The Local Langlands Conjecture for GSp(4)
Wee Teck Gan, Shuichiro Takeda

TL;DR
This paper proves the local Langlands conjecture specifically for the group GSp(4) over non-archimedean local fields of characteristic zero, confirming a major theoretical prediction in number theory.
Contribution
It establishes the local Langlands correspondence for GSp(4), a significant advancement in understanding automorphic forms and representation theory for this group.
Findings
Confirmed the local Langlands conjecture for GSp(4)
Established explicit correspondence between representations and Langlands parameters
Advanced the theoretical framework for automorphic forms on GSp(4)
Abstract
We prove the local Langlands conjecture for where is a non-archimedean local field of characteristic zero.
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsAdvanced Algebra and Geometry · Algebraic Geometry and Number Theory · Geometry and complex manifolds
The Local Langlands Conjecture for
Wee Teck Gan and Shuichiro Takeda
Mathematics Department, University of California, San Diego, 9500 Gilman Drive, La Jolla, 92093
Abstract.
We prove the local Langlands conjecture for where is a non-archimedean local field of characteristic zero.
2000 Mathematics Subject Classification:
11F27, 11F70, 22E50
1. Introduction
Let be a non-archimedean local field of characteristic [math] and residue characteristic . Let be the Weil group of and let be the Weil-Deligne group. It was shown by Harris-Taylor [HT] and Henniart [He1] that there is a natural bijection between the set of equivalence classes of irreducible smooth representations of and the set of conjugacy classes of -parameters for , i.e. admissible homomorphisms
[TABLE]
This bijection satisfies a number of natural conditions which determine it uniquely.
For a general connected reductive group over , which we assume to be split for simplicity, Langlands conjectures that there is a surjective finite-to-one map from the set of (equivalence classes of) irreducible smooth representations of to the set of (equivalence classes of) admissible homomorphisms
[TABLE]
where is the Langlands dual group of and the homomorphisms are taken up to -conjugacy. This leads to a partition of the set of equivalence classes of irreducible representations of into a disjoint union of finite subsets, which are the fibers of the map and are called -packets. Again, this map is supposed to preserve natural invariants which one can attach to both sides. These natural invariants are the -factors, -factors and -factors. Unfortunately, on the representation theoretic side, one only has a general theory of these invariants for generic representations of .
The purpose of this paper is to prove the local Langlands conjecture for .
Main Theorem
There is a surjective finite-to-one map
[TABLE]
with the following properties:
(i) is a (essentially) discrete series representation of if and only if its -parameter does not factor through any proper Levi subgroup of .
(ii) For an -parameter , its fiber can be naturally parametrized by the set of irreducible characters of the component group
[TABLE]
This component group is either trivial or equal to . When , exactly one of the two representations in is generic and it is the one indexed by the trivial character of .
(iii) The similitude character of is equal to the central character of (via local class field theory). Here, is the similitude character of .
(iv) The -parameter of is equal to . Here, is the similitude character of , and we have regarded as both a character of and a character of by local class field theory.
(v) Suppose that is a generic representation or a non-supercuspidal representation. Then for any irreducible representation of , with -parameter , we have:
[TABLE]
Here the functions on the RHS are the local factors of Artin type associated to the relevant representations of , whereas those on the LHS are the local factors of Shahidi, as defined in §4.
(vi) Suppose that is a non-generic supercuspidal representation. For any irreducible supercuspidal representation of with -parameter , let denote the Plancherel measure associated to the family of induced representations on , where we have regarded as a representation of the Levi subgroup . Then is equal to
[TABLE]
(vii) An -packet contains a generic representation if and only if the adjoint -factor is holomorphic at . Here, denotes the adjoint representation of on the complex Lie algebra . Moreover, contains an essentially tempered generic representation if and only if is an essentially tempered -parameter, i.e. has bounded image in . In this case, the generic representation in the packet is unique and is indexed by the trivial character of .
(viii) The map is uniquely determined by the properties (i), (iii), (v) and (vi), with in (v) and (vi).
To the best of our knowledge, for non-generic supercuspidal representations, the theory of local -factors, -factors and -factors of pairs has not been fully developed and so at this point, it is not possible for us to say anything regarding these in part (v) of the theorem. However, if one assumes the existence of a theory of -factors satisfying the usual properties (such as those listed as the “Ten Commandments” in [LR]), then we can show that (v) holds for all representations, in which case the map will be uniquely characterized by (iii) and (v) (with in (v)).
In any case, we substitute the (as yet non-existent) theory of -factors of pairs by the Plancherel measure. The Plancherel measure in (vi) is a coarser invariant than the -factor, but has the advantage that it is defined for all representations. For generic representations, the identity in (vi) follows from results of Shahidi [Sh] and Henniart [He2]. Thus, (vi) shows that the Plancherel measure is an invariant of a supercuspidal -packet. Moreover, it turns out that this coarser invariant is sufficient to distinguish between the non-generic supercuspidal representations of , and this gives the characterization of by (i), (iii), (v) and (vi).
The proof of the theorem relies on the local Langlands correspondence for and and a consideration of the following two towers of dual pairs:
[TABLE]
Thus, it relies on a study of the local theta correspondence arising between and the orthogonal similitude groups associated to quadratic spaces of rank or with trivial discriminant and the accidental isomorphisms:
[TABLE]
as well as the analogs for the inner forms.
The first attempt to define -packets for is the paper [V] of Vigneras, who considered the case . She defined her -packets via theta lifts from various forms of (including those whose discriminant is not trivial). However, she did not check that her -packets exhaust . In a more recent paper [Ro2], B. Roberts has given a more detailed treatment of Vigneras’ construction of local -packets (including the case ).
Our construction of the local -packets, on the other hand, relies only on the theta lifting from to , and . It works for any and also enables us to show that our packets exhaust all representations. We shall also show that, starting from a given -parameter, the -packets constructed by Vigneras and Roberts agree with ours. As a consequence, we show that when , their construction exhausts all irreducible representations of . We also mention that, considering only those representations of whose central character is a square, a section of our map was constructed by Jiang-Soudry [JSo]. More precisely, for the split group , they constructed an injective (but definitely not surjective unless ) map from verifying the conditions (i), (iv) and (v) of our main theorem.
Let us mention the various key ingredients used in the proof of the theorem. The first is a paper of Muić-Savin [MS] which studies the theta lifting of generic discrete series representations for isometry groups and relates the non-vanishing of these theta lifts to the local -functions of Shahidi. The second ingredient is a paper of Kudla-Rallis [KR] which proves the conservation conjecture for the first occurrences of supercuspidal representations of symplectic groups in orthogonal towers. In particular, their results imply that every non-generic supercuspidal representations of can be obtained as a theta lift from the anisotropic . The third is a recent result of Henniart [He2] which shows that the local Langlands correspondence for respects the twisted exterior square -factor. Finally, we have a companion paper [GT4] in which we determine completely the three local theta correspondences mentioned above. The detailed study of these theta correspondences is necessary to supplement the results of [KR] and [MS], particularly for the non-discrete series representations.
The crucial results of [KR], [MS] and [He2] are reviewed in Section 5, after we introduce some basic facts about the theta correspondence for similitudes in Sections 2 and 3 and recall Shahidi’s construction of certain -functions in Section 4. Our construction of the -packets and the proof of exhaustion are given in Section 7, where we also verify the relation of genericity and the adjoint -factor. The preservation of local factors and Plancherel measures is demonstrated in Sections 8 and 9 respectively, and the characterization of is given in Section 10. In Section 11, we reconcile our construction with that given by Vigneras and Roberts. Finally, in Section 12, we give a global application: using the results of this paper, we establish the strong lifting of generic cuspidal representations from to .
We conclude this introduction with a number of subsequent developments:
- (i)
In a sequel [GT2] to this paper, we deduce the local Langlands correspondence for from the results of this paper;
- (ii)
In another sequel [GTW] to this paper, by the first author and W. Tantono, the local Langlands correspondence is extended to the unique inner form of ;
- (iii)
An -packet is supposed to be “stable” and should satisfy some character identities relative to (twisted) endoscopic transfers. Our method, unfortunately, does not shed any light on these harmonic analytic issues. However, another sequel [CG] to this paper, by P. S. Chan and the first author, establishes these properties of the -packets constructed here using the Arthur-Selberg trace formula;
- (iv)
In a recent paper [DR], Debacker and Reeder have given a construction of -packets associated to certain tamely ramified -parameters of an arbitrary reductive group . The elements in their -packets are all depth zero supercuspidal representations. One can ask whether their packets agree with ours in the case . This is shown to be the case in the UCSD PhD thesis of J. Lust;
- (v)
A theory of - and -factors for (for or ) which is valid for all representations, including the non-generic ones, is being developed by N. Townsend in his UCSD PhD thesis.
Acknowledgments: We thank Dipendra Prasad, Brooks Roberts, Gordan Savin and Marie-France Vigneras for their interests, suggestions and encouragements. We also thank Wilhelm Zink for sending us a copy of his letter [Z] to Vigneras. W. T. Gan’s research is partially supported by NSF grant DMS-0500781.
2. Similitude Theta Correspondences
In this section, we shall describe some basic properties of the theta correspondence for similitude groups. The definitive reference for this subject matter is the paper [Ro1] of B. Roberts. However, the results of [Ro1] are not sufficient for our purposes and need to be somewhat extended.
Consider the dual pair ; for simplicity, we assume that is even. For each non-trivial additive character , let be the Weil representation for , which can be described as follows. Fix a Witt decomposition and let be the parabolic subgroup stabilizing the maximal isotropic subspace . Then
[TABLE]
where . The Weil representation can be realized on the Schwartz space and the action of is given by the usual formulas:
[TABLE]
where is the quadratic character associated to and is the natural symplectic form on . To describe the full action of , one needs to specify the action of a Weyl group element, which acts by a Fourier transform.
If is an irreducible representation of (resp. ), the maximal -isotypic quotient has the form
[TABLE]
for some smooth representation of (resp. ). We call the big theta lift of . It is known that is of finite length and hence is admissible. Let be the maximal semisimple quotient of ; we call it the small theta lift of . Then it was a conjecture of Howe that
- •
is irreducible whenever is non-zero.
- •
the map is injective on its domain.
This has been proved by Waldspurger when the residual characteristic of is not and can be checked in many low-rank cases, regardless of the residual characteristic of . If the Howe conjecture is true in general, our treatment in the rest of the paper can be somewhat simplified. However, because we would like to include the case in our discussion, we shall refrain from assuming Howe’s conjecture in this paper.
With this in mind, we take note of the following result which was shown by Kudla [K] for any residual characteristic :
Proposition 2.1**.**
(i) If is supercuspidal, is irreducible or zero.
(ii) If for two supercuspidal representations and , then .
One of the main purposes of this section is to extend this result of Kudla to the case of similitude groups.
Let and be the similitude factors of and respectively. We shall consider the group
[TABLE]
where is the subgroup of consisting of elements such that is in the image of . In fact, for the cases of interest in this paper (see the next section), is surjective, in which case .
The group contains the subgroup
[TABLE]
The Weil representation extends naturally to the group via
[TABLE]
where
[TABLE]
Note that this differs from the normalization used in [Ro1]. Observe in particular that the central elements act by the quadratic character .
Now consider the (compactly) induced representation
[TABLE]
As a representation of , depends only on the orbit of under the evident action of . For example, if is surjective, then is independent of . For any irreducible representation of (resp. ), the maximal -isotypic quotient of has the form
[TABLE]
where is some smooth representation of (resp. ). Further, we let be the maximal semisimple quotient of . Note that though may be reducible, it has a central character given by
[TABLE]
The extended Howe conjecture for similitudes says that is irreducible whenever is non-zero, and the map is injective on its domain. It was shown by Roberts [Ro1] that this follows from the Howe conjecture for isometry groups, and thus holds if . In any case, we have the following lemma which relates the theta correspondence for isometries and similitudes:
Lemma 2.2**.**
(i) Suppose that is an irreducible representation of a similitude group and is a constituent of the restriction of to the isometry group. Then implies that .
(ii) Suppose that
[TABLE]
Suppose further that for each constituent in the restriction of to , is irreducible and the map is injective on the set of irreducible constituents of . Then there is a uniquely determined bijection
[TABLE]
such that for any irreducible summand in the restriction of to the relevant isometry group,
[TABLE]
One has the analogous statement with the roles of and exchanged.
(iii) If is a representation of (resp. ) and the restriction of to the relevant isometry group is , then as representations of (resp. ),
[TABLE]
In particular, is admissible of finite length. Moreover, if for each , then
[TABLE]
Proof.
(i) Without loss of generality, suppose that is a representation of . As a representation of ,
[TABLE]
Hence the result follows by Frobenius reciprocity.
(ii) This is [Ro1, Lemma 4.2], taking note of the results of [AP], where it was shown that restrictions of irreducible representations from similitude groups to isometry groups are multiplicity-free.
(iii) By symmetry, let us suppose that is a representation of . Then we have the following sequence of -equivariant isomorphisms:
[TABLE]
where denotes the full linear dual of . Thus, we have an -equivariant isomorphism of -smooth vectors
[TABLE]
Note that since has a central character, the subspace of -smooth vectors in is the same as its subspace of -smooth vectors. In other words, the contragredient of as a representation of is the same as its contragredient as a representation of . Using the fact that the ’s are admissible of finite length, the above identity implies that is admissible of finite length. The desired result then follows by taking contragredient. Moreover, if is semisimple for each , then it is clear from the above that is semisimple as well. ∎
Proposition 2.3**.**
Suppose that is a supercuspidal representation of (resp. ). Then we have:
(i) is either zero or is an irreducible representation of (resp. ).
(ii) If is another supercuspidal representation such that , then .
Proof.
(i) Without loss of generality, suppose that is a supercuspidal representation of and is nonzero. By Lemma 2.2(iii), if , we have:
[TABLE]
By Lemma 2.2(ii) and Prop. 2.1, we see that any irreducible constituent of satisfies:
[TABLE]
Thus we see that is irreducible. This proves (i).
(ii) Prop. 2.1(ii) implies that if , then . Since and must have the same central character, we see that for some quadratic character . Moreover, it is easy to see that
[TABLE]
Hence we would be done if we can show that for any quadratic character ,
[TABLE]
Of course, the implication is clear from the above. To show the converse, let us set
[TABLE]
and let be the analogous group of quadratic characters. As we noted above,
[TABLE]
and we need to show the reverse inclusion. Now the size of the group is equal to the number of irreducible constituents in . By Lemma 2.2(ii), however, the number of irreducible constituents in and are equal. Hence
[TABLE]
as desired.
∎
3. Theta Correspondences for
In this section, we specialize to the cases of interest in this paper. Let be a (possibly split) quaternion algebra over and let be its reduced norm. Then is a rank 4 quadratic space. We have an isomorphism
[TABLE]
via the action of the latter on given by
[TABLE]
Moreover, an element of of determinant is given by the conjugation action on . An irreducible representation of is thus of the form where are representations of with equal central character. Moreover, the action of on representations of is given by . One of the dual pairs which will be of interest to us is . In the companion paper [GT4], we determine the associated theta correspondence completely. Some of the results are summarized in Theorem 5.6 below.
Now consider the rank 6 quadratic space:
[TABLE]
where is the hyperbolic plane. Then one has an isomorphism
[TABLE]
To see this, note that the quadratic space can also be described as the space of -Hermitian matrices with entries in , so that a typical element has the form
[TABLE]
equipped with the quadratic form . The action of on this space is given by
[TABLE]
Observe that an irreducible representation of is of the form where is a representation of and is a square root of the central character of . The similitude factor of is given by , where is the reduced norm on the central simple algebra . Thus,
[TABLE]
We can now consider the theta correspondence in this case. Since we only need to consider when is split, we shall suppress from the notations. Thus we specialize the results of the previous section to the case when and is the split quadratic space of dimension , so that is surjective and the induced Weil representation is a representation of . In fact, we shall only consider the theta correspondence for . There is no significant loss in restricting to because of the following lemma:
Lemma 3.1**.**
Let (resp. ) be an irreducible representation of (resp. ) and suppose that
[TABLE]
Then the restriction of to is irreducible. If is the unique non-trivial quadratic character of , then does not participate in the theta correspondence with .
Proof.
The analogous result for isometry groups is a well-known result of Rallis [R, Appendix] (see also [Pr1, §5, Pg. 282]). The lemma follows easily from this and we omit the details. ∎
We now collect some results concerning the theta correspondence for . Firstly, we have:
Theorem 3.2**.**
Let be an irreducible representation of . Then is irreducible or zero as a representation of . Moreover, if , then .
Proof.
For supercuspidal representations, the result follows by Prop. 2.3 and the previous lemma. For non-supercuspidal representations, the result follows by the explicit determination of the theta correspondence for given in the companion paper [GT4, Thm.8.3]. ∎
Suppose now that (resp. ) is the unipotent radical of a Borel subgroup of (resp. ) and (resp. ) is a generic character of (resp. ). One may compute the twisted Jacquet module . The following lemma (see [GT1, Prop. 7.4] and [MS, Prop. 4.1]) describes the result:
Lemma 3.3**.**
As a representation of ,
[TABLE]
In particular, if is an irreducible generic representation of , then is nonzero.
Finally we describe the functoriality of the above theta correspondence for unramified representations. The -group of is and so an unramified representation of corresponds to a semisimple class in . On the other hand, the -group of is the subgroup of given by
[TABLE]
which is isomorphic to the group . There is a natural map
[TABLE]
given by
[TABLE]
where is the similitude factor. The following is shown in the companion paper [GT4, Cor. 12.3]:
Proposition 3.4**.**
Let be an unramified representation of corresponding to the semisimple class . Then is the unramified representation of corresponding to the semisimple class .
4. On Certain -functions
In this section, we introduce certain -functions, -factors and -factors which we will need. These local factors were defined by Shahidi [Sh]. To specify them more precisely, we need to consider certain representations of the relevant -groups.
Recall that we have an inclusion of -groups:
[TABLE]
The projection of onto the first factor thus defines a natural 4-dimensional representation of (one of the half-spin representations of ) whose restriction to is the natural 4-dimensional representation of . Following a terminology common in the literature, we call this representation the Spin representation of and .
Now, corresponding to the inclusion , one has a map of -groups
[TABLE]
Indeed, one has the map
[TABLE]
given by
[TABLE]
and the map is simply the restriction of this map to the subgroup . Similarly, corresponding to the inclusion , we have a map of -groups
[TABLE]
and a commutative diagram
[TABLE]
We regard as a 5-dimensional (resp. 6-dimensional) representation of (resp. and call it the standard representation.
Thus, for a representation of , one expects to be able to define a standard degree 5 -function and a degree 4 Spin -function . More generally, for representations of and of , one expects to have the -functions
[TABLE]
which are associated to the representations and respectively.
Similarly, given a representation of , one expects to have the degree 6 -function and the degree 4 -function . More generally, for representations of and of , one expects to have the -functions
[TABLE]
Moreover, if we regard as a representation of , then should be nothing but the Rankin-Selberg -function of the representation of . Further, one expects that
[TABLE]
where the -function on the RHS is the twisted exterior square -function.
In the important paper [Sh], the above -functions and their associated -factors were defined by Shahidi when the representation or is generic. More precisely, suppose that
- •
is the Levi subgroup of a maximal parabolic subgroup ;
- •
is an irreducible generic representation of ;
- •
the adjoint action of the dual group on decomposes as , where each is a maximal isotypic component for the action of the central torus in .
Then Shahidi defined the local factors , , which satisfy
[TABLE]
In Table 1, we list the various -functions that we will use, and the data which are used in their definition via the Shahidi machinery.
Though Shahidi defined the -functions in Table 1 only for generic representations, the definition can be extended to non-generic non-supercuspidal representations of the simple factors of the groups which occur in the table. This uses the Langlands classification, which says that every irreducible admissible representation can be expressed as the unique quotient of a standard module, i.e. one induced from a non-negative twist of a discrete series representation of a Levi subgroup. For the groups occurring in Table 1, their simple factors have proper Levi subgroups which are essentially products of ’s, so that a discrete series representation of such a proper Levi subgroup is generic. Moreover, the restriction of each representation to a proper Levi subgroup of decomposes into the sum of irreducible constituents, all of which appears in the setup of Shahidi’s theory. Thus, one may extend the definition of the local factors to all non-generic non-supercuspidal representations of each simple factor of by multiplicativity (with respect to the standard module under consideration). Thus, the local factors given in Table 1 are defined except when the representation of is non-generic supercuspidal.
The -function in (a) of Table 1 is the main one which intervenes in our main theorem. Hence, we shall simplify notations by writing it as , suppressing the mention of . The same comment applies to the - and -factors.
Finally, we note the following two lemmas:
Lemma 4.1**.**
Let be an irreducible generic representation of which we may identify with a representation of via the isomorphism
[TABLE]
Then we have:
[TABLE]
where the -function on the LHS (resp. RHS) is that in (d) (resp. (f)) of the above table. Moreover, one has the analogous identity for the - and -factors.
Lemma 4.2**.**
Suppose that is an irreducible generic representation of a similitude group or and is an irreducible constituent of the restriction of to the isometry group or . Then one has:
[TABLE]
Both of these lemmas follow from the characterization of Shahidi’s local factors given in [Sh, Thm. 3.5].
5. The Results of Kudla-Rallis, Muić-Savin and Henniart
In this section, we review some crucial general results of Kudla-Rallis [KR], Muić-Savin [MS] and Henniart [He2] before specializing them to the cases of interest in this paper.
Let be the 2n-dimensional symplectic vector space with associated symplectic group and consider the two towers of orthogonal groups attached to the quadratic spaces with trivial discriminant. More precisely, let
[TABLE]
and denote the orthogonal groups by and respectively. For an irreducible representation of , one may consider the theta lifts and to and respectively (with respect to a fixed non-trivial additive character ). Set
[TABLE]
Then Kudla and Rallis [KR, Thms. 3.8 & 3.9] showed:
Theorem 5.1**.**
(i) For any irreducible representation of ,
[TABLE]
(ii) If is a supercuspidal representation of , then
[TABLE]
If we specialize this result to the case and take into account the results of the companion paper [GT4], we obtain:
Theorem 5.2**.**
Let be an irreducible representation of . Then one has the following two mutually exclusive possibilities:
(A) participates in the theta correspondence with , where is non-split;
(B) participates in the theta correspondence with .
Another way of describing this result is that one of the following two possibilities holds:
(I) participates in the theta correspondence with either or (but necessarily not both);
(II) does not participate in the theta correspondence with or , in which case it must participate in the theta correspondence with .
Proof.
Theorem 5.1(i) implies that any representation participates in the theta correspondence with at most one of or . Hence it remains to show that any does participate in the theta correspondence with or . If is supercuspidal, this is an immediate consequence of Theorem 5.1(ii). For generic representations, it follows by Lemma 3.3 that has nonzero theta lift to ; in particular, this implies the theorem for essentially discrete series representations which are not supercuspidal, since these are generic. For the remaining non-generic representations, the result follows by an explicit determination of theta correspondences for given in the companion paper [GT4], especially [GT4, Thms. 8.1 and 8.3]. ∎
Now we come to the results of Muić-Savin [MS]. In the setting above, they considered a discrete series representation of which is generic with respect to a character and determine the value of . Similarly, starting with a discrete series representation of , one may define analogously. Here is the result of Muić-Savin:
Theorem 5.3**.**
(i) Suppose that is a discrete series representation of which is generic with respect to a character .
- (a)
If the standard -factor has a pole at , then and all the irreducible constituents (as -modules) of are discrete series representations which are generic with respect to .
- (b)
If the standard -factor does not have a pole at , then and has a unique -generic constituent. Moreover, this -generic constituent is a discrete series representation.
(ii) Suppose that is a discrete series representation of which is generic with respect to a character .
- (a)
If the standard -factor has a pole at , then and all the irreducible constituents of are discrete series representations which are generic with respect to .
- (b)
If the standard -factor does not have a pole at , then and has a unique -generic -constituent. Moreover, this -generic constituent is a discrete series representation.
Corollary 5.4**.**
Let be a (necessarily generic) discrete series representation of and suppose that the twisted exterior square -function has a pole at . Then the theta lift of to is a nonzero generic discrete series representation.
Proof.
Let be an irreducible constituent of the restriction of to , so that is -generic with respect to some . By Lemmas 4.1 and 4.2, the standard -function of is
[TABLE]
and thus has a pole at . By Thm. 5.3(ii)(a), the theta lift of to is nonzero and all its constituents are -generic discrete series representations. This shows that is a nonzero generic discrete series representation, which is irreducible by Thm. 3.2. ∎
Finally, in our application of Cor. 5.4 later on, we need the following crucial result of Henniart [He2]:
Theorem 5.5**.**
The local Langlands correspondence for respects the twisted exterior square -function. In other words, if is an irreducible representation of with -parameter and is a 1-dimensional character of , then
[TABLE]
where the -function on the LHS is that of Shahidi.
The dichotomy given in Theorem 5.2 is qualitative in nature, but can be made more concrete by the explicit determination of the three theta correspondences in question. This detailed study of theta correspondences is given in the companion paper [GT4]. We conclude this section by assembling the above results together with those of [GT4].
Theorem 5.6**.**
(i) The set of irreducible representations of which are of Type (A) is precisely
[TABLE]
More precisely, under the theta correspondence for , the map
[TABLE]
defines a bijection
[TABLE]
Moreover, the image of the subset of ’s, with , is precisely the subset of non-generic supercuspidal representations of . The other representations in the image are the non-discrete series representations in [GT4, Table 1, NDS(c)].
(ii) The theta correspondence for defines an injection
[TABLE]
The image is disjoint from and consists of:
- (a)
the generic discrete series representations (including supercuspidal ones) such that has a pole at .
- (b)
the non-discrete series representations in **[GT4, Table 1, NDS(b, d,e)]**.
Moreover, the images of the representations ’s, with discrete series representations of , are precisely the representations in (a).
(iii) The theta correspondence for defines an injection
[TABLE]
Moreover, the representations of which are of Type (II), i.e. those not accounted for by (i) and (ii) above, are
- (a)
the generic discrete series representations whose standard factor is holomorphic at . The images of these representations under the above map are precisely the discrete series representations of such that has a pole at .
- (b)
the non-discrete series representations in **[GT4, Table 1, NDS(a)]**. The images of these under the above map consists of non-discrete series representations such that
[TABLE]
for an irreducible two dimensional and a character .
(iv) If a representation of with central character participates in the theta correspondence with , so that
[TABLE]
then has a nonzero theta lift to . If is the small theta lift of to , with a representation of , then
[TABLE]
The proof of Theorem 5.6 is given in [GT4], where more complete and explicit information can be found.
6. Langlands Parameters
In this section, we record some facts about the Langlands parameters for . Given such a parameter
[TABLE]
we may consider its composition with the similitude factor to obtain a 1-dimensional character of . We call the similitude character of .
Now consider the composite of with the inclusion
[TABLE]
to obtain an -parameter for . We note the following crucial lemma (cf. also [V, §7]), which is the analog of Thm. 5.6(iii) for -parameters.
Lemma 6.1**.**
The map gives an injection
[TABLE]
The discrete series -parameters of which are in the image of the map are precisely those such that has a pole at .
Proof.
We shall prove the lemma in the context of . For any , we shall show that the natural map
[TABLE]
is injective. This injectivity is equivalent to the following assertion: if is an -parameter, with a -dimensional vector space over , and and are two nondegenerate symplectic forms on which are preserved by up to a similitude character , then and are conjugate by an element of which centralizes .
We now prove the above statement. Since is semisimple, we may write
[TABLE]
where is irreducible and is its multiplicity space. Since , we see that, for each , either , or else for some . Corresponding to these two possibilities, we may decompose as:
[TABLE]
Since the nondegenerate forms and remain nondegenerate on each summand above, we are reduced to showing the statement on each summand.
First examine . Since is irreducible, it follows by Schur’s lemma that there is a unique (up to scaling) bilinear form on which is -equivariant with similitude character . Any such nonzero form is necessarily nondegenerate and has a sign . Thus, giving a nondegenerate -equivariant symplectic form on is equivalent to giving a nondegenerate bilinear form on of sign . But any two such forms on are conjugate under , which commutes with the action of . This proves the statement for the summand .
Now examine the summand . In this case, the subspaces and are necessarily totally isotropic (with respect to any -equivariant symplectic form). Moreover, there is a unique -equivariant pairing on , up to scaling. Hence, giving a nondegenerate -equivariant symplectic form on is equivalent to giving a nondegenerate bilinear form on . But any two such are conjugate under the natural action of , which commutes with the action of . This proves the statement for the summand .
Finally, suppose that is such that is irreducible as a -dimensional representation. Then it is not difficult to see that has a pole at if and only if contains the trivial representation as a summand. In other words, the action of via preserves a non-zero symplectic form up to scaling by the character . This symplectic form is necessarily nondegenerate, so that factors through after conjugation, for otherwise its kernel would be a non-trivial submodule.
∎
The theory of endoscopy shows that has a unique endoscopic group which is isomorphic to . The dual group of is
[TABLE]
so that there is a distinguished conjugacy class of embeddings of dual groups
[TABLE]
This gives rise to a natural map
[TABLE]
We say that an -parameter is endoscopic if it is in the image of this map. More concretely, is endoscopic if
[TABLE]
Note that the ’s are not necessarily distinct and may be reducible. Observe further that the outer automorphism group of acts on via . It is clear that the natural map above descends to give
[TABLE]
The following lemma is the analog of Thm. 5.6(i) and (ii) for -parameters.
Lemma 6.2**.**
(i) The map
[TABLE]
is injective.
(ii) If is a discrete series parameter of , then is either endoscopic or is irreducible as a 4-dimensional representation. If is endoscopic, then with irreducible, and the component group is . Otherwise, is trivial.
(iii) If is a non-discrete series parameter of , then is either endoscopic or with and . The component group is trivial unless is endoscopic with irreducible, in which case .
In particular, under the injection of Lemma 6.1, the non-endoscopic -parameters consists precisely of those pairs which arise as in Thm. 5.6(iii)(a) and (b).
Proof.
(i) By Lemma 6.1, it suffices to show that the natural map
[TABLE]
is injective. This is a simple exercise which we leave to the reader.
(ii) If is irreducible, then the centralizer in of the image of is , so that is trivial. If is reducible, then the hypothesis that does not factor through any proper parabolic subgroup implies that does not stabilize any nonzero isotropic subspaces. Thus cannot contain any 1-dimensional summand and any 2-dimensional summand must be a nondegenerate symplectic subspace. This shows that with irreducible 2-dimensional and . Moreover, , for otherwise would stabilize a 2-dimensional isotropic subspace. Thus, has the form given in (i). Moreover,
[TABLE]
with sitting diagonally as a subgroup. Thus, ,
(iii) If stabilizes an isotropic line which affords the character , then must stabilize another line which has nonzero pairing with and affords the character . The subspace spanned by and supports a 2-dimensional submodule of with . It follows that is endoscopic. On the other hand, if stabilizes an isotropic plane, but not a line, then
[TABLE]
for an irreducible 2-dimensional and with . If , then is endoscopic. If , then is non-endoscopic of the type given in (ii). We leave the determination of in the various cases to the reader. ∎
Remarks: In a letter [Z] to Vigneras (dated Nov. 25, 1984 and mentioned at the end of the introduction of [V]), W. Zink gave an argument that there do not exist primitive symplectic representations of . However, this is not the case, as one can construct such L-parameters when the residue characteristic of is ; see [GT2, Prop. 5.3].
7. Construction of -Packets and Exhaustion
In this section, we shall construct the map , show that it is surjective and verify (i), (ii), (iii), (iv) and (vii) of the Main Theorem. Since we are working with reductive but non-semisimple groups , let us mention that for the rest of the paper, by a discrete series (resp. tempered) representation of , we mean a representation which is equal to a unitary discrete series (resp. tempered) representation after twisting by a 1-dimensional character.
Definition of the Map
According to Thm. 5.2, the irreducible representations of fall into two disjoint families of Type (I) or (II).
Type I: If is of Type (I), then there is an irreducible representation of (where is possibly split) such that
[TABLE]
By the Jacquet-Langlands correspondence and the local Langlands correspondence for , each gives rise to an irreducible 2-dimensional representation of such that . We define to be the parameter
[TABLE]
By Thm. 5.6(i, ii) and Lemma 6.2, we see that is a discrete series parameter iff is a discrete series representation.
Type II: If is of type (II), then the theta lift of to is nonzero. Regarding as a quotient of , we may write
[TABLE]
Note that the central character of is necessarily equal to . Then we set
[TABLE]
where is the Langlands parameter of . We need to show that factors through the inclusion .
If is a discrete series representation as in Thm. 5.6(iii)(a), then is a discrete series representation of such that has a pole at . By Lemma 6.1, we conclude that factors through and is a discrete series parameter. On the other hand, if is a non-discrete series representation as in Thm. 5.6(iii)(b), then we have
[TABLE]
One may consider the nondegenerate symplectic form which is totally isotropic on and and such that the pairing between and is given by the natural map
[TABLE]
It is clear that this last map is -equivariant, so that is a non-discrete series -parameter of . Thus, we see again that is a discrete series parameter iff is a discrete series representation.
Using the fact that the theta correspondence preserves central characters and the basic properties of the local Langlands correspondence for , one easily checks that the central character of is equal to the similitude character of , and for any character , one has
[TABLE]
We have thus completed the definition of the map and verified properties (i), (iii) and (iv) of the Main Theorem.
Observe that, in Thm. 5.2, there is another partition of the set of irreducible representations of into two sets , namely those of Type (A) or (B). One could have defined the map using this partition as follows.
Type A: If is of Type (A), then where now is the quaternion division algebra. One then defines as in the Type (I) case above.
Type B: If is of Type (B), then the theta lift of to is nonzero and has the form . One then defines following the Type (II) case above.
The only potential difference in these two definitions is for those representations which have nonzero theta lifts to . However, it follows from Thm. 5.6(iv) that the two definitions are in fact the same.
Surjectivity and Fibers
For a given -parameter
[TABLE]
with , we must now determine the fiber of the map over , and in particular show that it is non-empty. From the construction of , observe that the parameter is endoscopic if and only if is of Type I. Thus, we see that if is endoscopic, and otherwise. We want to show that , and we consider the endoscopic and non-endoscopic cases separately.
**Endoscopic case: ** If is endoscopic, then gives rise to an -parameter of and thus determines a representation of , where has -parameter . If the ’s are irreducible, then is also an -parameter of and thus determines a representation of , with the Jacquet-Langlands lift of . By Thm. 5.6(i) and (ii), both and have nonzero theta lifts to , and it follows from the construction of that
[TABLE]
where the latter representation is regarded as zero if one of the ’s is reducible. In view of Lemma 6.2, we see that . Moreover, when , we have , which has two irreducible characters and . In that case, we set
[TABLE]
Note that the representation is generic, whereas is non-generic.
Non-Endoscopic case: If is non-endoscopic with similitude character , then is as described in Lemma 6.2 or Thm. 5.6(iii). In this case, Thm. 5.6(iii) implies immediately that is non-empty, so that . Indeed, if is the representation of with -parameter , then the representation of has nonzero theta lift to by Thm. 5.6(iii), so that
[TABLE]
Thus, for each -parameter , we have determined the fiber , which consists of one or two representations parametrized by the irreducible characters of . Indeed, in the companion paper [GT4, Prop. 13.1], we explicitly write down the -parameters of all non-supercuspidal representations. We should mention that for a non-discrete series parameter , a construction of the -packet has been carried out by B. Roberts and R. Schmidt in their monograph [RS] using the results of Sally-Tadic [ST]. The reader can refer to the table in [RS, §A.5] for the explicit description. Using [GT4, Prop. 13.1], it is easy to check that our definition of agrees with theirs.
We have thus verified property (ii) in the Main Theorem.
Genericity
Let us conclude this section by verifying property (vii) of the Main Theorem, which relates generic representations and the adjoint -factor. For those -packets which contain a non-supercuspidal representation, this is a straightforward calculation using the companion paper [GT4] and the structure of principal series representations of [ST]. The verification in this case has been carried out in a recent paper of Asgari-Schmidt [AS]. However, we give an independent proof in [GT4, Prop. 13.2], as our verification is much more concise than that of [AS].
Suppose on the other hand that contains only supercuspidal representations. From our construction of the discrete series -packets, it is clear that such an -packet always contains a generic supercuspidal representation, since it contains an element which is the theta lift of a (necessarily generic) supercuspidal representation of or . Thus, we need to show that is holomorphic at . But we know that is either an irreducible 4-dimensional representation of , or else where are irreducible 2-dimensional representations of with . Thus, up to a twist by an unramified character, is of Galois type and has bounded image in . Thus, is a representation of with bounded image. It follows that is holomorphic in .
8. Preservation of Local Factors
In this section, we shall check that the local Langlands correspondence we defined in the previous section respects the -, - and -factors of pairs for generic representations and non-generic non-supercuspidal representations. This will prove property (v) of the Main Theorem. More precisely, we have:
Theorem 8.1**.**
Let be an irreducible generic or non-supercuspidal representation of and an irreducible representation of . Suppose that and are the -parameters of and respectively. Then
[TABLE]
Proof.
The proof is entirely similar to that of [MS, Prop. 5.4]. By using the multiplicativity of -factors and the explicit determination of local theta correspondence given in the companion paper [GT4], we are reduced by a standard argument to the case when and are generic supercuspidal representations.
We first consider the case when the theta lift of to is zero, so that its theta lift to is nonzero supercuspidal. By [MS, Lemma 5.2], one can find a totally imaginary number field such that for some place . We may consider the split group over . By [Sh, Prop. 5.1], one can find a globally generic cuspidal representation of such that and for all other finite places , is unramified. By [GRS], the global theta lift of to is nonzero and cuspidal (since its local component does not participate in the local theta correspondence with ). Similarly, let be a cuspidal representation of such that and is unramified for all finite places .
By the functoriality of the theta correspondence for unramified representations at finite places (see Prop. 3.4), we see that for all finite ,
[TABLE]
Here, if we regard as a representation of , then the -factor is simply the Rankin-Selberg -factor (cf. (e) of Table 1). Moreover, the theta correspondences over have been completely determined by Adams-Barbasch [AB]. Though they work with isometry groups, there is no subtlety in passing from isometry groups to similitude groups over . Thus, from their results, one sees that the theta correspondence from to is functorial with respect to the inclusion
[TABLE]
Thus the above identities of local factors also hold over the archimedean places of .
Now by [Sh, Thm. 3.5 (3.14)], we see that for some finite set of places including all the archimedean ones, we have
[TABLE]
Taking everything into account, one deduces that at
[TABLE]
From the definition of the parameter and the fact that the local Langlands correspondence for respects local factors of pairs, we conclude that
[TABLE]
Then as in [MS, Cor. 5.1], since one knows that -functions of discrete series representations are holomorphic in , we deduce the desired identity of -factors from that of the -factors. From this, the identity of -factors also follows. This proves the theorem for those supercuspidal which do not lift to .
Consider now the case when the theta lift of to is non-zero. Thus, there is a supercuspidal representation of such that . Now we can globalize as above and repeat the same argument to get
[TABLE]
From this, the desired identities of -factors and -factors follow. The theorem is proven. ∎
Proposition 8.2**.**
Suppose that one has a theory of -factors for all irreducible representations of satisfying the expected properties, for example those listed in [LR, Thm. 4]. Then the conclusion of Theorem 8.1 holds for all irreducible representations.
Proof.
We shall only give a sketch proof. As before, one is reduced to the case when is a non-generic supercuspidal representation and is a supercuspidal representation of . In this case, for a representation of (where is the quaternion division algebra over ).
Now as in the proof of Theorem 8.1, let be a totally imaginary number field with . Let be another finite place of and let be the quaternion algebra over ramified at precisely and . Pick an irreducible representation of of dimension , so that is a representation of . One can then find a cuspidal representation of whose local components at and are and respectively. Moreover, by the Jacquet-Langlands correspondence, one knows that is cuspidal and hence the local components of are generic at all places outside of .
One can show that the global theta lift of to is nonzero, irreducible and cuspidal (cf. [GT3]). Moreover, the local component of is at , non-supercuspidal at and generic at all other places. Thus, one knows the desired equalities of local factors at every place except , and the same argument as in the proof of the Theorem 8.1 gives the desired result at . ∎
One can also consider the standard -function of . By the same argument as above, one has the following theorem; we omit the details.
Theorem 8.3**.**
Let be a generic representation or a non-supercuspidal representation of and a representation of . Suppose that and are the -parameters of and respectively. Then
[TABLE]
If one has a theory of these standard -factors for all irreducible representations of , then the above identities hold for all irreducible representations.
9. Conservation of Plancherel Measure
In this section, we prove (vi) of the Main Theorem, which expresses the Plancherel measure in terms of the product of various gamma factors of the corresponding -parameters. Let us briefly recall the definition of the relevant Plancherel measure.
Let be an irreducible representation of and a representation of , so that is a representation of
[TABLE]
Now is the Levi factor of a maximal parabolic subgroup of . One can thus form the generalized principal series representation
[TABLE]
where is the determinant character of . If is the opposite parabolic, then we similarly have the induced representation . The additive character determines a Haar measure on , which induces a dual measure on . Then there is a standard intertwining operator
[TABLE]
Then the composite is a scalar operator on and the Plancherel measure is the scalar-valued meromorphic function defined by
[TABLE]
When and are both supercuspidal, the analytic properties of determine the points of reducibility of the principal series . More precisely, we have [Si, §5.3-5.4]:
Proposition 9.1**.**
Suppose that is a unitary supercuspidal representation of . (i) On the imaginary axis , is holomorphic and .
(ii) If , then and is irreducible for all .
(iii) If , then there is a unique real such that is reducible. Moreover, if and only if , in which case is the unique pole of on the positive real axis.
When is a generic (not necessarily supercuspidal) representation, Shahidi showed that the meromorphic function can be expressed as a product of gamma factors. As a result, he was able to determine with great precision the unique point of reducibility in (iii) of the above proposition when . Let us recall his results for the case at hand.
The parabolic subgroup gives rise to a dual parabolic subgroup in the dual group and we have:
[TABLE]
Under the adjoint action of , the Lie algebra of the unipotent radical decomposes as with
[TABLE]
where denotes the relevant standard representation and is the similitude character of . On the opposite nilpotent radical , the adjoint action of is the dual representation . Now we have:
Proposition 9.2**.**
Suppose that is a generic representation of . Then
[TABLE]
which is in turn equal to
[TABLE]
Proof.
The first equality is a result of Shahidi [Sh, Thm. 3.5]. The second equality follows from Thm. 8.1 and a result of Henniart [He2, Thm. 1.4]. Indeed, Henniart showed that the local Langlands correspondence for respect the twisted symmetric square epsilon-factors up to multiplication by a root of unity . Hence
[TABLE]
which is in turn equal to
[TABLE]
This explains why the root of unity disappears. We thank Henniart for explaining this point to us. ∎
After these preliminaries, the main result of this section is:
Theorem 9.3**.**
Let be an -packet of such that is non-generic supercuspidal and is a generic discrete series representation. Then for any supercuspidal representation of ,
[TABLE]
In particular, if is also supercuspidal, then is reducible if and only if is reducible.
Corollary 9.4**.**
Property (vi) of the Main Theorem holds.
The rest of the section is devoted to the proof of Thm. 9.3, which is similar to the proof of [MS2, Prop. 2.1]. We know by the results of §7 that the representations and can be obtained as theta lifts from and respectively, where is the quaternion division algebra over . Thus, we have
[TABLE]
where is a representation of with the Jacquet-Langlands lift on . Moreover, we know that .
Now choose a number field such that for two places and , one has . Let be the quaternion division algebra over which is ramified precisely at and . One can find a cuspidal representation of such that for or . If denotes the Jacquet-Langlands lift of to , then we have for or . We may now consider the global theta lifts and of and to . We have:
Lemma 9.5**.**
The global theta lifts and are nonzero irreducible cuspidal representations of . Moreover, for or ,
[TABLE]
and for all or