arXiv:1511.06209v2 [math.

NT] 10 Dec 2015

EXCEPTIONAL ZEROS OF L-SERIES AND

BERNOULLI-CARLITZ NUMBERS
` TUAN NGO DAC, AND FLORIC TAVARES RIBEIRO
BRUNO ANGLES,
Abstract. Bernoulli-Carlitz numbers were introduced by L. Carlitz in 1935,
they are the analogues in positive characteristic of Bernoulli numbers. We
prove a conjecture formulated by F. Pellarin and the first author on the nonvanishing modulo a given prime of families of Bernoulli-Carlitz numbers. We
then show that the exceptional zeros of certain L-series are intimately connected to the Bernoulli-Carlitz numbers.

With an appendix by B. Angl`es, D. Goss, F. Pellarin, F. Tavares Ribeiro

Contents
1.
2.
3.
4.
5.
6.
A.

Introduction
1
Proof of Theorem 1.1
3
Exceptional zeros and eigenvalues of certain K-endomorphisms
11
Answer to Problem 1 for q = p
16
Some hints for Problem 1 for general q.
22
An example
27
Appendix: The digit principle and derivatives of certain L-series, by B.
Angl`es, D. Goss, F. Pellarin and F. Tavares Ribeiro
27
B. Table
31
References
32

1. Introduction
Recently, M. Kaneko and D. Zagier have introduced
the Q-algebra of finite mulQ
p Fp
tiple zeta values which is a sub-Q-algebra of A := p Fp (p runs through the prime
numbers). This algebra of finite multiple zeta values contains the following elements:
 !

Bpk
A,
k 2, Z(k) =
k
p
where Bn denotes the nth Bernoulli number. It is not known that the algebra
of finite multiple zeta values is non-trivial. In particular, it is an open problem to
prove that Z(k) 6= 0 for k 3, k 1 (mod 2) (observe that Z(k) = 0 if k 2, k 0
(mod 2)). This latter problem is equivalent to the following:
Date: December 11, 2015.
1

` TUAN NGO DAC, AND FLORIC TAVARES RIBEIRO

BRUNO ANGLES,

Conjecture 1. Let k 3 be an odd integer. Then, there exist infinitely many

primes p such that Bpk 6 0 (mod p).
Let k 3 be an odd integer. M. Kaneko ([14]) remarked that, viewing the Bpk s
as being random
P modulo p when p varies through the prime numbers, taking into
account that p p1 diverges, then it is reasonable to expect that there exist infinitely
many prime numbers p such that Bpk 0 (mod p).
Let Fq be a finite field having q elements, q being a power of a prime number
p, and let be an indeterminate over Fq . In 1935, L. Carlitz has introduced the
analogues of Bernoulli numbers for A := Fq [] ([9]). The Bernoulli-Carlitz numbers,
BCn K := Fq (), n N, are defined as follows:
- BCn = 0 if n 6 0 (mod q 1),
- for n 0 (mod q 1), we have:
A (n)
BCn
,
=
(n)

en

where (n) A is the Carlitz factorial ([13], chapter 9, paragraph

e is the
P 9.1),
Carlitz period ([13], chapter 3, paragraph 3.2), and A (n) := aA,a monic a1n
K := Fp (( 1 )) is the value at n of the Carlitz-Goss zeta function. The BernoulliCarlitz numbers are connected to Taelmans class modules introduced in [19] (see for
example [21] and [6]). L. Carlitz established a von-Staudt result for these numbers
([13], chapter 9, paragraph 9.2), and as an easy consequence, we get that if P is a
monic irreducible polynomial in A, then BCn is P -integral for 0 n q deg P 2.
It is natural to ask if Conjecture 1 is valid in the carlitzian context. In this paper,
we prove a stronger result which answers positively to a Conjecture formulated in
[3]:
Theorem 1.1. Let N 2 be an integer, N 1 (mod q 1). Let q (N ) be the sum
of the digits in base q of N. Let P A be a monic irreducible polynomial of degree
(N )1
d such that q d > N. If d q q1 N, then:
BCqd N 6 0 (mod P ).
The above Theorem is linked with the study of exceptional zeros of certain Lseries introduced in 2012 by F. Pellarin ([15]), but from a slightly different point
of view. More precisely, let N be as above and for simplicity we assume that
q (N ) q, let t be an indeterminate over K , lets consider:
X X aN
1
A[[ ]] ,
LN (t) :=
a(t)
t
d0 aA+,d

where A+,d is the set of monic elements in A of degree d. It was already noticed
by F. Pellarin ([16]) that such L-series can be related with Andersons solitons
and should play an important role in the arithmetic theory of function fields. Let
C be the completion of a fixed algebraic closure of K . Then, one can show
that LN (t) converges on {x C , v (x) < 0}, where v is the valuation on
C normalized such that v () = 1. Furthermore, one can easily see that the
j
elements of S := {q , j Z, q j N } are zeros of the function LN (t). We call the
zeros of LN (t) which belong to {x C , v (x) < 0} \ S the exceptional zeros of
LN (t). Lets briefly describe the case q = p. In this case, the exceptional zeros of
LN (t) are simple, belong to Fp (( 1 )) and are the eigenvalues of a certain K-linear

EXCEPTIONAL ZEROS OF L-SERIES AND BERNOULLI-CARLITZ NUMBERS

(N )

endomorphism t of a finite dimensional K-vector space H((N ) ) connected to

the generalization of Taelmans class modules introduced in [5]. The proof of the
fact that the exceptional zeros are simple and real uses combinatorial techniques
introduced by F. Diaz-Vargas ([12]) and J. Sheats ([18]). Furthermore, if pd > N,
then:
p (N )p Q
Qd1
k
pl
pn nl
(1) p1
d
l=0
n=0,n6=l ( )
(N )
= det (p Id t |H((N ) ) ),
BCpd N
d
K
(N )(p N )
Pk
where (.) is the Carlitz factorial, and N = l=0 nl pl , nl {0, . . . , p 1}. Since
(N )
the eigenvalues of t are exactly in this situation the exceptional zeros of LN (t),
we also obtain another proof of Theorem 1.1 as a consequence of the fact that:
(N )

det (ZId t

K[Z]

|H((N ) ) ) Fp [Z, ],

and therefore (P is a monic irreducible polynomial of degree d):

d

(N )

det (p Id t
K

(N )

|H((N ) ) ) det (Id t

K

|H((N ) ) ) (mod P ).

Lets observe that Theorem 1.1 implies the following (see [3], page 248):
X X (a )N
6= 0,
a
d0 aA+,d

where a denotes the derivative of a and N 1 (mod q 1). In the appendix of

this paper, we discuss a digit principle for such Euler type sums.
We mention that the construction of Kaneko-Zagiers objects in the positive
characteristic world is the subject of a forthcoming work of F. Pellarin and R.
Perkins ([17]), they prove, in this context, that the algebra of finite multiple zeta
values is non-trivial. In this situation, it would be very interesting to examine the
validity of Conjecture 1 for Bernoulli-Goss numbers (see [2] for a special case).
2. Proof of Theorem 1.1
2.1. Notation. Let Fq be a finite field having q elements and let p be the characteristic of Fq . Let be an indeterminate over Fq and let A = Fq [], K = Fq (), K =
Fq (( 1 )). Let C be the completion of a fixed algebraic closure of K . Let v :
C Q {+} be the valuation on C normalized such that v () = 1. For
d N, let A+,d be the set of monic elements in A of degree d.
2.2. The L-series LN (t).
Let N 1 be an integer. Let t be an indeterminate over C . Let Tt be the Tate
algebra in the variable t with coefficients in C . Lets set:
X X a(t)N
T
LN (t) =
t .
a
d0 aA+,d

Then, we can write:

LN (t) =

X
i0

Observe that 0,N (t) = 1.

i,N (t)i , i,N (t) Fq [t].

` TUAN NGO DAC, AND FLORIC TAVARES RIBEIRO

BRUNO ANGLES,

Lemma 2.1. We have:

i 0, degt i,N (t) N (Max{
In particular LN (t) is an entire function.

q (N )
log(i)
, 0} + [
] + 1).
log(q)
q1

(N )

q
] N. This Lemma is a consequence of the proof of [3], Lemma
Proof. Let u = [ q1
7. We give a proof for the convenience of the reader. We will use the following
elementary fact ([3], Lemma 4):
Let s P
1 be an integer and let t1 , . . . , ts be s indeterminate over Fp . If d(q 1) > s
then aA+,d a(t1 ) a(ts ) = 0.
If a is a monic polynomial in A, we will set:
a
1
1
< a > = deg a 1 + Fq [[ ]].

P
a(t)N
Let Sd := aA+,d a . Observe that:

degt Sd = dN.

We have :

1 X
a(t)N < a >1
.
d
aA+,d
P
Observe that (p-adically) 1 = n0 (q 1)q n . For m 0, set:
Sd =

ym =

m
X

(q 1)q n .

n=0

Then:

ym 1 (mod q m+1 ), q (ym ) = (m + 1)(q 1).

Therefore:
v (

aA+,d

a(t)N < a >1

aA+,d

a(t)N < a >ym ) q m+1 ,

where v is the -adic valuation on Tt such that v () = 1. Thus, if q (N ) +

(m + 1)(q 1) < d(q 1), we get:
X
a(t)N aym = 0.
aA+,d

We therefore get, if d u + 2 :

v (Sd ) d + q du1 .

This implies that LN (t) is an entire function. Let j such that tj appears in i,N (t).
Let x = [ Nj ]. Let d be minimal such that tj comes from Sd . We must have d x
and j dN. Furthermore, if d u + 2, we have:
In particular:
Therefore:

i d + q du1 .

i d + q du1 q du1 .
d Max{

log(i)
, 0} + u + 1.
log(q)


EXCEPTIONAL ZEROS OF L-SERIES AND BERNOULLI-CARLITZ NUMBERS

2.3. The two variable polynomial BN (t, ).

Let N 2, N 1 (mod q 1). If q (N ) = 1, we set BN (t, ) = 1. Lets assume
that q (N ) 6= 1 and lets set s = q (N ) 2. Let t1 , . . . , ts be s indeterminates over
C . Let Ts be the Tate algebra in the indeterminates t1 , . . . , ts with coefficients in
C . Let : Ts Ts be the continuous morphism of Fq [t1 , . . . , ts ]-algebras such
that c C , (c) = cq . For i = 1, . . . , s, we set:
Y
ti
(ti ) =
(1 qj )1 Ts ,

j0
where is a fixed (q 1)th-root of in C . Set:
Y
j

e =
(1 1q )1 .
j1

Set:

Ls =

X X a(t1 ) a(ts )
T
s .
a
d0 aA+,d

We also set:

Ls (t1 ) . . . (ts )
Ts .

e
Then, by [5], Lemma 7.6 (see also [3], Corollary 21), Bs Fq [t1 , . . . , ts , ] is a monic
Pq (N ) en
sq
. Write N = n=1
q , e1 e2 eq (N ) .
polynomial in of degree r = q1
We set:
BN (t, ) = Bs |ti =tqei Fq [t, ].
s1

Bs = (1) q1

We observe that BN (t, ) is a monic polynomial in such that deg BN (t, ) = r.

Lemma 2.2. Let N 2, N 1 (mod q 1). Then:

1) BN (tp , p ) = BN (t, )p .
2) BqN (t, ) = BN (tq , ).
3) We have:
BN (t, ) ( t)r r(tq t)( t)r1

4) If N 0 (mod p), then BN (t, ) Fp [tp , ].

(mod (tq t)2 Fp [t, ]).

Proof. Recall that:

BN (t, ) =
where N =

Pk

l=0

(1)

q (N )1
q1

k
X X a(t)N Y
l
(tq )nl ,
a
d0 aA+,d

l=0

nl q l , n0 , . . . , nk {0, . . . q 1}. Observe that:

X X a(t)N
1
Fp [t][[ ]].
a

d0 aA+,d

Thus:
BN (t, ) Fp [t, ].
Thus we get assertion 1). Assertion 2) is a consequence of the definition of BN (t, ).
Let Fq . By [4], theorem 2.9, we have:
(t) |t= = expC (

e
),

` TUAN NGO DAC, AND FLORIC TAVARES RIBEIRO

BRUNO ANGLES,

where expC : C C is the Carlitz exponential ([13], chapter 3, paragraph 3.2).

Now, by [15] Theorem 1, we get:
X X a()
X X a(t)N

e
|t= =
=
.

e
a
a
)
( ) expC (
d0 aA
d0 aA
+,d

+,d

But observe that:

expC (
Thus:
Observe that

d
dt BN (t, )
k
Y

n0 (1)r+1

BN (t, ) |t= = ( )r .

is equal to:

(tq )nl

l=0

e q1
)
= ( ).

X X

d
N 1
dt (a(t))a(t)

d0 aA+,d

a(t)N

d
dt ((t))

(t)

Thus we get assertion 4). Since for q (N ) = q, we have BN (t) = 1. We get:

d
BN (t, ) |t= = 0.
dt
This concludes the proof of the Lemma.
Fq ,

Lemma 2.3. Let N 2, N 1 (mod q 1). Then degt BN (t, ) p if r 1

and the total degree in t, of BN (t, ) is less than or equal to Max{rN + r 2, 0}.
Furthermore BN (t, ) (as a polynomial in t) is a primitive polynomial.
Proof. Recall that if r = 0 then BN (t, ) = 1. Lets assume that r 1. Observe
that by Lemma 2.2,we have:
BN (t, 0) (t)r1 (t + r(tq t)) (mod (tq t)2 Fp [t]).
In particular degt BN (t, ) p. Let x C such that v (x) >
X X a(x)N
=
a
d aA+,d

Write N =

Pk

l=0

P monic irreducible in A

(1

1
N .

Then:

P (x)N 1
) C
.
P

nl q l , nl {0, . . . , q 1}, nk 6= 0. For l = 0, . . . , k, we have:

l

(tq ) |t=x C
.

Therefore:
BN (t, ) |t=x 6= 0.

This implies that, if x C is a root of BN (t, ) then v (x)

C [t] :
BN (t, ) =

m
Y

j=1

1
N

< 0. Write in

(t xj ), Fp [] \ {0}, deg r 1, x1 , . . . , xm C ,

where m = degt N (t, ). Then:

r = (1)m

m
Y

j=1

xj .

EXCEPTIONAL ZEROS OF L-SERIES AND BERNOULLI-CARLITZ NUMBERS

Therefore:
deg r

m
.
N

We finally get:
degt BN (t, ) (r deg )N rN.
Since BN (t, ) is a monic polynomial in , the total degree in t, of BN (t, ) is less
than or equal to degt BN (t, ) + r 2.
Write:
BN (t, ) = F, Fp [] \ {0},
where F is a primitive polynomial (as a polynomial in t). In particular must
divide r and BN (1, ) in Fp []. By Lemma 2.2, this implies that F

p.
Remark 2.4. Let f () be a monic irreducible polynomial in Fp []. Let P1 , . . . , Pm
be the monic irreducible polynomials in A such that f () = P1 Pm . We can order
them such that if d = deg f, then for i = 1, . . . , m, we have:
Pi = i1 (P1 ),
where : A A is the morphism of Fp []-algebras such that x Fq , (x) = xp ,
and m = [Fpd Fq : Fp ]. Let N 1, and lets set:
N (f ) =

m
Y

i=1

We have:

(Pi () Pi (t)N ) f () Fp [t, ].

degt N (f ) = N deg f,
log(p)
).
log(q)
If f1 , . . . , fn are n irreducible monic polynomials in Fp [], we set:
deg N (f ) deg f deg P1 deg (f )(1

N (f1 fm ) =
Thus in Fp [t](( 1 )):
LN (t) =
P

n
Y

l=1

N (fl ) Fp [t, ].

d0 aFp []+,d
N (a)
a

N (a)
.
a

Observe that
converges on {x C , v (x) >
d0
aFp []+,d
does not vanish. Therefore on {x C , v (x) > 1
N } :

1
N }

and

l
k Y
Y
1
tq
N (a)
= q (N )q BN (t, )
(1 qj )nl ,
a

q1
l=0 j0
d0 aFp []+,d
Pk
where N 1 (mod q 1), q (N ) q, N = l=0 nl q l , nl {0, . . . , q 1}, nk 6= 0.

Let s 2, s 1 (mod q 1). Recall that we have set:

Ls (t1 ) . . . (ts )
Ts ,

e
where Ts is the Tate algebra in the Q
indeterminates t1 , . . . , ts with coefficients in
C , and for i = 1, . . . , s, (ti ) = j0 (1 tqij )1 . For m N, we denote by
BCm K the mth Bernoulli-Carlitz number ([13], chapter 9, paragraph 9.2).
s1

Bs = (1) q1

` TUAN NGO DAC, AND FLORIC TAVARES RIBEIRO

BRUNO ANGLES,

Proposition 2.5.
1)Let N 1, N 1 (mod q 1), q (N ) q. Recall that r =
such that q d > N, then we have the following equality in C :
d

BN (, q )

Qk

l=0

Qd1

n=0,n6=l (

ql

qn

)nl

= (1)r

q (N )q
q1 .

Let d 1

BCqd N
,
(N )(q d N )

P
where (.) is the Carlitz factorial, and N = kl=0 nl q l , nl {0, . . . , q 1}.
2) Let N 2, N 1 (mod q 1). Let P be a monic irreducible polynomial in
A of degree d 1 such that q d > N. Then BCqd N 0 (mod P ) if and only if
BN (, ) 0 (mod P ).
Proof.
1) The first assertion of the Proposition is a consequence of the proof of Theorem 2
in [3]. For the convenience of the reader, we give the proof of this result. Let s q,
s 1 (mod q 1). Then ([5], Lemma 7.6), we have that Bs Fq [t1 , . . . , ts , ] is a
sq
.
monic polynomial in of degree q1
Let : Ts Ts be the continuous morphism of Fq [t1 , . . . , ts ]-algebras such that
x C , (x) = xq . Since q = , we have:
((ti )) = (ti )(ti ), i = 1, . . . , s.
Let d 1, we get:
s1

(1) q1

d (Ls )
d (Bs )
(t
)
.
.
.
(t
)
=
.
Q
1
s
d
d1
ql
ql

eq
l=0 (t1 ) (ts )

Recall that, by formula (24) in [15], we have:

eq
.
(ti )(ti ) |t=ql =
Dl
ql

Let l1 , . . . , ls N, we get:
s1

(1) q1

l1

ls

ls
l1
d (Bs )(t1 q ) (ts q )
d (Ls )
(t1 q )(t1 ) . . . (ts q )(ts ) = Qd1
.
d
q
ql
ql

e
l=0 (t1 ) (ts )

Now, let N 1 such that q (N ) = s (observe that N 1 (mod q 1) and

P
q (N ) q). Write N = kl=0 nl q l , n0 , . . . nk {0, . . . , q 1}, nk 6= 0. Let d k + 1.
We get:
P
P
a(t)N
d
Qk
s1
l
l
l
u0
aA+,u
d
BN (t, q )
aq
( l=0 ((tq q )(tq ))nl ) = Qk Qd1
.
(1) q1
qd
ql
q n nl

We get:

l=0

Qk

l=0

BN (t, q )

Qd1

n=0,n6=l

(tql

qn )nl

|t= = (1)

q (N )q
q1

n=0,n6=l (t

BCqd N
.
(N )(q d N )

2) The result is well-known for q (N ) = 1 (this is a consequence of the definition

of the Bernoulli-Carlitz numbers and [13], Lemma 8.22.4). Thus, we will assume
q (N ) q. The assertion is then a consequence of the fact that:
d

BN (, ) BN (, q ) (mod P ).


EXCEPTIONAL ZEROS OF L-SERIES AND BERNOULLI-CARLITZ NUMBERS

We have already mentioned that Bs Fq [t1 , . . . , ts , ] is a monic polynomial in

sq
([5], Lemma 7.6). Lets observe that we have:
of degree q1
Lemma 2.6. For s 2q 1, s 1 (mod q 1), we have:

Bs (t1 , . . . , ts(q1) , 0, . . . , 0) = ( t1 ts(q1) )Bs(q1) (t1 , . . . , ts(q1) ).

More generally, if is in the algebraic closure of Fq in C , let P be the monic

irreducible polynomial in A such that P () = 0. Let s 1 (mod q1), s q+q d 1,
where d is the degree of P. Write s = s (q d 1). We have:
Bs (t1 , . . . , ts ) , , . . . , ) = (P P (t1 ) P (ts ))Bs (t1 , . . . , ts ).

Proof. The polynomial Bs (t1 , . . . , ts(q1) , 0, . . . , 0) is equal to:

Y
s1 (t1 ) . . . (ts(q1) )
P (t1 ) P (ts(q1) ) 1
()(1) q1
) .
(1

e
P
P monic irreducible in A,
P 6=

The proof of the second assertion of the Lemma is similar, using [4], Theorem 2.9,
and the properties of Gauss-Thakur sums ([22]).

Lemma 2.7.
B2q1 =

1i1 <...<iq 2q1

ti1 tiq .

Proof. Let T2q1 (K ) be the Tate algebra in the variable t1 , . . . , t2q1 with coefficients in K . Then:
1
T2q1 = A[t1 , . . . , t2q1 ] N,

where N = {f T2q1 , v (f ) 2}. Let : A A[t1 , . . . , t2q1 ]{ } be the

morphism of Fq -algebras give by = (t1 ) (t2q1 ) + . Then by the
results in [5], we have:
N = exp (T2q1 ),
T

and H := A[t1 ,...,t2q1

is a free Fq [t1 , . . . , t2q1 ]-module of rank one generated
2q1 ]N
1
by . Furthermore:
Bs =

det

Fq [t1 ,...,t2q1 ][Z]

(ZId |H Fq [t1 ,...,t2q1 ] Fq [t1 ,...,t2q1 ][Z] ) |Z= .

Now, observe that:

P
1
1i1 <...<iq 2q1 ti1 tiq
( )

The Lemma follows.

(mod A[t1 , . . . , t2q1 ] N ).



For m = (m1 , . . . , md ) Nd , we set m0 := s (m1 + + md ), and:

s (m) =

d Y
XY

u=1 iJu

tui ,

F F
where the sum runs through the disjoint unions J1 Jk {1, . . . , s} such that
| Ju |= mu , u = 1, . . . , d. Notice in particular that s (m) = 0 if m1 + + md > s,
that is, if m0 < 0. To give an example, the above lemma shows that B2q1 =
((q)).

` TUAN NGO DAC, AND FLORIC TAVARES RIBEIRO

BRUNO ANGLES,

10

Lemma 2.8. Let m Nd . We have :

s (m) |ts =0,...,ts(q2) =0 = s(q1) (m).

In particular, if m0 < q 1, we have:

s (m) |ts =0,...,ts(q2) =0 = 0.

Proof. This is a straight computation.

Let : Fp [t1 , . . . , ts ] N {+} be the function given by:

- if f = 0, (f ) = +,
P
- if f 6= 0, f =
j1 ,...,js tj11 tjss , j1 ,...,js Fp , then (f ) = Inf{j1 + . . . +
js , j1 ,...,js 6= 0}.
Lets write:
r
X
Bi,s ri ,
Bs =
i=0

where Bi,s Fp [t1 , . . . , ts ] is a symmetric polynomial, and B0,s = 1.

Proposition 2.9. For i = 1, . . . , r, we have:

(Bi,s ) i(q 1) + 1.
Proof. By Lemma 2.7, this is true for r = 1, thus we can assume that r 2. The
proof is by induction on r. Recall that by Lemma 2.2, we have:
Bs |t1 =...=ts =0 = r .

Thus, for i = 1, . . . , r, we can write:

X
Bi,s =
xi,m s (m), xi,m Fp ,
mS

where S = {(m1 , . . . , ms ) Ns , 1 m1 + + md s}. Set:

X
g
xi,m s (m).
B
i,s = Bi,s
mS,m0 <q1

Then:

g
(Bi,s B
i,s ) r(q 1) + 2.
Therefore we have to prove:
g
(B
i,s ) i(q 1) + 1.
Observe that, by Lemma 2.8, we have:

g
Bi,s |ts =...=ts(q2) =0 = B
i,s |ts =...=ts(q2) =0 =

xi,m s(q1) (m).

mS,m0 q1

By Lemma 2.6, we have:

Bs |ts =...=ts(q2) =0 = ( t1 ts(q1) )Bs(q1) .

We therefore get, for i = 1, . . . , r :

g
B
i,s |ts =...=t
=0 = Bi,s(q1) t1 ts(q1) Bi1,s(q1) ,
s(q2)

where we have set Br,s(q1) = 0. Now, by the induction hypothesis:

Thus:

(Bi,s(q1) t1 ts(q1) Bi1,s(q1) ) i(q 1) + 1.

(Bi,s ) i(q 1) + 1.

EXCEPTIONAL ZEROS OF L-SERIES AND BERNOULLI-CARLITZ NUMBERS

11


Corollary 2.10. Let N 1 (mod q 1), q (N ) 2q 1. Then a Fq [], we
have:
BN (t, ) |t=a 6= 0.
Proof. By Proposition 2.9, we have:
The Corollary follows easily.

BN (t, ) r t(t, )r .

Proof of Theorem 1.1

We can assume that q (N ) q. By Lemma 2.3, the total degree in t, of BN (t, )
(N )q
is strictly less than (r + 1)N, where r = q q1 . Now, by Corollary 2.10:
BN (, ) 6= 0.

Furthermore, deg BN (, ) < (r +1)N. Thus if P is a monic irreducible polynomial

in A such that deg P (r + 1)d, we have:
BN (, ) 6 0

(mod P ).

We conclude the proof of the Theorem by Proposition 2.5.

3. Exceptional zeros and eigenvalues of certain K-endomorphisms
3.1. The L-series LN (t).
Let N 1 be an integer. Recall that
X X a(t)N
X
LN (t) =
=
i,N (t)i , i,N (t) T
t .
a
d0 aA+,d

i0

Let d 1 be an integer, we set for k = (k0 , . . . , kd1 ) Nd :

- (k) = d,
- | k |= k0 + + kd1 ,
Qd1
- if a = a0 + a1 + + ad1 d1 + d , ai Fq , i = 0, . . . , d 1, ak = i=0 aki i .

Lets begin by a simple observation. Let d 1 and let k = (k0 , . . . , kd1 ) Nd .

Let N 1 be an integer, we get:
X
X
X
a(t)N ak =
C(N, m)ak am tm1 +2m2 ++dmd ,
aA+,d

aA+,d mNd+1 ,|m|=N

where:

N!
Fp .
m0 ! md !
Recall that by Lucas Theorem C(N, m) 6= 0 if and only if there is no carryover
p-digits in the sum N = m0 + + md . Furthermore, recall that, for n N,
P
n 6= 0 if and only if n 0 (mod q 1) and n 1. Thus, for m Nd+1 ,
PFq
am = 0 unless (m0 , . . . , md1 ) ((q 1)(N \ {0}))d and in this latter case
PaA+,d m
= (1)d .
aA+,d a
Thus for d, N 1, k {0, . . . , q 1}d, we denote by Ud (N, k) the set of elements
m Nd+1 such that:
- there is no carryover p-digits in the sum N = m0 + + md ,
- for n = 0, . . . , d 1, mn kn (q 1)N.
C(N, m) =

` TUAN NGO DAC, AND FLORIC TAVARES RIBEIRO

BRUNO ANGLES,

12

For m Ud (N, k), we set:

deg m = m1 + 2m2 + + dmd .

An element m Ud (N, k) is called optimal if deg m = Max{deg n, n Ud (N, k)}.

If Ud (N, k) 6= , the greedy element of Ud (N, k) is the element m = (m0 , . . . , md )
Ud (N, k) such that (md , . . . , m1 ) is largest lexicographically.
Let k Nd , d 1. For n = 0, . . . , d 1, let kn {0, . . . , q 1} be the least
integer such that kn + kn (q 1)(N \ {0}). We set:
= (k0 , . . . , kd1 ).
k
We get:

a(t)N ak = (1)d

C(N, m)tdeg m .

mUd (N,k)

aA+,d

Let N 1 be an integer and let q (N ) be the sum of digits of N in base q. Then

we can write in a unique way:
q (N )

N=

n=1

We set:

q en , e1 e2 eq (N ) .

r = Max{0, [
Lemma 3.1. We have:
i 0, i,N (t) =

q (N ) q
]} N.
q1

(1)(k) Ck

aA+,(k)

(k)+w(k)=i

a(t)N ak Fp [t],

|k|!
Fp , w(k) = kd1 + + (d 1)k1 + dk0 , for
where Ck = (1)|k| k0 !k
d1 !
k = (k0 , . . . , kd1 ).

Proof. Let a A+,d . We have:

1
1 X
1
= d
Ck ak w(k) ,
a

d
kN

|k|

where Ck = (1)

|k|!
k0 !kd1 !

Fp , w(k) = kd1 + + (d 1)k1 + dk0 . Thus:

X a(t)N
X
(1)d X
1
=
C
a(t)N ak .
k
w(k)
a
d

d
aA
aA
kN

+,d

Therefore:
i,N (t) =

(1)(k) Ck

(k)+w(k)=i

+,d

aA+,(k)

a(t)N ak Fp [t].


Lemma 3.2. Let j Z. Then LN (t) |t=qj = 0 if and only if N 1 (mod q 1),
and q j N > 1.
Proof. This
P comes
P from the following facts:
- n 1, d0 aA+,d a1n 6= 0,
P
P
- for n 0, d0 aA+,d an = 0 if and only if n 1, n 0 (mod q 1).

EXCEPTIONAL ZEROS OF L-SERIES AND BERNOULLI-CARLITZ NUMBERS

13

We will need the following Lemma in the sequel:

P
Lemma 3.3. Let F (t) = n0 n (t) 1n Fq [t][[ 1 ]], n (t) Fq [t], and let R
such that F (t) converges on {x C , v (x) }. Let M 1 and set FM (t) =
PM
1
n=0 n (t) n K [t]. Let R, . Suppose that FM (t) has exactly k 1
zeros in C with valuation . Then either F (t) has k zeros with valuation or F (t)
has at least degt FM (t) + 1 zeros with valuation > .
Proof. Lets assume that the side of the Newton polygon of FM (t) corresponding to
the k zeros of valuation is not a portion of a side of the Newton polygon of F (t),
then F (t) has a side of slope < with end point of abscissa k > degt FM (t).
Thus the Newton polygon of F (t) delimited by the vertical axis of abscissas 0 and
k has only sides of slope . Thus F (t) has k zeros of valuation .

3.2. An example.
For the convenience of the reader, we treat a basic example: N = 1. We set
d
0 = 1 and for d 1, d = ( q ) ( q ).

Lemma 3.4. Let d 0. Then:

X 1
1
= .
a
d

aA+,d

Proof. This is a well-known consequence of a result of Carlitz ([13], Theorem 3.1.5).

Lets give a proof for the convenience of the reader. We can assume that d 1.
Set:
Y
ed (X) =
(X a) A[X].
aA,deg a<d

Then ([13], Theorem 3.1.5):

ed (X) =

d
X

Dd

qi
i=0 Di di

Xq ,

q
where D0 = 1, and for i 1, Di = (q )Di1
. Now observe that:

X 1
d ))
|X=0 =
.
d
ed (X )
a

d
dX (ed (X

aA+,d

Since ed ( ) = Dd ([13], Corollary 3.1.7), we get the desired result.

Lemma 3.5. Let d 0. then:
X a(t)
aA+,d

(t ) (t q
=
a
d

Proof. We can assume that d 1. Set:

S=

d1

X a(t)
.
a

aA+,d

Then for i = 0, . . . , d 1, we have:

S |t=qi = 0.

Furthermore, by Lemma 3.4, S has degree d in t and the coefficient of td is

Lemma follows.

1
d .

The


14

` TUAN NGO DAC, AND FLORIC TAVARES RIBEIRO

BRUNO ANGLES,
d

q
Lemma 3.6. The edge points of the Newton polygon of L1 (t) are (d, q qq1
), d N.

Proof. Lets write:

L1 (t) =

Sd (t),

d0

where:
Sd (t) =

X a(t)
.
a

aA+,d

Let d N and let d > d. Let x K be the coefficient of td in Sd (t). Then, by

Lemma 3.5, we get:

v (x) v (d ) q d q d 1 > q
Thus, if we write:
L1 (t) =

d0

qd 1
.
q1

d td K [[t]], d K , d N,

by the above observation and again by Lemma 3.5, we get:

d 0, v (d ) = q

qd 1
.
q1

This latter Lemma implies the following formula due to F. Pellarin ([15], Theorem 1):
Proposition 3.7. Let C
be a fixed (q 1)th root of . Set:
Y
j

e =
(1 1q )1 C
.
j1

Then:

( t)L1 (t) =
Proof. We observe that:

e Y
(1 qj ).

j0

n 1, L1 (t) |t=qn = 0,

L1 () = 1.
By Lemma 3.6, the entire function (t )L1 (t) has simple zeros in K and if
x K is such a zero, v (x) {q i , i N}. Thus, there exists C
such that:
Y
t
(t )L1 (t) =
(1 qj ).

j0
But, observe that:

Therefore:

e Y
t
(1 qj ) |t= = .

j1
=

EXCEPTIONAL ZEROS OF L-SERIES AND BERNOULLI-CARLITZ NUMBERS

15

3.3. Eigenvalues and Bernoulli-Carlitz numbers.

In this paragraph, we slightly change our point of view. Let t be an indeterminate
over C and let : C [[ 1t ]] C [[ 1t ]] be the continuous (for the 1t -adic topology)
morphism of C -algebras such that (t) = tq . We first recall some consequences of
the work of F. Demeslays (see the appendix of [5] or [11]) generalizing the work of
L. Taelman ([20]).
P
Let N 1, N 1 (mod q1), q (N ) q. Write N = kl=0 nl q l , nl {0, . . . , q
1}, l = 0, . . . , k, and nk 6= 0. We set B = K[t]. Let (N ) : B B{} be the
morphism of K-algebras given by:
(N )

k
Y
l
= ( (q t)nl ) + t.
l=0

Since t is transcendental over Fq , there exists a unique power series exp(N )

K(t){{}} such that:
exp(N ) 1 (mod ),
(N )

exp(N ) t = t
One can easily see that:
exp(N ) =

exp(N ) .

X Qk (Qj1 (ql tqn ))nl

n=0
l=0
j .
Qj1
q n tq j )
(t
n=0
j0

In particular exp(N ) induces a continuous K-linear endomorphism of K(( 1t )) which

is an isometry on a sufficiently small neighborhood of zero (for the 1t -adic topology).
Lets set:
K(( 1t ))
H((N ) ) =
.
(B + exp(N ) (K(( 1t )))
Then H((N ) ) is a finite K-vector space and a B-module via . Lets denote by
[H((N ) )]B the monic generator (as a polynomial in t) of the Fitting ideal of the
B-module H((N ) ), i.e.:
(N )

[H((N ) )]B = det (ZId t

K[Z]

|H((N ) ) ) |Z=t .

As in [5], Proposition 7.2, one can prove that:

dimK H((N ) ) =

where:

q (N ) q
,
q1

1
B,
{x K(( )), exp(N ) (x) B} = q (N )q Q
k
t
ql )nl

(
t q1
l=0

j1

j
1
(1 t1q )1 Fp [[ ]] ,
t

1
q

( ) =
(1 qj )1 A[[ ]] .
t
t
j0
ql

Furthermore, if we set:
LN (t) =

P (t) monic irreducible polynomial of Fq [t]

B
]B
[ P (t)B
B
[(N ) ( P (t)B
)]B

` TUAN NGO DAC, AND FLORIC TAVARES RIBEIRO

BRUNO ANGLES,

16

then, by the appendix of [5], LN (t) converges in K(( 1t )), and:

[H((N ) )]B
t

q (N )q
q1

Qk

l=0

(ql )nl

= LN (t).

Now, one can compute LN (t) as in [5], paragraph 5.3, and we get:
LN (t) =

X X

d0 aA+,d

1
aN
A[[ ]] .
a(t)
t

Therefore:
[H((N ) )]B = BN (, t).
We warn the reader not to confuse BN (, t) and BN (t, ), here and in the sequel
of the paper, since we will be interested in those two polynomials. Recall that r =
q (N )q
q1 . Let 1 (N ), . . . , r (N ) C be the eigenvalues (counted with multiplicity)
(N )

of the K-endomorphism of H((N ) ): t

BN (, ) =

. We get:

r
Y

j=1

( j (N )).

aN
aA+,d a(t)

A[[ 1t ]]. By Lemma 2.1, LN (t) conPk

verges on {x C , v (x) < 0}. Lets write N = l=0 nl q l , nl {0, . . . , q1}, nk 6=
i
0. Set S = {q , i k}. Then, by Lemma 3.2, the elements of S are zeros of LN (t)
The elements of S are called the trivial zeros of LN (t). A zero of LN (t) which
does belong to {x C , v (x) < 0} \ S will be called an exceptional zero of
LN (t). It is clear that the exceptional zeros of LN (t) are roots of BN (, t) with the
same multiplicity. Our aim in the remaining of the article is to study the following
problem:
Recall that LN (t) =

d0

Problem 1.
(N )
Let N 2, N 1 (mod q 1), q (N ) q. Then all the eigenvalues of t
1
(N )
(viewed as a K-endomorphism of H( )) are simple and belong to Fp (( )).
(N )

Theorem 1.1 implies that is not an eigenvalue of t . We presently do not

(N )
know whether another trivial zero of LN (t) can be an eigenvalue of t .On the
other side, the above problem implies that the exceptional zeros of LN (t) are simple.
Observe that, by Lemma 2.2, the above Problem has an affirmative answer for
q q (N ) 2q 1.
4. Answer to Problem 1 for q = p
In this section we give an affirmative answer to Problem 1 in the case q = p.
By Proposition 2.5 and Proposition 4.6 below, this implies Theorem 1.1. For the
convenience of the reader, we have tried to keep the text of this section as selfcontained as possible.
In this section q = p.

EXCEPTIONAL ZEROS OF L-SERIES AND BERNOULLI-CARLITZ NUMBERS

17

4.1. Preliminaries.
Lemma 4.1 and Proposition 4.2 below are slight generalizations of the arguments
used in the proof of Theorem 1 in [12].
Lemma 4.1. Let d, N 1 and k = (k0 , . . . , kd1 ) {0, . . . , p 1}d. We assume
Pi1
that | k | p (N ). For 1 i d, we set : i = n=0 kn . We also set 0 = 0 and
d+1 = p (N ). Let m = (m0 , . . . , md ) Nd+1 be the element defined as follows:
n+1

n = 0, . . . , d, mn =

pel .

l=n +1

Then:

m Ud (N, k).
Furthermore m is the greedy element of Ud (N, k). In particular Ud (N, k) 6= if and
only if | k | p (N ).
Proof. Observe that d =| k | p (N ). Thus m is well-defined. It is then straightforward to verify that m Ud (N, k) and that m is the greedy element of Ud (N, k).
Now assume that Ud (N, k) 6= . Let m Ud (N, k). Then:
This implies:

n = 0, . . . , d 1, p (mn ) kn

(mod p 1).

n = 0, . . . , d 1, p (mn ) kn .

Thus:

p (N )

d1
X

n=0

p (mn ) | k | .


| p (N ). Then
Proposition 4.2. Let d, N 1 and k Nd . We assume that | k

Ud (N, k) contains a unique

optimal element which is equal to the greedy element of
N k

In particular P
Ud (N, k).
aA+,d a(t) a 6= 0 if and only if | k | p (N ).

Let m Ud (N, k)

Proof. Let u = (u0 , . . . , ud ) be the greedy element of Ud (N, k).

such that m 6= u. We will show that m is not optimal.
Write cn = p (mn ), n = 0, . . . d 1. Then :
n = 0, . . . , d 1, cn kn , cn kn (mod p 1).

For n = 0, . . . , d 1, there exist fn,1 fn,cn such that we can write in a

unique way:
cn
X
pfn,l .
mn =
l=1

Case 1) There exists an integer j, 0 j d 1, such that cj > kj .

Let m Nd+1 be defined as follows:
- mn = mn for 0 n d 1, n 6= j,
Pkj fj,l
p .
-mj = l=1
-md = N m0 md1 = md + mj mj .
Then m Ud (N, k) and:
deg m = deg m + (d j)(mj mj ) > deg m.

Thus m is not optimal.

` TUAN NGO DAC, AND FLORIC TAVARES RIBEIRO

BRUNO ANGLES,

18

Case 2) For n = 0, . . . , d 1, cn = kn .
Let j {0, . . . d 1} be the smallest integer such that mj 6= uj . Then, by the
construction of u, we have:
mj > u j .
Thus there exists an integer l such that the number of times pl appears in the sum
of mj as kj powers of p is strictly greater than the number of times it appears in the
sum of uj as kj powers of p. Also, there exists an integer v such that the number
of times pv appears in the sum of uj as kj powers of p is strictly greater than the
number of times it appears in the sum of mj as kj powers of p. Thus there exists
an integer t > j such that pv appears in the sum of mt as p (mt ) powers of p. We
observe that, by the construction of u, we can choose v and l such that v < l. Now
set:
- for n = 0, . . . , d, n 6= j, n 6= t, mn = mn ,
- mj = mj pl + pv ,
- mt = mt pv + pl .
and:
Let m = (m0 , . . . , md1 ) Nd . Then m Ud (N, k)
deg m =

d
X
l=0

Thus m is not optimal.

lml = deg m + (t j)(pl pv ) > deg m.



We have the following key result:

Proposition 4.3. Let d, N 1 and k Nd . We assume that p (N ) p and that
d(p 1) p (N ) p. Then:
X
a(t)N ak N d.
N (d 1) < degt
aA+,d

| d(p 1).
Proof. It is clear that degt aA+,d a(t)N ak N d. Observe that | k
By Proposition 4.2, we have:
Let m be the greedy element of Ud (N, k).
degt

a(t)N ak =

d
X

n=0

aA+,d

nmn = dN

d
X

nmdn .

n=1

By Lemma 4.1, we have:

n = 0, . . . , d 1, mn kn pek0 ++kn ,
where we recall that:
p (N )

N=

pen .

n=1

Let l = p (N ) 1. Observe that:

ep (N )p(p1)t el 1 t.

Since d(p 1) p (N ) p, we get:

n = 1, . . . , d, k0 + + kdn (p 1)(d n + 1) p (N ) p (p 1)(n 1).
Thus:
n = 1, . . . , d, mdn (p 1)pel n ,

EXCEPTIONAL ZEROS OF L-SERIES AND BERNOULLI-CARLITZ NUMBERS

19

Therefore:
d
X

n=1

nmdn (p 1)pel

Recall that if x R \ {1}, we have:

d
X

nxn1 =

n=1

Thus:
(p 1)
Now:

d
X

npn =

n=1

npn .

n=1

1 xd+1 + (d + 1)(x 1)xd

.
(x 1)2

p pd (d + 1)(p 1)pd
p
<
.
p1
p1

d
X

nmdn <

n=1

Thus:

d
X

d
X

p el
p .
p1

nmdn < 2pel .

n=1

But 2pel N since l = p (N ) 1. Thus:

degt

aA+,d

a(t)N ak = dN

d
X

n=1

nmdn > dN N.


4.2. Newton polygons of truncated L-series.

For i, j 0, we set:
X
a(t)i Fp [t].
Sj (i) =
aA+,j

Note that by Proposition 4.2, we have Si (N ) 6= 0 for i = 0, . . . , r.

Proposition 4.4. We have:
i = 0, . . . , r, degt i,N (t) = degt Si (N ).
Proof. We recall that:
r = Max{[

p (N ) p
], 0} N.
p1

We can assume that r 1. By Proposition 4.3, for i = 0, . . . , r,

X
Max{degt
a(t)N ak , w(k) + (k) = i}
aA+,(k)

is attained for a unique k which is (0, . . . , 0) Ni . It remains to apply Lemma

3.1.

We set:
r (N ) =

r
X
i=0

i (N )i K [t].

` TUAN NGO DAC, AND FLORIC TAVARES RIBEIRO

BRUNO ANGLES,

20

Lets write:
N=

k
X
l=0

nl pl , n0 , . . . , nk {0, . . . p 1}, nk 6= 0.

Proposition 4.5. We have degt r (N ) = degt Sr (N ). Furthermore the edge points

of the Newton polygon of r (N ) are:
(degt Sj (N ), j), j = 0, . . . , r.
Proof. We can assume that r 1. By Proposition 4.2, we have Ur (N, (p 1, . . . , p
1)) 6= . Let m Nr+1 be the optimal element of Ur (N, (p 1, . . . , p 1)) given by
Proposition 4.2 and Lemma 4.1. For j = 0, . . . , r, let m(j) = (m0 , . . . , mj1 , N
Pj1
j+1
. Then, again by Proposition 4.2 and Lemma 4.1, m(j) is the
n=0 mn ) N
optimal element of Uj (N, (p 1, . . . , p 1)). Therefore, degt Sj (N ) = deg m(j), j =
0, . . . , r. For j = 0, . . . , r, we have:
pk+1 > N
Now let j {0, . . . , r 1}, we have:

j1
X

mn > n k p k .

n=0

pk+1 > deg m(j + 1) deg m(j) = N

j
X

mn > n k p k .

n=0

Thus, by Proposition 4.4, we get degt r (N ) = degt Sr (N ). Furthermore, we observe that for j {1, . . . , r 1}, we have:
deg m(j) deg m(j 1) > deg m(j + 1) deg m(j).
Thus, one easily sees that the edge points of the Newton polygon of r (N ) are
(deg m(j), j), j = 0, . . . , r.

4.3. A positive answer to Problem 1.
Let N 2 be an integer, N 1 (mod p 1). Recall that:
N=

k
X
l=0

nl pl , n0 , . . . , nk {0, . . . p 1}, nk 6= 0.

Recall that p (N ) = n0 + + nk and r = M ax{

total degree in t, of the polynomial BN (t).

p (N )p
p1 , 0}.

Let bN N be the

Proposition 4.6. The polynomial BN (t, ) has only one monomial of total degree
bN , which is of the form tbN . Furthermore:
bN = degt Sr (N ).
Proof. We can assume that r 1. First lets observe that:
Y

j1

(1 1p )1 BN (t, ) (1)

p (N )1
p1

where:
N =

k Y
Y

l=0 j0
r
X
i=0

(1

ri (N )i .

tp nl
)
pj

(mod

1
1
Fp [t][[ ]]),

EXCEPTIONAL ZEROS OF L-SERIES AND BERNOULLI-CARLITZ NUMBERS

21

Let N Fp [t, ] be uniquely determined by the congruence:

N

k Y
Y

tp
N
(1 pj )nl

l=0 j0

(mod

1
1
Fp [t][[ ]]).

(mod

1
1
Fp [t][[ ]])

Let i {0, . . . , r}. A monomial in the product

i (N )i

k Y
Y

l=0 j0

is of the form:

(1

tp nl
)
pj

t
, j degt i (t), i, pk .

Thus the total degree of a monomial in

i tj

i (N )i

k Y
Y

l=0 j0

(1

tp nl
)
pj

(mod

1
1
Fp [t][[ ]])

is less than or equal to pk i + degt i (t). To conclude the proof of the Proposition, we
use the same arguments as that used in the proof of Proposition 4.5. By Proposition
4.2, we have Ur (N, (p 1, . . . , p 1)) 6= . Let m Nr+1 be the optimal element
of Ur (N, (p 1, . . . , p 1)) given by Proposition 4.2 and Lemma 4.1. For j =
P
0, . . . , r, let m(j) = (m0 , . . . , mrj1 , N rj1
mn ) Nrj+1 . Then, again
n=0
by Proposition 4.2 and Lemma 4.1, m(j) is the optimal element of Urj (N, (p
1, . . . , p 1)). For j = 0, . . . , r, we have:
N
Now let j {0, . . . , r 1}, we have:

rj1
X

mn > p k .

n=0

deg m(j) deg m(j + 1) = N

rj1
X

mn > p k .

n=0

Thus, by Proposition 4.3 and Proposition 4.4, Max{pk i + degt i (t), i = 0, . . . , r}

is attained exactly at i = 0. Again by Proposition 4.4, this implies that the total
degree in t, of N is equal to degt Sr (N ) and that N (t) has only one monomial of
total degree degt Sr (N ) which is of the form tdegt Sr (N ) . The Proposition follows. 
Theorem 4.7. Let N 2, N 1 (mod p 1). The polynomial BN (, t) (viewed
as a polynomial in t) has r simple roots and all its roots are contained in Fp (( 1 )) \
i
{p , i Z}.
Proof.
Recall that BN (t, ) is a monic polynomial in such that deg BN (t, ) = r. We
can assume that r 1. By Proposition 4.6, the leading coefficient of BN (t, ) as a
polynomial in t is in F
p and:
bN = degt BN (t, ) > r.
i

Let S = {p , i k}. Then if C is a zero of LN (t) and 6 S, must be a

zero of BN (t, ). Observe that by Proposition 4.5 and Proposition 4.6, we have:
degt r (N ) = degt BN (t, ).

22

` TUAN NGO DAC, AND FLORIC TAVARES RIBEIRO

BRUNO ANGLES,

By Proposition 4.5, the zeros in C of r (N ) are not in S. By Lemma 2.1 and

Lemma 3.3, r BN (t, ) and r (t) have the same Newton polygon. Thus, by the
proof of Proposition 4.5 and the properties of Newton polygons ( [13], chapter 2),
we get in K [t] :
r
Y
Pj (t),
BN (t, ) =
j=1

where F
p , Pj (t) is an irreducible monic element in K [t], Pj (t) 6= Pj (t) for
j 6= j . Furthermore each root of Pj (t) generates a totally ramified extension of
K and pk+1 > degt Pj (t) > nk pk . Also note that degt Pj (t) 6 0 (mod pk ) and
degt Pj (t) 1 (mod p 1). Observe that if x iZ (Fp (( 1pi ))ab )perf , then there
m
exist l 0, m Z, d 1, p 1 mod d, such that v (x) = dp
l . Thus, Pj (t) has no
1
ab perf
roots in iZ (Fp (( pi )) )
.
Write:
r
X
j (t)j , j (t) Fp [t].
r BN (t, ) =
j=0

Observe that 0 (t) = 1 and by the above discussion, r BN (t, ) and r (t) have
the same Newton polygon (as polynomials in t). Now, by Proposition 4.5 and
Proposition 4.6, we get:
degt r (t) = degt r (t).
We deduce that:
i = 0, . . . , r, degt i (t) = degt i (t).
By the proof of Proposition 4.5, for i {1, . . . , r 1}, degt i+1 (t) degt i (t) <
degt i (t) degt i1 (t). Thus the edges of the Newton polygon of r BN (t, )
viewed as polynomial in 1 are (i, degt (i (t))), i = 0, . . . , r.

5. Some hints for Problem 1 for general q.
In this section q is no longer assumed to be equal to p.
5.1. The work of J. Sheats.
For N, d 1, we set Ud (N ) = Ud (N, (q 1, . . . , q 1)). Thus:
X
X
C(N, m)tdeg m .
a(t)N = (1)d
Sd (N ) :=
aA+,d

mUd (N )

J. Sheats proved ([18], Lemma 1.3) that if Ud (N ) 6= , Ud (N ) has a unique optimal

element and it is the greedy element of Ud (N ). In particular Ud (N ) 6= Sd (N ) 6=
0. Observe that if m = (m0 , . . . , md ) Ud (N ), then (m0 , . . . , md2 , md1 + md )
Ud1 (N ). In particular Ud (N ) 6= Ud1 (N ) 6= .
Proposition 5.1.
1) Let d 1 such that Ud (N ) 6= . Then, for j {1, . . . , d 1}, we have:
degt Sj (N ) degt Sj1 (N ) > degt Sj+1 (N ) degt Sj (N ).

2) Let d 1 such that Ud+1 (N ) 6= . Let m be the greedy element of Ud+1 (N ).

Then:
degt Sd (N ) > N (d 1),
and:
degt Sd (N ) degt Sd1 (N ) > md+1 .

EXCEPTIONAL ZEROS OF L-SERIES AND BERNOULLI-CARLITZ NUMBERS

23

Proof.
1) Observe that this assertion is a consequence of the proof of [18], Theorem 1.1
(see pages 127 and 128 of [18]).
2) Let m = (m0 , . . . , md+1 ) be the greedy element of Ud+1 (N ). Define m =
(m0 , . . . , md ) Ud (N md+1 ). Then:
md 0

(mod q 1), md q 1.

Furhermore, observe that m is the greedy element of Ud (N md+1 ). By [18],

Lemma 1.3 and Proposition 4.6, we get:

degt Sd (N md+1 ) > (N md+1 )(d 1).

Let m = (m0 , . . . , md1 , md + md+1 ) Ud (N ). We have:

degt Sd (N ) m1 + + (d 1)(md1 ) + d(md + md+1 ).
Thus:
degt Sd (N )
>

degt Sd (N md+1 ) + dmd+1

(N md+1 )(d 1) + dmd+1 = (d 1)N + md+1 .

Thus:
degt Sd (N ) > N (d 1),

degt Sd (N ) degt Sd1 (N ) degt Sd (N ) (d 1)N > md+1 .

To conclude this paragraph, we recall the following crucial result due to G. Bockle
([8], Theorem 1.2):
Sd (N ) 6= 0 d(q 1) Min{q (pi N ), i N}.
An integer N 1 will be called q-minimal if:
[

q (pi N )
q (N )
] = Min{[
], i N}.
q1
q1

5.2. Consequences of Sheats results.

Let N 1, and write:
X
LN (t) =
i,N (t)i , i,N (t) Fq [t].
i0

Proposition 5.2. Let d 1 such that Ud+1 (N ) 6= . Set:

d (t) =

d
X
i=0

i,N (t)i K [t].

Then degt d (t) = degt Sd (N ) and the edge points of the Newton polygon of d (t)
are:
(degt Si (N ), i), i = 0, . . . , d.
Proof. The proof uses similar arguments as that used in the proof of Proposition
4.5. Let j 0, then (see Lemma 3.1), we have:
X
X
(1)(k) Ck
a(t)N ak .
j,N (t) =
(k)+w(k)=j

aA+,(k)

24

` TUAN NGO DAC, AND FLORIC TAVARES RIBEIRO

BRUNO ANGLES,

Observe that:

degt

aA+,(k)

a(t)N ak (k)N.

Thus, for j = 0, . . . , d, by Proposition 5.1, assertion 2), we get:

degt j,N (t) = degt Sj (N ).
In particular, again by Proposition 5.1, assertion 2), we have:
degt d (t) = degt Sd (N ).
Finally, by Proposition 5.1, assertion 1), (degt Si (N ), i), i = 0, . . . , d, are the edge
points of the Newton polygon of d (t).

Lemma 5.3. We assume that N is q-minimal, N 1 (mod q1). We also assume
(N )q
that r 1 (recall that r = q q1 ).
Pk
1) Write N = l=0 nl q l , n0 , , nk {0, . . . , q 1}, nk 6= 0. Then there exists an
integer 0 m k such that nm 6 0 (mod p).
2) Let n = Max{l, 0 l k, nl 6 0 (mod p)}. Let m = (m0 , . . . , mr+1 ) Ur+1 (N )
be the greedy element, then:
mr+1 = q n .
3) Assume that n < k. Then, for i {1, . . . , r 1}, we have:
degt Si (N ) degt Si1 (N ) > pq k .

Proof.
1) Lets assume that the assertion is false. Then:
q
q (N )
q ( N ) =
.
p
p
This contradicts the q-minimality of N.
2) Observe that:
i 0, q (pi N ) = q (pi N pi q n ) + q (pi ).

Therefore, N q n is q-minimal. Thus, by Bockles result: Ur+1 (N q n ) 6= . This

easily implies that there exits n = (n0 , . . . , nr+1 ) Ur+1 (N ) such that:
nr+1 = q n .

We have:
1 + (q 1)(r + 1) = q (N ) =
Furthermore:

r
X
i=0

and

q (mi ) 0

r
X
i=0

Thus:

r+1
X

q (mi ).

i=0

(mod q 1),

q (mi ) (q 1)(r + 1).

q (mr+1 ) = 1.

EXCEPTIONAL ZEROS OF L-SERIES AND BERNOULLI-CARLITZ NUMBERS

25

This implies that mr+1 is a power of q and since m is the greedy element of
Ur+1 (N ), we also have: mr+1 | q n . Since there is no carryover p-digits in the sum
m0 + + mr+1 , by the definition of n, we deduce that mr+1 = q n .
3) Let m = (m0 , . . . , mr1 , mr + mr+1 ) Ur (N ). If n is the greedy element of
Ur (N ) then:
nr mr + mr+1 .
Since m is the greedy element of Ur+1 (N ), we have:
m0 m1 mr .

Since there is no carryover p-digits in the sum m0 + + mr+1 , and n < k, this
implies that:
mr =

k
X
l=0

Thus:

jl q l , jl {0, . . . , q 1}, jk 6= 0, jk 0 (mod p).

mr > pq k .

It remains to apply Proposition 5.1.

5.3. Zeros of BN (, t).

The following theorem implies in particular Theorem 1.1 in the case where N is
q-minimal.
Theorem 5.4. We assume that N is q-minimal, N 1 (mod q 1). We also
assume that r 1.
1) BN (, ) 6= 0 and the zeros of BN (t, ) are algebraic integers (i.e. they are
integral over A). Furthermore: (r 1)N < degt BN (t, ) < rN.
2) BN (, t) has only simple roots and its roots belong to Fp (( 1 )) \ {}.
Proof. The proof is in the same spirit as that of the proofs of Proposition 4.6 and
Theorem 4.7. P
k
Write N = l=0 nl q l , where n0 , . . . , nk {0, . . . , q 1}, and nk 6= 0. Recall that
q (N )
r + 1 = [ q1 ]. Let n 0 be the integer such that n = Max{l, nl 6 0 (mod p)} (see
Lemma 5.3). Let N (t) Fp [t, 1 ] be the polynomial determined by the congruence:
N (t) r (t)
where r (t) =

Pr

l=0

k Y
Y

(1

j=0 i0

tq nj
)
qi

(mod

1
r+1

1
Fp [t][[ ]]),

l,N (t)l . We can write:

N (t) =

r
X
l=0

l (t)l , l (t) Fp [t].

Note that l (t) is a Fp -linear combination of terms of the form

xl,j,u = lu,N (t)l+u tj u , j q k u.

By Proposition 5.2, we have:

l = 0, . . . , r, degt l,N (t) = degt Sl (N ).

1) Case n = k.
By Proposition 5.1 and Lemma 5.3 :
l = 0, . . . , h 1, degt hl,N (t) < degt h,N (t) q k l.

26

` TUAN NGO DAC, AND FLORIC TAVARES RIBEIRO

BRUNO ANGLES,

Thus, if l 6= h or u 6= 0, we get:

l = 0, . . . , h, degt xl,j,u < degt l,N (t).

Therefore:
l = 0, . . . , r, degt l (t) = degt Sl (N ).
2) Case n < k.
As in the proof of the case n = k, we get by Proposition 5.1 and Lemma 5.3:
l = 0, . . . , r 1, degt l (t) = degt Sl (N ).

Furthermore, by Proposition 5.1 and Lemma 5.3, we have for l 2:

degt Srl (N ) < degt Sr1 (N ) pq k (l 1),

Thus, for u 2:

degt Sr (N ) degt Sr1 (N ) > q n .

degt xr,j,u degt Sru (N ) + q k u < degt Sr (N ).

Now, observe that:

k Y
Y

j=0 i0

Thus:
Thus we get:

1 X ql
tq nj
1
nl t
i )
q

l=0
j

(1

(mod

1
1
Fp [t][[ ]]).
2

degt xr,j,1 degt Sr1 (N ) + q n < degt Sr (N ).

degt r (t) = degt Sr (N ).

Now, observe that:

q (N )1
q1

r BN (t, )
N
1qj )
j1 (1

(1)
Q

(mod

1
r+1

1
Fp [t][[ ]]).

We easily deduce that:

r
X
r BN (t, ) =
l (t)l , l (t) Fp [t], degt l (t) = degt Sl (N ), l = 0, . . . , r.
l=0

Observe that, by Proposition 5.1, we have degt Sr (N ) > N (r 1), and it is obvious
that degt Sr (N ) < rN. Now, we get assertion 1) and 2) by the same reasoning as
that used in the proof of Theorem 4.7.

Corollary 5.5. We assume that N is q-minimal, N 1 (mod q 1). We also
i
assume that r 1. Then, BN (t, ) has at most one zero in {q , i Z}.
i

Proof. Lets assume that BN (t, ) has a zero {q , i Z}. Let n be the integer
introduced in the proof of Theorem 5.4. Then:
i

= q , k i > n,

where q k N < q k+1 . Thus n < k, and therefore, by Lemma 5.3, we must have:
degt Sr (N ) degt Sr1 (N ) = q i .

By the proof of Theorem 5.4, we have:

BN (t, ) =

!
i
tq
1 F (t),

EXCEPTIONAL ZEROS OF L-SERIES AND BERNOULLI-CARLITZ NUMBERS

27

Pr1
where F (t) = l=0 l (t)l , l (t) Fp [t], degt l (t) = degt Sl (N ), l = 0, . . . , r 1.
i
Furthermore 0 (t) = 1.This implies that the zeros of F (t) are not in {q , i
Z}.

Corollary 5.6.
1) The polynomial Bs is square-free, i.e. Bs is not divisible by the square of a
non-trivial polynomial in Fq [t1 , . . . , ts , ].
l

2) For all l, n N, Bs is relatively prime to (tq1 q ) (observe that Bs is a

symmetric polynomial in t1 , . . . , ts ).
3) For all monic irreducible prime P of A, Bs is relatively prime to P (t1 ) P (ts )
P.
Proof. Let N = q e1 + . . . + q es , 0 e1 < e2 < . . . < es . Then:
BN (t, ) = Bs |ti =tqei .
We observe that N is q-minimal. Thus we can apply Theorem 5.4. This Theorem
i
and its proof imply that BN (t, ) is square-free and has no roots in {q , i Z}.
This proves 1) and 2).
Let P be a monic irreducible polynomial in A. Suppose that that P (t1 ) P (ts )P
and Bs are not relatively prime.Then P (t)N P and BN (t, ) are not relatively
prime. But, by the proof of Theorem 5.4, if C is a root of BN (t, ), then:
v () >

1
.
N

Now, observe that if C is a root of P (t)N P, then v () =

to a contradiction.

1
N .

This leads


Note that assertion 1) of the above Corollary gives the cyclicity result implied
by [7], Theorem 3.17, but by a completely different method.
6. An example
We study here an example of an N which is not q-minimal, so that our method
does not apply. We choose q = 4, and N = 682 = 2 + 2 4 + 2 42 + 2 43 + 2 44 .
We get lq (N ) = 10 = 3q 2 so that deg (BN (t, )) = 2. Moreover, lq (pN ) = 5 so
that N is not q-minimal. By using the table of section B, we get :


BN (t, ) = 2 + t10 + t34 + t40 + t130 + t136 + t160 + t514 + t520 + t544 + t640


+ t170 + t554 + t650 + t674 + t680 .

The Newton polygon of BN (t, ) has then the end points (0, 2), (640, 1), (680, 0).
1
and 40 distinct
We deduce that BN (t, ) has 640 distinct zeroes of valuation 640
1
zeros of valuation 40 . Similarly, BN (, t) has two zeros of respective valuations
40 and 640. In particular, we still have that BN (t, ) has no zero of the form
i
q , i Z, and an affirmative answer to Problem 1.
A. Appendix: The digit principle and derivatives of certain L-series,
by B. Angl`
es, D. Goss, F. Pellarin and F. Tavares Ribeiro
We keep the notation of the article.

28

` TUAN NGO DAC, AND FLORIC TAVARES RIBEIRO

BRUNO ANGLES,

Pr
i
Let N be a positive integer. We consider its base-q
Pr expansion N = i=0 ni q ,
with ni {0, . . . , q 1}. We recall that q (N ) = i=0 ni and the definition of the
Carlitz factorial :
Y
(N ) =
Dini A+ ,
i0

j1

where [k] = q if k > 0 and Dj = [j][j 1]q [1]q

for j > 0.
It is easy to see (the details are in A.1, A.2 and A.3) that, if we denote by a
d
a of a A with respect to , the series
the derivative d
X X aN
a

k1 aA+,k

converges in K to a limit that we denote by N . In particular, if n = q j with

j > 0, we will see (Proposition A.4) that
X 1
Dj 1qj
and qj =

e
.
1 =
[k]
[j]
k1

Our aim is to prove the following:

Theorem A.1. If N q is such that N 1 (mod q 1) and q (N ) q, then

n
r 
N
(N ) Y qk k
,
= N
q

e
([ Nq ])
k=1

s1

where for x R, [x] denotes the integer part of x, and where N = (1) q1 BN (, ).
Our Theorem A.1 can be viewed as a kind of digit principle for the values j in
the sense of [10].
The plan of this appendix is the following. In A.1, we recall the first properties
of Anderson and Thakur function . In A.2 we discuss the one-digit case of our
Theorem, while the general case is discussed in A.3.
A.1. The Anderson-Thakur function. Recall that Tt denotes the Tate algebra
over C in the variable t, C : A A{ } is the Carlitz module ([13], chapter 3), in
other words, C is the morphism of Fq -algebras given by C = + , and
X 1
expC =
i Tt {{ }}
Di
i0

is the Carlitz exponential. In particular, we have the following equality in Tt {{ }} :

expC = C expC .

Let us choose a (q 1)-th root q1 of in C and set:

Y
j

e = q1
(1 1q )1 C
.
j1

We recall the Anderson-Thakur function ([1], proof of Lemma 2.5.4):

1

Y
t
q1

T
(t) =
1 qj
t .

j0

EXCEPTIONAL ZEROS OF L-SERIES AND BERNOULLI-CARLITZ NUMBERS

29

To give an idea of how to compute expC (f ) for certain f in Tt , we verify here that
 X

j

eq

e
=
expC
t
Dj (qj t)
j0

is a well defined element of Tt . Indeed, for j 0 :

!


j

eq
q
j
.
v
=q j+1
q1
Dj (qj t)

P
qj
Therefore j0 D (e qj t) converges in Tt .
j
We will need the following crucial result in the sequel:

Proposition A.2. We have the following equality in Tt :




e
.
(t) = expC
t

Proof. It is a consequence of the formulas established in [15]. We give details for

the convenience of the reader. Let us set



e
.
F (t) = expC
t
We observe that:




( t + t)e

= expC
C (F (t)) = expC
t
t




te

e
= expC (e
) + expC
= t expC
= tF (t).
t
t
Therefore:
(F (t)) = (t )F (t).
But we also have:
((t)) = (t )(t).
Finally, we get:


F (t)
F (t)
=
.

(t)
(t)
It is a simple and well-known consequence of a ultrametric variant of Weierstrass
preparation Theorem that {f Tt , (f ) = f } = Fq [t]. Since T
t , we have then:
F (t)
Fq [t].
(t)

Now observe that

X
X
e
j
F (t) = expC
=
j+1 tj
t
j+1
j0

and that, for all j 0, v (j+1 ) = j + 1

j0

q
q1 .

This implies v ( F(t)

1) > 0. By

the definition of (t), we also have v ( (t)

1) > 0. Thus:


F (t)
v
1 > 0.
(t)

Since

F (t)
(t)

Fq [t], we get (t) = F (t).

` TUAN NGO DAC, AND FLORIC TAVARES RIBEIRO

BRUNO ANGLES,

30

Notice that (t) defines a meromorphic function on C without zeroes. Its only
2
poles, simple, are located at t = , q , q , . . .. As a consequence of Proposition A.2,
we get:
Corollary A.3. Let j 0 be an integer, then:

(t q )(t) |t=qj =

eq
.
Dj

A.2. The one digit case. Let us consider the following L-series:
X X a(t)
L(t) = L1 (t) =
Tt .
a
k0 aA+,k

By Proposition 3.7, we have the following equality in Tt (see [15], Theorem 1):
1
L(t)(t)
=
.

e
t
This implies that L(t) extends to an entire function on C (see also Lemma 2.1 or
[3, Proposition 6]). We set:
X X a (t)
Tt ,
L (t) =
a
k0 aA+,k

d
a(t) of a(t) with respect to t. The derivawhere a (t) denotes the derivative dt
d
tive dt inducing a continuous endomorphism of the algebra of entire functions
C C , L (t) extends to an entire function on C . Thus, for j 0 an inP
P
qj
teger, k1 aA+,k a a converges in K and we have:

q j =

X X aqj
= L (t) |t=qj .
a

k1 aA+,k

Proposition A.4. The following properties hold:

(1) We have:
X 1
.
1 =
[k]
k1

(2) Let j 1 be an integer, then:

q j =
Proof.

(q j ) 1qj

e
.
[j]

(1) It is well known that, for n > 0, Dn =

a
a

1
[n]

aA+,n

=
from which the first formula follows.
(2) By [13, Remark 8.13.10], we have:
aA+,n

L(t) |t=qj = 0.

Thus:

qj = L (t) |t=qj =
But,
L(t)
j
t q

L(t)
| qj .
t qj t=

(t q )(t)

1
.
t

a. Therefore,

EXCEPTIONAL ZEROS OF L-SERIES AND BERNOULLI-CARLITZ NUMBERS

31

It remains to apply Corollary A.3.


P
1
Remark A.5. The transcendence over K of the bracket series 1 = i1 [i]
was first obtained by Wade [23]. The transcendence of 1 directly implies the
transcendence of
e.

A.3. The several digits case. As a consequence of [5], Lemma 7.6 (see also [3],
N
P
P
has a zero of order at least
Corollary 21), the series LN (t) = d0 aA+,d a(t)
a
N at t = . Thus,
e N (t) =
L

X X a (t)N
a
d1 aA+,d

defines an entire function on C such that

e N ().
N = L

Pr
Proof of Theorem A.1. Recall that N = i=0 ni q i , is the q-expansion of N . We
set s = q (N ). We can assume that s q by Proposition A.4. From the definition
of BN (t, ) in 2.3, we have :
(1)

s1
q1

r
Y

BN (t, ) = LN (t)

qi ni

(t )

i=0

Since
e N () =
N = L

LN (t)
qi ni
i=0 (t )

Qr

e1 A[t].

|t=

we obtain, by Corollary A.3 and our previous discussions:

N =

Qr

e
q
i=0 ( Di

)ni

Now, by Proposition A.4, we have, for all k 1, Dk = [k]qk

eq
Theorem by using the fact that:

. We obtain the

Y
(N )
=
[k]nk .
([ Nq ])q
k1

B. Table
We give an explicit expression of the polynomials Bs for s {q, 2q 1, 3q 2}.
sq
. One obtains the corresponding
We recall that Bs is monic of degree r = q1

` TUAN NGO DAC, AND FLORIC TAVARES RIBEIRO

BRUNO ANGLES,

32

expressions for BN (t, ) if q (N ) = s by evaluating the variables ti s as in 2.3.

Bq
B2q1

B3q2

= 1,
=

i1 <<iq

= 2

ti1 tiq ,
2q1
Y

i1 <<i2q1 j=1

q
Y

i1 <<iq j=1

t2ij

tij +

q
Y

i1 <<iq j=1

q1
Y

m1 <<mq1 ,mj 6=ij j=1

tij +
tij +

2q
Y

i1 <<i2q j=1

tij .

One easily computes the discriminant of B3q2 from this table. It is then an
easy computation to prove that BN (, t) has only simple roots for all N such that
q (N ) = 3q 2.
References
[1] G. Anderson, D. Thakur, Tensor powers of the Carlitz module and zeta values, Annals of
Mathematics 132 (1990), 159-191.
[2] B. Angl`
es, M. Ould Douh, Arithmetic of units in Fq [T ], Publications math
ematiques de
Besancon Fascicule 2 (2012), 5-17.
[3] B. Angl`
es, F. Pellarin, Functional identities for L-series values in positive characteristic,
Journal of Number Theory 142 (2014), 223-251.
[4] B. Angl`
es, F. Pellarin, Universal Gauss-Thakur sums and L-series, Inventiones mathematicae
200 (2015), 653-669 .
[5] B. Angl`
es, F. Pellarin, F. Tavares Ribeiro, with an appendix by F. Demeslay, Arithmetic of
positive characteristic L-series values in Tate algebras , to appear in Compositio Mathematica,
http://dx.doi.org/10.1112/S0010437X15007563.
[6] B. Angl`
es, L. Taelman, with an appendix by V. Bosser, Arithmetic of characteristicc p special
L-values, Proceedings of the London Mathematical Society 110 (2015), 1000-1032.
[7] B. Angl`
es, F. Tavares-Ribeiro, Arithmetic of function fields units, arXiv:1506.06286 (2015).
F2 [x,y]
[8] G. B
ockle, The distribution of the zeros of the Goss zeta-function for A = (y 2 +y+x
3 +x+1) ,
Mathematische Zeitschrift 275 (2013), 835-861.
[9] L. Carlitz, On certain functions connected with polynomials in a Galois field, Duke Mathematical Journal 1 (1935), 137-168.
[10] K. Conrad. The digit principle. Journal of Number Theory 84, October 2000, 230-257.
[11] F. Demeslay, A class formula for L-series in positive characteristic, arXiv:1412.3704 (2014).
[12] F. Diaz-Vargas, Riemann Hypothesis for Fp [T ], Journal of Number Theory 59 (1996), 313318.
[13] D. Goss, Basic Structures of Function Field Arithmetic, Springer, 1996.
[14] M. Kanekos seminar talk in Caen (september 4th 2015) based on a forthcoming work of M.
Kaneko and D. Zagier on finite multiple zeta values.
[15] F. Pellarin, Values of certain L-series in positive characteristic, Annals of Mathematics 176
(2012), 2055-2093 .
[16] F. Pellarin, Private communication to the first author.
[17] F. Pellarin, R. Perkins, On finite zeta values in positive characteristic, work in progress.
[18] J. Sheats, The Riemann Hypothesis for the Goss Zeta Function for Fq [T ], Journal of Number
Theory 71 (1998), 121-157.
[19] L. Taelman, A Dirichlet unit theorem for Drinfeld modules, Mathematische Annalen 348
(2010), 899-907.
[20] L. Taelman, Special L-values of Drinfeld modules, Annals of Mathematics 175 (2012), 369391.
[21] L. Taelman, A Herbrand-Ribet theorem for function fields, Inventiones mathematicae 188
(2012), 253-275.

EXCEPTIONAL ZEROS OF L-SERIES AND BERNOULLI-CARLITZ NUMBERS

33

[22] D. Thakur, Gauss sums for Fq [t], Inventiones mathematicae 94 (1988), 105-112 .
[23] L. I. Wade. Certain quantities transcendental over GF(pn , x), Duke Math. J. 8 (1941), 701720.
Universit
e de Caen Normandie, Laboratoire de Math
ematiques Nicolas Oresme, CNRS
UMR 6139, Campus II, Boulevard Mar
echal Juin, B.P. 5186, 14032 Caen Cedex, France.
E-mail address: bruno.angles@unicaen.fr, tuan.ngodac@unicaen.fr, floric.tavares-ribeiro@unicaen.fr
Department of Mathematics, The Ohio State University, 231 W . 18th Ave., Columbus,
Ohio 43210
E-mail address: goss@math.ohio-state.edu
Institut Camille Jordan, UMR 5208 Site de Saint-Etienne, 23 rue du Dr. P. Michelon,
42023 Saint-Etienne, France
E-mail address: federico.pellarin@univ-st-etienne.fr