Finite Fields and Their Applications 46 (2017) 255–270

Contents lists available at ScienceDirect

Finite Fields and Their Applications

www.elsevier.com/locate/ffa

n × 1 local gamma factors and Gauss sums ✩

Chufeng Nien a,b,∗,1
a
Department of Mathematics, National Cheng Kung University, Tainan 701,
Taiwan
b
National Center for Theoretical Sciences, Taipei, 106, Taiwan

a r t i c l e i n f o a b s t r a c t

Article history: This paper relates n × 1 local gamma factors over finite
Received 31 August 2016 fields with Gauss sums and characterizes F× q -distinguished
Received in revised form 6 April
2017
characters of F×
q 2 in terms of special values of twisted local
Accepted 10 April 2017 gamma factors.
Available online 27 April 2017 © 2017 Elsevier Inc. All rights reserved.
Communicated by Igor Shparlinski

MSC:
primary 20C33
secondary 11L05

Keywords:
Generic representation
Whittaker models
Gamma factors
Local converse theorem
Distinguished representation
Abelian Gauss sums

✩
The work is supported by MOST 105-2115-M-006-010-MY2.
* Correspondence to: Department of Mathematics, National Cheng Kung University, Tainan 701, Taiwan.
E-mail address: nienpig@mail.ncku.edu.tw.
1
I am very thankful to the reviewers for their valuable comments and suggestions, which improved this
paper in a great deal.

http://dx.doi.org/10.1016/j.ffa.2017.04.005
1071-5797/© 2017 Elsevier Inc. All rights reserved.
256 C. Nien / Finite Fields and Their Applications 46 (2017) 255–270

1. Introduction

Let F denote a p-adic field and Fq a finite field of q elements. For a finite group G,
denote by G  the set of isomorphism classes of irreducible representations of G.
In representation theory, it is natural to ask how much invariants of an irreducible
representation can tell us about the nature of the representation. One might ask for a
(minimal) set of invariants which determine a representation uniquely (up to equiva-
lence); or, more generally, for a set of invariants which together determine whether the
representation belongs to some particular class of representations.
In the case of the group GLn (Fq ), taking inspiration from the Local Converse Theorem
for p-adic groups, in [11] the author considered this question in terms of invariants known
as twisted gamma factors. More precisely, the author proved that an irreducible cuspidal
representation π of GLn (F ), with n  2, is determined (up to isomorphism) by its central
character together with the set of twisted gamma factors
  n 
γ(π × τ, ψ) | τ ∈ Gt , 1  t  , (1.1)
2

where Gt is the set of irreducible generic representations of GLt (Fq ), and ψ is a fixed
non-trivial additive character of Fq .
This is precisely an analogue of the so-called Jacquet Conjecture for p-adic groups,
recently proved by Jacquet–Liu in [9] and, independently, by Chai in [2].
In this paper we prove two further results on twisted gamma factors. The first relates
an n × 1 gamma factor to an abelian Gauss sum. Write N for the reduced norm map
from Fqn to Fq , and tr for the reduced trace.

Theorem 1.1. Let π be an irreducible cuspidal representation of GLn (Fq ), with n  2

and write ηχ for the regular character of F×
q n corresponding to π in Green’s construction.
Then for any character of F× q

γ(π × χ, ψ)

= (−q −1 χ(−1))n−1 ψ(trσ −1 )ηπ (σ)χ(N (σ)).
σ∈F×
qn

We hope to be able to use this theorem to help determine whether the bound [ n2 ]
in (1.1) is sharp: one could check whether, for n = 4, 5, the set {γ(π×χ, ψ) | χ a character
of F×q } alone is sufficient to determine a unique cuspidal representation π (with given
central character) of GLn (Fq ). This should also shed light on the analogous question
for p-adic groups, where Adrian–Liu–Stevens–Tam in [1] claimed that the bound [ n2 ] is
indeed sharp when n is prime but there are no current results for composite n.
The second result concerns the class of distinguished representations. Put Gn =
GLn (Fq2 ) and Hn = GLn (Fq ). Recall that an irreducible representation π of Gn is
said to be Hn -distinguished if
 C. Nien / Finite Fields and Their Applications 46 (2017) 255–270 257

HomHn (π, 1) = {0},

where 1 denotes the trivial representation of Hn .

Converse results for distinguished representations in the p-adic case, due to Ok [10] and
strengthened by Hakim–Offen [8], suggest that one should be able to determine whether
a cuspidal irreducible representation π of GLn (E) is GLn (F )-distinguished by its twisted
gamma factors by GLr (F )-distinguished representations of GLr (E) only, where E is a
quadratic extension of F . More precisely, the symmetry properties of the Bessel function
(see Proposition 2.6) in the description of the twisted gamma factor suggest the following
conjecture, as well as its analogue for p-adic groups, where Ψ denotes a fixed additive
character of Fq2 which is trivial on Fq .

Conjecture 1.2 (n × [ n2 ] converse problem on distinction). For n  2, let π be an irre-

ducible cuspidal representation of Gn whose central character restricts trivially to F× q .
If γ(π × ρ, Ψ) = q −r(n−r−1) for all GLr (Fq )-distinguished, generic representations ρ of
GLr (Fq2 ), for r = 1, . . . , [ n2 ], then π is GLn (Fq )-distinguished.

In future work, the author hopes to prove this conjecture, having already verified the
case n = 2. Here, we prove an analogue in the case n = 1:

Theorem 1.3 (1 × 1 converse theorem on distinction). Suppose that ρ is a non-trivial

character of F× ×
q 2 . Then ρ is Fq -distinguished if and only if γ(ρχ, Ψ) = 1, for all
× −1 ×
Fq -distinguished χ = ρ of Fq2 .

2. Notation and known results

2.1. Whittaker models

For the rest of the paper, all groups are considered over a finite field Fq unless state
otherwise.
We fix, once and for all, a non-trivial additive character ψ of Fq . Let Un be the
unipotent radical of the standard Borel subgroup Bn of GLn , and Zn the center of GLn .
Denote by Pn the mirabolic subgroup of GLn , consisting of matrices in GLn with last
row equal to (0, . . . , 0, 1). A character ψ  of Un is called non-degenerate if
n−1

ψ  (u) = ψ ai ui,i+1 , for u = (ui,j ) ∈ Un ,
i=1

for some ai ∈ F×
q . Also, we denote by ψn , the standard non-degenerate character given
by
n−1

ψn (u) = ψ ui,i+1 , for u = (ui,j ) ∈ Un .
i=1
258 C. Nien / Finite Fields and Their Applications 46 (2017) 255–270

Let π be an irreducible representation of GLn . Given a non-degenerate character ψ 

of Un , we call π ψ  -generic if

Un ψ ) = 0.
dim HomGLn (π, IndGLn

It is well known that if π is ψ  -generic, then it is also ψ  -generic for any other non-
degenerate character ψ  of Un . Therefore, we will use the term “genericity” instead of
“ψ  -genericity” from now on.
In the following, we recall some essential results about Whittaker models.

Theorem 2.1 (Multiplicity one or uniqueness of Whittaker model, [5]). Let π be an irre-
ducible representation of GLn . Then

Un ψn )  1.
dim HomUn (π|Un , ψn ) = dim HomGLn (π, IndGLn

When π is generic, the above Hom-space is of dimension one. Let ψn ∈ HomUn (π|Un ,
ψn ) be a nonzero Whittaker functional of π. Then for v ∈ Vπ , define Wv (g) := ψn (π(g)v),
which is called the Whittaker function attached to the vector v and belongs to the induced
representation IndGLUn ψn . By Theorem 2.1, the subspace generated by all Whittaker
n

functions Wv (g) is unique and will be denoted by W(π, ψn ). This space is called the
Whittaker model of π.
For Wv ∈ W(π, ψn ), we denote Wv the function on GLn given by

Wv (g) = Wv (wn (t g −1 )), g ∈ GLn ,

where wn is the longest Weyl element of GLn , with 1’s on the second diagonal and zeros
elsewhere. Then

Wv ∈ W(π̃, ψn−1 ),

where π̃ denotes the representation contragredient to π. Moreover,


(Rg · Wv )(x) := Wv (xg) = Wπ( t g −1 )·v (x), for g, x ∈ GLn (Fq ),

where Rg denotes the right-translation operator by g on functions. Also, denote by

Wv  = Rwn · Wv . (2.1)

It is known that any irreducible cuspidal representation of GLn is generic.

2.2. Twisted gamma factors over finite fields

Roditty in her thesis worked on a finite field analogue of the local converse theorem and
established the following functional equations. Note that in the theorem, the conditions
of cuspidality and n > t are imposed.
 C. Nien / Finite Fields and Their Applications 46 (2017) 255–270 259

Theorem 2.2 ([12, Theorem 5.1 and 5.4] or [11, Theorem 2.10]). Let π be an irre-
ducible cuspidal representation of GLn (Fq ) and τ an irreducible generic representation
of GLt (Fq ), with n > t. Then there exists a complex number γ(π × τ, ψ) such that

  m 0 0
γ(π × τ, ψ)q tj Wπ x In−t−j−1 0 Wτ (m)
m∈Ut \GLt x∈Mn−t−j−1,t 0 0 Ij+1

  0 In−t−j 0
= Wπ 0 0 Ij Wτ (m),
m∈Ur \GLt y∈Mt,j m 0 y

for all 0  j  n − t − 1, Wπ ∈ W(π, ψn ), Wτ ∈ W(τ, ψt−1 ).

Corollary 2.3. γ(π × τ, ψ)γ(π × τ , ψ −1 ) = q −t(n−t−1) .

Proof. By applying the above functional equation twice,

γ(π × τ, ψ)q tj γ(π × τ , ψ −1 )q t(n−t−j−1) = 1. 2

Definition 2.4. Let π be an irreducible generic representation of GLn (Fq ). We call a

function B a Bessel function of π with respect to ψn , if B satisfies the following two
conditions:

(1) B ∈ W(π, ψn ).
(2) B(u1 gu2 ) = ψn (u1 u2 )B(g), for all g ∈ GLn (Fq ), u1 , u2 ∈ Un .

Furthermore, if B is normalized so that B(In ) = 1, then B is called a normalized Bessel

function of π with respect to ψn .

We denote by Bπ,ψn the unique normalized Bessel function of π with respect to ψn .

An explicit expression for the normalized Bessel function of an irreducible, generic rep-
resentation of GLn (Fq ) is given by the following.

Proposition 2.5 ([3, Proposition 4.5] or [12, Lemma 6.1.1]). Let π be an irreducible
generic representation of GLn (Fq ) and χπ its character. Define

−1

B(g) = |Un (Fq )| ψn (u−1 )χπ (gu), for g ∈ GLn (Fq ).
u∈Un

Then B = Bπ,ψn .

Moreover, normalized Bessel functions possess nice symmetry, which is the key to
verifying the finite field analogue of Jacquet’s conjecture on the local converse theorem
in [11].
260 C. Nien / Finite Fields and Their Applications 46 (2017) 255–270

Proposition 2.6 ([12, Proposition 6.1.2]). Let π be an irreducible generic representation

of GLn (Fq ). Then

Bπ,ψn (g −1 ) = Bπ,ψn (g), g ∈ GLn (Fq ).

Let n = n1 + n2 + · · · + nk be a partition of n into positive terms, and λi ∈ F×

q , for
i = 1, . . . k. Denote by
⎛ ⎞
0 ... 0 λ1 I n 1
⎜ ..
..
.
..
. ⎟
⎜ . 0 ⎟
gn1 ,...,nk (λ1 , . . . , λk ) = ⎜ ⎟. (2.2)
⎝ ..
.
..
. .. ⎠
0 .
λk Ink 0 ... 0

Let An be the Levi subgroup of the standard Borel subgroup Bn of GLn , consisting of
diagonal matrices in GLn (Fq ). Let W (GLn ) denote the Weyl group of GLn .

Lemma 2.7 ([4]). Let π be an irreducible, generic representation of GLn (Fq ) and B its
normalized Bessel function. If B(aw) = 0, for a ∈ An , w ∈ W (GLn ), then

aw = gn1 ,...,nk (λ1 , . . . , λk ) for some λi ∈ F×

q and ni  0, i = 1, . . . , k.

In particular, if B(p) = 0, for some p ∈ Pn , then p ∈ Un .

Proposition 2.8 ([12, Lemma 6.1.4]). Let π be an irreducible cuspidal representation of

GLn (Fq ) and τ be an irreducible generic representation of GLr (Fq ), r < n. Then
  
0 In−r
γ(π × τ, ψ) = Bπ,ψ Bτ,ψ−1 (m).
m 0
m∈Ur \GLr

Lemma 2.9. Let π be an irreducible generic representation of GLn (Fq ) and B be the
normalized Bessel function of π with respect to ψn . Denote by B̄ the composition of
complex conjugation with B. Then B s is the normalized Bessel function for the irreducible
generic representation π̃ with respect to ψn−1 .

Proof. Recall in Eq. (2.1)

B  (g) = (Rwn · B̃)(g) = B(wn (t g −1 )wn ) for g ∈ GLn (Fq ).

Since

B  (u1 gu2 )
= B([wn (t u−1 t −1
1 )wn ][wn ( g )wn ][wn (t u−1
2 )wn ])

= ψn (wn (t u−1 t −1 t −1
1 )wn )ψn (wn ( u1 )wn )B(wn ( g )wn )
= ψn−1 (u1 )ψn−1 (u2 )B  (g), for all u1 , u2 ∈ Un (Fq ), g ∈ GLn (Fq ),
 C. Nien / Finite Fields and Their Applications 46 (2017) 255–270 261

and B  ∈ W(π̃, ψn−1 ), then B  = Bπ,ψn−1 by the uniqueness of the normalized Bessel
functions. Therefore, by Lemma 2.7, the support of B  is contained in Un (Fq )AUn (Fq ),
where A is the set of elements of the form gn1 ,...,nk (λ1 , . . . , λk ) defined in Eq. (2.2). Then

B  (gn1 ,...,nk (λ1 , . . . , λk )) = B(wn (t gn1 ,...,nk (λ1 , . . . , λk )−1 )wn )

= B(gn1 ,...,nk (λ1 , . . . , λk )−1 )
= B(g
s n ,...,n (λ1 , . . . , λk )),
1 k

where the last equality follows Proposition 2.6. Note that B

s also satisfies conditions with
respect to ψn−1 in Definition 2.4 and the support of B̄ is in Un (Fq )AUn (Fq ). Therefore
B = Bs is the normalized Bessel function for π̃ with respect to ψ −1 . 2
n

Lemma 2.10. Let π be an irreducible cuspidal representation of GLn (Fq ) and χ a char-
acter of F×
q . Then

γ(π × χ, ψ) = χ(−1)n−1 γ((π ⊗ χ) × 1, ψ),

where π ⊗ χ := π ⊗ χ ◦ det.

Proof. By Proposition 2.8,

γ(π × χ, ψ)
  
0 In−1
= Bπ χ(m)
m 0
m∈GL1 (Fq )

     
n−1 0 In−1 0 In−1
= χ(−1) Bπ χ det
m 0 m 0
m∈GL1 (Fq )

  
n−1 0 In−1
= χ(−1) Bπ⊗χ
m 0
m∈GL1 (Fq )

= χ(−1)n−1 γ((π ⊗ χ) × 1, ψ). 2

3. The n × 1 gamma factors over finite fields

In this section, we want to relate γ(π × ρ, ψ) with abelian Gauss sums, where π is
an irreducible cuspidal representation of GLn (Fq ) and ρ is an irreducible representation
of GL1 (Fq ), for n  2. The computation is based on Green’s character formula for
irreducible cuspidal representations of GLn (Fq ) and Proposition 2.8. (We refer to [6] and
[3] for details.)
262 C. Nien / Finite Fields and Their Applications 46 (2017) 255–270

3.1. Green’s construction

Let Fq denote the algebraic closure of Fq and in the following the field Fqd is the
unique field extension of Fq with index d such that Fq ⊂ Fqd ⊂ Fq .
Let f (t) be an irreducible polynomial over Fq of degree d  1 with leading coefficient
1 such that f (t) = t. For each such f we fix a matrix Lf ∈ GLd (Fq ) whose characteristic
polynomial is

det(tId − Lf ) = f (t).

For any u  1, we denote by

⎛ ⎞
Lf 1 0 ··· 0
⎜ .. .. ⎟
⎜ 0 Lf 1 . . ⎟
Lf,u = ⎜ . ⎟,
⎝ . ..
.
..
.
..
. ⎠
. 0
0 ... ... 0 Lf

a square matrix of size du. Let ρ = (r1 , r2 , . . . , rs ) be a partition with s parts, which
are written in decreasing order r1  r2  · · ·  rs . Let |ρ| = r1 + r2 + · · · + rs . Denote
by
⎛ ⎞
Lf,r1 0 ··· 0
⎜ .. .. ⎟
⎜ 0 Lf,r2 . . ⎟
Lρ (f ) = ⎜ . ⎟,
⎝ . ..
.
..
. ⎠
. 0
0 ... 0 Lf,rs

a square matrix of size d|ρ|. Given a set of irreducible polynomials f1 , . . . , fM with leading
coefficient 1 and a set of partitions ρ1 , . . . , ρM such that

M
i=1 |ρi | deg fi = n,
(3.1)
fi (t) = t, for all i = 1, . . . , M,

consider the matrix of GLn (Fq ) of the following form:

⎛ ⎞
Lρ1 (f1 ) 0 ··· 0
⎜ .. ⎟
⎜ 0 Lρ2 (f2 ) . 0 ⎟
A=⎜ ⎟. (3.2)
⎝ .. ..
.
..
. ⎠
. 0
0 ... 0 LρM (fM )

Note that its characteristic polynomial is given by

|ρ1 | |ρ |
det(tIn − A) = f1 · · · fMM .
 C. Nien / Finite Fields and Their Applications 46 (2017) 255–270 263

By the theory of rational canonical form ([7]), the set of matrices in the form of Eq. (3.2)
for all possible choices of (fi , ρi ), with distinct irreducible polynomials fi (t) = t and
partitions ρi satisfying Eq. (3.1), consists a complete system of representatives of the
conjugacy classes of GLn (Fq ). We denote by [f1ρ1 f2ρ2 · · · fM
ρM
] the conjugacy class corre-
sponding to the set {(fi , ρi ) | 1  i  M }.
Let η be a character of F× q n satisfying the following properties:

m
−1
ηq = 1, for all positive m < n.

Such a character η is called regular and it defines an irreducible cuspidal representation

π = πη of GLn (Fq ) whose character χπ is given as follows

(P1) χπ (c) = 0, if the class c is not given by a power of a single polynomial which is
irreducible over Fq .
(P2) Suppose that the class c has the form c = [f ρ ], where f is an irreducible polynomial
over Fq of degree d, ρ a partition of n/d having s parts, and σ ∈ Fqd is one of the
roots of f . By Eq. (6.1) in [4],
d−1
χπ (c) = (−1)s+n Φs (q d )[η(σ) + η(σ q ) + · · · + η(σ q )],

where Φs (t) = (t − 1)(t2 − 1) · · · (ts−1 − 1), for s > 1, and Φ1 (t) = 1.

In the above construction, η1 and η2 produce the same character χπ if and only if
t
η1 = η2q for some t  1.

3.2. Gauss sum formula

Before we prove Theorem 1.1, we introduce some auxiliary lemmas.

Write A ∼ B, if A, B ∈ GLn (Fq ) are conjugate, and let
 
0 In−1
g(m, u) := u,
m 0

where m ∈ F×
q , u ∈ Un .

Lemma 3.1. Each g(m, u) is conjugate to a matrix in

 
g = g( t ) | t = (t1 , . . . , tn ) ∈ Fnq , t1 = 0 ,
 
0 In−1
where g( t ) := . Moreover, for each t , there are |Un−1 (Fq )| elements
t1 (t2 . . . , tn )
in
264 C. Nien / Finite Fields and Their Applications 46 (2017) 255–270

G := {g(m, u) | m ∈ F×
q , u ∈ Un−1 },

which are conjugate to g( t ).

Proof. Throughout, we think of Un−1 (Fq ) embedded in Un (Fq ) via

u → diag(u, 1).

We claim that the conjugation action of the group Un−1(Fq ) on the set G is simple (i.e.
stabilizers are all trivial) and the set g is a set of representatives for the orbits. Write
e1 , . . . , en for the standard basis of column vectors, Vi for the linear span of e1 , . . . , ei ,
V0 = {0} and p for the projection onto Vn−1 with the kernel containing en . Consider any
fixed g = g(m, u) ∈ G and note that we have

p(gek ) ∈ ek−1 + Vk−2 , for 1 < k  n. (3.3)

We put en = en and define recursively

ek−1 = p(gek ) for n  k > 1.

Note that Eq. (3.3) implies that ek ∈ ek + Vk−1 , for n − 1  k  1, so that the change of
basis matrix h from the standard basis to e1 . . . , en lies in Un−1 (Fq ). By construction,
h−1 gh ∈ g. Moreover, distinct g( t ) ∈ g have distinct characteristic polynomials, so they
are not conjugate. Thus we have proved that the set g is a set of representatives for the
orbits in G under Un−1 (Fq ). Finally, we need to check that the stabilizer of any g( t ) is
trivial. Write u = (ui,j ) for an element of u ∈ Un−1 (Fq ) so that ui,j = 0, for i > j and
for i < n = j. By comparing the coefficients in the first n − 1 rows of ug( t ) = g( t )u,
we get

ui,j = ui+1,j+1 , for i < j, 2  j < n
ui,n = 0.

Then we see that u is the identity. 2

Lemma 3.2. Let π be an irreducible cuspidal representation of GLn (Fq ) and ηπ the regular
character of F×q n , which corresponds to π in Green’s construction. Let χ be a character
of F×
q . Then the regular character corresponding to π ⊗ χ := π ⊗ χ ◦ det is given by
ηπ ⊗ χ ◦ N , where N is the reduced norm map from Fqn → Fq .

Proof. Note that ηπ ⊗ χ ◦ N is also a regular character. Let π  be the irreducible cuspidal
representation of GLn (Fq ), which corresponds to the regular character ηπ ⊗ χ ◦ N . Con-
sider both the trace characters χ1 of π ⊗ χ and χ2 of π  in Green’s construction. Recall
that χi is only non-vanishing on matrices of the form Lρ (f ), where f is an irreducible
 C. Nien / Finite Fields and Their Applications 46 (2017) 255–270 265

polynomial over Fq of degree d, ρ a partition of n/d having s parts, and σ ∈ Fqd is one
of the roots of f . By Green’s construction,

χ1 (Lρ (f ))
d−1
= (−1)s+n Φs (q d )[η(σ) + η(σ q ) + · · · + η(σ q )]χ(det Lρ (f ))
d−1
= (−1)s+n Φs (q d )[η(σ) + η(σ q ) + · · · + η(σ q )]χ(N (σ))
= χ2 (Lρ (f )).

Therefore the regular character corresponding to π ⊗ χ is given by ηπ ⊗ χ ◦ N . 2

Proof of Theorem 1.1. By Propositions 2.5 and 2.8,

  
0 In−1
γ(π × 1, ψ) = Bπ,ψ
m 0
m∈GL1

−1
 
= |Un (Fq )| ψn (u−1 )χπ (g(m, u)).
m∈GL1 u∈Un

Recall that by (P1) we only need to consider that kind of matrix which is conjugate
to Lρ (f ) for some irreducible polynomial f (t) = t and some partition ρ. Assume that
g(m, u) is conjugate to some Lρ (f ). Since each g(m, u) ∼ g( t ), the partition ρ must
have only 1 part. Thus, writing d = deg(f ) and for σ any root of f , property (P2) gives
us
d−1
χπ (g(m, u)) = (−1)1+n [η(σ) + η(σ q ) + · · · + η(σ q )].

The matrix Lρ (f ) is conjugate in GLn (Fq ) to

⎛A Id 0 ... 0 ⎞
f
⎜ .. .. ⎟
⎜ 0 Af Id . . ⎟
⎜ ⎟
A=⎜
⎜
..
.
..
.
..
.
..
. ⎟
0 ⎟,
⎜ ⎟
⎝ .. ..
. ⎠
. Af Id
0 ... ... 0 Af

d−1
where Af = diag(σ, σ q , · · · , σ q ), so that g(m, u) and A have the same characteristic
polynomial. Comparing the coefficients of t and the constant terms in their characteristic
polynomials, we deduce that


n−1
σ 1+q+···+q
n−1

n−1
N (σ)
−m(u1,2 + · · · + un−1,n ) = (−1) n−1
= (−1) n−1
,
i=0
σ qi i=0
σ qi
266 C. Nien / Finite Fields and Their Applications 46 (2017) 255–270

and −m = (−1)n N (σ), where N denotes the reduced norm map from Fqn to Fq . Then


n−1
1 
n−1
σ −q = −trσ −1 ,
i
u1,2 + · · · + un−1,n = − i = −
i=0
σq i=0

where tr is the reduced trace map from Fqn to Fq .

By Lemma 3.1, each g( t ) is conjugate to |Un−1 (Fq )| elements in G, so

γ(π × 1, ψ)
    
−1 −1 0 In−1
= |Un (Fq )| ψn (u )χπ u
m 0
m∈GL1 (Fq ) u∈Un (Fq )

−1

= (−1)n−1 |Un−1 (Fq )||Un (Fq )| ψ(trσ −1 )ηπ (σ)
σ∈F×
qn

= (−1)n−1 q −(n−1) ψ(trσ −1 )ηπ (σ).
σ∈F×
qn

Let Bπ = Bπ,ψn , Bχ = Bχ,ψ1 and Bπ⊗χ = Bπ⊗χ,ψn be the normalized Bessel functions
for π, χ and π ⊗ χ respectively. By Lemma 2.10,

γ(π × χ, ψ)
= χ(−1)n−1 γ((π ⊗ χ) × 1, ψ)

= (−q −1 χ(−1))n−1 ψ(trσ −1 )ηπ (σ)χ(N (σ)),
σ∈F×
qn

where ηπ⊗χ = ηπ ⊗ χ ◦ N follows Lemma 3.2. 2

3.3. The 1 × 1 local converse theorem over finite fields

Let ρ be a character of F×
q and ψ be a non-trivial additive character of Fq . Let

S(Fq ) = {f : Fq → C}.

The Fourier transform2 of φ ∈ S(Fq ) in terms of ψ is given by


φ̂(x) = q − 2
1
φ(y)ψ(yx).
y∈Fq

The analogue for the finite field case of Tate’s local functional equation, the 1 × 1 local
gamma factor, is given by [12] in the following theorem.

2
In this paper we use the normalized Fourier transform and gamma factor in Eq. (3.4), which is different
from the notations in [12].
 C. Nien / Finite Fields and Their Applications 46 (2017) 255–270 267

Theorem 3.3 ([12], local gamma factor for GL1 (Fq )). Let φ ∈ S(Fq ). Given any fixed
non-trivial additive character ψ of Fq and ρ a character of F×
q , the following functional
equation holds:
  
φ(x)ρ(x) + q − 2 φ(0) φ̂(x)ρ−1 (x),
1
γ(ρ, ψ) ρ(x) = (3.4)
x∈F×
q x∈F×
q x∈F×
q


with γ(ρ, ψ) = q − 2 ψ(x)ρ−1 (x).
1
x∈F×
q

Lemma 3.4 (1 × 1 local converse theorem over finite fields). Given ρ, τ characters of F×
q ,
if γ(ρχ, ψ) = γ(τ χ, ψ), for any character χ of F× , then ρ ∼
= τ .
q

Proof. It is well known that


ψ(x)(ρ−1 χ−1 )(x)| = q 2 if ρχ = 1;
1
(1) | x∈F×
 q
(2) x∈F×
q
ψ(x)(ρ−1 χ−1 )(x) = −1 if ρχ = 1. 2

3.4. F×
q -distinguished characters

A character χ of F× × ×
q 2 is Fq -distinguished if the restriction of χ to Fq is trivial. Define

T := {χ ∈ F
× ×
q 2 | χ is Fq -distinguished}.

2πi
Let Ψ be any fixed nontrivial additive character of Fq2 such that Ψ|Fq = 1. Let μ = e q2 −1
be a primitive (q 2 −1)-th root of unity and α be a generator of F×
q 2 . For t ∈ {1, · · · , q −1},
2

define

χt : F × ×
q 2 → C by χt (α ) = μ .
k kt

Theorem 3.5 (1 × 1 converse theorem on distinction). Assume that ρ is a non-trivial

character of F×
q 2 . Then the following conditions are equivalent.

(1) ρ = χt , where t = r(q − 1) for some r ∈ N.

(2) ρ is F×
q -distinguished.
(3) γ(ρπ, Ψ) = 1, for all F×q -distinguished π, satisfying π = ρ
−1
.

Proof. Note that Fq2 = {aα + b | a, b ∈ Fq }, where α is a generator of F× q 2 . It is clear

that Condition (1) and (2) are equivalent.
Next we show that Condition (1) implies Condition (3). Let ρ = χt , where t = r(q −1)
for some r ∈ N. Then

γ(ρ, Ψ) = q −1 Ψ(g)ρ−1 (g)
g∈F×
q2
268 C. Nien / Finite Fields and Their Applications 46 (2017) 255–270

2
q −1
−1
=q Ψ(αk )ρ−1 (αk )
k=1


q 
q−1
= q −1 Ψ(αk+s(q+1) )μ−(k+s(q+1))t
k=0 s=1

q−1  q q−1
= + q −1 Ψ(αk+s(q+1) )μ−kt
q s=1 k=1

q−1 
q 
q−1
= + q −1 μ−kt Ψ(αk+s(q+1) ).
q s=1
k=1

For k = 1, . . . , q, the map

a → Ψ(αk a), for a ∈ F×

q

defines an additive character, which is non-trivial, since {1, αk } is a basis for Fq2 over Fq .
Hence

q−1  q
γ(ρ, Ψ) = + q −1 μ−kt (−1)
q
k=1

q−1 μ−(q+1)t − μ−t

= − q −1 = 1.
q μ−t − 1

To show that Condition (3) implies (1) and (2), we assume that

γ(χt χr(q−1) , Ψ) = 1, for all r = 1, . . . , q

where t ∈ {1, . . . , q 2 − 1} and q − 1  t. Then


q
γ(χt χr(q−1) , Ψ) = q. (3.5)
r=1

On the other hand,


q
γ(χt χr(q−1) , Ψ)
r=1

 −1
q q 2

−1
=q Ψ(αk )μ−k[r(q−1)+t]
r=1 k=1
2
q −1 
q
= q −1 Ψ(αk )μ−kt μ−k[r(q−1)]
k=1 r=1
 C. Nien / Finite Fields and Their Applications 46 (2017) 255–270 269

 
q
= q −1 Ψ(αk )μ−kt μ−k[r(q−1)]
k=1,...,q 2 −1:q+1k r=1


q−1 
q
−1 −m(q+1)t
+q Ψ(α m(q+1)
)μ μ−m(q+1)[r(q−1)]
m=1 r=1

 μ−k(q+1)(q−1) − μ−k(q−1)
= q −1 Ψ(αk )μ−kt
μ−k(q−1) − 1
k=1,...,q 2 −1:q+1k
 q−1 

−1 m(q+1) −m(q+1)t
+q Ψ(α )μ q
m=1
2
q −1 
q−1
−1
= −q Ψ(αk )μ−kt + (1 + q −1 ) Ψ(αm(q+1) )μ−m(q+1)t
k=1 m=1
2
q −1 
q−1
−1 −kt −1
= −q k
Ψ(α )μ + (1 + q ) μ−m(q+1)t
k=1 m=1
2
q −1
−1 μ−q(q+1)t − μ−(q+1)t
= −q Ψ(αk )μ−kt + (1 + q −1 )
k=1
μ−(q+1)t − 1
2
q −1
−1
= −q Ψ(αk )μ−kt .
k=1

q2 −1
Since χt is non-trivial, | k=1 Ψ(αk )μ−kt | = q. This contradicts Eq. (3.5). Hence Con-
dition (3) implies (1) and (2). 2

References

[1] Moshe Adrian, Baiying Liu, Shaun Stevens, Kam-Fai Tam, On the sharpness of the bound for the
local converse theorem of p-adic GL(prime), 2017, submitted for publication.
[2] J. Chai, Bessel functions and local converse conjecture of Jacquet, J. Eur. Math. Soc. (JEMS)
(2017), in press.
[3] S.I. Gelfand, Representations of the full linear group over a finite field, Mat. Sb. (N. S.) 83 (125)
(1970) 15–41.
[4] S.I. Gelfand, Representation of the general linear group over a finite field, in: Proc. Summer School
of the Bolya-Janos Math. Soc., Budapest, 1971, Halsted, New York, 1975.
[5] I.M. Gelfand, M.I. Graev, Construction of irreducible representations of simple algebraic groups
over a finite field, Dokl. Akad. Nauk SSSR 147 (1962) 529–532 (in Russian).
[6] J.A. Green, The characters of the finite general linear groups, Trans. Am. Math. Soc. 80 (1955)
402–447.
[7] Robert E. Hartwig, Roth’s removal rule and the rational canonical form, Am. Math. Mon. 103 (4)
(1996) 332–335.
[8] Jeffrey Hakim, Omer Offen, Distinguished representations of GL(n) and local converse theorems,
Manuscr. Math. 148 (1–2) (2015) 1–27.
[9] H. Jacquet, Baiying Liu, On the local converse theorem for p-adic GLn , Am. J. Math. (2017), in
press.
270 C. Nien / Finite Fields and Their Applications 46 (2017) 255–270

[10] Y. Ok, Distinction and gamma factors at 1/2: supercuspidal case, Ph.D. thesis, Columbia University,
ProQuest LLC, Ann Arbor, MI, 1997.
[11] Chufeng Nien, A proof of the finite field analogue of Jacquet’s conjecture, Am. J. Math. 136 (3)
(2014) 653–674.
[12] Edva-Aida Roditty, On gamma factors and Bessel functions for representations of general linear
groups over finite field, M.s.c. Thesis, Tel-Aviv University, 2010.