H}$
Let be a good lattice in such that

$L$
. Then I is contained in $\mathscr{V}$
$G_{L,0}+\subseteq I$
. Let
$G_{L}$
be a generalized lattice model with respect to defined
$(M_{B}[g], \mathscr{S}(B))$ $B$
by
$B:=B(L, L^{\prime}):=L^{*}\otimes L^{\prime}\cap L\otimes L^{\prime*}$ .
for some good lattice in . It is clear that is a lattice in . $L^{\prime}$ $\mathscr{V}^{\prime}$

$B$ $\mathscr{W}:=\mathscr{V}\otimes \mathscr{V}^{\prime}$
Let $c_{B}(g, g)$ denote the cocycle defined in (3.3.a) with respect to $M_{B}[g]$ . From
(1.5.c), we see that . It is clear that . Hence the $c_{B}|_{K_{B}\times K_{B}}=1$ $l_{\mathscr{V}^{\prime}}(G_{L})\subseteq K_{B}$
mapping : ’
by $\beta_{I}(g):=1$ for $g\in I$ is a splitting of
$\beta_{I}$
$I\rightarrow C$
$ c_{B}|_{l_{\mathscr{V}}’(I)\times l_{\mathscr{V}(I)}},\cdot$
Therefore the map : defined by is an injective

$\tilde{\beta}_{I}$
$ g\mapsto$ $(l_{\mathscr{V}^{\prime}}(g), M_{B}[l_{\mathscr{V}^{\prime}}(g)])$
homomorphism. Although the lattice depends on the good lattice in , it $B$ $L^{\prime}$ $\mathscr{V}^{\prime}$
turns out that the splitting only depends on the parity of the dimensions of $\tilde{\beta}_{I}$
’
and
$L^{\prime*}/L^{\prime}$
, which is the same for all good lattices in
$ L^{\prime}/L^{\prime*}\varpi$
(cf. [Pnl]). $\mathscr{V}$
Therefore depends only on but not on . We know that $I=G_{x_{0}}$ where

$\tilde{\beta}_{I}$
$\mathscr{V}^{\prime}$ $L^{\prime}$
is the barycenter of some Weyl chamber

$x_{0}$ in the Bruhat-Tits building of . $C_{0}$ $G$
It is known that if . Therefore we will identify with

$G_{x,0}+\subseteq I$ $x\in C_{0}$ $G_{x,0}+$
i.e., we regard
$\beta_{I}(G_{x,0}+)$ as a subgroup of via the splitting . For a $G_{x,0}+$
$\tilde{G}$
$\beta_{I}$
general point in the building, there exists an element $g\in G$ such that $g.y\in C_{0}$ .
$y$
Therefore $gG_{y,0}+g^{-1}=G_{g.y,0}+\subseteq I$ . From now on, when we mention or $G_{x,0}+$
, we always restrict ourselves to the situation that

$G_{L,0}+$ or $G_{x,0}+\subseteq I$ $G_{L,0}+\subseteq I$
and regard and as subgroups of . $G_{X},0+$ $G_{L},0+$

$\tilde{G}$
An irreducible admissible representation $(\pi, V)$ of is said to be of depth $\tilde{G}$
zero if is not trivial for some good lattice in

$V^{G_{L,0+}}$
(such that ). $L$ $\mathscr{V}$
$G_{L,0}+\subseteq I$
It is clear that this definition is independent of the choice of an Iwahori subgroup

. This can be seen as follows. Suppose that , I2 are two Iwahori subgroups
$I$ $I_{1}$
of . It is known that there is an element $g\in G$ such that $I_{1}=gI_{2}g^{-1}$ .

$G$
Therefore $gG_{L,0}+g^{-1}=G_{g.L,0}+\subseteq I2$ if . Hence is nontrivial $G_{L,0}+\subseteq I_{1}$

$V^{\beta_{l_{1}}(G_{L,0+})}$
if and only if is nontrivial. $V^{\beta_{l_{2}}(G_{gL,0+})}$
Suppose that is not trivial and is defined with respect to $M_{B}[g]$

$V^{G_{L,0+}}$ $\tilde{G}$
where $B:=B(L, L^{\prime})$ for some . Then : by $L^{\prime}$

$\beta$
$G_{L}\rightarrow\tilde{G}_{L}$
$g\mapsto(l_{\mathscr{V}^{\prime}}(g) , M_{B}[l_{\mathscr{V}^{\prime}}(g)])$
extends . So we can regard $\beta_{I}$

as a subgroup of . We say that contains a $G_{L}$
$\tilde{G}$
$\pi$
minimal $K$-type if is nontrivial and contains . $(G_{L}, \zeta)$

$V^{G_{L,0+}}$ $\zeta$
4. Preservation of depth zero representations.

In this section we prove our first main result, which indicates that the depth
zero representations are preserved by the local theta correspondence.
4.1.
Let $(G, G^{\prime}):=(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$
be a reductive dual pair as usual. Suppose
that $L$
(resp. $L^{\prime}$
) is a good lattice in (resp. ). Define $\mathscr{V}$

$B:=B(L, L^{\prime}):=L^{*}\otimes L^{\prime}\cap L\otimes L^{\prime*}$ . (4.1.a)

’
It is clear that $B$
is a lattice in $\mathscr{W}:=\mathscr{V}\otimes \mathscr{V}$
and is equal to $L\otimes L^{\prime}+$
806 S.-Y. PAN
$L^{*}\varpi\otimes L^{\prime*}$
. Hencehave we $ B(L, L^{\prime})^{*}=L^{*}\otimes L^{\prime}+L\otimes L^{\prime*}=L\varpi^{-1}\otimes L^{\prime}\cap$
$L^{*}\otimes L^{\prime*}$
. It is easy to check that $B(L, L^{\prime})$
is a good lattice in $\mathscr{W}$
i.e.,
$B(L, L^{\prime})^{*}\varpi_{F}\subseteq B(L, L^{\prime})\subseteq B(L, L^{\prime})^{*}$
.
Recall that is the stabilizer of in and is the subgroup of $K_{B}$ $B$ $Sp(\mathscr{W})$ $K_{B}^{\prime}$
elements such that $(g-1).B^{*}\subseteq B$ . It is easy to see that

$g$ is a subgroup of $G_{L,0}+$
. So is
$K_{B}^{\prime}$
. Recall that we fix Iwahori subgroups of re-
$G_{L,0^{+}}^{\prime},$ $I,$
$I^{\prime}$
$G,$ $G^{\prime}$
spectively and r q re that . We regard , as $e$ $ui$ $G_{L,0}+\subseteq I,$ $G_{L,0^{+}}^{\prime},\subseteq I^{\prime}$ $G_{L,0}+,$ $G_{L,0^{+}}^{\prime}$
subgroups of via identifying , with $Sp(\mathscr{W})$ $G_{L,0}+,$ $G_{L,0^{+}}^{\prime}$ $\beta_{I}(G_{L,0}+),$ $\beta_{I^{\prime}}(G_{L,0^{+}}^{\prime},)$
respectively.
PROPOSITION. Let $A$
be a good lattice in such that $\mathscr{W}$
$A^{*}\subseteq B(L, L^{\prime})^{*}$
. Then
the subspace $\mathscr{S}(A)_{B(L,L)^{*}}$ , of is fixed pointwise by
$\mathscr{S}(A)$ $G_{L,0}+$ and $G_{L,0^{+}}^{\prime},$
.
The proof of this proposition is postponed to subsection 4.10.
4.2.
The following is our main result of this section, whose proof will be
postponed to subsection 4. 12.
THEOREM. Let $(G, G^{\prime}):=(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$
be a reductive dual pair and be $\psi$
a nontrivial character of F. Suppose that $L$

is a good lattice in . Then $\mathscr{V}$
$\mathscr{S}^{G_{L,0+}}=\omega_{\psi}(\mathscr{H}^{\prime})$
. $(\sum_{L^{\prime}\in 9(\Gamma_{M}^{\prime})}\mathscr{S}^{B(L,L^{\prime})})$
(4.2.a)
where denotes the set of good lattices

$\ovalbox{\tt\small REJECT}(\Gamma_{M}^{\prime})$
$cont\underline{ain}ed$ in a fixed maximal good
lattice in is the Hecke algebra of
$\Gamma_{M}^{\prime}$ $\mathscr{V}^{\prime},\underline{\mathscr{H}^{\prime}}$ $U(\mathscr{V}^{\prime})$
and $(\omega\psi, \mathscr{S})$
is the Weil
representation of . $Sp(\mathscr{W})$
4.3.
COROLLARY.
Let be a reductive dual pair. Let $(G, G^{\prime}):=(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$
(resp. ) be an irreducible admissible representation of

$(\pi\underline{V},)$
(resp.
$(\pi^{\prime},$ $V^{\prime})$ $U(\mathscr{V})$
Suppose that is nontrivial for some good lattice in . Then there exists
$V^{G_{L,0+}}$ $L$ $\mathscr{V}$
a good lattice in such that is nontrivial. $L^{\prime}$ $\mathscr{V}^{\prime}$

$V^{\prime G_{L^{\prime},0+}^{\prime}}$
PROOF. Suppose that is nontrivial for some good lattice in . $V^{G_{L,0+}}$ $L$ $\mathscr{V}$
Let 77 be the projection where is the Weil representation. We $\mathscr{S}\rightarrow V\otimes_{C}V^{\prime}$ $\mathscr{S}$
have a nontrivial surjective map . Because we assume $\mathscr{S}^{G_{L,0+}}\rightarrow V^{G_{L,0+}}\otimes_{C}V^{\prime}$
that the space is nontrivial, by Theorem 4.2 there is an element $V^{G_{L,0+}}$

$f\in \mathscr{S}^{B(L,L^{\prime})}$
in such that $\Pi(f)$ is not zero. But $L^{\prime}$
$\Pi(f)$ is fixed $by,,G_{L,0}+G$ and , by Proposition 4.1. Therefore $\Pi(f)$ belongs $G_{L,0^{+}}^{\prime}$
to . Henc is also nontrivial.

$V^{G_{L,0+}}\otimes_{C}VL^{\prime},0+$ $e$
$\square $
4.4.
COROLLARY. Let be a reductive dual pair. Let $(G, G^{\prime}):=(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$
(resp. ) be an irreducible admissible representation of

$(\pi\underline{V},)$ (resp.
Then the depth of is zero if and only if the depth of is zero. $\pi$
$\pi^{\prime}$
As in subsection 3.3 we know an irreducible admissible represen-

PROOF.
tation of is of depth zero if and only if
$(\pi, V)$ is nontrivial for some $U(\mathscr{V})$
$V^{G_{L,0+}}$
good lattice in . Hence this corollary follows immediately from Corollary

$L$ $\mathscr{V}$
4.3. $\square $
4.5.
Now we start the preparation for the proofs of Proposition 4.1 and Theorem
4.2. First we need to introduce the Cayley transforms. Recall that $(\mathscr{V}, \langle , \rangle)$
is a non-degenerate -Hermitian space. Let be the space of elements $\epsilon$

$u(\mathscr{V})$
such that
$c\in End_{D}(\mathscr{V})$ for all . If is an ele- $\langle c.v, v^{\prime}\rangle+\langle v, c.v^{\prime}\rangle=0$
$v,$
$v^{\prime}\in \mathscr{V}$
$c$
ment in and $1+c$ is invertible, we define $u(c):=(1-c)(1+c)^{-1}$ . It is

$u(\mathscr{V})$
easy to check that $u(c)$ is an element in when it is defined. If $U(\mathscr{V})$ $u\in U(\mathscr{V})$
and $1+u$ is invertible, we define $c(u):=(1-u)(1+u)^{-1}$ . It is also clear that

$c(u)$ is an element in . For any two elements in , we define $u(\mathscr{V})$
$x,$ $y$
$\mathscr{V}$
:
$c_{x,y}$ by$\mathscr{V}\rightarrow \mathscr{V}$
. (4.5.a) $ c_{x,y}.v:=x\langle y, v\rangle-\epsilon y\langle x, v\rangle$
It is easy to check that belongs to for . If $1+c_{x,y}$ is in- $c_{x,y}$

$u(\mathscr{V})$
$x,$ $y\in \mathscr{V}$
vertible, define $u_{x,y}:=u(c_{x,y})$ i.e., $u_{x,y}=(1-c_{x,y})(1+c_{x,y})^{-1}$ . Then it is easy to

check that $1+u_{x,y}$ is invertible and $=c_{x,y}$ . The following lemma is from $c(u_{x,y})$
[Wp].
LEMMA. Suppose that is a good lattice in and are elements in $L$ $\mathscr{V}$
$x,$ $y$ .
$\mathscr{V}$
(i) If $ord_{L}(x)+ord_{L}(y)\geq-\kappa$ and $ord_{L^{*}}(x)+ord_{L^{*}}(y)\geq 1-\kappa$ , then $u_{x,y}$ is

deftned and belongs to . $G_{L},0+$
(ii) If $ord_{L}(x)+ord_{L^{*}}(y)\geq 1-\kappa$ and $ord_{L}(y)+ord_{L^{*}}(x)\geq 1-\kappa$ , then $u_{x,y}$
is deftned and belongs to . $G_{L,1}$
PROOF. Part (i) is lemme I.17 of [Wp]. Part (ii) can be proved similarly.
$\square $
4.6.
We fix a maximal good lattice $\Gamma_{M}^{\prime}$
and a minimal good lattice $\Gamma_{m}^{\prime}$
in $\mathscr{V}^{\prime}$
such
that . $\Gamma_{m}^{\prime}\subseteq\Gamma_{M}^{\prime}$
LEMMA. Let $L$

be a good lattice in $\mathscr{V}$
. Then $L^{*}\otimes\Gamma_{M}^{\prime}\cap L\otimes\Gamma_{m}^{\prime*}$

is a good
lattice in . $\mathscr{W}$
808 S.-Y. PAN
PROOF. From I.15 in [Wp], we know that there exists a decomposition

$\mathscr{V}=X_{1}\oplus X_{2}$
such that $L=(L\cap X_{1})\oplus(L\cap X_{2})$ and $ L^{*}=(L\cap X_{1})\oplus$
$(L\cap X_{2})\varpi^{-1}$
. There also exists a decomposition such $\mathscr{V}^{\prime}=Y_{1}\oplus Y_{2}\oplus Y_{3}\oplus Y_{4}$
that
$\Gamma_{m}^{\prime}=(\Gamma_{M}^{\prime}\cap Y_{1})\varpi\oplus(\Gamma_{M}^{\prime}\cap Y_{2})\oplus(\Gamma_{M}^{\prime}\cap Y_{3})\oplus(\Gamma_{M}^{\prime}\cap Y_{4})$
$\Gamma_{M}^{\prime}=(\Gamma_{M}^{\prime}\cap Y_{1})\oplus(\Gamma_{M}^{\prime}\cap Y_{2})\oplus(\Gamma_{M}^{\prime}\cap Y_{3})\oplus(\Gamma_{M}^{\prime}\cap Y_{4})$
$\Gamma_{M}^{\prime*}=(\Gamma_{M}^{\prime}\cap Y_{1})\oplus(\Gamma_{M}^{\prime}\cap Y_{2})\varpi^{-1}\oplus(\Gamma_{M}^{\prime}\cap Y_{3})\oplus(\Gamma_{M}^{\prime}\cap Y_{4})$
$\Gamma_{m}^{\prime*}=(\Gamma_{M}^{\prime}\cap Y_{1})\oplus(\Gamma_{M}^{\prime}\cap Y_{2})\varpi^{-1}\oplus(\Gamma_{M}^{\prime}\cap Y_{3})\varpi^{-1}\oplus(\Gamma_{M}^{\prime}\cap Y_{4})$

.
From the above decompositions it is easy to check that

$L\otimes\Gamma_{m}^{\prime*}\cap L^{*}\otimes\Gamma_{M}^{\prime}\subseteq L^{*}\otimes\Gamma_{m}^{\prime}+L\otimes\Gamma_{M}^{\prime*}=(L\otimes\Gamma_{m}^{\prime*}\cap L^{*}\otimes\Gamma_{M}^{\prime})^{*}$
.
Moreover we have
$L^{*}\otimes\Gamma_{m}^{\prime}+L\otimes\Gamma_{M}^{\prime*}\subseteq L^{*}\otimes\Gamma_{M}^{\prime}+L\otimes\Gamma_{M}^{\prime*}=L\varpi^{-1}\otimes\Gamma_{M}^{\prime}\cap L^{*}\otimes\Gamma_{M}^{\prime*}$
$\subseteq(L\otimes\Gamma_{m}^{\prime*}\cap L^{*}\otimes\Gamma_{M}^{\prime})\varpi^{-1}$
.
Hence $L^{*}\otimes\Gamma_{M}^{\prime}\cap L\otimes\Gamma_{m}^{\prime*}$

is a good lattice in $\mathscr{W}$
. $\square $
4.7.
LEMMA. Let be a maximal good lattice in . $\Gamma_{M}^{\prime}$
fixed $\mathscr{V}^{\prime}$
(i) Suppose that are two -modules in such that $M_{1},$ $M_{2}$ $\mathscr{O}$
$ M_{1}\subseteq M_{2}\subseteq$
, and
$M_{1}\varpi^{-1},$
. Then there exists a good $M_{1}\subseteq\Gamma_{M}^{\prime}$ $\langle M_{1} , M_{2}\rangle^{\prime}\subseteq \mathfrak{p}^{\kappa^{\prime}}$
lattice in such that and .

$L^{\prime}$ $\mathscr{V}^{\prime}$
$L^{\prime}\subseteq\Gamma_{M}^{\prime},$ $M_{1}\subseteq L^{\prime}$ $M_{2}\subseteq L^{\prime*}$
(ii) Let be a minimal good lattice in such that

$\Gamma_{m}^{\prime}$
. Suppose $\mathscr{V}^{\prime}$
$\Gamma_{m}^{\prime}\subseteq\Gamma_{M}^{\prime}$
that are two -modules in such that

$M_{1},$ , $M_{2}$ $\mathscr{O}$
$M_{1}\subseteq M_{2}\subseteq M_{1}\varpi^{-1}$
and . Then there exists a good

$M_{1}\subseteq\Gamma_{M}^{\prime},$ $M_{2}\subseteq\Gamma_{m}^{\prime*},$ $\langle M_{1} , M_{2}\rangle^{\prime}\subseteq \mathfrak{p}^{\kappa^{\prime}}$
lattice in such that and .

$\Gamma_{m}^{\prime}\subseteq L^{\prime}\subseteq\Gamma_{M}^{\prime},$ $M_{1}\subseteq L^{\prime}$ $M_{2}\subseteq L^{\prime*}$
(iii) Suppose that $M$ is -module in such that and $a\mathscr{O}$ $\mathscr{V}^{\prime}$
$\langle M, M\rangle^{\prime}\subseteq \mathfrak{p}^{\kappa^{\prime}}$
. Then $M$ is contained in .

$M\subseteq\Gamma_{M}^{\prime*}$ $\Gamma_{M}^{\prime}$
PROOF. Without loss of generality, we may assume that are lattices $M_{1},$ $M_{2}$
in . From the assumption we have

. So $M_{2}+\Gamma_{M}^{\prime*}\subseteq M_{1}\varpi^{-1}+\Gamma_{M}^{\prime*}\subseteq\Gamma_{M}^{\prime}\varpi^{-1}$
we may regard as an -subspace of $M_{2}/(M_{2}\cap\Gamma_{M}^{\prime*})$ $\simeq(M_{2}+\Gamma_{M}^{\prime*})/\Gamma_{M}^{\prime*}$ $f_{D}$
. Let (for some finite index set ) be a subset of

$\Gamma_{M}^{\prime}\varpi^{-1}/\Gamma_{M}^{\prime*}$
such $\{z_{i}\}_{i\in J}$ $J$ $M_{2}$
that the images of these in are linearly independent over . $z_{i}$

$M_{2}/(M_{2}\cap\Gamma_{M}^{\prime*})$ $f_{D}$
From the assumption we have . Therefore $\langle M_{2}, M_{2}\rangle^{\prime}\subseteq \mathfrak{p}^{\kappa^{\prime}-1}$

$\langle z_{i}\varpi, z_{j}\varpi\rangle^{\prime}\subseteq$
for any $j\in J$ . Therefore the set

$\mathfrak{p}^{\kappa^{\prime}+1}$
as subset of satisfies the

$i,$ $\{z_{i}\varpi\}_{i\in J}$ $\Gamma_{M}^{\prime}$
condition in [Wp] corollaire I.8 with $n=1$ . Thus there exists a -admissible $\Gamma_{M}^{\prime}$
decomposition where $X$, are totally isotropic and in

$\mathscr{V}^{\prime}=X\oplus \mathscr{V}^{\prime\circ}\oplus Y$ $Y$
duality, and a basis of $X$ such that for every . Note $\{v_{i}\}_{i\in J}$ $ v_{i}-z_{i}\varpi\in\Gamma_{M}^{\prime}\varpi$ $i\in J$
that a decomposition of is -admissible if and only if it is -admissible. $\mathscr{V}^{\prime}$

$\Gamma_{M}^{\prime}$ $\Gamma_{M}^{\prime*}$
Therefore from the choice of the set we have $\{z_{i}\}_{i\in J}$
$M_{2}+\Gamma_{M}^{\prime*}=(\Gamma_{M}^{\prime*}\cap X)\varpi^{-1}\oplus(\Gamma_{M}^{\prime*}\cap \mathscr{V}^{\prime\circ})\oplus(\Gamma_{M}^{\prime*}\cap Y)$

. (4.7.a)
From (4.7.a), we see that is a good lattice in . Define $(M_{2}+\Gamma_{M}^{\prime*})^{*}$

. Therefore It is clear that

$L^{\prime}:=(M_{2}+\Gamma_{M}^{\prime*})^{*}=M_{2}^{*}\cap\Gamma_{M}^{\prime}$ $L^{\prime*}=M_{2}+\Gamma_{M}^{\prime*}.$
and . Hence
$L^{\prime}\subseteq\Gamma_{M}^{\prime}$
. Since and , we
$L^{\prime}\subseteq M_{2}^{*}$ $M_{2}\subseteq L^{\prime*}$ $M_{1}\subseteq M_{2}^{*}$ $M_{1}\subseteq\Gamma_{M}^{\prime}$
have . Then satisfies all requirements.

$M_{1}\subseteq L^{\prime}$ $L^{\prime}$
For (ii), let be given as in the proof of (i). We only need to check that
$L^{\prime}$
. As in the previous paragraph, we have

$\Gamma_{m}^{\prime}\subseteq L^{\prime}$
Hence $L^{\prime*}=M_{2}+\Gamma_{M}^{\prime*}.$
because and
$L^{\prime*}\subseteq\Gamma_{m}^{\prime*}$
. Therefore . $M_{2}\subseteq\Gamma_{m}^{\prime*}$ $\Gamma_{M}^{\prime*}\subseteq\Gamma_{m}^{\prime*}$ $\Gamma_{m}^{\prime}\subseteq L^{\prime}$
Now we prove part (iii). Let be a set of vectors in $M$ such that their $\{x_{i}\}_{i\in J}$
images in the quotient are linearly independent

$\{\overline{x}_{i}\}_{i\in J}$ $(M+\Gamma_{M}^{\prime})/\Gamma_{M}^{\prime}\subseteq\Gamma_{M}^{\prime*}/\Gamma_{M}^{\prime}$
for some finite index set . Let , denote the form on the space by $J$ $\langle$
$\rangle^{\prime*}$
$\Gamma_{M}^{\prime*}/\Gamma_{M}^{\prime}$
(1.2.c). Because we assume that must be zero for all $\langle M, M\rangle^{\prime}\subseteq \mathfrak{p}^{\kappa^{\prime}},$ $\langle\overline{x}_{i},\overline{x}_{i}\rangle^{\prime*}$
. But we know the form ,

$i\in J$
is anisotropic because is a maximal good $\langle$
$\rangle^{\prime*}$
$\Gamma_{M}^{\prime}$
lattice. Hence all must be zero i.e., $M$ is contained in . $\overline{x}_{i}$ $\Gamma_{M}^{\prime}$ $\square $
4.8.
LEMMA. Let be a good lattice in and be an element in . Let $K$ be $A$ $\mathscr{W}$
$w$ $\mathscr{W}$
the subgroup of
’
of elements such that $(g-1).w$ belongs to A. Then the map
$K_{A}^{\prime}$
$g$
:
$\psi_{w}$
deftned by $g\mapsto\psi((1/2)\ll(g-1).w, w\gg)$ is a character of $K$.
$K\rightarrow C$
Let PROOF. be two elements in $K$. Then both $(g_{1}^{-1} - 1).w$ and $g_{1},$ $g_{2}$
$(g_{2} - 1).w$ belong to . It is clear that $\ll g.w,$ $w\gg=\ll(g - 1).w,$ because $A$ $ w\gg$
$w\gg=0$ . Now $g_{1}g_{2}-1$ is equal to

$\ll w,$ 1 $)(g_{2}-1)+(g_{1}-1)+(g_{2}-1)$ . $(g_{1}$ –
Therefore
$\psi_{w}(g_{1}g_{2})$
$=\psi(\frac{1}{2}\ll(g_{1}g_{2}-1).w,$ $w\gg)$
$=\psi(\frac{1}{2}\ll(g_{1}-1)(g_{2}-1).w,$ $w\gg)\psi(\frac{1}{2}\ll(g_{1}- 1).w,$ $w\gg)\psi(\frac{1}{2}\ll(g_{2}- 1).w,$ $w\gg)$ .
It is easy to compute that 1 $)(g_{2}- 1).w,$ $w\gg=\ll(g_{2}- 1).w,$ $(g_{1}^{-1}-1).w\gg$ . $\ll(g_{1}$ –
Hence $\psi((1/2)\ll(g_{1} - 1)(g_{2}- 1).w, w\gg)=1$ because both elements $(g_{2}-1).w$ ,

$(g_{1}^{-1}-1).w$ are in and is a good lattice. Therefore we have $A$ $A$
810 S.-Y. PAN
$\psi(\frac{1}{2}\ll(g_{1}g_{2}-1).w,$ $w\gg)=\psi(\frac{1}{2}\ll(g_{1}-1).w,$ $w\gg)\psi(\frac{1}{2}\ll(g_{2}-1).w,$ $w\gg)$ ,
that is, $\psi_{w}(g_{1}g_{2})=\psi_{w}(g_{1})\psi_{w}(g_{2})$ . It is clear that the map $\psi_{w}$

is continuous.
Hence $\psi_{w}$
is a character of K. $[$
4.9.
Let $K$ be a compact subgroup of $Sp(\mathscr{W})$
contained in for some good
$K_{A}^{\prime}$
lattice in . If is an element in
$A$ $\mathscr{W}$
$w$ $\mathscr{W}$
and $f$ is an element of , then we $\mathscr{S}(A)_{w}$
define
$f[w, K]:=\int_{K}\omega_{\psi}(k).fdk$ (4.9.a)
where is a Haar measure on $K$. Then it is clear that $f[w, K]$ belongs to
$dk$
. If $f[w, K]$ is not the zero vector, then $f[w, K]$ is fixed by $K$.
$\mathscr{S}(A)_{A^{*}+K.w}$
Moreover those $f[w, K]$ when runs over a basis of span the subspace of $f$ $\mathscr{S}(A)_{w}$
of functions with support in $A^{*}+K.w$ , that is, we have

$\mathscr{S}(A)^{K}$
$(\mathscr{S}(A)_{A^{*}+K.w})^{K}=\sum_{f\in \mathscr{S}(A)_{w}}Cf[w, K]$

(4.9.b)
(cf. [MVW], chapitre 5 section III.1). The following lemma, which is from
[MVW] chapitre 5 section III.3 plays an important role in the proofs of the main
results in subsection 4.13.
LEMMA. Let be an element in and $K$ be a compact subgroup of
$w$ . $\mathscr{W}$
$K_{A}^{\prime}$
Suppose that is a nonzero vector in $f$. Then $f[w, K]$ is nonzero if and only $\mathscr{S}(A)_{w}$
if is fixed by the subgroup

$f$ $K_{1}:=\{g\in K|g^{-1}.w\in A+w\}$ .
PROOF. First we prove that $f[w, K]$ is nonzero if and only if $f[w, K](w)$ is
nonzero. Now $f[w, K]$ is a mapping with support in $A^{*}+K.w$ . Suppose that
$f[w, K]$ is nonzero, that is, $f[w, K](w^{\prime})\neq 0$ for some element in $A^{*}+K.w$ . $w^{\prime}$
Write $w^{\prime}=a+k.w$ for some $a\in A$ ’, $k\in K$ . Now we have
$f[w, K](a+k.w)=\psi(\frac{1}{2}\ll k.w,$ $a\gg)\tilde{\rho}_{\psi}(a).(f[w, K](k.w))$
by (1.5.b). Hence $f[w, K](w^{\prime})\neq 0$ if and only if $f[w, K](k.w)\neq 0$ . Moreover

from (4.9.a) it is clear that $f[w, K](k.w)=f[w, K](w)$ . Hence $f[w, K]$ is nonzero
if and only if $f[w, K](w)$ is nonzero.
Now $f[w, K](w)=\int_{K}(\omega_{\psi}(k) .f)(w)dk$ is equal to $\int_{K}f(k^{-1}.w)dk$ because $K$ is
contained in . Thus $K_{A}^{\prime}$
$\int_{K}f(k^{-1}.w)dk=\int_{K_{1}}f(k^{-1}.w)dk$
because $f$ is supported in $A^{*}+w$ . Now $f(k^{-1}.w)=\psi((1/2)\ll(k-1).w, w\gg)$ .

$f(w)$ for $k\in K_{1}$ because $(k-1).w$ belongs to . Therefore we conclude that $A$
$f[w, K](w)=\int_{K_{1}}\psi_{w}(k)f(w)dk=(\int_{K_{1}}\psi_{w}(k)dk)f(w)$ (4.9.c)
where is defined in Lemma 4.8. By Lemma 4.8 we know that

$\psi_{w}$
is a $\psi_{w}$
character of . The integral in (4.9.c) is essentially equal to the

$K_{1}$ $\int_{K_{1}}\psi_{w}(k)dk$
sum of values of a character over all elements of a finite group. Therefore the
last integral in (4.9.c) is nonzero if and only if is trivial on . Hence $\psi_{w}$ $K_{1}$
$f[w, K]\neq 0$ if and only if is trivial on . $\psi_{w}$ $K_{1}$ $\square $
4.10.
We identify the space with $\mathscr{W}$
$Hom_{D}(\mathscr{V}, \mathscr{V}^{\prime})$
. Hence an element $w\in \mathscr{W}$
can be regarded as a homomorphism from to $\mathscr{V}$

. We use the notation $w.x$

to denote the image of under the map for $x$ $w$ $x\in \mathscr{V}$
. The following lemma is
[Wp] lemme II.4.
LEMMA. Let be a good lattice in be an element in $A$

be an $\mathscr{W},$
$g$ $K_{A}^{\prime},$
$w$
element in and denote the element $c(g)$ in

$\mathscr{W}$
. Suppose that $c.w$
$c$
belongs $5\mathfrak{p}(\mathscr{W})$
to ’. $A$
(i) For we have $(M_{A}[g].f)(w)=\psi(\ll w, c.w\gg)\tilde{\rho}_{\psi}(2c.w).f(w)$ .

$f\in \mathscr{S}(A)$
(ii) Suppose that $c=c_{x,y}$ for some . Then $c_{x,y}.w\gg=$ $x,$ $y\in \mathscr{V}$ $\ll w,$
. $-2Trd_{D/F}\langle w.x, w.y\rangle^{\prime}$
Let be an irreducible smooth representation of the Heisenberg group

$(\rho_{\psi}, \mathscr{S})$
associated to the character of $F$ of conductor

$H(\mathscr{W})$
. Suppose that is a $\psi$
$Q$
lattice in such that . Then we have

$\mathscr{W}$
for any . $Q\subseteq Q^{*}$ $\ll b,$ $b^{\prime}\gg\in \mathfrak{p}_{F}^{\lambda_{F}}$
$b,$ $b^{\prime}\in Q$
Therefore is a subgroup of . The representation

$Q\times \mathfrak{p}_{F}^{\lambda_{F}}$
$H(\mathscr{W})$ $(\rho_{\psi}|_{Q\times \mathfrak{p}_{F}^{\lambda_{F}}}, \mathscr{S})$
factors through the projection because is in the kernel of . $Q\times \mathfrak{p}_{F}^{\lambda_{F}}\rightarrow Q$ $\mathfrak{p}_{F}^{\lambda_{F}}$
$\psi$
The action of on the space will be also denoted by . If is self-dual,

$Q$ $\mathscr{S}$
$\rho_{\psi}$
$Q$
then it is not difficult to see that the representation of the additive group on $\rho_{\psi}$
$Q$
is equivalent to its regular representation.

$\mathscr{S}$
Suppose is a lattice in and contained in a good lattice . Then we

$Q$ $\mathscr{W}$ $A$
have . Hence acts on

$Q\subseteq Q^{*}$
as in the previous paragraph. From (i) $Q$ $\mathscr{S}(A)$
of Lemma 4.10 it is not difficult to check that
$\mathscr{S}(A)^{Q}=\mathscr{S}(A)_{Q^{*}}$
(4.10.a)
(a proof can be found in [Pn2]). Hence corollaire III.2 of [Wp] for $G=G_{L,0}+,$ $\theta$
trivial and $A$

as defined in II.2 can be rewritten as
$\mathscr{S}(A)^{G_{L,0+}}=\omega_{\psi}(\mathscr{H}^{\prime}).(\mathscr{S}(A)^{B_{M,N}^{*}})^{G_{L,0+}}$
(4. 10. b)
812 S.-Y. PAN
because where $B_{M,N}$ is as defined in [Wp].

$B_{M,N}^{*}\subseteq A\subseteq B_{M,N}$ Clearly, the
lattice in above expression is not essential. Hence we have
$A$
$\mathscr{S}^{G_{L,0+}}=\omega_{\psi}(\mathscr{H}^{\prime})$
. $(\mathscr{S}^{B_{M,N}^{*}})^{G_{L,0+}}$
(4.10.c)
where $\mathscr{S}$
is any model of the Weil representation.

4.11.
PROOF PROPOSITION 4.1. Let 0F denote . Suppose $B$ $B(L, L^{\prime})$ $f\in \mathscr{S}(B)_{B^{*}}$
and $g\in G_{L,0}+$ . We know that is in . We have $(\omega_{\psi}(g).f)(x)=$ $g$ $K_{B}^{\prime}$
$(M_{B}[g].f)(x)=\tilde{\omega}_{\psi}(g).(f(g^{-1}.x))$ from (1.5.c) for any . Now is $x\in \mathscr{W}$ $\tilde{\omega}_{\psi}(g)$
trivial because is in . Hence $g$ is not zero only if $g^{-1}.x$

belongs
$K_{B}^{\prime}$ $(\omega_{\psi}(g).f)(x)$
to . But
$B^{*}$
is stable by , so $B^{*}$
belongs to . Now suppose is $g$ $\omega_{\psi}(g).f$ $\mathscr{S}(B)_{B^{*}}$ $w$
in . Hence
$B^{*}$
1 .viz is in . Then $(g^{-1}$ – $)$ $B$
$(\omega_{\psi}(g).f)(w)=f((g^{-1}-1).w+w)=\psi(\frac{1}{2}\ll w,$ $(g^{-1}-1).w\gg)\tilde{\rho}((g^{-1}-1).w).f(w)$ .
Because $(g^{-1}-1).w$ is in $B,\tilde{\rho}((g^{-1}-1).w)$ becomes trivial. Moreover $w\in B^{*}$

and $(g^{-1}-1).w\in B$ imply that $\psi((1/2)\ll(g-1).w, w\gg)=1$ . Hence
$(\omega_{\psi}(g).f)(w)=f(w)$ for any $w\in B^{*}$
. Therefore is fixed pointwise by $\mathscr{S}(B)_{B^{*}}$
. By the remark in subsection 4.10 we conclude that

$G_{L,0}+$ is fixed ’
$\mathscr{S}(A_{0})_{B(L,L)^{*}}.$
pointwise by . By symmetry, , is also fixed pointwise by

$G_{L,0}+$ $\mathscr{S}(A_{0})_{B(L,L)^{*}}$
.
$G_{L,0^{+}}^{\prime},$ $\square $
4.12.
PROPOSITION. Let be a good lattice in
$L$
. Suppose that is in $\mathscr{V}$
$w$
$L^{*}\otimes\Gamma_{m}^{\prime*}\cap L\varpi^{-1}\otimes\Gamma_{M}^{\prime}$
and $f[w, G_{L,0}+]$ is nonzero for some where $f\in \mathscr{S}(A)_{w}$
$A:=L^{*}\otimes\Gamma_{M}^{\prime}\cap L\otimes\Gamma_{m}^{\prime*}$
in . Then belongs to
$\mathscr{W}$
for some good $w$ $B(L, L^{\prime})^{*}$
lattice $L^{\prime}$
such that .
$\Gamma_{m}^{\prime}\subseteq L^{\prime}\subseteq\Gamma_{M}^{\prime}$
We know that
PROOF. is a good lattice in by Lemma 4.6. Identify $A$ $\mathscr{W}$
with
$L^{*}\otimes\Gamma_{m}^{\prime*}\cap L\varpi^{-1}\otimes\Gamma_{M}^{\prime}$
. Let be $Hom_{\mathscr{O}}(L\varpi^{-\kappa}, \Gamma_{m}^{\prime*})\cap Hom_{\mathscr{O}}(L^{*}\varpi^{1-\kappa}, \Gamma_{M}^{\prime} )$ $x$
an element in be an element in . Then $ord_{L}*(x)\geq ord_{L}(x)\geq 0$ ,

$L,$ $y$
$L^{*}\varpi^{1-\kappa}$
$ord_{L}(y)\geq-\kappa$ , and $ord_{L^{*}}(y)\geq 1-\kappa$ . Then satisfy the condition in Lemma $x,$ $y$
4.5(i). Hence is defined and belongs to . Now

$u_{x,y}$ , so $G_{L,0}+$ $c_{x,y}.w\subseteq A$
$(\omega_{\psi}(u_{x,y}).f)(w)=\psi(\ll w, c_{x,y}.w\gg)f(w)$
by Lemma 4.10(i). Because we have $(u_{x,y}^{-1} - 1).w\in A$ and we assume that

$f[w, G_{L,0}+]$ is nonzero, we have $\psi(\ll w, c_{x,y}.w\gg)=1$ by Lemma 4.9. Hence we
have . Now
$\ll w,$
is in by Lemma 4.10(ii).
$c_{x,y}.w\gg\in \mathfrak{p}_{F}^{\lambda_{F}}$
$Trd_{D/F}(\langle w.x, w.y\rangle^{\prime})$ $\mathfrak{p}_{F}^{\lambda_{F}}$
Therefore is in from the definition in subsection 2.1.

$\langle w.x, w.y\rangle^{\prime}$ $\mathfrak{p}^{\lambda}=\mathfrak{p}^{\kappa+\kappa^{\prime}}$
Therefore $\langle w.x\varpi^{-}’, w.y\rangle^{\prime}\in \mathfrak{p}^{\kappa^{\prime}}$

. Because $x$ (resp. $y$ ) is arbitrary in $L$
(resp.
$L^{*}\varpi^{1-}’)$
, we have
$\langle w.L\varpi^{-\kappa}, w.L^{*}\varpi^{1-\kappa}\rangle^{\prime}\subseteq \mathfrak{p}^{\kappa^{\prime}}$
.
Now is a good lattice. Because is

$w.L^{*}\varpi^{1-\kappa}\subseteq w.Lm$ $’\subseteq w.L^{*}\varpi^{-}’$
because $L$ $w$
in , we have and w.Lm . Hence

$L^{*}\otimes\Gamma_{m}^{\prime*}\cap L\varpi^{-1}\otimes\Gamma_{M}^{\prime}$ $w.L^{*}\varpi^{1-\kappa}\subseteq\Gamma_{M}^{\prime}$ $’\subseteq\Gamma_{m}^{\prime*}$
by Lemma 4.7(ii) there exists a good lattice in such that , $L^{\prime}$ $\mathscr{V}^{\prime}$
$\Gamma_{m}^{\prime}\subseteq L^{\prime}\subseteq\Gamma_{M}^{\prime}$
and w.Lm . Therefore

$w.L^{*}\varpi^{1-\kappa}\subseteq L^{\prime}$
is in $’\subseteq L^{\prime*}$
$w$ $Hom_{\mathscr{O}}(L\varpi^{-}’, L^{\prime*})\cap$
which is exactly
$Hom_{\mathscr{O}}(L^{*}\varpi^{1-\kappa}, L^{\prime})$
. $[$ $L^{*}\otimes L^{\prime*}\cap L\otimes L^{\prime}\varpi^{-1}=B(L, L^{\prime})^{*}$
4.13.
PROOF 0F THEOREM 4.2. Now we begin to prove Theorem 4.2. The in-
clusion
$\omega_{\psi}(\mathscr{H}^{\prime})$
. $(,\sum_{L\subseteq\Gamma_{M}^{\prime}}\mathscr{S}(A)_{B(L,L^{\prime})^{*}})\subseteq \mathscr{S}(A)^{G_{L,0+}}$
is an easy consequence of and the fact that ) and ) $Pro\underline{pos}ition$ $4.1$ $U\overline{(\mathscr{V}}$ $U\overline{(\mathscr{V}}^{\prime}$
commute with each other in . We shall prove the opposite inclusion by $Sp(\mathscr{W})$
discussion according to the following three separate cases: (1) $L=L^{*};$ (2)
$L=L$ “ $\varpi;(3)L^{*}\varpi\neq L\neq L$ ’.
First suppose that we are in the first case, that is, is self-dual. Hence $L$ $L$
is a maximal good lattice in . Then $G_{L,0}+=\{g\in G|(g-1).L\subseteq L\varpi\}$ . Let $\mathscr{V}$
$M:=L\varpi,$ $N:=L$ and , $KM$ , be as defined in [Wp] I.15, I. 16, and $\mathscr{R}(L),$ $B_{M}$
$N,$ $N$
II.6. Clearly the pair $(M, N)$ belongs to . So $\mathscr{R}(L)$ $ B_{M,N}=L\varpi^{-1}\otimes\Gamma_{M}^{\prime}+L\otimes$
for a fixed maximal good lattice

$\Gamma_{M}^{\prime*}=L\varpi^{-1}\otimes\Gamma_{M}^{\prime}$
in . It is easy to $\Gamma_{M}^{\prime}$
check that $K_{M,N}=G_{L,0}+$ in this case. Now we have for $(g-1).B_{M,N}\subseteq L\otimes\Gamma_{M}^{\prime}$
all $g\in G_{L,0}+$ . Clearly is a good lattice in , and $A:=L\otimes\Gamma_{M}^{\prime}=B(L, \Gamma_{M}^{\prime})$ $\mathscr{W}$
$G_{L,0}+$
is a subgroup of . Let be an element in be in . Then we have

$K_{A}^{\prime}$
$x$ $L,$ $y$
$L\varpi^{1-\kappa}$
$ord_{L}(x)\geq 0$ and $ord_{L}(y)\geq 1-\kappa$ . Therefore satisfy the condition in Lemma $x,$ $y$
4.5(i). Hence is defined and belongs to $u_{x,y}$. Let be an element in $G_{L,0}+$ $w$
$B_{M,N}$ . Then belongs to . Therefore $\omega_{\psi}(u_{x,y}).f=\psi(\ll w, c_{x,y}.w\gg)f$ by

$c_{x,y}.w$
$A$
Lemma 4.10(i) for . Suppose that $f[w, G_{L,0}+]$ is nonzero. Then $f\in \mathscr{S}(A)$
$\psi(\ll w, c_{x,y}.w\gg)=1$ by Lemma 4.9. Hence we have . Now $\ll w,$ $c_{x,y}.w\gg\in \mathfrak{p}_{F}^{\lambda_{F}}$
by Lemma 4.10(ii). From subsection 2.1 we know

$Trd_{D/F}(\langle w.x, w.y\rangle^{\prime})\in \mathfrak{p}_{F}^{\lambda_{F}}$
that . Hence we have . Therefore

$\langle w.x, w.y\rangle^{\prime}\in \mathfrak{p}^{\lambda}=\mathfrak{p}^{\kappa+\kappa^{\prime}}$ $\langle w.x\varpi^{-}’, w.y\rangle^{\prime}\in \mathfrak{p}^{\kappa^{\prime}}$
is contained in
$\langle w.L\varpi^{-}’, w.Lm1-\kappa\rangle^{\prime}$
. Since is in , we have $\mathfrak{p}^{\kappa^{\prime}}$
$w$ $L\varpi^{-1}\otimes\Gamma_{M}^{\prime}$
. Hence by Lemma 4.7(i) we know that there exists a good lattice

$w.L\varpi^{1-\kappa}\subseteq\Gamma_{M}^{\prime}$
in
$L^{\prime}$
such that and w.Lm

. Hence $L^{\prime}\subseteq\Gamma_{M}^{\prime}$
$’\subseteq L^{\prime*}$
$w\in L\otimes L^{\prime*}=B(L, L^{\prime})^{*}$
because $L=L^{*}$ in this case. Therefore we have proved that if is in $B_{M,N}$ and $w$
814 S.-Y. PAN
$f[w, G_{L,0}+]$ is nonzero, then $w$ must be in $B(L, L^{\prime})^{*}$

. So we conclude
$(\mathscr{S}(A)_{B_{M,N}})G_{L,o+}$ $=$ $\sum$ $\sum$ $f[w,$ $G_{L,0^{+}}]$ $\subseteq$
$\sum$ $\mathscr{S}(A)_{B(L,L^{\prime})}*$ .
$w\in B_{M,N}f\in \mathscr{S}(A)_{\mathcal{W}}$ $L^{\prime}\subseteq\Gamma_{M}^{\prime}$
Therefore by (4.10.c) and [Wp] corollaire III.2 we have
$\mathscr{S}^{G_{L,0+}}=\omega\psi(\mathscr{H}^{\prime}).((\mathscr{S}^{B_{M,N}^{*}})^{G_{L,0+}})\subseteq\omega\psi(\mathscr{H}^{\prime})$
. $(,\sum_{L\subseteq\Gamma_{M}^{\prime}}\mathscr{S}^{B(L,L^{\prime})})$
.
Secondly, suppose that $ L=L^{*}\varpi$ . The proof for this case is very similar to
the proof for the first case. Now again $G_{L,0}+=\{g\in G|(g-1).L\subseteq L\varpi\}$ in this
case. Let $M=N:=L$ . We have where is a fixed max- $(M, N)\in \mathscr{R}(\Gamma_{M})$ $\Gamma_{M}$
imal good lattice in containing . Now we have $B_{M,N}=L^{*}$ OX $\mathscr{V}$

$L$ $\Gamma_{M}^{\prime}+$
for a fixed maximal good lattice

$L^{*}\otimes\Gamma_{M}^{\prime*}=L^{*}\otimes\Gamma_{M}^{\prime*}$
in . And we also $\Gamma_{M}^{\prime}$
have $K_{M,N}=G_{L,0}+$ . Clearly I is a good lattice in . Then $:=L\otimes\Gamma_{M}^{\prime*}$ $\mathscr{W}$

$G_{L,0}+$
is a subgroup of . Let be an element in be an element in $K_{A}^{\prime}$

$x$ $L,$ $y$
$L\varpi^{-}’.$
Then satisfy the conditions in Lemma 4.5(i). Hence

$x,$ $y$ is defined and $u_{x,y}$
belongs to . By the same argument in the first case we can prove that
$G_{L,0}+$
. Now w.Lm ’ is contained in because $w\in B_{M,N}$ .

$\langle w.L\varpi^{-}’, w.Lm -\kappa\rangle^{\prime}\subseteq \mathfrak{p}^{\kappa^{\prime}}$ $\Gamma_{M}^{\prime*}$
Therefore we get w.Lm by Lemma 4.7(iii). Let , which is of $’\in\Gamma_{M}^{\prime}$ $L^{\prime}:=\Gamma_{M}^{\prime}$
course a good lattice in . Hence is in because now $\mathscr{V}^{\prime}$
$w$ $L^{*}\otimes L^{\prime}=B(L, L^{\prime})^{*}$
’
$L^{*}=L\varpi^{-1}$
in this case. Therefore we have . $\mathscr{S}^{G_{L,0+}}\subseteq\omega\psi$ $(\mathscr{H}$ $)$ $(\sum_{L^{\prime}\subseteq\Gamma_{M}^{\prime}}\mathscr{S}^{B(L,L^{\prime})})$
by the same argument in the first case.

Finally we suppose that $L^{*}\varpi\neq L\neq L$ ’. Now we let $M:=L^{*}\varpi,$ $N:=L$ .
Then we know that and $(M, N)\in \mathscr{R}(\Gamma_{M})$
$B_{M,N}=L\varpi^{-1}\otimes\Gamma_{M}^{\prime*}\cap L^{*}\otimes\Gamma_{M}^{\prime}\varpi^{-1}$
. (4.12.a)
Also we have $K_{M,N}=G_{L,1}$ . Suppose that $g$ is an element in . Then from $G_{L,1}$
(4.12.a), we have $(g-1).B_{M,N}\subseteq L$ OX $\Gamma_{M}^{\prime*}\cap L^{*}$

OX . Let be the good lattice
$\Gamma_{M}^{\prime}$ $A$
. Hence
$L\otimes\Gamma_{M}^{\prime*}\cap L^{*}\otimes\Gamma_{M}^{\prime}$
$(g-1).B_{M,N}\subseteq A\subseteq A^{*}\subseteq B_{M,N}$
for all $g\in G_{L,1}$ We identify $B_{M,N}$ with . Hence $G_{L,1}$ is a subgroup of $K_{A}^{\prime}$
.
. Let be an element in , and be
$Hom_{\mathscr{O}}(L^{*}\varpi^{1-\kappa}, \Gamma_{M}^{\prime*})\cap Hom_{\mathscr{O}}(L\varpi^{1-\kappa}, \Gamma_{M}^{\prime} )$ $x$ $L$ $y$
an element in . Then satisfy the condition in Lemma 4.5(ii). Hence $L\varpi^{1-\kappa}$

$x,$ $y$
is defined and belongs to

$u_{x,y}$ . By the same argument in the first case $G_{L,1}$
we can prove that . We know that . $\langle w.L\varpi^{1-\kappa}, w.Lm -\kappa\rangle^{\prime}\in \mathfrak{p}^{\kappa^{\prime}}$ $w.L\varpi^{1-\kappa}\subseteq\Gamma_{M}^{\prime}$
Therefore by Lemma 4.7(i) we have w.Lm is contained in

’
for some good $L^{\prime\prime*}$
lattice in such that . Let be a minimal good lattice in

$L^{\prime\prime}$ $\mathscr{V}^{\prime}$
$L^{\prime\prime}\subseteq\Gamma_{M}^{\prime}$ $\Gamma_{m}^{\prime}$
such that . Hence is contained in . Let be an element in

$\Gamma_{m}^{\prime}\subseteq L^{\prime\prime}$ $w.L\varpi^{-}’$ $\Gamma_{m}^{\prime*}$
$x$
and be an element in
$ L^{*}\varpi$
$y$ . Then again satisfy the condition in $L^{*}\varpi^{1-\kappa}$
$x,$ $y$
Lemma 4.5(ii). Hence is defined and belongs to . By the same ar-$u_{x,y}$ $G_{L,1}$
gument in the first case we can prove that . Since $\langle w.L^{*}\varpi^{1-\kappa}, w.L^{*}\varpi^{1-\kappa}\rangle^{\prime}\subseteq \mathfrak{p}^{\kappa^{\prime}}$
$w$in $B_{M,N}$ , wwee have . Therefore we have $w.L^{*}\varpi^{1-\kappa}\subseteq\Gamma_{M}^{\prime*}$ $w.L^{*}\varpi^{1-\kappa}\subseteq\Gamma_{M}^{\prime}$
by Lemma 4.7(iii). Therefore we have proved that if $w\in B_{M,N}$ and

is nonzero for some
$f[w, G_{L,1}]\in \mathscr{S}(A)$ , then must be in $f\in \mathscr{S}(A)_{w}$ $w$
for some minimal good lattice . Therefore

$L^{*}\otimes\Gamma_{m}^{\prime*}\cap L\otimes\Gamma_{M}^{\prime}\varpi^{-1}$ $\Gamma_{m}^{\prime}\subseteq\Gamma_{M}^{\prime}$
where . Hence by (4. 10.a)

$(\mathscr{S}(A)_{B_{M,N}})^{G_{L,1}}\subseteq \mathscr{S}(A)_{R}$ $R$ $:=L^{*}\otimes\Gamma_{m}^{\prime*}\cap L\otimes\Gamma_{M}^{\prime}\varpi^{-1}$
we have , or
$(\mathscr{S}(A_{1})^{B_{M,N}^{*}})^{G_{L,1}}\subseteq \mathscr{S}(A_{1})^{R(L,\Gamma_{m}^{\prime})^{*}}$
$(\mathscr{S}^{B_{M,N}^{*}})^{G_{L,1}}\subseteq \mathscr{S}^{R(L,\Gamma_{m}^{\prime})^{*}}$
where is any model of the Weil representation of

$S$
Now by Proposition
).
4.11 we know that if $w\in R$ and $f[w, G_{L,0}+]\in \mathscr{S}(A_{0})$ is nonzero for some
’
, then
$f\in \mathscr{S}(A)_{w}$ for some good lattice $w\in B(L, L^{\prime})$ $L^{\prime}$
in such that .
Hence we have proved
$(\mathscr{S}(A)_{B_{M,N}})^{G_{L,0+}}=((\mathscr{S}(A)_{B_{M,N}})^{G_{L,1}})^{G_{L,0+}}\subseteq(,\sum_{\Gamma_{m}\subseteq\Gamma_{M}^{\prime}}\mathscr{S}(A)_{R})^{G_{L,0+}}$
$\subseteq,\sum_{L\subseteq\Gamma_{M}^{\prime}}\mathscr{S}(A)_{B(L,L^{\prime})^{*}}$
.
Therefore by (4.10.b) and [Wp] corollaire III.2 we have
$\mathscr{S}(A)^{G_{L,0+}}\subseteq\omega_{\psi}(\mathscr{H}^{\prime}).(\mathscr{S}(A)_{B_{M,N}})^{G_{L,0+}}\subseteq\omega_{\psi}(\mathscr{H}^{\prime})$
. $(,\sum_{L\subseteq\Gamma_{M}^{\prime}}\mathscr{S}(A)^{B(L,L^{\prime})})$
.
Hence the theorem is pprrooved. $\square $
5. Correspondence of depth zero minimal $K$ -types.

In section 4 we proved that the depth zero representations are preserved by
the local theta correspondence. In this section we want to investigate how the
depth zero minimal $K$-types of the paired representations are related.
5.1.
Let $L$
(resp. $L^{\prime}$
) be a good lattice in $\mathscr{V}$
(resp. $\mathscr{V}^{\prime}$
). Recall that $B:=$

$B(L, L^{\prime})=L^{*}\otimes L^{\prime}\cap L\otimes L^{\prime*}$
and $B^{*}=L^{*}\otimes L^{\prime}+L\otimes L^{\prime*}$
from subsection 4.1.
Define
$B\varpi^{r}:=L^{*}\otimes L^{\prime}\varpi^{r}\cap L\otimes L^{\prime*}\varpi^{r}$
$B^{*}\varpi^{r}:=L^{*}\otimes L^{\prime}\varpi^{r}+L\otimes L^{\prime*}\varpi^{r}$

816 S.-Y. PAN
for any integer . Clearly $r$

$B\varpi^{r}$
and $B^{*}\varpi^{r}$
are in . As usual let
(!) $F$ -lattices $\mathscr{W}$
$K_{B}$
denote the stabilizer of in $B$ $Sp(\mathscr{W})$

and $K_{B}^{\prime}$
be as defined in (1.5.a). Define
$K_{B}^{\prime\prime}:=\{g\in K_{B}|(g-1).B^{*}\varpi^{-1}\subseteq B^{*}\}=\{g\in K_{B}|(g-1).B\subseteq B\varpi\}$ . (5.1.a)
It is easy to see that . We know that $B^{*}/B$ is a nondegenerate $K_{B}^{\prime\prime}\subseteq K_{B}^{\prime}\subseteq K_{B}$
symplectic space over . It is not difficult to check that the quotient is $f_{F}$ $K_{B}/K_{B}^{\prime}$
isomorphic to the finite symplectic group $Sp(B^{*}/B)$ .

Define the maps and from to $Z$ as follows: $ord_{B}(w):=m$ if $ord_{B}$ $ord_{B^{*}}$ $\mathscr{W}$
$w\in B\varpi^{m}-B\varpi^{m+1}$
, and $ord_{B^{*}}(w):=m$ if . $w\in B^{*}\varpi^{m}-B^{*}\varpi^{m+1}$
be elements in
LEMMA. such that $ord_{B}(x)+ord_{B^{*}}(y)\geq 1-\lambda$
Let $x,$ $y$
$\mathscr{W}$
and $ord_{B}(y)+ord_{B}*(x)\geq 1-\lambda$ where 2 is deftned in subsection 2.1. Then is $u_{x,y}$
deftned and belongs . $K_{B}^{\prime\prime}$
PROOF. From the condition in the lemma, it is clear that $1+c_{x,y}$ is in-
vertible. Hence is well-defined. Let . Then $c_{x,y}.b=x\ll y,$ $b\gg+$
$u_{x,y}$
$b\in B^{*}\varpi^{-1}$
$y\ll x,$. It is clear that $ b\gg$

by the condition in the lemma. $ c_{x,y}.B^{*}\subseteq B^{*}\varpi$
Therefore $(u_{x,y}-1).B^{*}\varpi^{-1}=-2c_{x,y}(1+c_{x,y})^{-1}.B^{*}\varpi^{-1}=-2c_{x,y}.B^{*}\varpi^{-1}\subseteq B$
’.
Thus belongs to .
$u_{x,y}$ $[$
$K_{B}^{\prime\prime}$
5.2.
LEMMA. Let be a character of $F$ with conductor . Suppose that $\psi$

$w\in B\varpi^{-1}$
and $\psi(\ll w, c(k).w\gg)=1$ for all . Then belongs to . $k\in K_{B}^{\prime\prime}$ $w$ $B^{*}$
PROOF. Here we consider a new dual pair in where $(U(\mathscr{V}_{a}), U(\mathscr{V}_{a}^{\prime}))$ $Sp(\mathscr{W})$
with the form ,

$\mathscr{V}_{a}:=\mathscr{W}$
and with the form , given by $\langle$
$\rangle_{a}:=\ll,$ $\gg$ $\mathscr{V}_{a}^{\prime}:=F$ $\langle$ $\rangle_{a}$
. Let and . It is clear that

$\langle t_{1}, t_{2}\rangle_{a}^{\prime}:=t_{1}t_{2}$
is equal to $\kappa_{a}:=\lambda$ $\kappa_{a}^{\prime}:=0$ $\ll,$ $\gg$
, OX , . Identify
$\langle$ $\rangle_{a}$
with $\langle$
. Suppose that
$\rangle_{a}^{\prime}$
is any $B\varpi^{-1}$
$Hom_{\mathscr{O}}$ $(B^{*}\varpi^{1-\lambda}, \mathscr{O})$ $x_{0}$
element in . Let $x:=x_{0}$ and . Then satisfy the condition in

$B$ $y:=x_{0}\varpi^{1-\lambda}$
$x,$ $y$
Lemma 5.1. Therefore is defined and belongs to . Now $=c_{x,y}$ $u_{x,y}$

$c(u_{x,y})$
from subsection 4.5. Then $\psi(\ll w, c_{x,y}.w\gg)=1$ by the assumption. Therefore

$\ll w,$
. Then we get by Lemma 4. 10(ii). Hence
$c_{x,y}.w\gg\in \mathfrak{p}_{F}^{\lambda_{F}}$ $\langle w.x, w.y\rangle_{a}^{\prime}\in \mathfrak{p}_{F}^{\lambda_{F}}$
is in . So
$\langle w.x_{0}\varpi^{1-\lambda}, w.x_{0}\varpi^{1-\lambda}\rangle_{a}^{\prime}$
. Now $\mathfrak{p}_{F}^{\lambda_{F}}\varpi^{1-\lambda}=\mathfrak{p}_{F}$ $(w.x_{0}\varpi^{1-\lambda})^{2}\in \mathfrak{p}_{F}\subseteq \mathfrak{p}$
by the assumption that

$w.x_{0}\varpi^{1-\lambda}\in(!)$
. Therefore is in $w\in B\varpi^{-1}$ $w.x_{0}\varpi^{1-\lambda}$
$\mathfrak{p}$
because is a prime ideal of the ring . Since

$\mathfrak{p}$
is arbitrary in , we conclude $\mathscr{O}$
$x_{0}$
$B$
that is in . Hence
$w$ belongs to . $Hom_{\mathscr{O}}(B\varpi^{-\lambda}, \mathscr{O})$
$[$ $w$ $B^{*}$
5.3.
PROPOSITION. Let $(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$

be a reductive dual pair and $L$
(resp. $L^{\prime}$
)
be a good lattice in (resp. $\mathscr{V}$
).
(i) If $w\in B\varpi^{-1}$

, then stabilizes the space
$\mathscr{S}(B)_{w}$
.
(ii) If $w\in B\varpi^{-1}$
and $K_{B}^{\prime\prime}$
fixes an nonzero element in $\mathscr{S}(B)_{w}$

, then $w$ belongs
to $B$
”.
(iii) .
$\mathscr{S}(B)^{K_{B}^{\prime}}=\mathscr{S}(B)^{K_{B}^{\prime\prime}}=\mathscr{S}(B)_{B^{*}}$
(iv) acts on
$K_{B}/K_{B}^{\prime}$
and the action is isomorphic to $\mathscr{S}(B)_{B^{*}}$ $\overline{\omega}_{\overline{\psi}}$
where $\overline{\omega}_{\overline{\psi}}$
is
the Weil representation of $Sp(B^{*}/B)$ .
PROOF. If belongs to and belongs to , then $k^{-1}.w$ is in $B+w$
$k$ $K_{B}^{\prime\prime}$
$w$
$B\varpi^{-1}$
by the definition of . Let be a nonzero element in . Now we have $K_{B}^{\prime\prime}$

$f$ $\mathscr{S}(B)_{w}$
$\omega_{\psi}(k).f=\psi(\ll w, c(k).w\gg)f$ from Lemma 4.10(i). Hence (i) is proved.

If fixes a nonzero element in
, then from (i) we know that $f$ $\mathscr{S}(B)_{w}$
$\psi(\ll w, c(k).w\gg)=1$ for all . So (ii) follows from Lemma 5.2. $k\in K_{B}^{\prime\prime}$
Now we prove (iii). It is known that is contained in $\mathscr{S}(B)^{K_{B}^{\prime}}$ $\mathscr{S}(B)^{K_{B}^{\prime\prime}}$
because is a subgroup of . Let $K_{B}^{\prime\prime}$
be the reductive dual pair $K_{B}^{\prime}$ $(U(\mathscr{V}_{a}), U(\mathscr{V}_{a}^{\prime}))$
considered in the proof of Lemma 5.2. Then is just the group of two $U(\mathscr{V}_{a}^{\prime})$
elements and is contained in the center of . Let $ M:=B^{*}\varpi$ and $Sp(\mathscr{W})$
$N:=B$ . Then the group $K_{M,N}$ defined in [Wp] I.16 is equal to , and the $K_{B}^{\prime\prime}$
lattice $B_{M,N}$ defined in [Wp] II.6 is equal to . Let $H:=K_{M,N}$ and 0 be the $B\varpi^{-1}$
trivial representation of $H$. Then corollaire III.2 of [Wp] becomes $\mathscr{S}(B)^{K_{B}^{\prime\prime}}=$
. Now by part
$\omega_{\psi}(\mathscr{H}_{a}^{\prime}).(\mathscr{S}(B)_{B\varpi^{-1}} )^{K_{B}^{\prime\prime}}=(\mathscr{S}(B)_{B\varpi^{-1}} )^{K_{B}^{\prime\prime}}$ $(\mathscr{S}(B)_{B\varpi^{-1}})^{K_{B}^{\prime\prime}}\subseteq \mathscr{S}(B)_{B^{*}}$
(ii). On the other hand, if is in is in and is in , then it is $k$ $K_{B}^{\prime},$

$w$ $B^{*}$
$f$ $\mathscr{S}(B)_{w}$
clear that $\omega_{\psi}(k).f=\psi((1/2)\ll(k-1).w, w\gg)f=f$ since $(k- 1).w\in B$ . There-

fore is contained in
$\mathscr{S}(B)_{B^{*}}$
. Thus we have proved that $\mathscr{S}(B)^{K_{B}^{\prime}}$
$\mathscr{S}(B)^{K_{B}^{\prime}}\subseteq \mathscr{S}(B)^{K_{B}^{\prime\prime}}\subseteq \mathscr{S}(B)_{B^{*}}\subseteq \mathscr{S}(B)^{K_{B}^{\prime}}$

. (5.3.a)
Hence (iii) is proved.

From (1.5.c), it is easy to see that acts on the space . Let be $K_{B}$ $\mathscr{S}(B)_{B^{*}}$
$\Omega$
the map from to given by $f\mapsto f(0)$ where is defined in subsection

$\mathscr{S}(B)_{B^{*}}$ $S$ $S$
1.5. Obviously is an isomorphism of vector spaces. Now we have $\Omega$
$\Omega(\omega_{\psi}(k).f)$
for any $k\in K_{B}$ by
$=(\omega_{\psi}(k).f)(0)=\tilde{\omega}_{\psi}(k).(f(0))=\tilde{\omega}_{\psi}(k).(\Omega(f))$
the definition of and (1.5.c). Since acts trivially on , we can regard

$\Omega$
$K_{B}^{\prime}$ $\mathscr{S}(B)_{B^{*}}$
as a representation of
$\mathscr{S}(B)_{B^{*}}$
. It is clear that is an isomorphism of $K_{B}/K_{B}^{\prime}$
$\Omega$
-representations.
$K_{B}/K_{B}^{\prime}$
lm
5.4.
LEMMA.
Let (resp. ) be a good lattice in (resp. $L$ $L^{\prime}$ $\mathscr{V}$
) and
$B:=B(L, L^{\prime})$ . Then we have and $l_{\mathscr{V}^{\prime}}(G_{L})\cap K_{B}^{\prime}=l_{\mathscr{V}^{\prime}}(G_{L,0}+)$ $l_{\mathscr{V}}(G_{L}^{\prime},)$ $\cap K_{B}^{\prime}=$
.
$l_{\mathscr{V}(G_{L^{\prime},0^{+}}^{\prime}}$
$)$
PROOF. From the discussion in [Wp] I. 15, we know that there exists a
decomposition $\mathscr{V}=X\oplus Y$ such that $L=L^{X}\oplus L^{Y}$ and $L’’=L^{X}\oplus L^{Y}\varpi^{-1}$ ,
818 S.-Y. PAN
where $L^{X}:=L\cap X,$ $L^{Y}:=L\cap Y$ Let . $\mathscr{V}^{\prime}=X^{\prime}\oplus Y^{\prime}$

be the similar decom-
position such that and
$L^{\prime}=L^{\prime X^{\prime}}\oplus L^{\prime Y^{\prime}}$ $L’=L^{\prime X^{\prime}}\oplus L^{\prime Y^{\prime}}\varpi^{-1}$
where $L^{\prime X^{\prime}}$
$:=$
$L^{\prime}\cap X^{\prime},$
$L^{\prime Y^{\prime}}:=L^{\prime}\cap Y^{\prime}$
. Then we have
$B^{*}=$ ( $L^{X}$
OX
$L^{\prime X^{\prime}}$
) $\oplus$
( $L^{X}$
OX
$L^{\prime Y^{\prime}}$
) $\varpi^{-1}\oplus$
( $L^{Y}$
OX
$L^{\prime X^{\prime}}$
) $\varpi^{-1}\oplus(L^{Y}\otimes L^{\prime Y^{\prime}})\varpi^{-1}$

,
$B=(L^{X}\otimes L^{\prime X^{\prime}})\oplus(L^{X}\otimes L^{\prime Y^{\prime}})\oplus(L^{Y}\otimes L^{\prime X^{\prime}})\oplus(L^{Y}\otimes L^{\prime Y^{\prime}})\varpi^{-1}$
.
It is easy to check that $l_{\mathscr{V}^{\prime}}(G_{L,0}+)\subseteq K_{B}^{\prime}$

Hence . . On $l_{\mathscr{V}^{\prime}}(G_{L,0}+)\subseteq l_{\mathscr{V}^{\prime}}(G_{L})\cap K_{B}^{\prime}$
the other hand, if $g\in G_{L}$ and , then we have $(g-1).L^{X}\subseteq L^{X}\varpi\oplus L^{Y}$
$l_{\mathscr{V}^{\prime}}(g)\in K_{B}^{\prime}$
and $(g-1).L^{Y}\subseteq L^{X}\varpi\oplus L^{Y}\varpi$

. Hence $g\in G_{L,0}+$ . Therefore $l_{\mathscr{V}^{\prime}}(G_{L,0}+ )=$
. The proof for

$l_{\mathscr{V}^{\prime}}(G_{L})\cap K_{B}^{\prime}$
is similar.
$l_{\mathscr{V}}(G_{L}^{\prime},)\cap K_{B}^{\prime}=l_{\mathscr{V}}(G_{L,0^{+}}^{\prime},)$ $\square $
5.5.
It is easy to check that $B^{*}/B$ is isomorphic to $(l^{*}\otimes_{d}l^{\prime})\times(l^{\prime*}\otimes_{d}l)$
where
$l,$
are defined in subsection 1.2, and the quotient
$l^{*},$ $l^{\prime},$
$l^{\prime*}$
$G_{L}/G_{L,0}+$ is isomorphic
to $U(l)\times U(l^{*})$ , and , is isomorphic to $G_{L}^{\prime},/G_{L,0^{+}}^{\prime}$ $U(l^{\prime})\times U(l^{\prime*})$
.
PROPOSITION. The pair $(G_{L}/G_{L,0}+, G_{L}^{\prime},/G_{L,0^{+}}^{\prime},)$ (possibly reducible) re-

is $a$
ductive dual pair in the finite symplectic group $Sp(B^{*}/B)$ .
PROOF. Now (resp. ) is an -Hermitian (resp. -Hermitian) space over $\mathscr{V}$

$\epsilon$
$\epsilon^{\prime}$
$D$ with . We know that (resp. ) is a (non-degenerate) (resp.

$\epsilon\epsilon^{\prime}=-1$ $l$ $l^{*},$ $l^{\prime},$
$l^{\prime*}$
$\mu$
$\mu’$
, $\mu^{\prime},$
-Hermitian space over for some (resp.
$\mu^{\prime*})$
) equal to 1 or $f_{D}$ $\mu$
$\mu^{*},$ $\mu^{\prime},$ $\mu^{\prime*}$
-1. To prove this theorem, we need to check that $\mu\mu’’=-1,$ and $\mu^{*}\mu^{\prime}=-1$
$\ll,$ $\gg_{b^{*}}=Trd_{f_{D}/f_{F}}((\langle, \rangle_{l}*\otimes\overline{\tau}0\langle, \rangle_{l}^{\prime},)\oplus(\langle, \rangle_{l}\otimes\overline{\tau}0\langle, \rangle_{l}^{\prime},*))$

(5.5.a)
where $\overline{\tau}$
denotes the involution of over induced from i.e., $f_{D}$ $f_{F}$ $\tau$
$\overline{\tau}$
is given by
$\overline{\tau}(\Pi_{\mathscr{O}}(t)):=\Pi_{\mathscr{O}}(\tau(t))$
for . Consider two elements $t\in(!)$ $w_{1}:=l_{1}^{*}\otimes l_{1}^{\prime}+l_{1}\otimes l_{1}^{\prime*}$
,
$w_{2}:=l_{2}^{*}\otimes l_{2}^{\prime}+l_{2}\otimes l_{2}^{\prime*}$
in where ’, and $B^{*}$ $l_{i}\in L,$ $l_{i}^{*}\in L$ $1_{i}^{\prime}\in L^{\prime}$ $l_{i}^{\prime*}\in L^{\prime*}$
. Now
$\ll\Pi_{B^{*}}$ $(w_{1} )$
, $\Pi_{B^{*}}(w_{2})\gg_{b^{*}}$
$=\Pi_{\mathscr{O}_{F}}(\ll w_{1}, w_{2}\gg\varpi_{F}^{1-\lambda_{F}})$
$=\Pi_{\mathscr{O}_{F}}(\ll l_{1}^{*}\otimes l_{1}^{\prime}, l_{2}^{*}\otimes l_{2}^{\prime}\gg\varpi_{F}^{1-\lambda_{F}})+\Pi_{\mathscr{O}_{F}}$

( $\ll l_{1}$
OX $l_{1}^{\prime*},$ $l_{2}^{*}\otimes l_{2}^{\prime}\gg\varpi_{F}^{1-\lambda_{F}}$
)
$+\Pi_{\mathscr{O}_{F}}(\ll l_{1}^{*}\otimes l_{1}^{\prime}, l_{2}\otimes l_{2}^{\prime*}\gg\varpi_{F}^{1-\lambda_{F}})+\Pi_{\mathscr{O}_{F}}$
( $\ll l_{1}$
OX $l_{1}^{\prime*},$ $l_{2}\otimes l_{2}^{\prime*}\gg\varpi_{F}^{1-\lambda_{F}}$
)
$=\Pi_{\mathscr{O}_{F}}(\ll l_{1}^{*}\otimes l_{1}^{\prime}, l_{2}^{*}\otimes l_{2}^{\prime}\gg\varpi_{F}^{1-\lambda_{F}})+\Pi_{\mathscr{O}_{F}}$
( $\ll l_{1}$
).
The last equality is due to the facts $\Pi_{\mathscr{O}_{F}}(\ll l_{1}\otimes l_{1}^{\prime*}, l_{2}^{*}\otimes l_{2}^{\prime}\gg\varpi_{F}^{1-\lambda_{F}})\in \mathfrak{p}_{F}^{\lambda_{F}}$
and
. We have
$\Pi_{\mathscr{O}_{F}}(\ll l_{1}^{*}\otimes l_{1}^{\prime}, l_{2}\otimes l_{2}^{\prime*}\gg\varpi_{F}^{1-\lambda_{F}})\in \mathfrak{p}_{F}^{\lambda_{F}}$
$\Pi_{\mathscr{O}_{F}}(\ll l_{1}^{*}\otimes l_{1}^{\prime}, l_{2}^{*}\otimes l_{2}^{\prime}\gg\varpi_{F}^{1-\lambda_{F}})$
$=\Pi_{\mathscr{O}_{F}}(Trd_{D/F}(\langle l_{1}^{*}, l_{2}^{*}\rangle\tau(\langle l_{1}^{\prime}, l_{2}^{\prime}\rangle^{\prime}))\varpi_{F}^{1-\lambda_{F}})$
$=Trd_{f_{D}/f_{F}}(\Pi_{\mathscr{O}}(\langle l_{1}^{*}, l_{2}^{*}\rangle\tau(\langle l_{1}^{\prime}, l_{2}^{\prime}\rangle^{\prime})\varpi^{1-\lambda}))$
$=Trd_{f_{D}/f_{F}}(\Pi_{\mathscr{O}}(\langle l_{1}^{*}, l_{2}^{*}\rangle\varpi^{1-\kappa})\overline{\tau}(\Pi_{\mathscr{O}}(\langle l_{1}^{\prime}, l_{2}^{\prime}\rangle^{\prime})\varpi^{-\kappa^{\prime}})))$
$=Trd_{f_{D}/f_{F}}(\langle\Pi_{L^{*}}(l_{1}^{*}), \Pi_{L^{*}}(l_{2}^{*})\rangle_{l^{*}}\overline{\tau}(\langle\Pi_{L^{\prime}}(l_{1}^{\prime}), \Pi_{L^{\prime}}(l_{2}^{\prime})\rangle_{l}^{\prime},))$

.
Similarly, we also have
$\Pi_{\mathscr{O}_{F}}$
( $\ll l_{1}$
)
$=Trd_{f_{D}/f_{F}}(\langle\Pi_{L}(l_{1}), \Pi_{L}(l_{2})\rangle_{l}\overline{\tau}(\langle\Pi_{L^{\prime*}}(l_{1}^{\prime*}), \Pi_{L^{\prime*}}(l_{2}^{\prime*})\rangle_{l}^{\prime},*))$
.
Hence (5.5.a) is true.

Now we check $\mu\mu’’=-1$ and . All the information is in the $\mu^{*}\mu^{\prime}=-1$
table in subsection 1.2. First suppose that $D=F$ . Then $f_{D}=f_{F},$ , $\mu=\mu^{*}=\epsilon$
. Hence
$\mu^{\prime}=\mu^{\prime*}=\epsilon^{\prime}$
. If $D$ is a unramified quadratic $\mu\mu^{\prime*}=\mu^{*}\mu^{\prime}=\epsilon\epsilon^{\prime}=-1$
extension of $F$, then is a quadratic extension of . $f_{D}$ $f_{F},$ $\mu=\mu^{*}=\epsilon,$ $\mu^{\prime}=\mu^{\prime*}=\epsilon^{\prime}$
Again . If is a ramified quadratic extension of , then

$D$ $F$
$\mu\mu^{\prime*}=\mu^{*}\mu^{\prime}=\epsilon\epsilon^{\prime}=-1$
$f_{D}=f_{F},$ , and . Note

$\mu’=(-1)^{\kappa+1}\epsilon,$ $\mu=(-1)^{\kappa}\epsilon,$ $\mu^{\prime*}=(-1)^{\kappa^{\prime}+1}\epsilon^{\prime}$ $\mu^{\prime}=(-1)^{\kappa^{\prime}}\epsilon^{\prime}$
that is odd in this case. Thus is even. Hence

$\kappa+\kappa^{\prime}=2\lambda_{F}+1$ $\kappa+\kappa^{\prime}+1$
. If $D$ is a quaternion algebra of $F$, then again

$\mu\mu^{\prime*}=\mu^{*}\mu^{\prime}=$ $(-1)^{\kappa+\kappa^{\prime}+1}\epsilon\epsilon^{\prime}=-1$
. Without loss of generality, we assume that is even and

$\kappa+\kappa^{\prime}=2\lambda_{F}+1$
is $\kappa$
$\kappa^{\prime}$
odd. Then , and . Therefore $\mu^{*}=-\epsilon,$ $\mu=\epsilon,$

$\mu^{\prime*}=\epsilon^{\prime}$ $\mu^{\prime}=-\epsilon^{\prime}$ $\mu\mu^{\prime*}=\epsilon\epsilon^{\prime}=-1$
and .
$\mu^{*}\mu^{\prime}=(-\epsilon)(-\epsilon^{\prime})=-1$ $\square $
Suppose are two -Hermitian spaces in the same Witt series. Let ,
$\mathscr{V},$
$\epsilon$ $L$
be good lattices in
$L^{\prime\prime}$
respectively. Then it is clear that the spaces $L‘‘/L$ $\mathscr{V},$

and (resp. and ) are in the same Witt tower.

$L^{\prime\prime*}/L^{\prime\prime}$ $ L/L^{*}\varpi$ $ L^{\prime\prime}/L^{\prime\prime*}\varpi$
5.6.
In this subsection, we prove our second main result of this paper.
THEOREM. Let be a reductive dual pair. Let $(G, G^{\prime}):=(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$
(resp.
$(\pi\underline{V},)$ ) be an irreducible admissible representation of (resp.
of depth zero such that two representations are paired by the local theta
correspondence. Suppose that has a minimal $K$-type . Then has $\pi$ $(G_{L}, \zeta)$
$\pi^{\prime}$
$a$
minimal $K$-type such that and are paired in the theta correspondence $(G_{L}^{\prime},, \zeta^{\prime})$ $\zeta$
$\zeta^{\prime}$
for the dual pair in $Sp(B^{*}/B)$ and the Weil representation $(G_{L}/G_{L,0}+ , G_{L}^{\prime},/G_{L,0^{+}}^{\prime},)$
on where $B:=B(L, L^{\prime})$ is deftned in subsection 4.1.

$\mathscr{S}^{K_{B}^{\prime}}$
820 S.-Y. PAN
PROOF. By Theorem 4.2 we know that
$\mathscr{S}(B)^{G_{L,0+}}=\omega_{\psi}(\mathscr{H}^{\prime})$
. $(\sum_{L^{\prime}\in 9(\Gamma_{M}^{\prime})}\mathscr{S}(B)_{B(L,L^{\prime})^{*}})$
where is the set of good lattices contained and

is the $i\underline{n\Gamma}_{M}^{\prime}$ $\mathscr{S}(B)$
generalized lattice model of the Weil representation of with respect to the $Sp(\mathscr{W})$
good lattice . Let 77 be the projection . Let$B$

denote $\mathscr{S}(B)\rightarrow V\otimes_{C}V^{\prime}$ $W\subseteq V$
the representation space of . Because is nontrivial, we have a nontrivial $\zeta$

$V^{G_{L,0+}}$
surjective map . Then there is a nonzero element $\mathscr{S}(B)^{G_{L,0+}}\rightarrow V^{G_{L,0+}}\otimes_{C}V^{\prime}$
, f r some good lattice

$f\in \mathscr{S}(B)_{B(L,L)^{*}}$ such that $\Pi(f)$ is not zero and $o$ $L^{\prime}\subseteq\Gamma_{M}^{\prime}$
$\Pi(f)$ is in the space . But we know that $\Pi(f)$ is fixed by , by $W\otimes_{C}V^{\prime}$ $G_{L,0^{+}}^{\prime}$
Proposition 4.1. Therefore $\Pi(f)$ is in . Hence is not $V^{G_{L,0+}}\otimes_{C}V^{\prime G_{L^{\prime},0+}^{\prime}}$

trivial. Let be an irreducible representation of , such that

$(\zeta^{\prime}, W^{\prime})$
$G_{L}^{\prime},/G_{L,0^{+}}^{\prime}$
$\Pi(f)$ has a nonzero image in . We know $W\otimes_{C}W^{\prime}$ $(G_{L}/G_{L,0}+ , G_{L}^{\prime},/G_{L,0^{+}}^{\prime}, )$
forms a finite reductive dual pair in $Sp(B^{*}/B)$ from Proposition 5.5. By

Proposition 5.3(iii) and 5.3(iv), we know that and the action $\mathscr{S}(B)_{B^{*}}=\mathscr{S}(B)^{K_{B}^{\prime}}$
of on is the Weil representation of $Sp(B^{*}/B)$ with respect to the

$K_{B}/K_{B}^{\prime}$ $\mathscr{S}(B)_{B^{*}}$
nontrivial character of . Now we have a nontrivial map from to $\overline{\psi}$
$f_{F}$ $\mathscr{S}(B)_{B^{*}}$
. Hence the two representations

$W\otimes_{C}W^{\prime}$
and are paired by the theta $\zeta$
$\zeta^{\prime}$
correspondence for the reductive dual pair . $(G_{L}/G_{L,0}+ , G_{L}^{\prime},/G_{L,0^{+}}^{\prime}, )$ $\square $
6. Correspondence of Iwahori-spherical representations.

We begin to give a few applications of Theorem 5.6. An irreducible ad-
missible representation of a -adic reductive group is called Iwahori-spherical if it $p$
admits a nontrivial vector fixed by an Iwahori subgroup. The following theorem

generalizes [Ab] corollaire 4.3, where the result for the cases of unramified re-
ductive dual pairs is proved.
THEOREM. Let be a reductive dual pair. Let $(G, G^{\prime}):=(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$ $I$
(resp. ) be an Iwahori subgroup of

$I^{\prime}$
(resp. ) , and (resp. $U(\mathscr{V})$ $U\underline{(\mathscr{V}^{\prime}}$

$)$
$(\pi\underline{V},)$
be an irreducible admissible representation of

$(\pi^{\prime}, V^{\prime}))$
(resp. ) such $U(\mathscr{V})$ $U(\mathscr{V}^{\prime})$
that are paired by local theta correspondence. Then

$\pi,$
$\pi^{\prime}$
is nontrivial if and $V^{I}$
only if is nontrivial i.e., an irreducible Iwahori-spherical representations

$V^{\prime I^{\prime}}$
corresponds to an irreducible Iwahori-spherical representation.

PROOF. First we assume that is nontrivial. Therefore there is a good $V^{I}$
lattice in such that $L$

is nontrivial. Let be an irreducible sub-
$\mathscr{V}$ $V^{G_{L,0+}}$ $\zeta$
representation of $G_{L}/G_{L,0}+$ occurring in . By Theorem 5.6, there exists a $V^{G_{L,0+}}$
good lattice in such that is nontrivial and there exists an irre-

$L^{\prime}$ $\mathscr{V}^{\prime}$ $V^{\prime G_{L^{\prime},0+}}$
ducible subrepresentation of , on such that are paired $\zeta^{\prime}$
$G_{L}^{\prime},/G_{L,0^{+}}^{\prime}$
$\zeta,$
$\zeta^{\prime}$
by the theta correspondence for the finite reductive dual pair $(G_{L}/G_{L,0}+$ ,
. Because is nontrivial, , as a representation of $G_{L}/G_{L,0}+$ , can be
$G_{L}^{\prime},/G_{L,0^{+}}^{\prime},)$
$V^{I}$ $\zeta$
chosen to be an irreducible subrepresentation occurring in the induced repre-

sentation triv where is the Levi factor of the Borel subgroup $I/G_{L,0}+$
$Ind_{T}^{G_{L}/G_{L,0+}}$ $T$
of $G_{L}/G_{L,0}+$ and “triv” denote the trivial character of . By the induction $T$
principle of the theta correspondence for finite reductive dual pairs (cf. [AMR],
th\’eor\‘em 3.7), the representation must be a subquotient of , triv $\zeta^{\prime}$
$Ind_{T}^{G_{L^{\prime}}^{\prime}/G_{L^{\prime},0+}^{\prime}}$
for
the trivial character of the Levi factor of the Borel subgroup of , $T^{\prime}$
$I^{\prime}/G_{L,0^{+}}^{\prime}$
of
. By Frobenius reciprocity, the representation
$G_{L}^{\prime},/G_{L,0^{+}}^{\prime},$
restricting to $\zeta^{\prime}$
the
Borel subgroup , contains a trivial representation. Therefore
$I^{\prime}/G_{L,0^{+}}^{\prime}$
,
as a representation of , contains the trivial representation. Hence $I^{\prime}/G_{L,0^{+}}^{\prime},$
$V^{\prime I^{\prime}}$
is nontrivial. Therefore we have proved one direction of the theorem. By

symmetry the other direction is also true. $\square $
7. Correspondence of unipotent representations.

Our next consequence concerns the unipotent representations of p-adic
reductive groups defined by G. Lusztig. First we recall some definitions for
representations of finite reductive groups. Let $H$ be a finite reductive group, $T$
be a maximal torus in $H$, and 0 be a character of . P. Deligne and G. Lusztig $T$
define in [DL] a virtual character of $H$. An irreducible representation of $R_{T}^{H}(\theta)$
$H$ is called unipotent if its character occurs as a constituent of (triv) for some $R_{T}^{H}$
$T$
. Let be a reductive -adic group occurring as a member a reductive
$U(\mathscr{V})$ $p$ $\underline{of}$
dual pair. An irreducible admissible representation $(\pi, V)$ of is called $U(\mathscr{V})$
unipotent if is nontrivial for some good lattice in

$V^{G_{L,0+}}$
and contains $L$ $\mathscr{V}$ $V^{G_{L,0+}}$
a unipotent representation of the reductive finite group $G_{L}/G_{L,0+}$ (cf. [Lt], 1.5).
THEOREM. Let be a reductive dual pair. Let
$(G, G^{\prime}):=(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$
(resp.
$(\pi\underline{V},)$ ) be an irreducible admissible representation of
$(\pi^{\prime},$ $V^{\prime})$
(resp. $U(\mathscr{V})$
such that
and are paired by the local theta correspondence. For
$\pi$
$\pi^{\prime}$
simplicity, we assume that . Suppose that the dual pair $\epsilon^{\prime}=1$

$(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$
satisftes one of the following conditions:

(i) $D$ is $F$, the characteristic of is large (see the proof ), and is of $f_{F}$ $U(\mathscr{V}^{\prime})$
type , or (cf. [Tt], pp. 60-66) for some $m$ ,

$D_{m},$ $2D_{m}$ $2D_{m}^{\prime}$
(ii) $D$ is an unramifted quadratic extension of $F$, no other restriction,

(iii) $D$ is a ramifted quadratic extension of $F$, the characteristic of is large, $f_{F}$
and the dimensions of are even, $\mathscr{V},$

(iv) $D$ is a central quaternion algebra, the characteristic of is large, and $f_{F}$
is of type for some integer $m$ .

$U(\mathscr{V})$ $2D_{m}^{\prime\prime}$
Then is unipotent if and only if

$\pi$
is unipotent. $\pi^{\prime}$
PROOF. Suppose that $\pi$

is unipotent representation. So $V^{G_{L,0+}}$
is nontrivial
822 S.-Y. PAN
for some good lattice in . Let be a unipotent subrepresentation of $L$ $\mathscr{V}$ $\zeta$
$G_{L}/G_{L,0}+$ on the space . By Theorem 5.6, we know that there exists a $V^{G_{L,0+}}$
good lattice in such that is nontrivial and there is a sub-

representation of , on such that are paired by the theta

$\zeta^{\prime}$
$G_{L}^{\prime},/G_{L,0^{+}}^{\prime}$
$\zeta,$
$\zeta^{\prime}$
correspondence for the finite reductive dual pair . J. $(G_{L}/G_{L,0}+, G_{L}^{\prime},/G_{L,0^{+}}^{\prime},)$
Adams and A. Moy ([AM], theorem 3.5) prove that a unipotent representation
corresponds to a unipotent representation for the following finite reductive dual
pairs:
(1) $(Sp(v), O(v^{\prime}))$ where the dimension of is even and the characteristic of $v^{\prime}$
is large enough such that every maximal torus in ( $Sp(v)$ or

$f_{F}$
) $T$ $O(v^{\prime})$
satisfies the condition that $T/Z$ has at least two regular orbits under the
Weyl group action where $Z$ is the center of the group,
(2) $(U(v), U(v^{\prime}))$ , a dual pair of finite unitary groups.
Therefore we only need to check that the dual pair under $(G_{L}/G_{L,0}+ , G_{L}^{\prime},/G_{L,0^{+}}^{\prime}, )$
the restrictions in (i), (ii), (iii), (iv) are a sum of irreducible reductive dual pairs of
the above two types. First suppose now we are in the case (i) i.e., is of $U(\mathscr{V}^{\prime})$
type , or
$D_{m},$ for some $m$ . Then it is clear that both
$2D_{m}$
and
$2D_{m}^{\prime}$ $U(l^{\prime})$ $U(l^{\prime*})$
are orthogonal groups from the table in subsection 1.2, and both and $l$ $l^{*}$
are even-dimensional for any good lattice in . Therefore $(G_{L}/G_{L,0}+$ , $L$ $\mathscr{V}^{\prime}$
is a dual pair of a sum of two irreducible dual pairs satisfying (1).

$G_{L}^{\prime},/G_{L,0^{+}}^{\prime},)$
To check the cases (ii), (iii), (iv) is analogous. So we just sketch the proof.
If we are in case (ii), then all groups $U(l),$ are unitary $U(l^{*}),$ $U(l^{\prime}),$ $U(l^{\prime*})$
groups. If we are in case (iii), then exact one of $U(l),$ (resp. ) $U(l^{*})$ $U(l^{\prime}),$ $U(l^{\prime*})$
is an orthogonal group. And this orthogonal group is of even variables if and

only if the dimension of (resp. ) is even. For case (iv), $U(l)$ is an or- $\mathscr{V}$
thogonal group. This orthogonal group is of even variables if and only if $U(\mathscr{V})$
is of type for some $m$ . Hence this theorem is proved.

$2D_{m}^{\prime\prime}$ $\square $
8. Admissible splitting of metaplectic covers.

From this section to the end of this paper we will assume that $D$ is
commutative. It is known that most reductive dual pairs are split. In this
section, we want to discuss the issue of splitting of the metaplectic covers of
reductive dual pairs in this section.
8.1. Ranga-Rao cocycle and its splitting.

Let be a reductive dual pair such that $D$ is commutative and
. Let $\mathscr{W}=X+Y$ be a complete polarization. Let $M_{Y}[g]$ be the

$\mathscr{W}:=\mathscr{V}\otimes \mathscr{V}^{\prime}$
action of the Schr\"odinger model. The cocycle defined as in (3.3.a) with $c_{Y}(g, g^{\prime})$
respect to $M_{Y}[g]$ is called the Ranga-Rao cocycle and it computed explicitly in $\underline{is}$
[RR]. An explicit admissible splitting (if it exists) of with respect to the $U(\mathscr{V})$
Ranga-Rao cocycle is given in [K1] theorem 3.1 by S. Kudla. We will use the
notation to denote this splitting. $\beta^{Y}(g)$
8.2. Splitting with respect to a generalized lattice model.

Let (resp. ) be a good lattice in
$L$
(resp. ). Define $B:=B(L, L^{\prime})$ as
$L^{\prime}$ $\mathscr{V}$
in subsection 4.1. Let denote the cocycle with respect to the generalized $c_{B}(g, g^{\prime})$
lattice model of the Weil representation of . An admissible

$(M_{B}[g], \mathscr{S}(B))$ $Sp(\mathscr{W})$
splitting of the cocycle $\beta^{L}$
, is given in [Pnl]. We shall $c_{B}|_{l_{\mathscr{V}}’(U(\mathscr{V}))\times l_{\mathscr{V}(U(\mathscr{V}))}}$
normahze such that where

$\beta^{L}$
is a character of inflated $\beta^{L}|_{U(\mathscr{V})_{L}}=\xi_{L}$ $\xi_{L}$ $U(\mathscr{V})_{L}$
from the character of OX of $U(l)\times U(l^{*})$ via the homomorphism $\xi$ $\xi^{*}$ $ U(\mathscr{V})_{L}\rightarrow$
$U(l)\times U(l^{*})$ and are defined as follows. $\xi,$ $\xi^{*}$
(i) If $D=F$ , let (resp. 4’) be the trivial character of $U(l)$ (resp. ). $\xi$
$U(l^{*})$
(ii) Suppose that $D$ is an unramified extension of . Let be the trivial $F$ $\xi$
character of $U(l)$ if is even-dimensional, be the sgn character of $l^{\prime*}$
$U(l)$ if is odd-dimensional. Let be the trivial character of $l^{\prime*}$

$\xi^{*}$
$U(l^{*})$
if is even-dimensional, be the sgn character of

$l^{\prime}$
if is odd- $U(l^{*})$
$l^{\prime}$
dimensional.
(iii) Suppose that $D$ is a ramified extension of . Then one of is $F$ $l,$
$l^{*}$
a quadratic space and the other is a symplectic space. When

$SU(l)/[U(l), U(l)]$ (resp. $SU(l^{*})/[U(l^{*}),$ ) is nontrivial, let $U(l^{*})]$ $\xi$
(resp. 4’) be a character of order two of $U(l)$ (resp. ) whose $U(l^{*})$
restriction to $SU(l)$ (resp. $SU(l^{*})$ ) is also nontrivial. Let be the $\xi,$ $\xi^{*}$
trivial characters otherwise.

From the definition, it is clear that , and . We also $\xi$ $=\xi^{-1}$ $\xi’=(\xi^{*})^{-1}$ $\xi_{L}=\xi_{L}^{-1}$
note here that the character depends only on the parity of the dimension of $\xi_{L}$
.
8.3. Depth and splitting.
Let be a good lattice in
$L$ $\mathscr{V}$
. The map $\tilde{\beta}^{L}$
: $U(\mathscr{V})$
$\rightarrow U\overline{(\mathscr{V}})$
given by
$\tilde{\beta}^{L}(g):=(l_{\mathscr{V}^{\prime}}(g),\beta^{L}(g)M_{B}[l_{\mathscr{V}^{\prime}}(g)])$
(8.3.a)
is a group homomorphism. Therefore, if is an irreducible admissible repre- $\pi$
sentation of , then is an irreducible admissible representation of

$U(\mathscr{V})$
. By 1.4.
$U(\mathscr{V})$ $(1.5.c)$ and 8.3. , we have $($ $b),$ $($ $a)$
$(\omega_{\psi}(\tilde{\beta}^{L}(k)).f)(w)=\xi_{L}(k)\tilde{\omega}_{\psi}(l_{\mathscr{V}^{\prime}}(k)).(f(l_{\mathscr{V}^{\prime}}(k)^{-1}.w))$
(8.3.b)
for $k\in U(\mathscr{V})_{L}$ and $f\in \mathscr{S}(B)$ .
PROPOSITION. Let $G:=U(\mathscr{V})$ . Suppose that the extension splits. $\tilde{G}\rightarrow G$
Let be an irreducible admissible representation of . Then is of depth zero

$\pi$
$\tilde{G}$
$\pi$
if and only if is of depth zero where is a good lattice in such that

$L$ $\mathscr{V}$
and I is a fixed Iwahori subgroup of .

$G_{L,0}+\subseteq I$ $G$
824 S.-Y. PAN
PROOF. Suppose that is of depth zero. Then has nontrivial vectors $\pi$ $\pi$
fixed by for some good lattice in such that

$\tilde{\beta}_{I}(G_{L_{1}},0+ )$
for a fixed $L_{1}$ $\mathscr{V}$
$G_{L_{1},0}+\subseteq I$
Iwahori subgroup I where is defined in subsection 3.3. Compare and , $\beta_{I}$ $\beta_{I}$
$\beta^{L}$
we see that we only need to prove that the restriction is trivial. From $\beta^{L}|_{G_{L_{1},0+}}$
subsection 8.2, we know that . Suppose that $I=G_{x_{0}}$ for some point $\beta^{L}|_{G_{L}}=\xi_{L}$
in the affine building of . From the definition in subsection 8.2, it is clear

$x_{0}$
$G$
that is trivial. Since ,

$\xi_{L}|_{G_{x_{0},0+}}$
, we see that is trivial. $G_{L_{1}}$ $0+\subseteq G_{x_{0}}$
$0+,$ $\beta^{L}|_{G_{L_{1},0+}}$
Hence has nontrivial vectors fixed by

$\pi$
if and only if has $\beta_{I}(G_{L_{1}},0+)$
nontrivial vectors fixed by , Hence is of depth zero if and only if $G_{L_{1}}$

$0+.$
$\pi$
is of depth zero. $\square $
8.4.
COROLLARY. Suppose that is a split reductive dual pair and $(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$
$D$ is commutative. Suppose that an irreducible admissible representation of $\pi$
corresponds to an irreducible admissible representation

$U(\mathscr{V})$
of with $\pi^{\prime}$
respect to the splittings and for good lattices in and re- $\beta^{L}$ $\beta^{L^{\prime}}$
$L,$
$L^{\prime}$ $\mathscr{V}$
spectively. Then is of depth zero if and only if is of depth zero.$\pi$

$\pi^{\prime}$
PROOF. The corollary follows from Corollary 4.4 and Proposition 8.3
immediately. $\square $
9. Correspondence of depth zero supercuspidal representations.

In this section, we consider the local theta correspondence of depth zero
supercuspidal representations of split reductive dual pairs.
9.1. Depth zero supercuspidal representations of classical groups.
Let be an irreducible depth zero supercuspidal representation of
$(\pi, V)$
$G:=U(\mathscr{V})$ with a minimal $K$-types for some good lattice in and an $(G_{L}, \zeta)$ $L$ $\mathscr{V}$
irreducible representation of . We will use the notation to denote $\zeta$

$G_{L}/G_{L,0}+$
$\tilde{\zeta}$
the representation of inflated from of . Then we know that $G_{L}$ $\zeta$

$G_{L}/G_{L,0}+$ $\zeta$
must be cuspidal and where -Ind” denotes the compact induction.

$\pi\simeq c- Ind_{G_{L}}^{G}\tilde{\zeta}$
“
$c$
On the other hand, any irreducible cuspidal representation of induces $G_{L}/G_{L,0}+$
an irreducible supercuspidal representation of (cf. [MP2], corollary 6.8). $G$
LEMMA. Suppose an irreducible supercuspidal representation $(\pi, V)$ of has $G$
a minimal $K$-type for some good lattice in . Then the lattice is $(G_{L}, \zeta)$ $L$ $\mathscr{V}$
$L$
unique up to equivalence.
PROOF. Recall that two lattices and in are said to be equivalent $L_{1}$ $L_{2}$ $\mathscr{V}$
if there exists an element $g\in G$ such that $L_{1}=g.L_{2}$ . Note that is a maximal $G_{L}$
open compact subgroup of . Suppose that has another minimal K-type $G$ $\pi$
where is an irreducible cuspidal representation. Moreover, by the

$(G_{M}, \zeta^{\prime})$ $\zeta^{\prime}$
associativity of the minimal $K$-types $G_{M}/G_{M,0}+$ must be isomorphic to $G_{L}/G_{L,0}+$ .

Then there exists an element $g\in G$ such that the vertex corresponding to $g.M$ and
the vertex corresponding to are in the same chamber. But surjects $L$ $G_{L}\cap G_{g.M}$
onto $G_{L}/G_{L,0}+$ and $G_{M}/G_{M,0}+$ by the associativity of minimal -types defined

$K$
in [MP1] 5.1. Then must be equal to $g.M$ . Therefore and $M$ are $L$ $L$
equivalent. $\square $
9.2. Depth zero supercuspidal representations of metaplectic covers.

Let be a split reductive dual pair that $D$ is com-
$(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$ $s\underline{uch}$
mutative. An irreducible admissible representation of is called a su- $\pi$ $U(\mathscr{V})$
percuspidal representation if is a supercuspidal representation of for $\underline{\pi\circ}\tilde{\beta}$

$U(\mathscr{V})$
some splitting : . Since for different splittings the rep-

$\tilde{\beta}$
$U(\mathscr{V})\rightarrow U(\mathscr{V})$ $\tilde{\beta}_{1},\tilde{\beta}_{2}$
resentations , are different up to a character of , we see the

$\pi\circ\tilde{\beta}_{1}$ $\pi\circ\tilde{\beta}_{2}$
$U(\mathscr{V})$
definition is independent of the choice of the splitting . As in subsection 9.1, an $\tilde{\beta}$
irreducible depth zero supercuspidal representation of $G:=U(\mathscr{V})$ with a minimal

$K$-type is isomorphic to . Therefore it is not difficult to see that
$(G_{L}, \zeta)$ $c- Ind_{G_{L}}^{G}\tilde{\zeta}$
if is an irreducible depth zero supercuspidal representation of with minimal

$\pi$
$\tilde{G}$
$K$-type , then is isomorphic to ( OX ) where

$(G_{L}, \zeta)$
is defined $\pi\circ\tilde{\beta}^{L}$
$c- Ind_{G_{L}}^{G}$ $\xi_{L}$
$\tilde{\zeta}$
$\xi_{L}$
in subsection 8.2.
9.3.
THEOREM. Suppose that is a split reductive dual $(G, G^{\prime}):=(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$
pair and $D$ is commutative. S ppose that ( (resp. ) is an irreducible $u$ $\pi\underline{V)},$ $(\pi^{\prime},$ $V^{\prime})$
admissible representation of (resp. ) such that the representation $U(\mathscr{V})$ $U(\mathscr{V}^{\prime})$
OX is a first occurrence in the correspondence. Suppose that is supercuspidal

$\pi$
$\pi^{\prime}$
$\pi$
with a minimal $K$-type . Then has a minimal $K$-type such that $(G_{L}, \zeta)$
$\pi^{\prime}$
$(G_{L}^{\prime},, \zeta^{\prime})$
OX is a first occurrence in the correspondence for the finite reductive dual pair
$\zeta$
$\zeta^{\prime}$
.
$(G_{L}/G_{L,0^{+}}, G_{L}^{\prime},/G_{L,0^{+}}^{\prime}, )$
PROOF. By Theorem 5.6, we know that has a minimal $K$-type $\pi^{\prime}$

$(G_{L}^{\prime},, \zeta^{\prime})$
such that and are paired by the theta correspondence for the finite reductive
$\zeta$
$\zeta^{\prime}$
dual pair . Suppose that OX is not a first occurrence.

$(G_{L}/G_{L,0}+, G_{L}^{\prime},/G_{L,0^{+}}^{\prime},)$
$\zeta$
$\zeta^{\prime}$
Then is not cuspidal by induction principle. Therefore the

$\zeta^{\prime}$
$G_{L}^{\prime},/G_{L,0^{+}}^{\prime},-$
representation contains a non-cuspidal component. Then can $V^{\prime G_{L^{\prime},0+}^{\prime}}$

$\pi^{\prime}\circ\tilde{\beta}^{L^{\prime}}$
not be an irreducible supercuspidal representation of by [MP2] corollary $U(\mathscr{V}^{\prime})$
6.8. But we know that must be an irreducible supercuspidal repre- $\pi^{\prime}\circ\tilde{\beta}^{L^{\prime}}$
sentation of by the induction principle of the local theta correspondence.

We get a contradiction. $\square $
9.4.
Next we want to consider the converse of Theorem 9.3. Suppose that we
826 S.-Y. PAN
have two finite reductive dual pairs $(U(v), U(v^{\prime*}))$ and $(U(v^{*}), U(v^{\prime}))$ satisfying
one of the following conditions:
(i) and are symplectic spaces, and are orthogonal $v$
$v^{*}$ $v^{\prime}$ $v^{\prime*}$
spaces,
(ii) all are Hermitian spaces with respect to a $v,$ $v^{*},$ $v^{\prime},$
$v^{\prime*}$
quadratic
extension of , $f_{F}$
(iii) and are symplectic spaces, and are orthogonal $v$

$v^{\prime}$
$v^{*}$ $v^{\prime*}$
spaces.
Then it is clear that there exists a field $D$ over and two spaces $F$ $\mathscr{V},$
, integers
, and good lattice

$\kappa,$
and (with respect to the integers
$\kappa^{\prime}$
) $L$ $L^{\prime}$
$\kappa,$
$\kappa^{\prime}$
in $\mathscr{V},$
respectively such that

$L/L^{*}\varpi\simeq v$ , $L^{*}/L\simeq v$ ’, $L^{\prime}/L^{\prime*}\varpi\simeq v^{\prime}$
, $L^{\prime*}/L^{\prime}\simeq v^{\prime*}$
,
as Hermitian spaces over . This is just an easy consequence of the table in $f_{D}$
subsection 1.2. Therefore, we have $G_{L}/G_{L,0}+\simeq U(v)\times U(v^{*})$ and $ G_{L}^{\prime},/G_{L,0^{+}}^{\prime},\simeq$
where $G:=U(\mathscr{V})$ and

$U(v^{\prime})\times U(v^{\prime*})$
. $G^{\prime}:=U(\mathscr{V}^{\prime})$
Given a character of , clearly there exists a character with conductoral $\emptyset$

$f$ $\psi$
exponent such that where is defined in subsection 1.5. Of course,

$\overline{\psi}=\phi$ $\overline{\psi}$
the character is not unique. $\psi$
9.5.
The following theorem is the converse of Theorem 9.3.
THEOREM. Suppose that is a split reductive dual $(G, G^{\prime}):=(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$
pair and $D$ is commutative. Suppose that are irreducible cuspidal repre- $\zeta,$
$\zeta^{\prime}$
sentations and OX is a first occurrence of theta correspondence for the finite $\zeta$
$\zeta^{\prime}$
reductive dual pair . Then OX is a first occurrence for $(G_{L}/G_{L,0}+ , G_{L}^{\prime},/G_{L,0^{+}}^{\prime}, )$ $\pi$
$\pi^{\prime}$
the reductive dual pair where is the irreducible admissible representation $(G, G^{\prime})$ $\pi$
of such that $\tilde{G}$
and is the irreducible admissible $\pi\circ\tilde{\beta}^{L}\simeq c- Ind_{G_{L}}^{G}(\xi_{L}\otimes\tilde{\zeta})$

$\pi^{\prime}$
representation of such that where are $\tilde{G}^{\prime}$
$\pi^{\prime}\circ\tilde{\beta}^{L^{\prime}}\simeq c- Ind_{G_{L}^{\prime}}^{G^{\prime}},$
$(\xi_{L^{\prime}}\otimes\tilde{\zeta}^{\prime})$
$\xi_{L},$ $\xi_{L^{\prime}}$
deftned in subsection 8.2.

and are irreducible cuspidal representations,
PROOF. Because $\zeta$
$\zeta^{\prime}$ $\xi_{L}\otimes\tilde{\zeta}$
and are representations inflated from irreducible cuspidal representations

$\xi_{L^{\prime}}\otimes\tilde{\zeta}^{\prime}$
of $G_{L}/G_{L,0}+$ and , respectively. By proposition 6.5 in [MP2], we know $G_{L}^{\prime},/G_{L,0^{+}}^{\prime}$
that both and are irreducible supercuspidal $c- Ind_{G_{L}}^{G}(\xi_{L}\otimes\tilde{\zeta})$ $c- Ind_{G_{L}^{\prime}}^{G^{\prime}},$
representations. Moreover it is clear that and are uniquely determined by $\pi$

$\pi^{\prime}$
the condition in the theorem. We know that has the minimal $K$-type . $\pi$ $(G_{L}, \zeta)$
Fix the group and let be a group in the Witt tower of . Suppose that $G$ $G^{\prime\prime}$ $G^{\prime}$
is a first occurrence for some irreducible admissible representation

$\pi\otimes\pi^{\prime\prime}$
of $\pi^{\prime\prime}$
. We know that
$\tilde{G}^{\prime\prime}$
must be of depth zero by Corollary 4.4. Therefore by $\pi^{\prime\prime}$
Theorem 9.3, has a minimal $K$-type for some good lattice in $\pi^{\prime\prime}$
$(G_{L}^{\prime\prime},, , \zeta^{\prime\prime})$
$L^{\prime\prime}$ $\mathscr{V}^{\prime\prime}$
where such that OX occurs in the theta correspondence for

$U(\mathscr{V}^{\prime\prime}):=G^{\prime\prime}$ $\zeta$
$\zeta^{\prime\prime}$
the dual pair . Moreover we know that OX is a $(G_{L}/G_{L,0}+ , G_{L}^{\prime\prime},,/G_{L,0^{+}}^{\prime\prime},,)$ $\zeta$

$\zeta^{\prime\prime}$
first occurrence. Write and $G_{L}^{\prime},/G_{L,0^{+}}^{\prime},\simeq U(l^{\prime})\times U(l^{\prime*})$ $ G_{L^{\prime}}^{\prime\prime},/G_{L,0^{+}}^{\prime\prime},,\simeq U(l^{\prime\prime})\times$
. Then
$U(l^{\prime\prime*})$
and (resp. and ) are in the same Witt $U(l^{\prime})$ $U(l^{\prime\prime})$ $U(l^{\prime*})$ $U(l^{\prime\prime*})$
tower. Therefore and because $\dim(l^{\prime})=\dim(l^{\prime\prime}),$ $\dim(l^{\prime*})=\dim(l^{\prime\prime*})$ $\zeta^{\prime}\simeq\zeta^{\prime\prime}$
now both OX and are first occurrences. Hence

$\zeta$
and .
$\zeta^{\prime}$ $\zeta\otimes\zeta^{\prime\prime}$ $G^{\prime}=G^{\prime\prime}$ $\pi^{\prime}=\pi^{\prime\prime}$
Now is an irreducible supercuspidal representation of

$\pi^{\prime}$
having a minimal K- $\tilde{G}^{\prime}$
type . Hence as in subsection 9.2.

$(G_{L^{\prime}}, \zeta^{\prime})$
$[$ $\pi^{\prime\prime}\circ\tilde{\beta}^{L^{\prime}}\simeq c- Ind_{G_{L}^{\prime}}^{G^{\prime}},$
Theorem 9.3 and Theorem 9.5 indicate that the theta correspondence of
depth zero supercuspidal representations for -adic reductive dual pairs can be $p$
completely described by the theta correspondence of cuspidal representations for

finite reductive dual pairs. Here we note that the theta correspondence of
cuspidal representations for finite reductive dual pair is one-to-one although the
general correspondence for finite reductive dual pair is not. In the following
subsections, we give a few examples to illustrate the nice relation of theta
correspondence of -adic and finite reductive dual pairs. These examples may $p$
not be entirely new but here we provide more precise information. All three
examples are pairs of orthogonal and symplectic groups. The characters $\xi_{L},$ $\xi_{L^{\prime}}$
are trivial in these situations, so we will just ignore them.

9.6.
EXAMPLE. This example is originally from [As] theorem 8.3. Assume that
is a character of $F$ with conductoral exponent
$\psi$
for some integer . $\mathfrak{p}_{F}^{\lambda_{F}}$
$\lambda_{F}$
Consider the dual pair where is a two-dimensional aniso- $(Sp_{4}(F), O(\mathscr{V}^{\prime}))$

tropic quadratic space over $F$ such that the corresponding quadratic form is split
over a unramified quadratic extension of $F$. This pair is unramified and split.
The sgn representation of is supercuspidal and first occurs in the cor- $O(\mathscr{V}^{\prime})$
respondence for this dual pair. Assume it corresponds to the representation

$(\pi, V)$ of $Sp_{4}(F)$ (under our choice of splitting). Let be the unique good $L^{\prime}$
lattice in . We know is self-dual and

, the aniso- $L^{\prime}$

$G_{L}^{\prime},/G_{L,0^{+}}^{\prime},\simeq O_{2}^{-}$ $(f_{F})$
tropic finite orthogonal group in two variables. The sgn representation factors
through the first congruence subgroup , and becomes a representation of $G_{L,0^{+}}^{\prime}$
the finite orthogonal group , still denoted by $sgn$ . For the dual pair $O_{2}^{-}(f_{F})$
the sgn representation corresponds to the representation

$(Sp_{4}(f_{F}), O_{2}^{-}(f_{F}))$ of $\theta_{10}$
, in the notation of [Sr]. By Corollary 4.3, there exists a good lattice

$Sp_{4}(f_{F})$ $L$
in such that
$\mathscr{V}$
is nontrivial. We know that $L^{*}=L\varpi^{m}$ where $m$ is 0 or 1

$V^{G_{L,0+}}$
depending on the parity of . By Theorem 9.5, we have . $\lambda_{F}$ $\pi\simeq c- Ind_{G_{L}}^{Sp_{4}(F)}\tilde{\theta}_{10}$
Note that is conjugate to when

$G_{L}$
is odd. This is the situation in $Sp_{4}(\mathscr{O}_{F})$ $\lambda_{F}$
[As]. But is not conjugate to when

$G_{L}$
is even. $Sp_{4}(\mathscr{O}_{F})$ $\lambda_{F}$
9.7.
EXAMPLE. Our next example is also originally from [As] but here we are
828 S.-Y. PAN
able to provide more information. Assume the character is of conductoral $\psi$
exponent . Consider the dual pair where is a two-

$(Sp_{4}(F), O(\mathscr{V}^{\prime}))$
dimensional anisotropic quadratic space over such that the corresponding $F$
quadratic form is split over a ramified quadratic extension of . We can identify $F$
with a ramified quadratic extension $E$ of with the quadratic form induced
$F$
from the norm of $E$ over . We know we have two ramified quadratic ex- $F$
tensions of , namely and where is a nonsquare unit in

$F$
. $F(\sqrt{\varpi})$ $F(\sqrt{\epsilon\varpi})$ $\epsilon$ $\mathscr{O}_{F}$
We know this pair is split but not unramified. Again the sgn representation
of is supercuspidal and first occurs in the correspondence with $Sp_{4}(F)$ .
$O(\mathscr{V}^{\prime})$
Assume it corresponds to the representation $(\pi, V)$ of $Sp_{4}(F)$ . Let be the $L^{\prime}$
unique good lattice in . We know that where $\mathscr{V}^{\prime}$
$G_{L}^{\prime},/G_{L,0^{+}}^{\prime},\simeq O_{1}\times O_{1}$ $O_{1}$
denote the orthogonal group in one variable i.e., the group of two elements.
Now the sgn representation factors through the subgroup , and is equivalent $G_{L,0^{+}}^{\prime}$
to the representation sgn sgn of the finite group . We know that $\otimes$ $O_{1}\times O_{1}$
has two irreducible $(q-1)/2$ -dimensional representations

$SL_{2}(f_{F})$ and , $\zeta_{1}$ $\zeta_{2}$
which are cuspidal. Now consider the theta correspondence for the dual pair
over the finite field with respect to a character . Note that the
$(SL_{2} (f_{F}) , O_{1} )$ $f_{F}$ $\emptyset$
character depends on the choice of . We know that the sgn representation of

$\emptyset$
$\varpi$
corresponds to
$O_{1}$
where is 1 or 2 depending on the character . By $\zeta_{i}$ $i$ $\emptyset$
Theorem 9.3, is nontrivial for the unique (up to equivalence) good lattice $V^{G_{L,0+}}$
$L$
in such that . We know that
$\mathscr{V}$
$L^{*}\varpi\subseteq L\subseteq L^{*}$ $ G_{L}/G_{L,0}+\simeq SL_{2}(f_{F})\times$
. Then the representation sgn sgn of

$SL_{2}(f_{F})$ , corresponds to the $\otimes$
$G_{L}^{\prime},/G_{L,0^{+}}^{\prime}$
representation of $G_{L}/G_{L,0}+$ where are 1 or 2. And we know that $i=j$ $\zeta_{i}\otimes\zeta_{j}$ $i,$ $j$
if and if . By Theorem 9.5, is isomorphic to

$E\simeq F(\underline{\sqrt{\varpi}),}$ $i\neq j$ $E\simeq F(\sqrt{\epsilon\varpi})$ $\pi$
.
$c- Ind_{G_{L}}^{Sp_{4}(F)}\zeta_{i}\otimes\zeta_{j}$
9.8.
EXAMPLE. Let be a four dimensional anisotropic quadratic space over $F$. $\mathscr{V}$
Let $G:=$ O4(F) denote the anisotropic orthogonal group on . Now we want $\mathscr{V}$
to determine the first occurrence of the sgn representation of $O_{4}(F)$ . There

is a unique good lattice in since is anisotropic. And $L$ $\mathscr{V}$ $\mathscr{V}$
$ G_{L}/G_{L,0}+\simeq$
. The sgn representation of O4(F) is of depth zero and, as a

$O_{2}^{-}(f_{F})\times O_{2}^{-}(f_{F})$
representation of $G_{L}/G_{L,0}+=U(l)\times U(l^{*})$ , isomorphic to sgn . We know $\otimes sgn$
that the first occurrence of sgn representation of is paired with of $O_{2}^{-}(f_{F})$ $\theta_{10}$
. Therefore by Theorem
$Sp_{4}(f_{F})$ , the first occurrence of sgn representation of $\underline{9.5}$
O4(F) is paired with where is the symplectic group $Sp_{8}(F)$ $c- Ind_{G_{L}^{\prime}}^{G^{\prime}},$ $\theta_{10}\otimes\theta_{10}$
$G^{\prime}$
over an eight-dimensional space over is a good lattice on such that $\mathscr{V}^{\prime}$
$F,$
both and are four-dimensional spaces over . The represen-

$L^{\prime*}/L^{\prime}$ $\underline{L^{\prime}/}L^{\prime*}\varpi$ $f_{F}$
tation is irreducible supercuspidal. We remark here that the

$c- Ind_{G_{L}^{\prime}}^{G^{\prime}},$ $\theta_{10}\otimes\theta_{10}$
first occurrence of the sgn character is already considered by S. Rallis in [R1].

10. “Dual pairs” of modular congruence groups.

10.1. Modular congruence groups.
So far we have seen that the local theta correspondence of depth zero
representations is closely connected with the theta correspondence for finite
reductive dual pairs. Similar results should hold for representations of positive
depths. Suppose that is an unramified reductive dual pair in $(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$
. Let (resp. ) be a lattice in

$Sp(\mathscr{W})$ $L$
(resp. ) such that (resp. $L^{\prime}$ $\mathscr{V}$
$L^{*}=L\varpi^{m}$
. Define
$L^{\prime*}=L^{\prime}\varpi^{m^{\prime}})$
. Then is a lattice in such that $A:=L\otimes L^{\prime}$ $A$ $\mathscr{W}$
. Fix a positive integer . Then we can define

$A^{*}=A\varpi^{m+m^{\prime}}$
(resp. )- $k$ $\epsilon$
$\epsilon^{\prime}$
Hermitian form , (resp. , ) on the free module $l_{k}:=L/L\varpi^{k}$ (resp.

$\langle$
$\rangle_{l_{k}}$
$\langle$
$\rangle_{l_{k}}$
over the (finite) commutative ring

$l_{k}^{\prime}:=L^{\prime}/L^{\prime}\varpi^{k})$
. We can also define a $\mathscr{O}/\mathfrak{p}^{k}$
skew-symmetric form on the free module over . It is $\ll,$ $\gg_{a_{k}}$ $a_{k}:=A/A\varpi_{F}^{k}$ $\mathscr{O}_{F}/\mathfrak{p}_{F}^{k}$
not difficult to verify that both are subgroups of . Moreover both $G_{L},$ $G_{L^{\prime}}$ $K_{A}$
$G_{L}/G_{L,(k-1)}+$ and , are subgroups of $K_{A}/K_{A,(k-1)}+$ where

$G_{L}^{\prime},/G_{L,(k-1)^{+}}^{\prime}$
$G_{L,(k-1)}+:=\{g\in U(\mathscr{V})|(g-1).L^{*}\subseteq L\varpi^{k-1}, (g-1).L\subseteq L^{*}\varpi^{k}\}$
and $K_{A,(k-1)}+$ is defined similarly. Therefore we have a “modular dual pair”

$(G_{L}/G_{L,(k-1)^{+}}, G_{L}^{\prime},/G_{L,(k-1)^{+}}^{\prime},)$ (10.1.a)
in $K_{A}/K_{A,(k-1)}+$ . We can also define the “Weil representation” of $K_{A}/K_{A,(k-1)}+$

as we do for a finite symplectic group. By restricting the “Weil representation”,
we can define a correspondence between some irreducible representations of the
modular congruence group $G_{L}/G_{L,(k-1)}+$ and some irreducible representations of
. Of course, we cannot expect the correspondence for a modular
$G_{L}^{\prime},/G_{L,(k-1)^{+}}^{\prime},$
congruence dual pair to be one-to-one. However this correspondence still

provides a way to classify irreducible representations of one group by the ir-
reducible representations of the other group. In fact, J. Shalika and S. Tanaka
[Tn] have implicitly used the pairs $(SL_{2}(Z/p^{k}Z), O_{2}^{+}(Z/p^{k}Z))$ and $(SL_{2}(Z/p^{k}Z)$ ,
$O_{2}^{-}(Z/p^{k}Z))$ to construct irreducible representations of the modular congruence
group $SL_{2}(Z/p^{k}Z)$ where $p\in Z$ is a prime number. But from their work, we
also know that these two pairs are not enough to construct all irreducible
representations of $SL_{2}(Z/p^{k}Z)$ unless $k=1$ .
10.2. Tanaka’s result.
Now we want to consider the third “modular dual pair”. It can be
constructed from the -adic reductive dual pair $p$
$(Sp(\mathscr{V}), O(\mathscr{V}^{\prime}))$
where both
spaces and are two-dimensional and the quadratic form on
$\mathscr{V}$
is not split
over but split over a ramified quadratic extension of . This

$F$ $F$
dual pair is
not unramified. We know that has a unique good lattice . $\mathscr{V}^{\prime}$ $L^{\prime}$
Let be a $L$
self-dual lattice in . Now is a good lattice in $\mathscr{V}$

$A:=L\otimes L^{\prime}$
. Again
$\mathscr{W}$
830 S.-Y. PAN
is a “dual pair” in $K_{A}/K_{A,(k-1)}+$ . The “Weil

$(G_{L}/G_{L,(k-1)}+, G_{L}^{\prime},/G_{L,(k-1)^{+}}^{\prime},)$
representation” of $K_{A}/K_{A,(k-1)}+$ can be realized on the space . If $\mathscr{S}(A_{0})_{A^{*}\varpi^{k-1}}$
$k=1$ , then we just get the finite dual pair $(SL_{2}(f) \times SL_{0}(f), O_{1}\times O_{1})$ , which is
not really interesting. However, if $k>1$ , then $G_{L}/G_{L,(k-1)}+$ is isomorphic
to and the “Schr\"odinger model of the Weil representation” of
$SL_{2}(\mathscr{O}_{F}/\mathfrak{p}_{F}^{k})$
$K_{A}/K_{A,(k-1)}+$ can be realized as the space of complex valued functions on the set
. By decomposing the “Weil representation” according to the

$\mathscr{O}_{F}/\mathfrak{p}_{F}^{k}\times \mathscr{O}_{F}/\mathfrak{p}_{F}^{k-1}$
representations of , which is a group with an abelian subgroup of $G_{L}^{\prime},/G_{L,(k-1)^{+}}^{\prime},$
index 2, we can construct all irreducible representations of that $SL_{2}(\mathscr{O}_{F}/\mathfrak{p}_{F}^{k})$
cannot be obtained from the two “modular dual pairs” considered in the previous
subsection. This is exactly what S. Tanaka has done in [Tn] although not
necessarily from this point of view.
11. Theta dichotomy for -adic unitary groups. $p$
Theta dichotomy is a very interesting phenomenon which concerns the first

occurrences in local theta correspondences. A result of theta dichotomy of
supercuspidal representations for reductive dual pairs of -adic unitary groups $p$
has been established by M. Harris, S. Kudla, S. Rallis and W. Sweet (cf. [HKS]).
The theta dichotomy depends on the splitting of the metaplectic covers. The
splittings used in [HKS] are the splittings given in [K1] i.e., the splittings with
respect to a Schr\"odinger model of the Weil representation. In this section, we
want to investigate the theta dichotomy for reductive dual pairs of unitary groups
by using the splittings with respect to a generalized lattice model. Under this
splitting, Theorems 9.3 and 9.5 can be applied and new results of theta di-
chotomy for finite reductive dual pairs and other -adic reductive dual pairs are $p$
obtained in the next two sections. In this section, let $D=E$ be a quadratic
extension of . $F$
11.1. Kudla and Rallis’ result.

Let $E:=F(\sqrt{\Delta})$ be a quadratic extension of . Let denote the $F$ $\epsilon_{E/F}$
quadratic character of associated to the extension $E/F$ i.e., $F^{\times}$

$\epsilon_{E/F}(t):=(\Delta, t)_{F}$
where denotes the Hilbert symbol of . Fix an -Hermitian space

$(, )_{F}$
over . $F$ $\epsilon$
$\mathscr{V}$
$E$
Let be an -dimensional -Hermitian space such that

. Denote the $m$

$\epsilon^{\prime}$
spaces , so that $\mathscr{V}^{\prime\pm}$
$\epsilon_{E/F}((-1)^{m(m-1)/2}\det(\mathscr{V}^{\prime\pm}))=\pm 1$
. (ll.l.a)
Let , $\{\mathscr{V}_{m_{i}^{+}}^{\prime+}\}$
be two Witt towers where
$\{\mathscr{V}_{m_{i}^{-}}^{\prime-} \}$
(resp. ) denotes $m_{i}^{+}$
$m_{i}^{-}$
the dimension of (resp. ). dual pair

$\mathscr{V}_{m_{i}^{+}}^{\prime+}$
, if a $\mathscr{V}_{m_{i}^{-}}^{\prime-}$
$Forthem_{i}^{\pm}$
$(U(\mathscr{V}), U(\mathscr{V}_{m_{i}^{\pm}}^{\prime\pm}))$
character $\underline{\chi}of$
such that
$E^{\times}$
is fixed, then a splitting : $\chi|_{F^{\times}}=\epsilon_{E/F}$
$\tilde{\beta}^{Y^{\pm}}$
$U(\mathscr{V})\rightarrow U(\mathscr{V})$
is determined in [K1]. Because is a character of order two, $\epsilon_{E/F}$
we will fix such that for the whole tower $\chi$

$\chi|_{F^{\times}}=\epsilon_{E/F}^{m_{0}^{\pm}}$
$\{\mathscr{V}_{m_{i}^{\pm}}^{\prime\pm}\}$
. The following
theorem is described in [HKS] p. 944.
THEOREM (Kudla-Rallis). Let be a reductive dual pair of $(U(\mathscr{V}), U(\mathscr{V}^{\prime\pm}))$
unit y groups. Suppose that

$ar$ are irreducible supercuspidal representations $\pi^{+},$ $\pi^{-}$
of such that
$U(\mathscr{V})$
. Let be the smallest number , $\pi^{+}\circ\tilde{\beta}^{Y^{+}}=\pi^{-}\circ\tilde{\beta}^{Y^{-}}$ $\ell_{0}^{\pm}(\chi)$
$m^{\pm}$
the dimension of , such that occurs in the theta correspondence for the dual
$\mathscr{V}^{\prime\pm}$ $\pi^{\pm}$
pair . Moreover, suppose that the parity of the dimension of

$(U(\mathscr{V}), U(\mathscr{V}^{\prime\pm}))$ $\mathscr{V}$
and the parity of the dimensions of are the same. Then $\mathscr{V}^{\prime\pm}$
$\ell_{0}^{+}(\chi)+\ell_{0}^{-}(\chi)=2n+2$
where $n$
is the dimension of $\mathscr{V}$
.
This theorem is called the preservation principle. A corollary of the theorem is
that there exist sequences of dimensions $ n=n_{0}<n_{1}<n_{2}<\cdots<n_{i}<\cdots$ and
irreducible supercuspidal representations of such $\pi=\pi_{0},$ $\pi_{1},$ $\pi_{2},$
$\ldots,$
$\pi_{i},$
$\ldots$
$U(\mathscr{V}_{i}^{\pm})$
that the dimension of is and OX occurs in the theta corre- $\mathscr{V}_{i}^{\pm}$

$n_{i}$ $\pi_{i}$ $\pi_{i+1}$
spondence for the dual pair with respect to the splittings , $(U(\mathscr{V}_{i}^{\pm}), U(\mathscr{V}_{i+1}^{\mp}))$
$\beta^{Y_{i}^{\pm}}$
. (Here the signs of these

$\beta^{Y_{i+1}^{\mp}}$
are alternating i.e., if , then $\mathscr{V}_{i}^{\pm}’ s$ $\mathscr{V}_{i}^{\pm}=\mathscr{V}_{i}^{+}$
.) If does not come from a smaller unitary group via the theta
$\mathscr{V}_{i+1}^{\mp}=\mathscr{V}_{i+1}^{-}$ $\pi$
correspondence, the sequence of dimensions will be $\{n_{i}\}$
$n,$ $n+2,$ $n+6,$ $n+12,$ $n+20,$ $\ldots,$ $n+i(i+1),$ $\ldots$ . (ll.l.b)

When the dimension of and the dimensions of are of opposite parity, the $\mathscr{V}$
similar statement in Theorem 11.1 is conjectured in [HKS] speculation 7.6. In

particular, the sequence of dimensions analogous to (ll.l.b) should be
$n,$ $n+1,$ $n+4,$ $n+9,$ $n+16,$ $\ldots,$
$n+i^{2},$ $\ldots$ . (ll.l.c)
11.2. Splittings of two related Witt towers.

Suppose that and are two -Hermitian spaces satisfying the con-$\mathscr{V}^{\prime+}$ $\mathscr{V}^{\prime-}$ $\epsilon^{\prime}$
vention (ll.l.a). Let (resp. ) denote the dimension of (resp. ). $m^{+}$ $m^{-}$
$\mathscr{V}^{\prime+}$ $\mathscr{V}^{\prime-}$
Let be a good lattice in

$L$
. Let for some good lattice $\mathscr{V}$
$B^{\pm}:=B(L, L^{\prime\pm})$
in
$L^{\prime\pm}-$
. Let denote the splitting defined in [Pnl] of the extension
$\beta_{\mathscr{V}}^{L},\pm$
with respect to the generalized lattice model

$U(\mathscr{V})\rightarrow U(\mathscr{V})$ $(M_{B}\pm[g], \mathscr{S}(B^{\pm} ))$
for the dual pair . It is known that the splitting does $(U(\mathscr{V}), U(\mathscr{V}^{\prime\pm}))$ $\beta_{\mathscr{V}}^{L},\pm$
not depend on the lattice . $L^{\prime\pm}$
Define a character of as follows. We have a homomorphism $\eta_{L}$

$U(\mathscr{V})_{L}$
. If is an unramified extension of , then both

$U(\mathscr{V})_{L}\rightarrow U(l)\times U(l^{*})$ $E$ $F$
$U(l),$ are finite unitary group. Let

$U(l^{*})$ (resp. ) be the character of order $\eta_{1}$ $\eta_{2}$
two of $U(l)$ (resp. ) if the group is not trivial, and the trivial character $U(l^{*})$
otherwise. If is a ramified extension of , then one of $U(l),$

$E$
is an $F$ $U(l^{*})$
orthogonal group and the other is a symplectic group. Let (resp. ) be $\eta_{1}$ $\eta_{2}$
832 S.-Y. PAN
the character of $U(l)$ (resp. ) of order two whose restriction to $SU(l)$ $U(l^{*})$
(resp. $SU(l^{*})$ ) is trivial if the group is a (nontrivial) orthogonal group, be the

trivial character otherwise. For both cases, let be the character of $\eta_{L}$
$U(\mathscr{V})_{L}$
lifted from by the above homomorphism. It is easy to see that the

$\eta_{1}\otimes\eta_{2}$
character factors through the determinant map $det$ :

$\eta_{L}$
$U(\mathscr{V})_{L}\rightarrow E^{(1)}$ $;=$
$\{t\in E|t\tau_{E}(t)=1\}$ . Moreover, we can check that the restriction of the sgn
character to is equal to . In fact, sgn is the only character of
$U(\mathscr{V})_{L}$ $\eta_{L}$
$U(\mathscr{V})$
whose restriction to is equal to if we are not in the case that $E$ is a

$U(\mathscr{V})_{L}$ $\eta_{L}$
ramified extension and is the good lattice such that the component of or- $L$
thogonal group of $U(l)\times U(l^{*})$ is trivial. For this exceptional case, there are
two such characters. One of them is the sgn character and the other is the trivial
character.
Recall that the function : for a dual pair is $\alpha_{\mathscr{V}^{\prime}}$
$U(\mathscr{V})$
$\rightarrow C’$ $(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$
defined by
$M_{Y}[l_{\mathscr{V}^{\prime}}(g)]\circ\Psi=\alpha_{\mathscr{V}^{\prime}}(g)\Psi\circ M_{B}[l_{\mathscr{V}^{\prime}}(g)]$
(11.2.a)
where is an isomorphism from

$\Psi$
to which intertwines the actions of $\mathscr{S}(B)$ $\mathscr{S}(Y)$
the Heisenberg group. We know that is independent of the choice of . $\alpha_{\mathscr{V}^{\prime}}$

$\Psi$
Let be the splitting in [K1] for the dual pair

$\beta^{Y^{\pm}}$
. Then $\beta_{\mathscr{V}^{\prime\pm},\chi}$
$(U(\mathscr{V}), U(\mathscr{V}^{\prime\pm}))$
we have
$\alpha_{\mathscr{V}^{\prime+}}(g)\beta_{\mathscr{V}^{+}}^{Y^{+}},(g)=\eta_{L}(g)\alpha_{\mathscr{V}^{\prime}}-(g)\beta_{\mathscr{V}^{\prime}}^{Y^{-}}-(g)$
(11.2.b)
for from [Pnl]. Since we know that

$g\in U(\mathscr{V})_{L}$ $sgn|_{U(\mathscr{V})_{L}}=\eta_{L}$ , we can and will
normalize the generalized model so that we have
$\alpha_{\mathscr{V}^{\prime}}+(g)\beta_{\mathscr{V}^{+}}^{Y^{+}},(g)=sgn(g)\alpha_{\mathscr{V}^{\prime}}-(g)\beta_{\mathscr{V}^{\prime}}^{Y^{-}}-(g)$
(11.2.c)
for $g\in U(\mathscr{V})$ .

11.3.
LEMMA. Let denote the
$\beta_{\mathscr{V}}^{L},\pm$
for the dual pair $\beta^{L}$ $(U(\mathscr{V}), U(\mathscr{V}^{\prime\pm}))$

. There
exists a character $\sigma$
of such that$U(\mathscr{V})$
$\beta_{\mathscr{V}^{+}}^{L},(g)=\sigma(g)\alpha_{\mathscr{V}^{\prime+}}(g)\beta_{\mathscr{V}^{+}}^{Y^{+}},(g)$
,
$\beta_{\mathscr{V}}^{L},-(g)=\eta_{L}(g)\sigma(g)\alpha_{\mathscr{V}^{\prime}}-(g)\beta_{\mathscr{V}^{\prime}}^{Y^{-}}-(g)$
.
for all $g\in U(\mathscr{V})_{L}$ .

PROOF. From [Pnl] subsection 3.4 we know that there exist characters
$\sigma^{+}$
, of such that
$\sigma^{-}$
, and
$U(\mathscr{V})$ $\beta_{\mathscr{V}^{+}}^{L},(g)=\sigma^{+}(g)\alpha^{+}(g)\beta_{\mathscr{V}^{+}}^{Y^{+}},(g)$ $\beta_{\mathscr{V}^{-}}^{L},(g)=$
. So we need to prove that

$\sigma^{-}(g)\alpha^{-}(g)\beta_{\mathscr{V}^{\prime}}^{Y^{-}}-(g)$
$\eta_{L}\sigma^{+}|_{U(\mathscr{V})_{L}}=\sigma^{-}|_{U(\mathscr{V})_{L}}$ . (11.3.a)
Let be the character defined in subsection 8.2 for the dual pair
$\xi_{L}^{\pm}$ $\xi_{L}$
. From subsection 8.2, we know that

and $\beta_{\mathscr{V}^{+}}^{L},|_{U(\mathscr{V})_{L}}=\xi_{L}^{+}$
And we know that the characters depends on only the

$\beta_{\mathscr{V}^{-}}^{L},|_{U(\mathscr{V})_{L}}=\xi_{L}^{-}.$
$\xi_{L}^{\pm}$
parity of dimensions of . The parity of the dimensions of and are $\mathscr{V}^{\prime\pm}$ $\mathscr{V}^{\prime+}$ $\mathscr{V}^{\prime-}$
the same, so . Hence . Therefore to prove

$\xi_{L}^{+}=\xi_{L}^{-}$ $\beta_{\mathscr{V}^{+}}^{L},|_{U(\mathscr{V})_{L}}=\beta_{\mathscr{V}}^{L},-|_{U(\mathscr{V})_{L}}$
(11.3.a), we need only to prove that

$\alpha_{\mathscr{V}^{+}}(g)\beta_{\mathscr{V}^{+}}^{Y^{+}},(g)=\eta_{L}(g)\alpha_{\mathscr{V}^{-}}(g)\beta_{\mathscr{V}^{\prime}}^{Y^{-}}-(g)$
for $g\in U(\mathscr{V})_{L}$ . Hence this lemma follows from (11.2.b). $\square $
In fact, under the normalization (11.2.c), we know that $\beta_{\mathscr{V}^{+}}^{L},=\beta_{\mathscr{V}}^{L}$

,- .
11.4. Another version of theta dichotomy for -adic unitary groups. $p$
The following theorem is another form of Theorem 11.1.

THEOREM. Let be a reductive dual pair of unitary groups $(U(\mathscr{V}), U(\mathscr{V}^{\prime\pm}))$
and be a good lattice in

$L$
. Suppose that are the irreducible super- $\mathscr{V}$
$\pi^{+},$ $\pi^{-}$
cuspidal represen tations of such that sgn . Suppose $U(\mathscr{V})$

$\pi^{+}\circ\tilde{\beta}^{L}=$ $\otimes(\pi^{-}\circ\tilde{\beta}^{L})$
that the parity of the dimension of and the parity of the dimensions of are $\mathscr{V}$
the same. Then

$\ell_{0}^{+}+\ell_{0}^{-}=2n+2$
where is the dimension of $n$

is the smallest dimension $\mathscr{V},$ $\ell_{0}^{\pm}$
of $\mathscr{V}^{\prime\pm}$
such that $\pi^{\pm}$
occurs in the theta correspondence for the pair $(U(\mathscr{V}), U(\mathscr{V}^{\prime\pm}))$

.
By Lemma 11.3, we see that
PROOF. and $\pi^{+}\circ\tilde{\beta}_{\mathscr{V}^{+}}^{Y},=\sigma\otimes(\pi^{+}\circ\tilde{\beta}_{\mathscr{V}^{+}}^{L},)$
. Therefore , from the

$\pi^{-}\circ\tilde{\beta}_{\mathscr{V}^{-}}^{Y},=sgn\otimes\sigma\otimes$ $(\pi^{-}\circ\tilde{\beta}_{\mathscr{V}^{-}}^{L}, )$ $\pi^{+}\circ\tilde{\beta}_{\mathscr{V}^{+}}^{Y},=\pi^{-}\circ\tilde{\beta}_{\mathscr{V}^{-}}^{Y}$
assumption of the theorem. Therefore the theorem follows from Theorem 11.1
immediately. $\square $
12. Theta dichotomy for finite reductive dual pairs.

The goal of this section is to establish certain results of theta dichotomy for
finite reductive dual pairs from Theorems 9.3, 9.5 and 11.4. First we introduce
related finite reductive dual pairs.
12.1.
Let $f$ be a finite field of odd characteristic and be either $f$ itself or a $d$
quadratic extension of $f$ . Let be 1 or -1. We know that maximal dimension $\epsilon$
of an anisotropic -Hermitian space is one (resp. zero; two) when is a quadratic

$\epsilon$
$d$
extension of $f$ (resp. $-1);(d,$ ). Fix a nontrivial character

$(d,$ $\epsilon)=(f,$ $\epsilon)=(f,$ $1)$
of $f$ . We shall consider the following related finite reductive dual pairs
$\psi$
$(U(v), U(v^{\prime}))$ and $(U(v), U(v’))$ :

834 S.-Y. PAN
is a quadratic extension of and the parities of the dimensions of

(i) $d$ $f$
$v^{\prime}$
and are different. $v^{\prime\prime}$
(ii) is a symplectic space and are even-dimensional quadratic spaces

$v$ $v^{\prime},$
$v^{\prime\prime}$
with Witt indices equal to half of the dimension and half minus one of
the dimension respectively.
(iii) is a symplectic space and are two quadratic spaces defined
$v$ $v^{\prime},$
$v^{\prime\prime}$
as follows. Let be a one-dimensional quadratic space and be $v_{1}^{\prime}$ $v_{1}^{\prime\prime}$
another one-dimensional quadratic space whose quadratic form is a

multiple of the form on by a nonsquare element in . Then $v_{1}^{\prime}$ $f^{\times}$
$v^{\prime}$
(resp. ) is an odd dimensional quadratic space in the Witt towers of $v^{\prime\prime}$
$v_{1}^{\prime}$
(resp. $v_{1}^{\prime\prime}$
).
(iv) is a quadratic space and $v$ $v^{\prime},$
$v^{\prime\prime}$
are symplectic spaces.

As usual, we allow some of the spaces $v,$
$v^{\prime},$
$v^{\prime\prime}$
to be the trivial space {0}.

12.2. Trivial examples.
Although the examples that we are going to consider are trivial, they are
important for the proof of our theorem in the next subsection. Let be the $\zeta$
trivial representation of a finite classical group $U(v)$ . If is the trivial space and $v$
OX is a first occurrence for the pair

$\zeta$
$\zeta^{\prime}$
, then by convention is also $(U(v), U(v^{\prime}))$ $\zeta^{\prime}$
the trivial representation. We know that is a cuspidal representation if and $\zeta^{\prime}$
only if the Witt index of is zero. Another example is that $U(v)$ is an or- $v^{\prime}$
thogonal group of one variable and sgn OX for some is a first occurrence for $\zeta^{\prime}$ $\zeta^{\prime}$
the pair . Then we know that is a two-dimensional symplectic

$(U(v), U(v^{\prime}))$ $v^{\prime}$
space.
12.3.
THEOREM. Suppose that and are related reductive dual pairs in $(v, v^{\prime})$ $(v, v^{\prime\prime})$
subsection 12.2. Let be an irreducible cuspidal representation of $U(v)$ . Let $\zeta$ $\ell_{0}^{\prime}$
(resp. ) be the smallest dimension of (resp. ) such that

$\ell_{0}^{\prime\prime}$
(resp. OX $sgn$ ) $v^{\prime}$ $v^{\prime\prime}$

$\zeta$ $\zeta$
occurs in the theta correspondence for the dual pair $(U(v), U(v^{\prime}))$ (resp.
with respect to a character of $f$ . Then
$(U(v), U(v^{\prime\prime})))$ $\emptyset$
$2n+1$ , for case (i);

$\ell_{0}^{\prime}+\ell_{0}^{\prime\prime}=\{$
$2n+2$ , for case (ii) and (iii);
$2n$ , for case (iv),
where $n$
is the dimension of $v$
.
PROOF. Let $F$ be a -adic field such that the residue field is isomorphic $p$ $f_{F}$
to $f$ . Suppose that we are in the first case in subsection 12.1 and is even i.e., $n$ $v$
is an even dimensional -Hermitian space over a quadratic extension of a finite $\epsilon$

$d$
field $f$ . From the discussion in subsection 9.4, we know that there exists an n-
dimensional -Hermitian space over an unramified quadratic extension $E$ of

$\epsilon$
$\mathscr{V}$
$F$
such that
(1) the Witt index of is equal to the Witt index of , and $v$
$\mathscr{V}$
(2) there exists a good lattice in such that and is trivial i.e., $L$ $\mathscr{V}$ $l\simeq v$ $l^{*}$
.
$U(\mathscr{V})_{L}/U(\mathscr{V})_{L,0^{+}}\simeq U(v)$
Let denote the group $G$

. Let be the representation of inflated from $U(\mathscr{V})$
$\tilde{\zeta}$
$G_{L}$
. We know that the representation

$\zeta$
of is irreducible su- $c- Ind_{G_{L}}^{G}(\xi_{L}\otimes\tilde{\zeta})$ $G$
percuspidal. Therefore the representation of such that $\pi^{+}$ $\tilde{G}$

$\pi^{+}\circ\tilde{\beta}^{L}=c-$
( OX ) is irreducible supercuspidal from the definition in subsection 9.2.

$Ind_{G_{L}}^{G}$ $\xi_{L}$
$\tilde{\zeta}$
Let be the representation of

$\pi^{-}$
that . It is $\tilde{G}$
$s\underline{uch}$
$\pi^{+}\circ\tilde{\beta}^{L}=sg-n$ $\otimes(\pi^{-}\circ\tilde{\beta}^{L})$
clear that ( OX ) where OX sgn denotes the repre-

$\pi^{-}\circ\tilde{\beta}^{L}=c- Ind_{G_{L}}^{G}$ $\xi_{L}$ $(\zeta\otimes sgn)$ $\zeta$
sentation of inflated from OX sgn of $U(v)$ . Let $G_{L}$

(resp. ) be an $\zeta$
$\mathscr{V}^{\prime+}$ $\mathscr{V}^{\prime-}$ $\epsilon^{\prime}-$
Hermitian space of even dimension (resp. ) defined in (ll.l.a) such that $m^{+}$ $m^{-}$
(resp.
$\pi^{+}$
) first occurs in the theta correspondence for the dual pair
$\pi^{-}$
(resp.
$(U(\mathscr{V}), U(\mathscr{V}^{\prime+}))$
and is paired with the representation $(U(\mathscr{V}), U(\mathscr{V}^{\prime-}))$
(resp.
$\pi^{\prime+}$
) with respect to a nontrivial character of such that

$\pi^{\prime-}$
. By $\psi$ $F$ $\overline{\psi}=\phi$
Theorem 11.4 we know that $m^{+}+m^{-}=2n+2$ . By Theorem 9.5 we know that

must be the representation
$\pi^{\prime+}$
for some good lattice in $c- Ind_{G_{L}^{\prime+}+}^{G^{\prime+}},(\xi_{L^{\prime+}}\otimes\tilde{\zeta}^{\prime+})$

$L^{\prime+}$
where
$\mathscr{V}^{\prime+}$
. Moreover we know that the restriction of

$G^{\prime+}:=U(\mathscr{V}^{\prime+})$
to $\tilde{\zeta}^{\prime+}$
, is trivial. So
$G_{L^{+},0^{+}}^{\prime+}$
is inflated from a representation OX of
$\tilde{\zeta}^{\prime+}$
$\zeta_{1}^{\prime+}$ $\zeta_{2}^{\prime+}$ $ U(l^{\prime+})\times$
. We can regard
$U((l^{\prime+})^{*})$
OX triv as a representation of $U(l)\times U(l^{*})$ . By $\zeta\simeq\zeta$
Theorem 9.3, we know that OX and triv must be two first occurrences $\zeta$ $\zeta_{2}^{\prime+}$ $\otimes\zeta_{1}^{\prime+}$
in the theta correspondence for the pairs $(U(l), U((l^{\prime+})^{*}))$ and $(U(l^{*}), U(l^{\prime+}))$
respectively. Because is a split ’-Hermitian space, we know that both $\mathscr{V}^{\prime+}$

$\epsilon$
$l^{\prime+}$
and are even-dimensional. Hence

$(l^{\prime+})^{*}$
must be trivial from subsection 12.2. $l^{\prime+}$
By same argument, we know that

$\underline{the}$
is the representation $\pi^{\prime-}$
$c- Ind_{G_{L^{-}}^{\prime-}}^{G^{\prime-}},(\xi_{L^{\prime}}$ -
$\otimes$
where
$(\zeta_{1}^{\prime-}\otimes\zeta_{2}^{\prime-}))$
and is a good lattice in such that
$G^{\prime-}:=U(\mathscr{V}^{\prime-})$ $L^{\prime-}$ $\mathscr{V}^{\prime-}$
and
$G_{L^{\prime-}}^{\prime-}/G_{L^{-},0^{+}}^{\prime-},\simeq U(l^{\prime-})$
(resp. ) is a representation of
$\times U((l^{\prime-})^{*})$ $\zeta_{1}^{\prime-}$ $\zeta_{2}^{\prime-}$
(resp.
$U(l^{\prime-})$
). We can regard sgn $U((l^{\prime-})^{*})$
triv as a rep- $\zeta\otimes$ $\simeq(\zeta\otimes sgn)\otimes$
resentation of $U(l)\times U(l^{*})$ . Therefore by Theorem 9.3, we know that

and triv
$(\zeta\otimes sgn)\otimes\zeta_{2}^{\prime-}$
must be two first occurrences in theta corre- $\otimes\zeta_{1}^{\prime-}$
spondence for the pairs $(U(l), U((l^{\prime-})^{*}))$ and respectively. $(U(l^{*}), U(l^{\prime-}))$
Because is a non-split even dimensional -Hermitian space, both

$\mathscr{V}^{\prime-}$
and $\epsilon^{\prime}$ $l^{\prime-}$
are odd-dimensional. Hence

$(l^{\prime-})^{*}$
must be one-dimensional from sub- $l^{\prime-}$
section 12.2. Now the two finite reductive dual pairs $(U(v), U((l^{\prime+})^{*}))$ ,
$(U(v), U((l^{\prime-})^{*}))$ are the related dual pairs case (i) in subsection 12.1 because
the parities of dimensions of and of are different. Therefore $(l^{\prime+})^{*}$ $(l^{\prime-})^{*}$
. Hence the proof of the theorem is complete

$\ell_{0}^{\prime}+\ell_{0}^{\prime\prime}=m^{+}+m^{-}-1=2n+1$
for case (i) such that the dimension of is even. When the dimension of is $v$ $v$
odd, the proof is almost the same. So we skip it.

836 S.-Y. PAN
Next we consider cases (ii). There exists an -dimensional -Hermitian space $n$ $\epsilon$
over a ramified quadratic extension $E$ of $F$ satisfying the condition (1) and (2)
$\mathscr{V}$
above i.e., $\dim(v)=\dim(\mathscr{V})$ and is a symplectic space. The proof for the $v\simeq l$
first case can be applied to this case. Now is a split orthogonal group of $(l^{\prime+})^{*}$
even number of variables, is a non-split orthogonal group of even number $(l^{\prime-})^{*}$
of variables and both are trivial from subsection 12.2. Hence $l^{\prime+},$ $l^{\prime-}$
$\ell_{0}^{\prime}+\ell_{0}^{\prime\prime}=$
$m^{+}+m^{-}=2n+2$ .
For case (iii), let be an $(n+1)$ -dimensional -Hermitian space over a $\mathscr{V}$
$\epsilon$
ramified quadratic extension $E$ of $F$ and be a good lattice in such that $L$ $\mathscr{V}$ $l\simeq v$
and is a one-dimensional quadratic space over $f_{E}=f_{F}$ . It is clear that

$l^{*}$
triv is an irreducible
$\zeta\otimes$
representation of $U(l)\times U(l^{*})$ and the induced $cuspid\underline{a1}$
representation ( OX ( OX triv)) is an irreducible supercuspidal represen-

$c$ - $Ind_{G_{L}}^{G}$ $\xi_{L}$ $\zeta$
tation of . Let be the spaces as above but they are assumed to be

$G$ $\mathscr{V}^{\prime+},$
odd-dimensional. By Theorem 11.4, we have $m^{+}\underline{+m}^{-}=2(n+1)+2=2n+4$ .

we should notice that (
$No\underline{w}$ ( OX OX triv)) OX $sgn=c- Ind_{G_{L}}^{G}(\xi_{L}$ OX $c- Ind_{G_{L}}^{G}$ $\xi_{L}$ $\zeta$
( OX $sgn$ ) . Similar to the previous two cases, we know that

$\zeta$ $)$ $U((l^{\prime+})^{*}),$ $U((l^{\prime-})^{*})$
are two orthogonal groups in odd variables and they satisfy the condition in (iii)
of subsection 12.1. We also know that and are two first $triv\otimes\zeta_{1}^{\prime+}$ $sgn\otimes\zeta_{1}^{\prime-}$
occurrences for the pairs $(U(l^{*}), U(l^{\prime+})),$ $(U(l^{*}), U(l^{\prime-}))$ respectively. Hence $l^{\prime+}$
is trivial and is two-dimensional. Thus $\ell_{0}^{\prime}+\ell_{0}^{\prime\prime}=m^{+}+m^{-}-2=2n+2$ .

$l^{\prime-}$
For case (iv), let be an -dimensional -Hermitian space over a ramified $\mathscr{V}$
$n$ $\epsilon$
quadratic extension of and be a good lattice in

$E$ $F$
such that and is $L$ $\mathscr{V}$ $l\simeq v$ $l^{*}$
trivial. Let be as given in previous cases. So we have

$\mathscr{V}^{\prime+},$
$m^{+}+m^{-}=$
$2n+2$ . Similar to the case (iii), we know that , are $U((l^{\prime+})^{*})$ $U((l^{\prime-})^{*})$
two symplectic groups. We also know that both are odd-dimensional if $l^{\prime+},$ $l^{\prime-}$
is odd-dimensional,
$\mathscr{V}$
is split even-dimensional and is non-split even- $l^{\prime+}$ $l^{\prime-}$
dimensional if is even-dimensional. Since $\mathscr{V}$
is trivial, we know that either $l^{*}$
both are one-dimensional, or

$l^{\prime+},$ $l^{\prime-}$
is trivial and is two-dimensional. $l^{\prime+}$ $l^{\prime-}$
Therefore $\ell_{0}^{\prime}+\ell_{0}^{\prime\prime}=m^{+}+m^{-}-2=2n+2-2=2n$ . $[$
12.4.
REMARK. is a unipotent cuspidal representation and we have
In fact, if $\zeta$
some restriction on the characteristic of , Theorem 12.3 can be figured out from $f$
[AM] theorem 4.1 and theorem 5.2 as follows.

(i) Suppose is an irreducible unipotent cuspidal representation of a unitary $\zeta$
group $U(v)$ . Let be the dimension of . Then $n=i(i+1)/2$ $n$ $v$
for some integer by G. Lusztig’s computation. By [AM] theorem $i$
4.1, is $\{(i-1)i/2, (i+1)(i+2)/2\}$ . Then $\{\ell_{0}^{\prime}, \ell_{0}^{\prime\prime}\}$
$\ell_{0}^{\prime}+\ell_{0}^{\prime\prime}=\frac{(i-1)i}{2}+\frac{(i+1)(i+2)}{2}=i^{2}+i+1=2n+1$
as we expect. Here we should notice that the Weil representation of

the finite symplectic group we use here is a little different from the Weil
representation used in [AM]. Therefore the theta correspondence for
the dual pairs of unitary groups is different by the sgn character (cf.
[AMR] introduction).
(ii) Similarly, if is an irreducible unipotent cuspidal representation of a
$\zeta$
symplectic group $U(v)$ , then $n=2i(i+1)$ for some . If the char- $i$
acteristic of is large enough and are even dimensional, from

$f$ $v^{\prime},$
$v^{\prime\prime}$
[AM] theorem 5.2, is $\{2i^{2},2(i+1)^{2}\}$ . Hence $\{\ell_{0}^{+}, \ell_{0}^{-}\}$
$\ell_{0}^{+}+\ell_{0}^{-}=2i^{2}+2(i+1)^{2}=4i(i+1)+2=2n+2$ .
(iii) If is an irreducible unipotent cuspidal representation of an or-

$\zeta$
thogonal group $U(v)$ , then $n=2i^{2}$ for some . Suppose that the $i$
characteristic of $f$ is large enough. Then from [AM] theorem 5.2,

is $\{2i(i-1), 2i(i+1)\}$ . Hence
$\{\ell_{0}^{+}, \ell_{0}^{-}\}$
$\ell_{0}^{+}+\ell_{0}^{-}=2i(i-1)+2i(i+1)=4i^{2}=2n$ .
If we accept the fact that an irreducible unipotent representation is paired

with an irreducible unipotent representation (under some restriction for the
symplectic-orthogonal pairs), then Theorem 12.3 provides another (independent)
proofs of theorem 4.1 and theorem 5.2 in [AM].
12.5.
REMARK. As in the remark after Theorem 11.1 for -adic unitary groups, $p$
Theorem 12.3 suggests that there exist a sequence of dimensions $n=n_{0}<n_{1}<$

$ n_{2}<\cdots<n_{i}<\cdots$ and irreducible cuspidal representations $\zeta=\zeta_{0},$ $\zeta_{1},$ $\zeta_{2},$
$\ldots,$
$\zeta_{i},$
$\ldots$
of $U(v_{i})$ such that occurs in the theta correspondence for the

$(\zeta_{i}\otimes sgn)\otimes\zeta_{i+1}$
dual pair $(U(v_{i}), U(v_{i+1}))$ . Suppose that does not come from a smaller group $\zeta$
via the theta correspondence. We have the following cases.

(i) Suppose that is a unitary group. So all are unitary
$U(v_{0})$ $U(v_{i})$
groups and the sequence of dimensions are
$n,$ $n+1,$ $n+3,$ $n+6,$ $n+10,$ $\ldots,$

$n+\frac{i(i+1)}{2},$
$\ldots$ . (12.5.a)
If $n=0$ , then is the trivial character and these dimensions are exactly
$\zeta$
the dimensions of where the finite unitary groups $U(v)$ contain $v$
unipotent cuspidal representations and are the only irreducible $\zeta_{i}$
unipotent cuspidal representations of the finite unitary groups.

(ii) Suppose that is an orthogonal group and is even. Then
$U(v_{0})$ $n$ $U(v_{i})$
are orthogonal groups and symplectic groups alternatively. The se-

838 S.-Y. PAN
quence of dimensions are

$n,$ $n,$ $n+2,$ $n+4,$ $n+8,$ $n+12$ ,
$n+18,$ $\ldots,$
$n+\frac{i(i+1)}{2}-\lceil\frac{i}{2}\rceil,$
$\ldots$ . (12.5.b)
If $n=0$ , these are exactly the dimensions of where the finite even $v$
orthogonal groups or symplectic groups $U(v)$ have unipotent cuspidal

representations. Moreover are the only irreducible unipotent cus-
$\zeta_{i}$
pidal representations of finite symplectic groups and finite even or-

thogonal groups. Suppose that is a symplectic group. Then $U(v_{0})$
the sequence of dimensions are

$n,$ $n+2,$ $n+4,$ $n+8,$ $n+12,$ $n+18$ ,
$n+24,$ $\ldots,$
$n+\frac{i(i+1)}{2}+\lceil\frac{i}{2}\rceil,$
$\ldots$ . (12.5.c)
This sequence is just the sequence in (12.5.b) with the index shifted by
one.
(iii) Suppose that $U(v_{0})$ is symplectic group and the other group is an odd
orthogonal group. Then by Theorem 12.3, the sequence of dimensions
are
$n,$ $n+1,$ $n+2,$ $n+5,$ $n+8$ ,
$n+13,$ $\ldots,$
$n+\frac{i(i+1)}{2}-\lfloor\frac{i}{2}\rfloor,$
$\ldots$ . (12.5.d)
For $n=0$ and even, these terms are exactly the dimensions of where
$i$
$v$
the finite odd orthogonal groups $U(v)$ have unipotent cuspidal rep-
resentations. As we know in [AM], the unipotent representations are
not preserved by the theta correspondence for the dual pairs of finite
symplectic groups and odd orthogonal groups. However, according to
the sequence of dimensions here, it should be reasonable to believe that
those representations of odd orthogonal groups in the chain should be
all unipotent cuspidal when $n=0$ in (12.5.d) although those repre-
sentations of symplectic groups are not unipotent. Suppose that $U(v_{0})$
is an orthogonal group and is odd. Then the sequence of di-

$n$
mensions are
$n,$ $n+1,$ $n+4,$ $n+7,$ $n+12$ ,
$n+17,$ $\ldots,$
$n+\frac{i(i+1)}{2}+\lfloor\frac{i}{2}\rfloor,$
$\ldots$ . (12.5.e)
12.6.
REMARK. It might be interesting to compare the chains of irreducible depth
zero supercuspidal representations of -adic unitary groups described in sub- $p$
section 11.1 with the chains of cuspidal representations of finite classical groups
described in subsection 12.5. However, here we will regard the representation $\pi_{i}$
in subsection 11.1 as a representation of instead of . Supp e that $U(\mathscr{V}_{i})$ $U(\mathscr{V}_{i})$ $os$
is an irreducible depth zero supercuspidal representation of

$\pi_{0}$
and $\tilde{G}:=U(\mathscr{V}_{0})$
does not come from a smaller unitary group via the theta correspondence. Let
$\pi_{0},$ be the chain of irreducible supercuspidal representations as
$\pi_{1},$
$\ldots,$
$\pi_{i},$
$\ldots$
in subsection 11.1. Since each is of depth zero, we know that there are a $\pi_{i}$
good lattice in 1 and a cuspidal representation of

$L_{i}$ $\zeta_{i}^{\prime}\otimes\zeta_{i}^{\prime\prime}$
$ G_{L_{i}}/G_{L_{i},0}+\simeq$
such that
$U(l_{i})\times U(l_{i}^{*})$ ( OX ) where is the representation $\pi_{i}\circ\tilde{\beta}^{L_{i}}=c- Ind_{G_{L_{i}}}^{G}$ $\xi_{L_{i}}$ $\zeta_{i}$ $\zeta_{i}$
of inflated from the representation

$G_{L_{i}}$
of . It is clear that $\zeta_{i}^{\prime}\otimes\zeta_{i}^{\prime\prime}$
$G_{L_{i}}/G_{L_{i},0}+$
the cuspidal representation (resp. ) of (resp. ) does not come $\zeta_{0}^{\prime}$ $\zeta_{0}^{\prime\prime}$
$U(l_{0})$ $U(l_{0}^{*})$
from a smaller group via the theta correspondence. Moreover we know that
$\zeta_{0}^{\prime},$
(resp.
$\zeta_{1}^{\prime},$
,...,
$\ldots,$ . . .) is the chain of cuspidal representations as
$\zeta_{i}^{\prime},$
$\ldots$
$\zeta_{0}^{\prime\prime},$ $\zeta_{1}^{\prime\prime}$ $\zeta_{i}^{\prime\prime},$
in subsection 12.5(i) starting from (resp. ). Of course, we must have $\zeta_{0}^{\prime}$ $\zeta_{0}^{\prime\prime}$
$\dim(\mathscr{V}_{i})=\dim(l_{i})+\dim(l_{i}^{*})$ . This can be easily verified according the following

two cases when the the dimension of and the the dimension of are of the $\mathscr{V}_{0}$ $\mathscr{V}_{0}^{\prime}$
same parity.
(i) Suppose that $E$ is an unramified quadratic extension of $F$. Then $U(l_{i})$
and are all finite unitary group. Thus $\dim(l_{i})+\dim(l_{i}^{*})=$

$U(l_{i}^{*})$
$\dim(l_{0})+i(i+1)/2+\dim(l_{0}^{*})+i(i+1)/2$ from (12.5.a). On the other

hand, $=\dim(\mathscr{V}_{0})+i(i+1)$ from 11.1. . Since
$\dim(\mathscr{V}_{i})$ $($
$b)$ $\dim(\mathscr{V}_{0})=$
$\dim(l_{0})+\dim(l_{0}^{*})$ , hence $\dim(\mathscr{V}_{i})=\dim(l_{i})+\dim(l_{i}^{*})$ as we expect.
(ii) Suppose that $E$ is a ramified quadratic extension of $F$. Then one
of is a finite orthogonal group and the other is a finite
$U(l_{i}),$ $U(l_{i}^{*})$
symplectic group. If the dimension of is even, then we have $\mathscr{V}_{0}$
$\dim(l_{i})+\dim(l_{i}^{*})=\dim(l_{0})+i(i+1)/2-\lceil i/2\rceil+\dim(l_{0}^{*})$ $+i(i+1)/2+$

from (12.5.b) and (12.5.c). If the dimension of
$\lceil i/2\rceil$
is odd, then $\mathscr{V}_{0}$
we have $\dim(l_{i})+\dim(l_{i}^{*})=\dim(l_{0})+i(i+1)/2-\lfloor i/2\rfloor+\dim(l_{0}^{*})+$

$ i(i+1)/2+\lfloor i/2\rfloor$ from (12.5.d) and 12.5. . In any cases, we have $($ $e)$
$\dim(\mathscr{V}_{i})=\dim(l_{i})+\dim(l_{i}^{*})$ as we expect again.
13. Theta dichotomy for -adic reductive dual pairs. $p$
An interesting phenomenon is that the theta dichotomy for irreducible depth

zero supercuspidal representations of -adic reductive dual pairs can be obtained $p$
from the theta dichotomy for irreducible cuspidal representations of finite re-
ductive dual pairs via Theorems 9.3, 9.5 and 12.3. This is the subject of this
section.
840 S.-Y. PAN
13.1.
First we consider theta dichotomy for reductive dual pairs $(U(\mathscr{V}), U(\mathscr{V}^{\prime\pm}))$
of unitary groups when the parity of the dimension of and the parity of the $\mathscr{V}$
dimension of are different. $\mathscr{V}^{\prime\pm}$
THEOREM. Let be a reductive dual pair of unitary groups. $(U(\mathscr{V}), U(\mathscr{V}^{\prime\pm}))$
Suppose that the parity

of the dimension of and the parity of the dimension of $\mathscr{V}$
are different.
that is an irreducible depth zero supercuspidal $\underline{S}uppose$

$\pi^{+}$
representations of with a minimal $K$-type ). Let be the irreducible $U(\mathscr{V})$ $\underline{(G_{L}},$

$\zeta$ $\pi^{-}$
depth zero supercuspidal representations of such that $U(\mathscr{V})$ $\pi^{+}\circ\tilde{\beta}^{L}=sgn\otimes$
. Then
$(\pi^{-}\circ\tilde{\beta}^{L})$
$\ell_{0}^{+}+\ell_{0}^{-}=2n+2$
where is the dimension of and is the smallest dimension

$n$
$\mathscr{V}$ $\ell_{0}^{\pm}$
of $\mathscr{V}^{\prime\pm}$
such that
occurs in the theta correspondence for the pair
$\pi^{\pm}$
.
PROOF. Let $G:=U(\mathscr{V})$ and . Because $G^{\prime\pm}:=U(\mathscr{V}^{\prime\pm})$ $\pi^{+}\circ\tilde{\beta}^{L}=sgn\otimes$
and has a minimal $K$-type

$(\pi^{-}\circ\tilde{\beta}^{L})$ $\pi^{+}$
, it is clear that has a minimal $(G_{L}, \zeta)$ $\pi^{-}$
$K$-type ( , sgn ). Because$G_{L}$

is an irreducible cuspidal representation of
$\otimes\zeta$ $\zeta$
$G_{L}/G_{L,0}+\simeq U(l)\times U(l^{*})$ , we can write where (resp. ) is an $\zeta\simeq\zeta_{1}\otimes\zeta_{2}$ $\zeta_{1}$ $\zeta_{2}$
irreducible cuspidal representation of $U(l)$ (resp. ). Hence clearly, sgn OX $U(l^{*})$ $\zeta$
is isomorphic to OX (sgn ) where the first sgn is a character of

$(sgn\otimes\zeta_{1})$ $\otimes\zeta_{2}$
$U(l)\times U(l^{*})$ , the second sgn is a character of $U(l)$ and the third one is a
character of . $U(l^{*})$
Suppose that . ) corresponds to an irreducible supercuspidal

$\pi^{+}$
$\underline{(res}p$ $\pi^{-}$
representation of (resp. of
$\pi^{\prime+}$
) in the theta correspondence

$U(\mathscr{V}^{\prime+})$ $\pi^{\prime-}$ $U(\overline{\mathscr{V}^{\prime-}})$
for the reductive dual pair (resp. ). Then by $(U(\mathscr{V}), U(\mathscr{V}^{\prime+}))$ $(U(\mathscr{V}),$ $U(\mathscr{V}^{\prime-}))$
Theorem 9.3, we know that has a minimal $K$-type such that $\pi^{\prime+}$
$(G_{L^{+}}^{\prime+},, \zeta^{\prime+})$
OX
$\zeta$
is a first occurrence. This means that

$\zeta^{\prime+}$
where (resp. $\zeta^{\prime+}\simeq\zeta_{1}^{\prime+}\otimes\zeta_{2}^{\prime+}$ $\zeta_{1}^{\prime+}$
is an irreducible cuspidal representation of

$\zeta_{2}^{\prime+})$
(resp. ), and $U(l^{\prime+})$ $U((l^{\prime+})^{*})$
(resp.
$\zeta_{1}\otimes\zeta_{2}^{\prime+}$
) is a first occurrence for the finite dual pair
$\zeta_{2}\otimes\zeta_{1}^{\prime+}$
$(U(l), U((l^{\prime+})^{*}))$ (resp. $(U(l^{*}),$ ). By the same reason, we also know that $U(l^{\prime+}))$
has a minimal $K$-type

$\pi^{\prime-}$
such that (resp. ) is an ir- $(G_{L^{\prime-}}^{\prime-}, \zeta_{1}^{\prime-}\otimes\zeta_{2}^{\prime-})$ $\zeta_{1}^{\prime-}$ $\zeta_{2}^{\prime-}$
reducible cuspidal representation of (resp. ), and (sgn ) $U(l^{\prime-})$ $U((l^{\prime-})^{*})$ $\otimes\zeta_{1}$ $\otimes\zeta_{2}^{\prime-}$
(resp. (sgn ) is a first occurrence for the finite dual pair

$\otimes\zeta_{2})\otimes\zeta_{1}^{\prime-}$
$(U(l), U((l^{\prime-})^{*}))$ (resp. $(U(l^{*}),$ ). $U(l^{\prime-}))$
If $E$
is an unramified quadratic extension of $F$, the related reductive dual
pairs
$\{(U(l), U((l^{\prime+})^{*})), (U(l^{*}), U(l^{\prime+}))\}$ and $\{(U(l), U((l^{\prime-})^{*})), (U(l^{*}), U(l^{\prime-}))\}$
(13.1.a)
are case (i) of subsection 12.1. If is a ramified quadratic extension of , then $E$ $F$
one of the related reductive dual pairs in (13.1.a) is case (iv) of subsection
12.1 and the other is either case (ii) or case (iii) depending the parity of the
’
dimension of . It clear that $n=\dim(l)+\dim(l^{*}),$ $\mathscr{V}$
$\ell_{0}^{+}=\dim(l^{\prime+})+\dim((l^{\prime+})$ $)$
and $\ell;=\dim(l^{\prime-})+\dim((l^{\prime-})^{*})$ . Therefore by Theorem 12.3, we have

2 $\dim(l)+1+2\dim(l^{*})+1$ , if is an unramified quadratic $E$
$\ell_{0}^{+}+\ell_{0}^{-}=\{$ extension;
2 $\dim(l)+2\dim(l^{*})+2$ , if is a ramified quadratic extension $E$
$=2n+2$ . $\square $
We have the following analogue of the result in subsection 12.6. $\underline{Su}ppose$
that is an irreducible depth zero supercuspidal representation of

$\pi_{0}$
and $U(\mathscr{V}_{0})$
does not come from a smaller unitary group via the theta correspondence. Let
$\mathscr{V}_{i},$
be as in subsection 12.6. Now we have the following two
$L_{i},$ $l_{i},$ $l_{i}^{*},$ $\zeta_{i},$ $\zeta_{i}^{\prime},$ $\zeta_{i}^{\prime\prime}$
cases.
(i) Suppose that $E$ is an unramified quadratic extension of $F$. Then $U(l_{i})$
and are all finite unitary group. Because now the dimensions of
$U(l_{i}^{*})$
are of opposite parity, we know that

$\mathscr{V}_{0},$ $\mathscr{V}_{1}$
. $\dim(\mathscr{V}_{1})=\dim(\mathscr{V}_{0})+1$
Therefore either $\dim(l_{0})=\dim(l_{1}^{*})$ or $\dim(l_{0}^{*})=\dim(l_{1})$ . If $\dim(l_{0})=$

, then
$\dim(l_{1}^{*})$ (resp. , . . . , , . . .) is the chain of $\zeta_{1}^{\prime},$ $\zeta_{2}^{\prime},$
$\ldots,$
$\ldots$
$\zeta_{0}^{\prime\prime},$ $\zeta_{1}^{\prime\prime}$ $\zeta_{i}^{\prime\prime}$
cuspidal representations as in subsection 12.5(i) starting from (resp. $\zeta_{1}^{\prime}$
. If $\dim(l_{0}^{*})=\dim(l_{1})$ , then
$\zeta_{0}^{\prime\prime})$
(resp. , . . . , , . . .) $\zeta_{0}^{\prime},$ $\zeta_{1}^{\prime},$
$\ldots,$
$\ldots$
$\zeta_{1}^{\prime\prime},$ $\zeta_{2}^{\prime\prime}$ $\zeta_{i}^{\prime\prime}$
is the chain of cuspidal representations as in subsection 12.5(i) starting

from (resp. ). In any case, we have $\dim(l_{i})+\dim(l_{i}^{*})=\dim(l_{0})+$
$\zeta_{0}^{\prime}$ $\zeta_{1}^{\prime\prime}$
$i(i+1)/2+\dim(l_{0}^{*})+(i-1)i/2$ from (12.5.a). Hence $\dim(\mathscr{V}_{i})=$
as we expect.
$\dim(\mathscr{V}_{0})+i^{2}$
(ii) Suppose that $E$ is a ramified quadratic extension of $F$. Then one of
$U(l_{i}),$ is a finite orthogonal group and the other is a finite
$U(l_{i}^{*})$
symplectic group. Define

$l_{i}$
, if $i$
is even; $l_{i}^{*}$
, if $i$
is even;
$l_{i}^{\prime}=\{$ $l_{i}’’=\{$
$l_{i}^{*}$
, if $i$
is odd, $l_{i}$
, if $i$
is odd.
Suppose that is a finite symplectic group and is a finite $U(l_{0})$ $U(l_{0}^{*})$
orthogonal group.
If
$(ii.a)$ is even-dimensional, then we have $\dim(l_{1}^{\prime})=\dim(l_{0})+1$ $\mathscr{V}_{0}$
and . More generally we have $\dim(l_{1}^{\prime\prime})=\dim(l_{0}^{*})$ $\dim(l_{i}^{\prime})=$
$\dim(l_{0})+i(i+1)/2-\lfloor i/2\rfloor$ from (12.5.d) and $\dim(l_{i}^{\prime\prime})=$
$\dim(l_{0}^{*})+i(i+1)/2-\lceil i/2\rceil$ from 12.5. . Hence $($

$b)$ $\dim(\mathscr{V}_{i})=$
$\dim(l_{i}^{\prime})+\dim(l_{i}^{\prime\prime})=\dim(l_{0})+\dim(l_{0}^{*})+i^{2}=\dim(\mathscr{V}_{0})+i^{2}$ as we
expect.
842 S.-Y. PAN
$(ii.b)$ If $\mathscr{V}_{0}$
is odd-dimensional, then we have $\dim(l_{1}^{\prime} )$ $=\dim(l_{0})$
and More generally we have

$\dim(l_{1}^{\prime\prime})=\dim(l_{0}^{*})+1$
. $\dim(l_{i}^{\prime})=$
$\dim(l_{0})+i(i-1)/2+\lceil(i-1)/2\rceil$ from (12.5.c) and $\dim(l_{i}^{\prime\prime})=$
$\dim(l_{0}^{*})+i(i+1)/2+\lfloor i/2\rfloor$ from (12.5.e). Hence $\dim(\mathscr{V}_{i})=$
$\dim(l_{i}^{\prime})+\dim(l_{i}^{\prime\prime})=\dim(l_{0})+\dim(l_{0}^{*})+i^{2}=\dim(\mathscr{V}_{0})+i^{2}$ as we
expect again.
13.2.
In this section, we discuss theta dichotomy for -adic reductive dual pairs $p$
of symplectic and orthogonal groups. Because we only consider split reductive

dual pairs right now, we shall assume that the dimensions of the quadratic spaces
are even. We have the following three types of related reductive dual pairs
and
of symplectic groups and orthogonal $(U(\mathscr{V}), U(\mathscr{V}^{\prime\prime}))$
groups.
(i) is a symplectic space and (resp.
$\mathscr{V}$
) is an even-dimensional $\mathscr{V}^{\prime}$ $\mathscr{V}^{\prime\prime}$
quadratic space whose Witt index is half (resp. half minus two) of its
dimension.
(ii) is a symplectic space, and is the quadratic spaces as follows.
$\mathscr{V}$
$\mathscr{V}^{\prime}$ $\mathscr{V}^{\prime\prime}$
Let be a two-dimensional anisotropic quadratic space. Then we

$\mathscr{V}_{0}^{\prime}$
can associate a quadratic extension of to . Let be $F(\sqrt{\Delta})$ $F$ $\mathscr{V}_{0}^{\prime}$ $\mathscr{V}_{0}^{\prime\prime}$
another two-dimensional anisotropic quadratic space with the quadratic

form equal to the multiple of the form on by an element $a\in F$ ’ $\mathscr{V}_{0}^{\prime}$
such that $(\Delta, a)_{F}=-1$ . Let and be the quadratic spaces in $\mathscr{V}^{\prime}$ $\mathscr{V}^{\prime\prime}$
the Witt towers of and respectively. We notice that the Witt $\mathscr{V}_{0}^{\prime}$ $\mathscr{V}_{0}^{\prime\prime}$
indices of and are half minus one of their dimensions. $\mathscr{V}^{\prime}$ $\mathscr{V}^{\prime\prime}$
(iii) is an even-dimensional quadratic space and

$\mathscr{V}$
is a symplectic $\mathscr{V}^{\prime}$
space.
13.3.
THEOREM. Suppose that and are one of the $(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$ $(U(\mathscr{V}), U(\mathscr{V}^{\prime\prime}))$
above three types of related reductive dual . Suppose that is an irreducible $p\underline{air}s$ $\pi$
depth zero supercuspidal representation of having a minimal pe . $U(\mathscr{V})$ $\underline{K}$

$(G_{L}, \zeta)$
Let be the irreducible depth zero supercuspidal representation of

$\pi^{s}$
such that $U(\mathscr{V})$
sgn . Let
$\pi^{s}\circ\tilde{\beta}^{L}=$
(resp. ) be the smallest dimension of (resp.
$\otimes(\pi\circ\tilde{\beta}^{L})$ $\ell_{0}^{\prime}$ $\ell_{0}^{\prime\prime}$
such that (resp.

$\mathscr{V}^{\prime\prime})$
) occurs in the theta correspondence for the pair $\pi$

$\pi^{s}$
. Then
$2n+4$ , for cases (i), (ii);

(13.3. a) $\ell_{0}^{\prime}+\ell_{0}^{\prime\prime}=\{$
$2n$ ,
for cases (iii),
where $n$
is the dimension of $\mathscr{V}$
.
PROOF. Let $G:=U(\mathscr{V}),$ $G^{\prime}:=U(\mathscr{V}^{\prime})$
and $G^{\prime\prime}:=U(\mathscr{V}^{\prime\prime})$
. First we con-
sider cases (i) and (ii) in subsection 13.2. We know that is isomorphic to $\pi\circ\tilde{\beta}^{L}$
where is the representation of

$c- Ind_{G_{L}}^{G}\tilde{\zeta}$
inflated from the representation $\tilde{\zeta}$
$G_{L}$ $\zeta$
of $G_{L}/G_{L,0}+$ . We can write where (resp. ) is an irreducible $\zeta=\zeta_{1}\otimes\zeta_{2}$ $\zeta_{1}$ $\zeta_{2}$
cuspidal representation of $U(l)$ (resp. ). Suppose that (resp. ) cor- $U(l^{*})$ $\underline{\pi}$
$\pi^{s}$
’
re p nds to the irreducible supercuspidal representation of
$s$ $o$ (resp. of $\pi^{\prime}$
$U(\mathscr{V}$ $)$
$\pi^{\prime\prime}$
in the theta correspondence for the dual pair

$U(\mathscr{V}^{\prime\prime}))$
(resp. $(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$
. By Theorem 9.3, we know that

$(U(\mathscr{V}), U(\mathscr{V}^{\prime\prime})))$
must have a minimal K- $\pi^{\prime}$
type for some good lattice in where (resp. ) is an

$(G_{L}^{\prime},, \zeta_{1}^{\prime}\otimes\zeta_{2}^{\prime})$
$\zeta_{1}^{\prime}$ $\zeta_{2}^{\prime}$
irreducible cuspidal representation of (resp. ) such that both , $U(l^{\prime})$ $U(l^{\prime*})$ $\zeta_{1}\otimes\zeta_{2}^{\prime}$
are first occurrences. By the same reason,

$\zeta_{2}\otimes\zeta_{1}^{\prime}$
must have a minimal K- $\pi^{\prime\prime}$
type ( OX ) for some good lattice

$G_{L^{\prime}}^{\prime\prime},,$
in where (resp.
$\zeta_{1}^{\prime\prime}$
) is an $\zeta_{2}^{\prime\prime}$
$L^{\prime\prime}$ $\mathscr{V}^{\prime\prime}$
$\zeta_{1}^{\prime\prime}$ $\zeta_{2}^{\prime\prime}$
irreducible cuspidal representation of (resp. ) such that both $U(l^{\prime\prime})$ $U(l^{\prime\prime*})$
OX
$(sgn\otimes\zeta_{1})$
( $sgn$ OX ) OX are first occurrences. If we are in case (i) or in $\zeta_{2}^{\prime},$ $\zeta_{2}$ $\zeta_{1}^{\prime}$
case (ii) with unramified, then the related dual pairs $(U(l), U(l^{\prime*}))$ and $F(\sqrt{\Delta})$
are case (iii) in subsection 12.1. Similarly, the dual pairs

$(U(l), U(l^{\prime\prime*}))$
and
$(U(l^{*}), U(l^{\prime}))$
are also case (iii) in subsection 12.1. If we are $(U(l^{*}), U(l^{\prime\prime}))$
in case (ii) and is a ramified quadratic extension of , then the related $F(\sqrt{\Delta})$ $F$
dual pairs $(U(l), U(l^{\prime*}))$ and are case (ii) in subsection 12.1. $(U(l), U(l^{\prime\prime*}))$
Similarly, the dual pairs and are also case (ii) in $(U(l^{*}), U(l^{\prime}))$ $(U(l^{*}), U(l^{\prime\prime}))$
subsection 12.1. Therefore by Theorem 12.3, we have

$\ell_{0}^{\prime}+\ell_{0}^{\prime\prime}=\dim(l^{\prime})+\dim(l^{\prime*})+\dim(l^{\prime\prime})+\dim(l^{\prime\prime*})=2\dim(l)+2+2\dim(l^{*})+2$
$=2n+4$ .
Hence the proof is complete for cases (i) and (ii). The proof of case (iii) is
similar. $\square $
It might be reasonable to believe that the theorem should also be true for
any irreducible supercuspidal representations or even for any irreducible ad-
missible representations. In fact, it has been conjectured by S. Kudla and D.
Prasad (cf. [Rb]) that (12.5.a) for type (iii) should be valid for any irreducible
admissible representations of . $U(\mathscr{V})$
13.4.
Theorem 13.3 suggests that there exist a sequence of dimensions
REMARK.
$ n=n_{0}<n_{1}<n_{2}<\cdots<n_{i}<\cdots$ and irreducible depth zero supercuspidal rep-
resentations of such that OX occurs in the $\pi=\pi_{0},$ $\pi_{1},$ $\pi_{2},$

$\ldots,$
$\pi_{i},$
$\ldots$
$U(\mathscr{V}_{i})$ $\pi_{i}^{s}$
$\pi_{i+1}$
theta correspondence for the pair where are symplectic $(U(\mathscr{V}_{i}), U(\mathscr{V}_{i+1} ))$ $U(\mathscr{V}_{i})$
and orthogonal groups alternatively and is the representation such that $\pi_{i}^{s}$
sgn . Suppose that

$\pi_{i}^{s}\circ\tilde{\beta}^{L_{i}}=$
does not come from a smaller group $\otimes(\pi_{i}\circ\tilde{\beta}^{L_{i}})$
$\pi_{0}$
via the theta correspondence and is an orthogonal group. The sequence $U(\mathscr{V}_{0})$
of the dimensions are

844 S.-Y. PAN
$n,$ $n,$ $n+4,$ $n+8,$ $n+16,$ $n+24,$ $\ldots,$

$n+i(i+1)+(-1)^{i}2\lceil i/2\rceil,$ $\ldots$ . (13.4.a)
We can write for some good lattice in where is $\pi_{0}\circ\tilde{\beta}^{L_{0}}=c- Ind_{G_{L_{0}}}^{G}\zeta_{0}$ $L_{0}$ $\mathscr{V}_{0}$ $\zeta_{0}$
the representation of inflated from the cuspidal representation OX of $G_{L_{0}}$ $\zeta_{0}^{\prime}$ $\zeta_{0}^{\prime\prime}$
. Then it is clear that

$G_{L_{0}}/G_{L_{0},0}+\simeq U(l_{0})\times U(l_{0}^{*})$and are cuspidal rep- $\zeta_{0}^{\prime}$ $\zeta_{0}^{\prime\prime}$
resentations of finite classical groups and do not come from smaller groups via
the theta correspondence. Then where is the represen- $\pi_{i}\circ\tilde{\beta}^{L_{i}}=c- Ind_{G_{L_{i}}}^{G_{i}}\zeta_{i}$ $\zeta_{i}$
tation of inflated from where (resp. ) is the -th cuspidal

$G_{L_{i}}$ $\zeta_{i}^{\prime}\otimes\zeta_{i}^{\prime\prime}$ $\zeta_{i}^{\prime}$ $\zeta_{i}^{\prime\prime}$ $i$
representation in the chain in subsection 12.5(ii) or 12.5(iii) starting from (resp. $\zeta_{0}^{\prime}$
.
$\zeta_{0}^{\prime\prime})$
13.5.
An immediate consequence of Theorem 13.3 is the following.
COROLLARY. Suppose that is a $2n$ -dimensional symplectic space. All $\mathscr{V}$
irreducible depth zero supercuspidal representations of occur in the local $Sp(\mathscr{V})$
theta correspondences with some where are $2n+2$ dimensional quadratic $O(\mathscr{V}^{\prime})$
spaces of Witt indices $n+1$ or $n-1$ .

Let be an irreducible admissible depth zero supercuspidal rep-
PROOF. $\pi$
resentation of and be as defined in Theorem 13.3. Because the $Sp(\mathscr{V})$ $\pi^{s}$
character sgn of is trivial, we know that . Suppose that are $Sp(\mathscr{V})$ $\pi=\pi^{s}$ $L_{1},$ $L_{2}$
two good lattices in . We know that the splittings are different up to $\mathscr{V}$ $\beta^{L_{1}},$ $\beta^{L_{2}}$
a character of . Since does not have any nontrivial character, we $Sp(\mathscr{V})$ $Sp(\mathscr{V})$
have . Hence we can fix a splitting for a good lattice in

$\beta^{L_{1}}=\beta^{L_{2}}$
and $\beta^{L}$ $\mathscr{V}$
regard as an irreducible depth zero supercuspidal representation of

$\pi$
. $Sp(\mathscr{V})$
Let (resp. ) be the quadratic space of even dimension

(resp. ) of $\mathscr{V}^{\prime\prime}$ $m^{\prime}$ $m^{\prime\prime}$
Witt index (resp. ) such that first occurs in the theta corre-
$m^{\prime}/2$ $m^{\prime\prime}/2-2$ $\pi$
spondence for the reductive dual pair (resp. $(Sp$ ( ), ))). $(Sp(\mathscr{V}), O(\mathscr{V}^{\prime}))$ $\mathscr{V}$ $O(\mathscr{V}^{\prime\prime}$
By Theorem 13.3, we know that . Therefore at least one of $m^{\prime}+m^{\prime\prime}=4n+4$
must be less than or equal to $2n+2$ .

$m^{\prime},$
$m^{\prime\prime}$
$[$
D. Prasad has conjectured in [Ps] corollary 1 that all irreducible admissible

representations of should occur in the correspondence with some $Sp(\mathscr{V})$ $O(\mathscr{V}^{\prime})$
where is $2n+2$ dimensional. Corollary 13.5 provides some support for his
conjecture.
References
[AM] J. Adams and A. Moy, Unipotent representations and reductive dual pairs over finite
fields, Trans. Amer. Math. Soc., 340 (1993), 309-321.
[As] C. Asmuth, Weil representations of symplectic -adic groups, Amer. J. Math., 101 (1979), $p$
885-909.
[Ab] A.-M. Aubert, Correspondance de Howe et sous-groupes parahoriques, J. Reine Angew.
Math., 392 (1988), 176-186.
[AMR] A.-M. Aubert, J. Michel and R. Rouquier, Correspondance de Howe pour les groupes
finis, Duke Math. J., 83 (1996), 353-397.
[BT] F. Bruhat and J. Tits, Sch\’emas en groupes et immeubles des groupes classiques sur un corps
locaj deuxi\‘eme partie: groupes unitaires, Bull. Soc. Math. France, 115 (1987), 141-195.
[DL] P. Deligne and G. Lusztig, Representations of reductive groups over finite fields, Ann. of
Math. (2), 103 (1976), 103-161.
[HKS] M. Harris, S. Kudla and W. Sweet, Theta dichotomy for unitary groups, J. Amer. Math.
Soc., 9 (1996), 941-1004.
[Hj] H. Hijikata, Maximal compact subgroups of some -adic classical groups, mimeo-
$p$
graphed notes, Yale University, 1964.

[Hw1] R. Howe, Invariant theory and duality for classical groups over finite fields with appli-
cations to their singular representation theory, preprint.
[Hw2] R. Howe, -series and invariant theory, In Automorphic forms, representations and
$0$ $L$ -
functions, Proc. Sympos. Pure Math., 33 (1979), 275-285.

[K1] S. Kudla, Splitting metaplectic covers of dual reductive pairs, Israel J. Math., 87 (1994),
361-401.
[Lt] G. Lusztig, Some examples of square integrable representations of semisimple -adic $p$
groups, Trans. Amer. Math. Soc., 277 (1983), 623-653.

[MVW] C. M\oe glin, M.-F. Vigneras and J.-L. Waldspurger, Correspondances de Howe sur un corps
-adiques, Lecture Notes in Math., 1291, Springer-Verlag, Berlin-Heidelberg-New York, 1987.
$p$
[Mn] C. Moen, The dual pair $(U(1),U(1))$ over a -adic field, Pacific J. Math., 158 (1993),
$p$
365-386.
[MP1] A. Moy and G. Prasad, Unrefined minimal $K$-types for -adic groups, Invent. Math., 116
$p$
(1994), 393-408.
[MP2] A. Moy and G. Prasad, Jacquet functors and unrefined minimal $K$-types, Comment.
Math. Helv., 71 (1996), 98-121.
[Pn1] S.-Y. Pan, Splittings of the metaplectic covers of some reductive dual pairs, Pacific J.
Math., 199 (2001), 163-227.
[Pn2] S.-Y. Pan, Depth preservation in local theta correspondence, Duke Math. J., 113 (2002),
531-592.
[Ps] D. Prasad, On the local Howe duality correspondence, Internat. Math. Res. Notices, 11
(1993), 279-287.
[Rl] S. Rallis, On the Howe duality conjecture, Compositio Math., 51 (l984), 333-399.
[RR] R. Ranga-Rao, On some explicit formulas in the theory of Weil representation, Pacific J.
Math., 157 (l993), 335-37l.
[Rb] B. Roberts, lecture notes, University of Maryland, l994.
[Sr] B. Srinivasan, Weil representations of finite classical groups, Invent. Math., 51 (l979),
l43-l53.
[Tn] S. Tanaka, Irreducible representations of the binary modular congruence group mod , $p^{\lambda}$
J. Math. Kyoto Univ., 7 (l967), l23-l32.

[Tt] J. Tits, Reductive groups over local fields, In Automorphic forms, representations and $L$ -
functions, Proc. Sympos. Pure Math., 33 (l979), 29-7l.

[Wp] J.-L. Waldspurger, Demonstration d’une conjecture de duality de Howe dans le case $p$ -
adiques, $p\neq 2$ , In Festschrift in honor of I. Piatetski-Shapiro, Israel Math. Conf. Proc., 2

(l990), 267-324.
Shu-Yen PAN
Department of Mathematics
National Cheng Kung University
Tainan City 701
TAIWAN
E-mail: sypan@mail.ncku.edu.tw

Local Theta Correspondence of Depth and Theta Dichotomy: Representations

Uploaded by

Copyright:

Available Formats

Local Theta Correspondence of Depth and Theta Dichotomy: Representations

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Local Theta Correspondence of Depth and Theta Dichotomy: Representations

Uploaded by

Copyright:

Available Formats

J. Math. Soc.

Local theta correspondence of depth zero representations and

(Received Feb. 25, 2000)

(resp. ) be a finite-dimensional nondegenerate -Hermitian (resp. -Hermitian)

space over $D$ where are 1 or -1 and . Let (resp. ) $\epsilon,$

denote the group of isometries of (resp. ). We can define a skew- $\mathscr{V}$

symmetric $F$-bilinear form on the space . Then the pair $\mathscr{W}:=\mathscr{V}\otimes_{D}\mathscr{V}^{\prime}$

forms a reductive dual pair (over

aplectic cover of . Let ( esp. ) denote the inverse of $Sp(\mathscr{W})$ $U(\mathscr{V}\underline{)}$$r$

with each other in

. restricting the Weil $representation\underline{o}f$

the local theta correspondence) between irreducible admissible representations of

2000 Mathematics Subject Primary 1lF27; Secondary $22E50,20C33$ .

and irreducible admissible representations of

) (cf. [MVW], [Wp]). $U\overline{(\mathscr{V}}^{\prime}$

To determine the explicit correspondence is extremely difficult. In fact very little

good lattice in i.e., a lattice such that where

{ for any $l\in L$ } and is an integer which will be specified in

$G_{L,0}+:=\{g\in G|(g-1).L^{*}\subseteq L, (g-1).L\subseteq L^{*}\varpi\}$ .

type for is a pair $G$

tation $(\pi, V)$ of contains a minimal $K$-type if $G$

An irreducible admissible representation containing a depth zero minimal K-type

: , the concept of minimal -types can be

Yu). In particular, an irreducible admissible representation $(\pi, V)$ of is said to $\tilde{G}$

. The following is our first main result.

THEOREM A. $\underline{Sup}pose$ and are irreducible admissible

good lattices in respectively, then is a reductive

THEOREM B. Let be a reductive dual pair. Let $(G, G^{\prime}):=(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$

that Then contains a minimal $K$-type

Let be a split dual pair i.e., a dual pair

explicit formula of the splittings with respect to a Schr\"odinger model of the

depending on a good lattice in

of is zero where is any irreducible admissible representation of

the correspondence of cuspidal representations for finite reductive dual pairs

THEOREM C. Suppose that is a split reductive dual $(G, G^{\prime}):=(U(\mathscr{V}), U(\mathscr{V}^{\prime}))$

pair such that $D$ is commutative.

for some good lattice in . Suppose that

cuspidal representation of for some splitting : $U(\mathscr{V})$

and OX is a first occurrence in the theta correspondence. Then has

correspondence for the reductive dual pair and $(G_{L}/G_{L,0}+ , G_{L}^{\prime},/G_{L,0^{+}}^{\prime}, )$

Then OX is a first occurrence for the reductive dual pair $\pi$

is the irreducible admissible representation of

( OX ), is the irreducible admissible representation of

that , and (resp. ) is a character of

(resp. ,) defined in subsection 8.2. $G_{L}^{\prime}$

A nice application of Theorem is that we can extend some of S. Kudla $C$

as follows. Let be an -Hermitian space over a quadratic extension $E$ of . $\mathscr{V}$

$(\pi, V)$ be an irreducible supercuspidal representation of and (resp. ) $U(\mathscr{V})$

. It is interesting that the sum $\ell++\ell-does$ not depend on although each of

0if $Z$ is symplectic;

4, if $Z$ is $p$ -adic orthogonal.

dual pair $(U(v), U(v^{\prime}))$ (resp. $(U(v),$ ). Then $U(v^{\prime\prime}))$

irreducible depth zero supercuspidal representations of with a minimal K-type $U(\mathscr{V})$

. Let be the irreducible depth zero supercuspidal representations of

where is the dimension of

(resp. ) such that (resp.

) occurs in the theta correspondence for the

reductive dual pair (resp. ).

exist a chain of irreducible (super)cuspidal representations such that any two

$G_{L,0}+:=\{g\in G|(g-1).L^{}\subseteq L, (g-1).L\subseteq L^{}\varpi\}$ .

$\langle\Pi_{L^{}}(w), \Pi_{L^{}}(w^{\prime})\rangle_{l^{*}}:=\Pi_{\mathscr{O}}(\langle w, w^{\prime}\rangle\varpi^{1-\kappa})$

nondegenerate skew-symmetric form on by (1.2.c). Let $H(b^{})$ be the $\ll,$ $\gg_{b}$

$Sp(b^{})$ . Let denote the representation of $H(B^{})$ inflated from