Abhandlungen aus dem Mathematischen Seminar der Universität Hamburg vol.

65, 1995,
pages 341-363

ON CHARACTERS OF IRREDUCIBLE UNITARY

REPRESENTATIONS OF GENERAL LINEAR GROUPS

Marko Tadić

Introduction
Founding harmonic analysis on classical simple complex groups, I.M. Gelfand and M.A.
Naimark in their classical book [GN] posed three basic questions: unitary duals, characters
of irreducible unitary representations and Plancherel measures.
In the case of reductive p-adic groups, the only series of reductive groups where unitary
duals are known are general linear groups. In this paper we reduce characters of irreducible
unitary representations of GL(n) over a non-archimedean local field F , to characters of
irreducible square-integrable representations of GL(m), with m ≤ n (we get an explicit
expression for characters of irreducible unitary representations in terms of characters of
irreducible square-integrable representations). In other words, we express characters of
irreducible unitary representations in terms of the standard characters. We get also a for-
mula expressing the characters of irreducible unitary representations in terms of characters
of segment representations of Zelevinsky (the formula for the Steinberg character of GL(n)
is a very special case of this formula).
The classification of irreducible square-integrable representations of GL(m, F )’s has re-
cently been completed ([Z], [BuK], [Co]). The characters of these representations are not
yet known in the full generality, although there exists a lot of information about them
([Ca2], [CoMoSl], [K], [Sl]).
Zelevinsky’s segment representations supported by minimal parabolic subgroups are
one dimensional, so their characters are obvious. Therefore, we get the complete formu-
las for characters of irreducible unitary representations supported by minimal parabolic
subgroups.
By the classification theorem for general linear groups over any locally compact non-
discrete field, any irreducible unitary representation is parabolically induced by a tensor
product of representations u(δ, n) where δ is an irreducible essentially square integrable
representation of some general linear group and n a positive integer (see the second section
for precise statements). Since there exists a simple formula for characters of parabolically
induced representations in terms of the characters of inducing representations ([D]), it is
enough to know the characters of u(δ, n)’s.
Our idea in getting the formula for characters of irreducible unitary representations
was to use the fact that unitary duals in archimedean and non-archimedean case can be
expressed in the same way. Using the fact that there also exists a strong similarity of
behavior of ends of complementary series, we relate in these two cases the formulas that

Typeset by AMS-TEX

1
2 MARKO TADIĆ

express characters of u(δ, n)’s in terms of standard characters. G. Zuckerman obtained the
formula for the trivial character (of any reductive group) in terms of standard characters
([V1], Proposition 9.4.16). Along this lines we obtain the formula for characters of u(δ, n)
in the non-archimedean case. Our method completely avoids the question of multiplicities.
It is interesting to note that the formulas for the trivial character in terms of standard
characters of GL(n), are very different in the complex and the p-adic case.
Denote by ν the character |det|F of GL(n). To each segment in irreducible cuspidal
representations of general linear groups [ρ, ν k ρ] = {ρ, νρ, ν 2 ρ, . . . , ν k ρ}, one can attach a
unique essentially square integrable subquotient of ρ×νρ×ν 2 ρ×· · ·×ν k ρ, which is denoted
by δ([ρ, ν k ρ]) (× denotes the J. Bernstein’s and A.V. Zelevinsky’s notation for the parabolic
induction in the setting of general linear groups, see the first section). The representations
u(δ, n)’s are convenient to write in the Langlands classification. In this classification, they
can be written as L(([νρ, ν d ρ], [ν 2 ρ, ν d+1 ρ], . . . , [ν n ρ, ν d+n−1 ρ])), where ρ is an irreducible
cuspidal representation of some general linear group (see the third section for all details
regarding notation). Denote by RF the algebra of Grothendieck groups of categories of
representations of finite length of all general linear groups over F (see the first section).
Now Theorem 5.4 says
Theorem. Let n, d ∈ Z, n, d ≥ 1. Let Wn be the group of permutations of the set
(d)
{1, 2, . . . , n}. Denote Wn = {w ∈ Wn ; w(i) + d ≥ i for all 1 ≤ i ≤ n}. Then we have the
following identity in RF
 
n
L(([νρ, ν d ρ], . . . , [ν n ρ, ν d+n−1 ρ])) = (−1)sgn(w) δ([ν i ρ, ν w(i)+(d−1) ρ]).
(d)
w∈Wn i=1

From this formula one can get simply, by multiplication of polynomials, the expression
for (character of) arbitrary irreducible unitary representation of GL(m, F ).
A similar approach to the characters of irreducible unitary representations of GL(n, R)
can be undertaken, since D. Vogan has obtained the formulas for ends of complementary
series in the real case which are similar to those ones in the complex case.
The ends of complementary series of GL(2n, C) seem to be known to several people
(up to a twist by a character, there is only one such end). Unfortunately, there is no
written reference for it. D. Vogan outlined to us a proof based on his paper [V2]. S.
Sahi’s paper [Sh1] seems to be the closest to the complete reference. He proved that there
are no more than two subquotients at the end ([Sh1], Theorem 3C). One subquotient is
more or less evident (reducibility is also evident). One needs only to see what is the other
subquotient. Sahi described to us an explicit intertwining giving the other subquotient
(subrepresentation in this case). In the appendix to our paper, we outline what is necessary
to get the complete argument from [Sh1]. Instead of writing down Vogan’s or Sahi’s
intertwining, we give a relatively simple argument based on the expression of the trivial
character in terms of the standard characters. This argument seems to us more in the
spirit of this paper and also the shortest since the necessary notations for such proof were
already introduced. We are thankful to both of them for communicating us their proofs.
We are thankful to D. Vogan for informing us of Proposition 9.4.16 of [V1] (he explained
it to us in the setting of GL(n, C)). A number of conversations with D. Miličić was very
 CHARACTERS OF UNITARY REPRESENTATIONS 3

helpful in the process of preparing of this paper. I. Mirković remarks helped a lot to make
this paper more readable.
Now we describe the content of this paper, according to sections. In the first section, we
introduce general notation for general linear groups over any local field. The second section
recalls the classification of the non-unitary and unitary dual. The third section gives a
more precise picture of these classifications in the non-archimedean case. In the fourth
section we reformulate the parameterization for GL(n, C) for the later use. We recall here
the formula for the character of the trivial representation in terms of standard characters.
In the fifth section we construct an isomorphism, using which we get the character formula
in the non-archimedean case. In the sixth section we consider the question of multiplicities
related to characters of irreducible unitary representations. In the appendix, we sketch the
argument which gives a complete written reference for the description of the ends of the
complementary series of GL(2n, C).
Questions of S.J. Patterson motivated our work on characters of GL(n). The other
motivation, besides the general concept of the harmonic analysis on general linear groups,
are possible applications of such character formulas in trace formulas where only unitary
representations show up.
We did this work during our stay in Göttingen as a guest of SFB 170. We want to thank
SFB 170 for their kind hospitality and support.

1. General notation for GL(n)

We denote in this paper the set of positive integers by N = {1, 2, 3, . . . }.
A locally compact non-discrete field is denoted by F . The modulus character
 of F is
denoted by | |F . It is determined by the condition that |x|F F f (xa)da = F f (a)da
for any continuous compactly supported complex-valued function f on F . In the above
formula, da denotes a Haar measure of the additive group (F, +) of the field. Note that
| |C is the square of the usual absolute value on C.
Set Gn = GL(n, F ) (n ∈ Z, n ≥ 0). We identify characters of F × with characters of Gn
using the determinant homomorphism. The character of Gn corresponding to | |F will be
denoted by ν.
Suppose that F is archimedean. Then we shall denote by gn , or simply by g, the Lie
algebra of Gn , where we consider Gn as a real Lie group. We denote by Kn , or simply by
K, a maximal compact subgroup in Gn . If F ∼ = R, then we can fix Kn to be the group
O(n) of all real orthogonal matrices in Gn . If F ∼ = C, then we can take Kn to be the
group U(n) of all unitary matrices in Gn . The choice of maximal compact subgroups is
not important for our purposes.
By an admissible module of Gn we shall mean an admissible representation of Gn
when F is non-archimedean. If F is archimedean, then by an admissible Gn -module we
shall mean an admissible (gn , Kn )-module. Admissible Gn -modules form a category in a
standard way.
The Grothendieck group of the category of all admissible Gn -modules of finite length,
will be denoted by RnF . The set G  n of all equivalence classes of irreducible Gn -modules
form a Z-basis of RnF . For an admissible Gn -module π of finite length, we denote the semi
4 MARKO TADIĆ

 n generates
simplification of π by s.s.(π) ∈ RnF . There is a natural partial order on RnF (G
F
the cone of positive elements in Rn ).
Let πi be an admissible Gni -modules of finite length, for i = 1, 2. We can consider
π1 ⊗ π2 as a Gn1 × Gn2 -module. We can identify Gn1 × Gn2 in an obvious way with a Levi
factor of the following parabolic subgroup
 
g1 ∗
; gi ∈ G n i .
0 g2
We denote by π1 × π2 the Gn1 +n2 -module parabolically induced by π1 ⊗ π2 from the above
parabolic subgroup. Then (π1 × π2 ) × π3 ∼
= π1 × (π2 × π3 ). Also χ(π1 × π2 ) ∼
= (χπ1 ) × (χπ2 ),
×
for any character χ of F .
Since the semi simplification of π1 × π2 depends only on semi simplifications of π1 and
π2 , we can lift × to a Z-bilinear mapping × : RnF1 × RnF2 → RnF1 +n2 . Set RF = ⊕n≥0 RnF .
One lifts × to an operation × : RF ×RF → RF in an obvious way. In this way, RF becomes
an associative graded ring. This ring is commutative. For a fixed character χ of F × , lift
π → χπ to a Z-linear map χ : RF → RF . Then this map on RF is an endomorphism of
the graded ring.
Natural orders on RnF ’s determine a natural order ≤ on RF in an obvious way. An
additive homomorphism φ : RF → RF is called positive if x ∈ RF and x ≥ 0 implies
φ(x) ≥ 0. In the sequel we shall denote RnF and RF often simply by Rn and R. We shall
use notation RnF and RF only when we shall work with more than one field F in the same
time.

2. Non-unitary and unitary dual

We say that a Gn -module is essentially square integrable if there exists a character of F ×
such that the module, after twisting with that character, becomes square integrable modulo
center. Denote by Du (resp. by D) the set of all square integrable classes (resp. essentially
square integrable classes) in ∪n≥1 G  n . For each δ ∈ D there exists a unique e(δ) ∈ R such
−e(δ)
that ν δ is unitarizable. We shall denote ν −e(δ) δ by δ u . Thus δ = ν e(δ) δ u , where
e(δ) ∈ R and δ u is unitarizable.
Denote by M (D) the set of all finite multisets in D. For d = (δ1 , δ2 , . . . , δk ) ∈ M (D),
chose a permutation p of {1, 2, . . . , k} such that e(δp(1) ) ≥ e(δp(2) ) ≥ · · · ≥ e(δp(k) ). Then
the module
λ(d) = δp(1) × δp(2) × · · · × δp(k)
has a unique irreducible quotient, whose class depends only on d (not on p as above). We
denote this irreducible quotient by L(d). This is the Langlands classification for the groups
Gn ([BoWh], [J]). It is easy to see that for a character χ of F × ,
χL(δ1 , δ2 , . . . , δk ) ∼
= L(χδ1 , χδ2 , . . . , χδk ).
Define on M (D) the structure of the additive semigroup by
(δ1 , δ2 , . . . , δk ) + (δ1 , δ2 , . . . , δk
) = (δ1 , δ2 , . . . , δk , δ1 , δ2 , . . . , δk
).
Then L(d + d ) is always a subquotient of L(d) × L(d ), for d, d ∈ M (D).
To shorten notation, we shall often denote s.s.(λ(d)) ∈ R simply by λ(d) ∈ R (this will
cause no confusion). Since λ(d) ∈ R, d ∈ M (D), form a basis of R, we have:
 CHARACTERS OF UNITARY REPRESENTATIONS 5

2.1. Proposition. Ring R is a polynomial ring over D.

This fact was first observed by A.V. Zelevinsky in the non-archimedean case ([Z], Corol-
lary 7.5).
 m will be denoted by G
The set of unitarizable classes in G
m . For δ ∈ Du and a positive
integer n set
n−1 n−3 n−1
u(δ, n) = L(ν 2 δ, ν 2 δ, . . . , ν − 2 δ).
2.2. Theorem. Set

B = {u(δ, n), ν α u(δ, n) × ν −α u(δ, n), δ ∈ Du , n ∈ N, 0 < α < 1/2}.

Then
(i) If σ1 , . . . , σk ∈ B, then σ1 × . . . × σk ∈ G
p for some p.
(ii) If π ∈ G
p , then there exist σ1 , . . . , σm ∈ B, unique up to a permutation, such that

π∼
= σ1 × . . . × σm .

Proof. In the non-archimedean case, this is Theorem 7.1 of [T4]. The proof there uses the
positivity of the Zelevinsky involution (see the next section). The positivity was announced
by J. Bernstein. Now there exist written proofs of this fact in preprints of A.-M. Aubert
([A]) and K.Procter ([P]). Let us note that a proof of the positivity is going to be contained
also in a forthcoming paper of P. Schneider and U. Stuhler (see 14.19 of [LRaSu]).
In [T3] we gave a proof in the archimedean case. This proof uses a result announced by
A.A. Kirillov for which there is no published proof yet. The complete proof is in [V1]. Our
notation in the archimedean case follows [T3]. For a unified point of view on the above
theorem, one can consult [T7].

3. Non-archimedean GL(n)
In this section we assume that F is non-archimedean.
Denote by C the set of all equivalence classes of irreducible cuspidal representation of
all general linear groups GL(n, F ), n ≥ 1. Take ρ ∈ C. Let k ∈ Z, k ≥ 0. The set

[ρ, ν k ρ] = {ρ, νρ, ν 2 ρ, . . . , ν k ρ}

is called a segment of irreducible cuspidal representations. We shall sometimes write a

segment
[ν k1 ρ, ν k2 ρ] = [k1 , k2 ](ρ)
(here k1 , k2 ∈ R such that k2 −k1 ∈ Z and k2 −k1 ≥ 0). Denote the set of all such segments
by S. We denote by M (S) the set of all finite multisets in S. We consider the partial order
on M (S) which is introduced in 7.1 of [Z].
For [ρ, ν k ρ] = {ρ, νρ, ν 2 ρ, . . . , ν k ρ} ∈ M (S), the representation ρ × νρ × ν 2 ρ × · · · × ν k ρ
has a unique irreducible subrepresentation, which we denote s([ρ, ν k ρ]), and a unique
irreducible quotient, which we denote δ([ρ, ν k ρ]). Representations s([ρ, ν k ρ]) will be called
6 MARKO TADIĆ

Zelevinsky’s segment representations. Representations δ([ρ, ν k ρ]) are essentially square

integrable.
The mapping s(∆) → δ(∆), ∆ ∈ S, can be in a unique way extended to a ring morphism
of R. A.V. Zelevinsky proved that this morphism is involutive ([Z], Proposition 9.12).
This morphism, denoted x → xt , is called the Zelevinsky involution. It carries irreducible
representations into irreducible ones ([A], [P]). This fact is equivalent to the positivity of
the involution.
Let a = (∆1 , ∆2 , . . . , ∆k ) ∈ M (S). Choose a permutation p of {1, 2, . . . , k} such that
e(δ(∆p(1) )) ≥ e(δ(∆p(2) )) ≥ · · · ≥ e(δ(∆p(k) )). Set
λ(a) = δ(∆p(1) ) × δ(∆p(2) ) × · · · × δ(∆p(k) ),
ζ(a) = s(∆p(1) ) × s(∆p(2) ) × · · · × s(∆p(k) ).
The representations λ(a) and ζ(a) are determined up to an isomorphism by p as above.
The representation ζ(d) has a unique irreducible subrepresentation which we denote by
Z(a). The representation λ(d) has a unique irreducible quotient which we denote by L(a).
If a = (∆1 , ∆2 , . . . , ∆k ) ∈ M (S), then the number k will be denoted by |a|. For
∆ = {ρ, νρ, . . . , ν  ρ} ∈ S we have in R

(3.1) Z(∆) = (−1)+1−|a| s.s.(λ(a)),
a∈M (S),a≤(ρ,νρ,...,ν  ρ)

(3.2) L(∆) = (−1)+1−|a| s.s.(ζ(a)).
a∈M (S),a≤(ρ,νρ,...,ν  ρ)

This is a special case of Proposition 9.13 of [Z].

Fix an irreducible cuspidal representation ρ of a general linear group. Let d and n be
positive integers. Denote a(1, d)(ρ) = [ν −(d−1)/2 ρ, ν (d−1)/2 ρ] ∈ S and
−(n−1)/2 1−(n−1)/2 (n−1)/2
a(n, d)(ρ) = (a(1, d)(ν ρ)
, a(1, d)(ν ρ)
, . . . , a(1, d)(ν ρ)
).
We take formally a(0, d)(ρ) to be the empty multiset. Thus, L(a(0, d)(ρ) ) = Z(a(0, d)(ρ) )
is the one-dimensional representation of G0 . This is identity of R. We take also a(n, 0)(ρ)
to be the empty multiset. So, again L(a(n, 0)(ρ) ) = Z(a(n, 0)(ρ) ) is identity in R.
Note that
(k1 +k2 )
(3.3) [ν k1 ρ, ν k2 ρ] = [k1 , k2 ](ρ) = a(1, k2 − k1 + 1)(ν ρ)/2
,
α
(3.4) a(1, d)(ν ρ)
= [−(d − 1)/2 + α, (d − 1)/2 + α](ρ) .
3.1. Remark. Proposition 2.1 says that R is a polynomial ring over {δ(∆), ∆ ∈ S} =
{L(a(1, d)(ρ) ); d ∈ N and ρ ∈ C}. Analogously, R is a polynomial ring over {s(∆), ∆ ∈
S} = {Z(a(1, d)(ρ) ); d ∈ N and ρ ∈ C}.

For α ∈ R, one has directly the following isomorphisms

ν α L(a(n, d)(ρ) ) ∼
α
(3.5) = L(a(n, d)(ν ρ) ),
ν α Z(a(n, d)(ρ) ) ∼
α
(3.6) = Z(a(n, d)(ν ρ) ).
 CHARACTERS OF UNITARY REPRESENTATIONS 7

3.2. Theorem. For positive integers n and d, and ρ ∈ C, we have

ν 1/2 L(a(n, d)(ρ) ) × ν −1/2 L(a(n, d)(ρ) ) =

L(a(n + 1, d)(ρ) ) × L(a(n − 1, d)(ρ) ) + L(a(n, d + 1)(ρ) ) × L(a(n, d − 1)(ρ) )

and

ν 1/2 Z(a(n, d)(ρ) ) × ν −1/2 Z(a(n, d)(ρ) ) =

Z(a(n + 1, d)(ρ) ) × Z(a(n − 1, d)(ρ) ) + Z(a(n, d + 1)(ρ) ) × Z(a(n, d − 1)(ρ) )

in R. Also
L(a(n, d)(ρ) ) = Z(a(d, n)(ρ) ).

Proof. Suppose that ρ is unitarizable. The last relation is just (iii) of Theorem A.9 of [T4].
For the proof of the first two relations, see Theorem 6.1 of [T5]. The case when ρ is not
necessarily unitarizable, follows from the fact that ν α , α ∈ R, is an automorphism of R
and the formulas (3.5) and (3.6).

3.3. Lemma. For positive integers n and d, and ρ ∈ C, we have
 t
Z(a(n, d)(ρ) ) = L(a(n, d)(ρ) ) = Z(a(d, n)(ρ) )

and  t
L(a(n, d)(ρ) ) = Z(a(n, d)(ρ) ) = L(a(d, n)(ρ) ).

Proof. The lemma follows from the last relation of the above theorem, and [Ro]. We shall
show here how one can get also the lemma directly from the relations in the above theorem.
It is enough to prove only one of the above two relations. We shall prove the first one, by
induction with respect to n. For n = 1, the relation holds by definition. It is enough to
consider only the case d ≥ 2.
Let n ≥ 1 and suppose that the formula holds for n ≤ n. We shall show that it holds
for n + 1. By Theorem 3.2 we have
1/2 −1/2
Z(a(n, d)(ν ρ)
) × Z(a(n, d)(ν ρ)
)=
Z(a(n + 1, d)(ρ) ) × Z(a(n − 1, d)(ρ) ) + Z(a(n, d + 1)(ρ) ) × Z(a(n, d − 1)(ρ) ).

After applying the Zelevinsky involution and the inductive assumption, we get
1/2 −1/2
L(a(n, d)(ν ρ)
) × L(a(n, d)(ν ρ)
)=
Z(a(n + 1, d)(ρ) )t × L(a(n − 1, d)(ρ) ) + L(a(n, d + 1)(ρ) ) × L(a(n, d − 1)(ρ) )

(if n = 1, then we take Z(a(n − 1, d)(ρ) ) and L(a(n − 1, d)(ρ) ) to be the identity of R, as
before). From Theorem 3.2 we know
1/2 −1/2
L(a(n, d)(ν ρ)
) × L(a(n, d)(ν ρ)
)=
L(a(n + 1, d) (ρ)
) × L(a(n − 1, d)(ρ) ) + L(a(n, d + 1)(ρ) ) × L(a(n, d − 1)(ρ) ).
8 MARKO TADIĆ

Taking the difference of these two relations, we get (Z(a(n + 1, d)(ρ) )t − L(a(n + 1, d)(ρ) )) ×
L(a(n − 1, d)(ρ) ) = 0. Since R has no non-trivial zero divisors, we get Z(a(n + 1, d)(ρ) )t =
L(a(n + 1, d)(ρ) ). This finishes the proof of the Lemma.

The formulas for the ends of the complementary series in the Theorem 3.2, give an
algorithm to express L(a(n, d)(ρ) ) in terms of the standard representations λ(a), a ∈ M (S).
The same application can be done for Z(a(n, d)(ρ) ) which are expressed in terms of ζ(a), a ∈
M (S). We shall illustrate this algorithm in the case of the Langlands classification. To
get the case of the Zelevinsky classification, one need only replace L by Z in the formulas
below.
We can write the formula for the ends of the complementary series as

L(a(n + 1, d)(ρ) ) = L(a(n − 1, d)(ρ) )−1


(ν 1/2 ρ) (ν −1/2 ρ)
L(a(n, d) ) × L(a(n, d) ) − L(a(n, d + 1) ) × L(a(n, d − 1) )
(ρ) (ρ)

in the field of fractions of R (i.e., in the field of rational functions corresponding to R).
We shall use these recursive relations for expressing characters of L(a(n, d)(ρ) ) in terms of
the standard characters, for n = 2 and n = 3. For n = 2 we have immediately the final
formula
1/2 −1/2
L(a(2, d)(ρ) ) = L(a(1, d)(ν ρ)
) × L(a(1, d)(ν ρ)
) − L(a(1, d + 1)(ρ) ) × L(a(1, d − 1)(ρ) ).
For n = 3 after a short calculation we get
−1
L(a(3, d)(ρ) ) =L(a(1, d)(νρ) ) × L(a(1, d)(ρ) ) × L(a(1, d)(ν ρ)
)
−1/2 −1/2
−L(a(1, d)(νρ) ) × L(a(1, d + 1)(ν ρ)
) × L(a(1, d − 1)(ν ρ)
)
1/2 1/2 −1
−L(a(1, d + 1)(ν ρ)
) × L(a(1, d − 1)(ν ρ)
) × L(a(1, d)(ν ρ)
)
1/2 −1/2
+L(a(1, d + 1)(ν ρ)
) × L(a(1, d + 1)(ν ρ)
) × L(a(1, d − 2)(ρ) )
1/2 −1/2
+L(a(1, d + 2)(ρ) ) × L(a(1, d − 1)(ν ρ)
) × L(a(1, d − 1)(ν ρ)
)
−L(a(1, d + 2)(ρ) ) × L(a(1, d)(ρ) ) × L(a(1, d − 2)(ρ) ).

4. GL(n, C)
In this section we assume F = C. Denote T = {z ∈ C; |z|C = 1}. Then C ∼ = T × R×+.
1/2
By | | we shall denote the usual absolute value on C, i.e. | | = | |C .
Each character χ of C× can be written in a unique way as a product χ(z) = (z/|z|)d φ(z)
where d ∈ Z and φ is trivial on T. We shall say that φ ∈ (C/T). This character will be
denoted by χ(d)(φ) . The character χ(d)(φ) determines via the determinant homomorphism
the character of Gn which we denote by χ(n, d)(φ) . Note that χ(1, d)(φ) = χ(d)(φ) . For



φ ∈ (C/T) and d ∈ Z we have χ(n, d )(φ ) χ(n, d)(φ) = χ(n, d + d )(φ φ) . In particular,
β
for β ∈ R we have ν β χ(n, d)(φ) = χ(n, d)(ν φ) . We take formally χ(0, d)(φ) to be the one-
dimensional representation of G0 . This is identity in R. Recall that R is a polynomial ring
over all χ(1, d)(φ) , where d ∈ Z and φ is a character of C× trivial on T.
 CHARACTERS OF UNITARY REPRESENTATIONS 9

The following proposition seems to be well known, but we could not find the complete
reference for it. Very close to a complete reference is [Sh1]. See the appendix for an
explanation of the proof.
4.1. Proposition. Let n, d ∈ Z, n ≥ 1, and let φ be a character of C× trivial on T. Then
we have in R

ν 1/2 χ(n, d)(φ) × ν −1/2 χ(n, d)(φ)

= χ(n + 1, d)(φ) × χ(n − 1, d)(φ) + χ(n, d + 1)(φ) × χ(n, d − 1)(φ)

(in the above formula for n = 1 we take χ(0, d)(φ) to be identity of R).

4.2. Remark. It is enough to prove the above equality only for one d and one φ. Other
equalities are obtained by twisting with characters, which are automorphisms of R.
Let Wn be the group of permutations of {1, 2, . . . , n}. The parity of a permutation
w ∈ Wn will be denoted by sgn(w). The group Wn acts on Cn by permuting coordinates:
w(x1 , . . . , xn ) = (xw−1 (1) , . . . , xw−1 (n) ). Denote

τ = (τ1 , τ2 , . . . , τn ) = ((n − 1)/2, (n − 3)/2, . . . , −(n − 1)/2).

Here τi = (n − 2i + 1)/2. Now we shall write a special case of the formula in Proposition
9.4.16 of [V1] for the character of the trivial representation. We are thankful to D. Vogan
for telling us of that formula and for writing for us the specialization of the formula for
GL(n, C). This proposition deals with unpublished results of G. Zuckerman.
4.3. Proposition. For n ∈ N we have in R the following identity
 (τ1 +τw(1) )/2 (τn +τw(n) )/2
χ(n, 0)(1C× ) = (−1)sgn(w) χ(τ1 − τw(1) )(ν )
× · · · × χ(τn − τw(n) )(ν )
.
w∈Wn

Now obviously for d ∈ Z and φ ∈ (C× /T)˜ we have

(4.1) χ(n, d)(φ) =

 (τ1 +τw(1) )/2 (τn +τw(n) )/2
(−1)sgn(w) χ(τ1 − τw(1) + d)(ν φ)
× · · · × χ(τn − τw(n) + d)(ν φ)
.
w∈Wn

5. Characters of non-archimedean GL(n)

In this section F will be non-archimedean.
F
Fix an irreducible cuspidal representation ρ of some GL(m, F ). Let R(ρ) be the subring
F
of R generated by
α
{Z(a(1, d)(ν ρ) ); d ∈ N and α ∈ R}.
F
Then R(ρ) is a polynomial ring over the above set (see Remark 3.1).
10 MARKO TADIĆ

Fix some φ ∈ (C× /T)˜. Consider the mapping

); d ∈ N and α ∈ R} → RC
α
ψ(ρ,φ) : {Z(a(1, d)(ν ρ)

α α
given by ψ(ρ,φ) (Z(a(1, d)(ν ρ) )) = χ(1, d)(ν φ) , when d ∈ N and α ∈ R. Then ψ(ρ,φ)
extends uniquely to a ring homomorphism Ψ(ρ,φ) : R(ρ) F
→ RC . Obviously, Ψ(ρ,φ) is in-
jective,
 since it maps different indetermine to different indetermine. For any finite sum
(ν α ρ)
d∈N,α∈R zd,α Z(a(1, d) ), where zd,α ∈ Z and zd,α = 0 can happen only for finitely
many pairs α ∈ R, d ∈ N, we have
 α  α
(5.1) Ψ(ρ,φ) ( zd,α Z(a(1, d)(ν ρ)
)) = zd,α χ(1, d)(ν φ)
.
d∈N,α∈R d∈N,α∈R

F
Note that the homomorphism Ψ(ρ,φ) does not respect the natural gradations of R(ρ) ⊆ RF
and RC . Nevertheless, Ψ(ρ,φ) has the following interesting property:
5.1. Lemma. Let n, d ∈ N and ρ ∈ C. Suppose n ≤ d. Then for any α ∈ R,
α α α
Z(a(n, d)(ν ρ) ) ∈ R(ρ)
F
and we have Ψ(ρ,φ) (Z(a(n, d)(ν ρ) )) = χ(n, d)(ν ρ) .
Proof. We prove the lemma by induction with respect to n.
α
For n = 1, Z(a(1, d)(ν ρ) ) ∈ R(ρ) F
is by the definition of R(ρ) F
. By the definition of
α α
Ψ(ρ,φ) , Ψ(ρ,φ) extends ψ(ρ,φ) , and further ψ(ρ,φ) (Z(a(1, d)(ν ρ) )) = χ(1, d)(ν φ) , for d ≥ 1
and α ∈ R. Thus, the statement of the lemma holds for n = 1.
Let n = 2. Assume d ≥ 2. Then we have from Theorem 3.2
α
Z(a(2, d)(ν ρ)
)
1/2 α −1/2 α α α
=Z(a(1, d)(ν ν ρ)
) × Z(a(1, d)(ν ν ρ)
) − Z(a(1, d + 1)(ν ρ)
) × Z(a(1, d − 1)(ν ρ)
).
α
This shows Z(a(2, d)(ν ρ)
) ∈ R(ρ)
F
. From Proposition 4.1 we have

α
χ(2, d)(ν φ)

1/2 α −1/2 α α α
=χ(1, d)(ν ν φ)
× χ(1, d)(ν ν φ)
− χ(1, d + 1)(ν φ)
× χ(1, d − 1)(ν φ)
.

Thus, since d ≥ 2 implies d − 1, d, d + 1 ≥ 1; we have from the above formulas

α 1/2 α −1/2 α
Ψ(ρ,φ) (Z(a(2, d)(ν ρ)
) = Ψ(ρ,φ) (Z(a(1, d)(ν ν ρ)
)) × Ψ(ρ,φ) (Z(a(1, d)(ν ν ρ)
))
α α
− Ψ(ρ,φ) (Z(a(1, d + 1)(ν ρ)
)) × Ψ(ρ,φ) (Z(a(1, d − 1)(ν ρ)
))
1/2 α −1/2 α α α α
= χ(1, d)(ν ν φ)
×χ(1, d)(ν ν φ)
− χ(1, d + 1)(ν φ)
× χ(1, d − 1)(ν φ)
= χ(2, d)(ν φ)
.
α α
Therefore, the statement holds for n = 2. Note that Ψ(ρ,φ) (Z(a(2, 1)(ν ρ) )) = χ(2, 1)(ν φ) .
Let n ∈ Z, n ≥ 2 and suppose that the statement of the lemma holds for n ≤ n. We
shall show now that it holds for n + 1. Fix some d ∈ Z, d ≥ n + 1 and α ∈ R. First, the
 CHARACTERS OF UNITARY REPRESENTATIONS 11

description of the composition series of representations ζ(a), a ∈ M (S), in Theorem 7.1 of

α
[Z], implies Z(a(n + 1, d)(ν ρ) ) ∈ R(ρ)
F
. Theorem 3.2 implies that we have in RF

1/2 α −1/2 α
Z(a(n, d)(ν ν ρ)
) × Z(a(n, d)(ν ν ρ)
)
α α α α
=Z(a(n + 1, d)(ν ρ)
) × Z(a(n − 1, d)(ν ρ)
) + Z(a(n, d + 1)(ν ρ)
) × Z(a(n, d − 1)(ν ρ)
).

Applying Ψ(ρ,φ) to the above identity, and using the inductive assumption (since d ≥ n,
d ≥ n − 1, d + 1 ≥ n and d − 1 ≥ n), we get
1/2 α −1/2 α
χ(n, d)(ν ν ρ)
× χ(n, d)(ν ν ρ)

α α α α
=Ψ(ρ,φ) (Z(a(n + 1, d)(ν ρ)
)) × χ(n − 1, d)(ν ρ)
+ χ(n, d + 1)(ν ρ)
× χ(n, d − 1)(ν ρ)
.

Proposition 4.1 implies

1/2 α −1/2 α
χ(n, d)(ν ν φ)
× χ(n, d)(ν ν φ)

α α α α
=χ(n + 1, d)(ν φ)
× χ(n − 1, d)(ν φ)
+ χ(n, d + 1)(ν φ)
× χ(n, d − 1)(ν φ)
.

Subtracting the above two formulas, we get

 α α  α
Ψ(ρ,φ) (Z(a(n + 1, d)(ν ρ)
)) − χ(n + 1, d)(ν φ)
× χ(n − 1, d)(ν φ)
= 0.

Since RC is a polynomial ring, it has no non-trivial zero divisors. Clearly, χ(n−1, d)(ν φ) =
α

α α
0. Thus Ψ(ρ,φ) (Z(a(n + 1, d)(ν ρ) )) = χ(n + 1, d)(ν φ) . This finishes the proof of the
lemma.

The last lemma and formulas (4.1) and (5.1) imply directly the following
5.2. Corollary. For 1 ≤ n ≤ d we have

 
n
(τi +τw(i) )/2
Z(a(n, d) (ρ)
) = (−1) sgn(w)
Z(a(1, τi − τw(i) + d)(ν ρ)
).

w∈Wn i=1

We shall write the above formula in a slightly different way:


Z(a(n, d)(ρ) ) = (−1)sgn(w)
w∈Wn

n
((τi +(d−1)/2)+(τw(i) −(d−1)/2))/2
Z(a(1, (τi + (d − 1)/2) − (τw(i) − (d − 1)/2) + 1)(ν ρ)
)
i=1
 
n
= (−1) sgn(w)
s([ν τw(i) −(d−1)/2 ρ, ν τi +(d−1)/2 ρ])
w∈Wn i=1
12 MARKO TADIĆ

by (3.3). Writing the last formula for ρ̃ and then taking the contragredient, gives

 
n
Z(a(n, d)(ρ) ) = (−1)sgn(w) s([ν −τi −(d−1)/2 ρ, ν −τw(i) +(d−1)/2 ρ])
w∈Wn i=1
 
n
= (−1)sgn(w) s([ν −(n+d)/2+i ρ, ν (d−n)/2+w(i)−1 ρ])
w∈Wn i=1

(here we have used the formula for the contragredient representation in the Zelevinsky
classification, Theorem 7.1 of [Z]). Thus

(5.2) ν (n+d)/2 Z(a(n, d)(ρ) ) = Z(([νρ, ν d ρ], [ν 2 ρ, ν d+1 ρ], . . . , [ν n ρ, ν d+n−1 ρ]))
 
n
sgn(w)
= (−1) s([ν i ρ, ν w(i)+(d−1) ρ]).
w∈Wn i=1

(d)
5.3. Theorem. Let n, d ∈ N. Denote Wn = {w ∈ Wn ; w(i) + d ≥ i for any 1 ≤ i ≤ n}.
Then we have

ν (n+d)/2 Z(a(n, d)(ρ) ) = Z(([νρ, ν d ρ], [ν 2 ρ, ν d+1 ρ], . . . , [ν n ρ, ν d+n−1 ρ]))

 
n
sgn(w)
= (−1) s([ν i ρ, ν w(i)+(d−1) ρ]),
w∈Wn
(d) i=1

where in above formula we take the terms of the form s([ν i ρ, ν j ρ]) with i > j to be the
identity of RF (i.e., they do not show up in the above formula).
Proof. For a segment ∆ = [σ, ν k σ] ∈ S set ∆− = [σ, ν k−1 σ] (if k = 0, then set ∆− = ∅).
Denote by D the endomorphism of the ring RF which satisfy D(s(∆)) = s(∆) + s(∆− )
(we take s(∅) to be identity of RF ). Then D is positive. See [Z] for more details regarding

D. One can easily see that D is injective. Further, let x ∈ RF , x = 0. Write D(x) = yn
where yn ∈ RnF . Chose m so that ym = 0 and yn = 0 for n < m. Then we call ym
the highest derivative of x and denote by h.d.(x). Note that h.d.(x1 × x2 ) = h.d.(x1 ) ×
h.d.(x2 ).
We shall prove now the theorem. Note that one needs to prove the formula only for
1 ≤ d ≤ n − 1. Fix such d and assume that the formula holds for d + 1 and n. So,

ν (n+d+1)/2 Z(a(n, d + 1)(ρ) ) = Z(([νρ, ν d+1 ρ], [ν 2 ρ, ν d+2 ρ], . . . , [ν n ρ, ν d+n ρ]))
 
n
sgn(w)
= (−1) s([ν i ρ, ν w(i)+d ρ]).
w∈Wn
(d+1) i=1

Suppose that ρ is a representation of GL(m, F ).

The highest derivative of the left-hand side is

ν (n+d)/2 Z(a(n, d)(ρ) ) = Z(([νρ, ν d ρ], [ν 2 ρ, ν d+1 ρ], . . . , [ν n ρ, ν d+n−1 ρ]))

 CHARACTERS OF UNITARY REPRESENTATIONS 13

([Z], Theorem 8.1). The degree is ndm.

(d+1) n n
For w ∈ Wn consider h.d.( i=1 s([ν i ρ, ν w(i)+d ρ])) = i=1 s([ν i ρ, ν w(i)+d−1 ρ]). If
(d) n
i ≤ w(i) + d for all 1 ≤ i ≤ n (i.e., if w ∈ Wn ), then the degree is i=1 (w(i) + d − i)m =
n (d+1) (d)
nmd. In general, the degree is m i=1 max{w(i) + d − i, 0}. So if w ∈ Wn \Wn , then
w(i0 ) + d − i0 ≤ −1 for some i0 ∈ {1, 2, . . . , n}. Then the degree is

n 
n
m max{w(i) + d − i, 0} > m w(i) + d − i = nmd.
i=1 i=1

This finishes the proof of the formula.

5.4. Theorem. Let n, d ∈ N and ρ ∈ C. Then
ν (n+d)/2 L(a(n, d)(ρ) ) = L(([νρ, ν d ρ], [ν 2 ρ, ν d+1 ρ], . . . , [ν n ρ, ν d+n−1 ρ]))
 
n
sgn(w)
= (−1) δ([ν i ρ, ν w(i)+(d−1) ρ]).
w∈Wn
(d) i=1

Proof. Applying the Zelevinsky involution to the formula in Theorem 5.3, one gets the
above formula, using Lemma 3.3.

At the first moment, our formulas for characters of irreducible unitarizable representa-
tions do not seem very similar to the formula for the Steinberg character. We shall show
now that the formula for the Steinberg character of Gn is a special case of our formulas.
Fix n ∈ N and ρ ∈ C.
By definition, w ∈ (Wn )−1 if and only if w(i) ≤ i + 1 for all 1 ≤ i ≤ n. Each
(1)

permutation can be written as a product of disjoint cycles (we shall write also cycles of
length 1). It is easy to write all such w. They are in bijection with all possible sequences
1 ≤ a1 < a2 < · · · < ak < n (k ∈ {0, 1, 2, . . . , n − 1}). To a such sequence is attached the
following element of (Wn )−1 :
(1)

(1 2 . . . a1 ) (a1 +1 a1 +2 . . . a2 ) . . . (ak−1 +1 ak−1 +2 . . . ak ) (ak +1 ak +2 . . . n),

written as a product of cycles. Note that the parity of the above permutation is (−1)n−k−1 .
Since (- -+1 . . . m−1 m)−1 = (m m−1 . . . -+1 -) for -, m ∈ N, - ≤ m, the formula in
Theorem 5.3 gives for d = 1
ν (n+1)/2 Z(a(n, 1)(ρ) ) = Z((νρ, ν 2 ρ, . . . , ν n ρ))
 
k
= (−1) n−k−1
s([ν ai +1 ρ, ν ai+1 ρ]),
1≤a1 <a2 <···<ak <n i=0

where we take in the formula a0 = 0 and ak+1 = n. Taking for ρ the unramified character
ν −(n+1)/2 , one gets the formula for the Steinberg character of Gn (see [Ca1]).
(d) (d)
5.5. Remark. Obviously, for d ≥ n − 1, Wn = Wn and card Wn = n!. An easy counting
(d) (d)
of permutations gives card Wn = d!(d + 1)n−d for 1 ≤ d ≤ n. For 1 ≤ d ≤ n − 2, Wn is
not a subgroup of Wn .
14 MARKO TADIĆ

6. On multiplicities related to characters (non-archimedean case)

In this section we assume that the field F is non-archimedean.
6.1. Proposition. Let n, d ∈ N and ρ ∈ C. Then ζ(a(n, d)(ρ) ) is a multiplicity one
representation, if and only if d = 1 or n ≤ 3. Further, λ(a(n, d)(ρ) ) is a multiplicity one
representation, if and only if d = 1 or n ≤ 3.
Proof. In the proof we shall consider only the case of ζ(a(n, d)(ρ) ). The other case follows
from this case using the fact that (ζ(a(n, d)(ρ) ))t = λ(a(n, d)(ρ) ) in R, and the positivity
of the Zelevinsky involution.
If d = 1, then we are in the regular situation where we always have multiplicity one. If
n = 1, then there is nothing to prove. The case n = 2 follows from Proposition 4.6 of [Z]
(see also Theorem 3.2). It remains to see the multiplicity one for n = 3 and d ≥ 2.
Using the highest derivatives, it is enough to prove the multiplicity one for ζ(a(3, 2)(ρ) )
(one proves this in the same way as Lemma 6.3 of [T5]). If a ∈ M (S) and a ≤ a(3, 2)(ρ) ,
then a consists of two or three segments. There are precisely six possibilities for such
a. Four of them consist of three segments each. Looking at the highest derivatives, the
representations corresponding to these segments have multiplicity one since ζ(a(3, 1)(ρ) ) is a
multiplicity one representation. The fifth representation is s(a(1, 4)(ρ) )×s(a(1, 2)(ρ) ). This
representation corresponds to a minimal among the six possible. Now (ii) of Proposition
3.5 in [T6] implies that it has multiplicity one. It remains to consider Z(a(2, 3)(ρ) ). From
Z(a(2, 3)(ρ) ) = s([ν −3/2 ρ, ν 1/2 ρ])×s([ν −1/2 ρ, ν 3/2 ρ])−s([ν −3/2 ρ, ν 3/2 ρ])×s([ν −1/2 ρ, ν 1/2 ρ])
we can compute the derivative. First

D(s([ν −3/2 ρ, ν 3/2 ρ]) × s([ν −1/2 ρ, ν 1/2 ρ])) = s([ν −3/2 ρ, ν 3/2 ρ]) × s([ν −1/2 ρ, ν 1/2 ρ])+
s([ν −3/2 ρ, ν 1/2 ρ]) × s([ν −1/2 ρ, ν 1/2 ρ])+
s([ν −3/2 ρ, ν 3/2 ρ]) × ν −1/2 ρ + s([ν −3/2 ρ, ν 1/2 ρ]) × ν −1/2 ρ.

Subtracting this from D(s([ν −3/2 ρ, ν 1/2 ρ]) × s([ν −1/2 ρ, ν 3/2 ρ])), we get

D(Z(a(2, 3)(ρ) )) = Z(a(2, 3)(ρ) ) + Z(([ν −3/2 ρ, ν −1/2 ρ], [ν −1/2 ρ, ν 3/2 ρ]))+
Z(([ν −3/2 ρ, ν −1/2 ρ], [ν −1/2 ρ, ν 1/2 ρ])).

The multiplicity of Z(([ν −3/2 ρ, ν −1/2 ρ], [ν −1/2 ρ, ν 1/2 ρ])) in

s([ν −3/2 ρ, ν −1/2 ρ]) × ν −1/2 ρ × ν 1/2 ρ

is one. This follows from the proof of Proposition 11.4 of [Z] (there, π(a ) is a multiplicity
one representation of length 3). Therefore, the multiplicity of

Z(([ν −3/2 ρ, ν −1/2 ρ], [ν −1/2 ρ, ν 1/2 ρ]))

in D(ζ(a(3, 2)(ρ) )) is one (i.e., Z(([ν −3/2 ρ, ν −1/2 ρ], [ν −1/2 ρ, ν 1/2 ρ])) ≤ D(ζ(a(3, 2)(ρ) ), but
2Z(([ν −3/2 ρ, ν −1/2 ρ], [ν −1/2 ρ, ν 1/2 ρ])) ≤ D(ζ(a(3, 2)(ρ) ) in R). This implies the multiplic-
ity one of Z(a(2, 3)(ρ) ) in ζ(a(3, 2)(ρ) ).
 CHARACTERS OF UNITARY REPRESENTATIONS 15

To finish the proof, we need to prove that ζ(a(n, d)(ρ) ) is not a multiplicity one rep-
resentation if d ≥ 2 and n ≥ 4. For this it is enough to prove that ζ(a(4, d)(ρ) ) is not a
multiplicity one representation for d ≥ 2. Using the highest derivatives, it is enough to see
that ζ(a(4, 2)(ρ) ) is not a multiplicity one representation. We shall show that Z(a(2, 4)(ρ) )
has at least multiplicity two in ζ(a(4, 2)(ρ) ). We have in R the following equalities

ζ(a(4, 2)(ρ) ) = s([ν −2 ρ, ν −1 ρ]) × s([ν −1 ρ, ρ]) × s([ρ, ν ρ]) × s([ν ρ, ν 2 ρ])
 
= s([ν −2 ρ, ν −1 ρ]) × s([ν −1 ρ, ρ]) × Z([ρ, ν ρ], [ν ρ, ν 2 ρ]) + s([ρ, ν 2 ρ]) × ν ρ .

From this one gets easily that in R we have

ζ(a(4, 2)(ρ) ) ≥ Z([ν −2 ρ, ν −1 ρ], [ν −1 ρ, ρ]) × Z([ρ, ν ρ], [ν ρ, ν 2 ρ])+

s([ν −2 ρ, ν −1 ρ]) × s([ν −1 ρ, ρ]) × s([ρ, ν 2 ρ]) × ν ρ.

We know

Z([ν −2 ρ, ν −1 ρ], [ν −1 ρ, ρ]) × Z([ρ, ν ρ], [ν ρ, ν 2 ρ]) =

L([ν −2 ρ, ν −1 ρ], [ν −1 ρ, ρ]) × L([ρ, ν ρ], [ν ρ, ν 2 ρ])

(see Theorem 3.2). Thus Z(a(2, 4)(ρ) ) = L(a(4, 2)(ρ) ) is a subquotient of

Z([ν −2 ρ, ν −1 ρ], [ν −1 ρ, ρ]) × Z([ρ, ν ρ], [ν ρ, ν 2 ρ]).

One has directly a(2, 4)(ρ) ≤ ([ν −2 ρ, ν −1 ρ], [ν −1 ρ, ρ], [ρ, ν 2 ρ], ν ρ). Now we see that the
multiplicity of Z(a(2, 4)(ρ) ) in ζ(a(4, 2)(ρ) is at least two. This finishes the proof of the
proposition.

Now we have a direct consequence of the above proof:
6.2. Corollary. Let n, d ∈ N, d ≥ 2, and ρ ∈ C. Then the representations ζ(a(4n, d)(ρ) )
and λ(a(4n, d)(ρ) ) have irreducible subquotients which have multiplicities at least 2n in
that representations.

6.3. Remark. One delicate point in the solution of the unitarizability problem in [T4] was
the proof that Z(a(n, d)(ρ) )’s are prime elements in R (Proposition 3.5 of [T4]). We present
here one very simple proof of this fact, based on the Gelfand-Kazhdan derivatives. The
proof here proceeds by induction on d. We shall assume that ρ is a representation of some
GL(p, F ). For n = 1, Z(a(n, 1)(ρ) ) is an essentially square integrable representation. Since
R is a polynomial ring over all essentially square integrable representations, Z(a(n, 1)(ρ) )
is prime. Suppose that we have proved that Z(a(n, d)(ρ) ) is prime for some d ≥ 1. Assume
that Z(a(n, d + 1)(ρ) ) = P1 × P2 . for some P1 , P2 ∈ R. For the proof, it is enough to
show that P1 or P2 must be in {±1}. Since R is integral and graded, P1 and P2 are
homogeneous with respect to the grading. Now on the level of the highest derivatives
−1/2
we have Z(a(n, d)(ν ρ)
) = h.d.(P1 ) × h.d.(P2 ). Thus, h.d.(P1 ) or h.d.(P2 ) is ±1 by the
inductive assumption. Assume h.d.(P2 ) ∈ {±1}. Without lost of generality we can assume
16 MARKO TADIĆ

−1/2
it is 1. Then h.d.(P1 ) = Z(a(n, d)(ν ρ)
). The theorem 8.1 of [Z] implies that if σ is an
irreducible representation of GL(m, F ) with m ≤ n(d + 1)p, whose highest derivative is
−1/2
Z(a(n, d)(ν ρ)
), then σ = Z(a(n, d + 1)(ρ) ). Thus m = n(d + 1)p, and the degree of P1
is m = n(d + 1)p. Therefore, the degree of P2 is 0. Since Z(a(n, d + 1)(ρ) ) is a primitive
element of the free abelian group R, P2 is also primitive. This implies P2 = 1. This finishes
the proof.

Appendix: Composition series of the ends

of the complementary series of GL(2n, C)
β β
For d ∈ Z and β ∈ C, we have denoted by χ(d)(ν ) the character χ(d)(ν ) (z) =
(z/|z|)d ν β (z) of C× . For x, y ∈ C such that x − y ∈ Z we define a character γ(x, y)
of C× by
(x+y)/2
γ(x, y)(z) = (z/|z|)x−y |z|x+y = χ(x − y)(ν )
.
Then β
χ(d)(ν )
= γ(β + d/2, β − d/2),
for d ∈ Z, β ∈ C. Note that γ(x, y) = γ(x , y  ) implies x = x and y = y  . Further

γ(x, y)γ(x , y  ) = γ(x + x , y + y  ),

0
γ(d/2, −d/2)(z) = (z/|z|)d = χ(d)(ν ) (z), d ∈ Z,
γ(β, β) = ν β , β ∈ C,
γ(p, q)(z) = z z̄ , p q
p, q ∈ Z.

For characters χ1 , χ2 of C× , the representation χ1 × χ2 is reducible if and only if

χ1 χ−1
2 = γ(p, q) where p, q ∈ Z and pq > 0. Thus, γ(x1 , y1 ) × γ(x2 , y2 ) is reducible if and
only if x1 − x2 ∈ Z and (x1 − x2 )(y1 − y2 ) > 0. In the case of reducibility, we have in the
Grothendieck group:

γ(x1 , y1 ) × γ(x2 , y2 ) = L(γ(x1 , y1 ), γ(x2 , y2 )) + γ(x1 , y2 ) × γ(x2 , y1 ).

For (χ1 , . . . , χn ) ∈ M (D) suppose that there exist 1 ≤ i < j ≤ n such that χi × χj is
reducible. Write χi = γ(xi , yi ), χj = γ(xj , yj ). Denote χi = γ(xi , yj ), χj = γ(xj , yi ). Then
we shall write

(χ1 , . . . , χi−1 , χi , χi+1 , . . . , χj−1 , χj , χj+1 , . . . , χn )

≺ (χ1 , . . . , χi−1 , χi , χi+1 , . . . , χj−1 , χj , χj+1 , . . . , χn ).

Generate by ≺ a minimal partial order ≤ on M (D). If (χ1 , . . . , χn ), (χ1 , . . . , χn ) ∈ M (D),

then L(χ1 , . . . , χn ) is a composition factor of χ1 × · · · × χn if and only if (χ1 , . . . , χn ) ≤
(χ1 , . . . , χn ).
Now we shall rewrite the formula in Proposition 4.3 for the trivial character 1GL(n,C) of
GL(n, C). We have in R
 CHARACTERS OF UNITARY REPRESENTATIONS 17
    n−3 n−3   
1GL(n,C) = χ(n, 0)(1C× ) = L γ n−1
2 , n−1
2 , γ 2 , 2 , . . . , γ − 2 ,− 2
n−1 n−1


= L (γ(τ1 , τ1 ), γ(τ2 , τ2 ), . . . , γ(τn , τn )) = (−1)sgn(w) γ(τ1 , τw(1) ) × · · · × γ(τn , τw(n) )
w∈Wn
 
n  
n
= (−1) sgn(w)
γ(τi , τw(i) ) = (−1) sgn(w)
γ((n − 2i + 1)/2, (n − 2w(i) + 1)/2)
w∈Wn i=1 w∈Wn i=1
 
n
= (−1)sgn(w) γ(i − (n + 1)/2, w(i) − (n + 1)/2)
w∈Wn i=1

(at the end we passed to the contragredients). From this one obtains that for x, y ∈ C
such that x − y ∈ Z, we have

(A.1) L((γ(x + 1, y + 1), γ(x + 2, y + 2), . . . , γ(x + n, y + n)))

 
n
sgn(w)
= (−1) γ(x + i, y + w(i)).
w∈Wn i=1

Set d−1/2 = (γ(−n, 0), γ(−n + 1, 1), . . . , γ(−1, n − 1)),

d1/2 = (γ(−n + 1, 1), γ(−n + 2, 2), . . . , γ(0, n)),
dGL(n+1) = (γ(−n, 0), γ(−n + 1, 1), . . . , γ(−1, n − 1), γ(0, n)),
dGL(n−1) = (γ(−n + 1, 1), . . . , γ(−1, n − 1)),
d− = (γ(−n, 1), γ(−n + 1, 2), . . . , γ(−1, n)),
d+ = (γ(−n + 1, 0), γ(−n + 2, 1), . . . , γ(0, n − 1)).

ν −1/2 χ(n, −n)(ν

0
)
Then = L(d−1/2 ),
0
ν 1/2 χ(n, −n)(ν )
= L(d1/2 ),
0
χ(n + 1, −n)(ν )
= L(dGL(n+1) ),
0
χ(n − 1, −n)(ν )
= L(dGL(n−1) ),
0
χ(n, −n − 1)(ν )
= L(d− ),
0
χ(n, −n + 1)(ν )
= L(d+ ).

In the following lemma we study some elements of R as polynomials in variables D (the

basis of R is λ(d), d ∈ M (D)).
A.1. Lemma. Let n ∈ N, n ≥ 2. Then
(i) Suppose that n is even. Then the monomial λ(d− + d+ ) has in the polynomial
ν −1/2 χ(n, −n)(ν ) × ν 1/2 χ(n, −n)(ν ) = L(d−1/2 ) × L(d1/2 ) coefficient equal to 1,
0 0
18 MARKO TADIĆ

0 0
and in the polynomial χ(n+1, −n)(ν ) ×χ(n−1, −n)(ν ) = L(dGL(n+1) )×L(dGL(n−1) )
coefficient is 0.
(ii) Suppose that n is odd. Then the monomial λ(d− + d+ ) has coefficient 0 in the
polynomial ν −1/2 χ(n, −n)(ν ) × ν 1/2 χ(n, −n)(ν ) = L(d−1/2 ) × L(d1/2 ), and −1 in
0 0

0 0
the polynomial χ(n + 1, −n)(ν ) × χ(n − 1, −n)(ν ) = L(dGL(n+1) ) × L(dGL(n−1) ).
(iii) If n = 3, then there does not exist d ∈ M (D) such that L(d) is unitarizable and
that
d− + d+ < d < d1/2 + d−1/2 (= dGL(n+1) + dGL(n−1) ).

For n = 3 there exists exactly one such d. That one is dGL(4) + (γ(−1, 1), γ(−2, 2)).
(iv) Suppose that n = 3. Then the monomial λ(dGL(4) + (γ(−1, 1), γ(−2, 2))) has co-
efficients equal to −2 in both polynomials ν −1/2 χ(3, −3)(ν ) × ν 1/2 χ(3, −3)(ν ) =
0 0

0 0
L(d−1/2 ) × L(d1/2 ) and χ(4, −3)(ν ) × χ(2, −3)(ν ) = L(dGL(4) ) × L(dGL(2) ).

Proof. First we shall see when the monomial λ(d− + d+ ) can show up in the polynomial
L(d−1/2 ) × L(d1/2 ). We shall use formula (A.1). Suppose that λ(d− + d+ ) is coming from a
monomial L(d−1/2 ) and L(d1/2 ) which correspond to permutations w1 and w2 respectively.
Write d− + d+ = (γ(x1 , y1 ), . . . , γ(x2n , y2n ). Note that yi − xi = n − 1 or n + 1.
Note that there is only one i with xi = −n. This, and the fact that yi − xi = n − 1 or
n + 1 imply w1 (1) = 2. Further, the fact that yi − xi = n − 1 or n + 1 implies w1 (2) = 1
or 3. If w1 (2) = 3, then w1−1 (1) ≥ 3. This shows that γ(−n − 1 + w1−1 (1), 0) appears in
the product. This contradicts the fact that yi − xi = n − 1 or n + 1. Continuing along this
line, one gets that w1 must be the product of cycles (1 2) (3 4) (5 6) . . . . If n is odd, then
we must have γ(−1, n − 1) in the product. This contradicts yi − xi = n − 1 or n + 1. Thus,
λ(d− + d+ ) does not show up in L(d−1/2 ) × L(d1/2 ). If n is even, then the only possibility
for w1 and w2 is w1 = w2 = (1 2) (3 4) (5 6) . . . (n−1 n). This implies that λ(d− + d+ )
has coefficient (−1)n = 1 in the polynomial L(d−1/2 ) × L(d1/2 ).
In a completely analogous way one proves that λ(d− +d+ ) has coefficient 0 in L(dGL(n+1) )
×L(dGL(n−1) ) if n is even, and (−1)n = −1 if n is odd. This finishes the proof of (i) and
(ii).
Take d ∈ M (D) so that L(d) is unitarizable and that d− +d+ < d < dGL(n+1) +dGL(n−1) =
d−1/2 + d1/2 . Write

d = (γ(x1 , y1 ), . . . , γ(x2n , y2n


)), dGL(n+1) + dGL(n−1) = (γ(x1 , y1 ), . . . , γ(x2n , y2n

)).

The first observation is that always n − 1 ≤ yi − xi ≤ n + 1. Only one γ(xi , yi ) has
xi = −n and also only one γ(xj , yj ) has xj = −n. After renumeration, we can assume
i = j = 1. Obviously y1 = −1. Thus y1 = 0 or 1.
Suppose that y1 = 0. Unitarizability of L(d) implies that d = dGL(n+1) + d for some
d ∈ M (S). Write d = (γ(u1 , v1 ), . . . , γ(un−1 , vn−1 ). There is only one γ(ui , vi ) with
ui = −n + 1. Now in the same way as before we conclude vi = 1 or 2. For the possibility
vi = 2 it must be n ≥ 3. If vi = 1, then unitarizability of L(d) implies d = dGL(n−1) .
Thus, d = dGL(n+1) + dGL(n−1) . Therefore in this situation vi = 2. The unitarizability
of L(d) implies d = (γ(−n + 1, 2), . . . , γ(−2, n − 1)) + (γ(−1, 1)). For n > 3, this can
 CHARACTERS OF UNITARY REPRESENTATIONS 19

not hold because it must be n − 1 ≤ vi − ui ≤ n + 1. For n = 3 one easily sees that

d− + d+ < dGL(4) + (γ(−1, 1), γ(−2, 2)) < dGL(4) + dGL(2) .
Suppose now y1 = 1. Then unitarizability of L(d) implies that d = d− + d for some d ∈
M (D). Write d = (γ(u1 , v1 ), . . . , γ(un , vn )). There is only one γ(ui , vi ) with ui = −n + 1.
If vi ≥ 1, then for γ(uj , vj ) for which vj = 0 one has uj ≥ −n + 2. Thus, vj − uj < n − 1.
This can not happen. Therefore, vi = 0. This implies that d = d+ . Thus, d = d− + d+ .
This completes the proof of (iii).
One gets (iv) directly from the formula (A.1).

is a subquotient of ν −1/2 χ(n, −n)(ν ) ×
0 0 0
A.2. Corollary. χ(n, −n−1)(ν ) ×χ(n, −n+1)(ν )
0
ν 1/2 χ(n, −n)(ν ) , of multiplicity one.
Proof. There exist integers pd
≥ 0 (d ∈ M (D)), so that L(d−1/2 )×L(d1/2 ) = L(dGL(n+1) )×

L(dGL(n−1) ) + pd
L(d ), where the sum runs over all d ∈ M (D), d < d−1/2 + d1/2 . We

need to show that pL(d− +d+ ) = 1. Recall that L(d ) = d

∈M (D),d

≤d
qd
,d

λ(d ), for
some qd
,d

∈ Z, where qd
,d
= 1. Thus

(A.2) L(d−1/2 ) × L(d1/2 )

 
= L(dGL(n+1) ) × L(dGL(n−1) ) + pd
qd
,d

λ(d ).
d
∈M (D),d
<d−1/2 +d1/2 d

∈M (D),d

≤d

Suppose n = 3. Now (i) and (ii) of the above lemma imply pd
qd
,d− +d+ = 0, for some
d ∈ M (D) such that d− + d+ ≤ d < d−1/2 + d1/2 . The above lemma ((iii)) implies
that the only possibility for d is d− + d+ . Now (i) and (ii) of the above lemma imply
pd− +d+ qd− +d+ ,d− +d+ = 1. Thus, pd− +d+ = 1.
Suppose now that n = 3. Then pL(dGL(4) )×γ(−1,1)×γ(−2,2) = 0 follows from (A.2) and
(iv) of the above lemma by a similar type of analysis as above. Now we can repeat the
above argument and get pd− +d+ = 1. This finishes the proof.

0 0
The multiplicity of the irreducible representation χ(n + 1, −n)(ν ) × χ(n − 1, −n)(ν )
in ν −1/2 χ(n, −n)(ν ) × ν 1/2 χ(n, −n)(ν ) is one. This follows directly from the fact that
0 0

L(d + d ) is always a subquotient of L(d) × L(d ), for d, d ∈ M (D) (this is Proposition

3.5 of [T3], one can see also the proof of Proposition 2.3 of [T6]; after minor changes this
proof applies also to the archimedean case). As noted by S. Sahi, one can also argue in
0
this case using the spherical representations (one needs to twist the situation by χ(n)(ν ) ).
Finally, Theorem 3C of [Sh1] implies that in R we have

ν −1/2 χ(n, −n)(ν

0 0
)
× ν 1/2 χ(n, −n)(ν )

0 0 0 0
= χ(n + 1, −n)(ν )
× χ(n − 1, −n)(ν )
+ χ(n, −n − 1)(ν )
× χ(n, −n + 1)(ν ) .

Multiplying the above relation with characters, one gets Proposition 4.1.

References
[A] Aubert, A.-M., Dualité dans le groupe de Grothendieck de la catégorie des représentations lisses
de longueur finie d’un groupe réductif p-adique, preprint (1994).
20 MARKO TADIĆ

[Be] Bernstein, J., P-invariant distributions on GL(N ) and the classification of unitary representa-
tions of GL(N ) (non-archimedean case), Lie Group Representations II, Proceedings, University
of Maryland 1982-83, Lecture Notes in Math. 1041, Springer-Verlag, Berlin, 1984, pp. 50-102.
[BoWh] Borel, A. and Wallach, N., Continuous cohomology, discrete subgroups, and representations of
reductive groups, Princeton University Press, Princeton, 1980.
[BuK] Bushnell, C.J. and Kutzko, P., The admissible dual of GL(N ) via compact open subgroups,
Princeton University Press, Princeton, 1993.
[Ca1] Casselman, W., The Steinberg character as a true character, Symp. Pure Math. 26,, Amer.
Math. Soc., Providence, Rhone Island, 1973, pp. 413-417.
[Ca2] Casselman, W., Characters and Jacquet modules, Math. Ann. 230 (1977), 101-105.
[Co] Corwin, L., A construction of the supercuspidal representations of GLn (F ), F p-adic, Tran.
Amer. Math. Soc. 337 (1993), 1-58.
[CoMoSl] Corwin, L., Moy, A. and Sally, P.J., Degrees and formal degrees for division algebras and GLn
over a p-adic field, Pacific J. Math. 141 (1990), no. 1, 21-45.
[D] van Dijk, G., Computation of certain induced characters of p-adic groups, Math. Ann. 199
(1972), 229-240.
[GN] Gelfand, I.M. and Naimark, M.A., Unitare Derstellungen der Klassischen Gruppen (German
translation of Russian publication from 1950), Akademie Verlag, Berlin, 1957.
[J] Jacquet, H, Generic representations, Non-Commutative Harmonic Analysis, Lecture Notes in
Math. 587, Spring Verlag, Berlin, 1977, pp. 91-101.
[K] Kutzko, P., Characters formulas for supercuspdal representations of GL ,  a prime, Amer. J.
Math. 109 (1987), 201-221.
[LRaSu] Laumon, G., Rapoport, M. and Stuhler, U., D-elliptic sheaves and the Langlands correspon-
dence, Invent. Math. 113 (1993), 217-338.
[Mi] Miličić, D., On C ∗ -algebras with bounded trace, Glasnik Mat. 8(28) (1973), 7-21.
[MœWd] Mœglin, C. and Waldspurger, J.-L., Sur l’involution de Zelevinsky, J. reine angew. Math. 372
(1986), 136-177.
[P] Procter, K., The Zelevinsky duality conjecture for GLn , thesis (King’s College, London) (1994).
[Ro] Rodier, F., Représentations de GL(n, k) où k est un corps p-adique, Séminaire Bourbaki no
587 (1982), Astérisque 92-93 (1982), 201-218.
[Sh1] Sahi, S., Jordan algebras and degenerate principal series, preprint.
[Sh2] Sahi, S., Letter.
[Sl] Sally, P.J., Some remarks on discrete series characters for reductive p-adic groups, Representa-
tions of Lie Groups, Kyoto, Hiroshima, 1986, 1988, pp. 337-348.
[Sn] Stein, E.M., Analysis in matrix spaces and some new representations of SL(N, C), Ann. of
Math. 86 (1967), 461-490.
[T1] Tadić, M., Unitary dual of p-adic GL(n), Proof of Bernstein Conjectures, Bulletin Amer. Math.
Soc. 13 (1985), no. 1, 39-42.
[T2] Tadić, M., Proof of a conjecture of Bernstein, Math. Ann. 272 (1985), 11-16.
[T3] Tadić, M., Unitary representations of general linear group over real and complex field, preprint
MPI/SFB 85-22 Bonn (1985).
[T4] Tadić, M., Classification of unitary representations in irreducible representations of general
linear group (non-archimedean case), Ann. Sci. École Norm. Sup. 19 (1986), 335-382.
[T5] Tadić, M., Topology of unitary dual of non-archimedean GL(n), Duke Math. J. 55 (1987),
385-422.
[T6] Tadić, M., Induced representations of GL(n, A) for p-adic division algebras A, J. reine angew.
Math. 405 (1990), 48-77.
[T7] Tadić, M., An external approach to unitary representations,, Bulletin Amer. Math. Soc. 28
(1993), no. 2, 215-252.
[V1] Vogan, D.A., Representations of real reductive groups, Birkhäuser, Boston, 1987.
[V2] Vogan, D.A., The unitary dual of GL(n) over an archimedean field, Invent. Math. 82 (1986),
449-505.
[V3] Vogan, D.A., Letter.
 CHARACTERS OF UNITARY REPRESENTATIONS 21

[Z] Zelevinsky, A.V., Induced representations of reductive p-adic groups II, On irreducible repre-
sentations of GL(n), Ann. Sci. École Norm Sup. 13 (1980), 165-210.

Department of Mathematics, University of Zagreb, Bijenička 30, 41000 Zagreb, Croatia

Current address: Sonderforschungsbereich 170, Geometrie und Analysis, Bunsenstraße 3-5, 37073
Göttingen, Germany