LXXVI.4 (1996)
Fermat quotient of cyclotomic units
by
Tsutomu Shimada (Yokohama)
Introduction. Let p be an odd prime number, K a finite extension of the field of rational numbers Q, O
Kthe ring of integers of K, and E
Kits group of units. Let N be the set of natural numbers, and Z and Z
pbe the ring of rational integers and the ring of p-adic integers, respectively.
We define
E
K(p
n) = {u ∈ E
K: u ≡ 1 mod p
n}, n ∈ N.
For each u ∈ E
K(p), we call u − 1
p mod p (∈ O
K/(p))
the Fermat quotient mod p of u. That is, for a unit u = 1 + px
u, x
u∈ O
K, x
umod p is the Fermat quotient mod p of u. From now on, we omit “mod p”
for simplicity.
We define a homomorphism ψ: E
K(p) → O
K/(p) by u = 1 + px
u7→
x
umod p. Let F(K) denote the set of all Fermat quotients of u ∈ E
K(p) or the image of ψ. Clearly, F(K) forms a subspace of F
p-vector space O
K/(p) where F
pdenotes the field with p elements, and kernel of ψ is E
K(p
2). So, we have
F(K) ∼ = E
K(p)/E
K(p
2) as F
p-vector spaces.
Now, the following two statements are well known: first, if ψ(u
1), . . . , ψ(u
s) are linearly independent over F
pthen u
1, . . . , u
s∈ F
K(p) are Z
p- independent, and secondly, the dimension of E
K(p
n)/E
K(p
n+1) over F
pequals the Z
p-rank of E
K(p) for sufficiently large n (see, Levesque [3] and Sands [4]). On the other hand, the Leopoldt conjecture states that Z
p-rank of E
K(p) equals Z-rank of E
K.
The aim of the present article is to study the dimension of F(K) over F
p, when K is a cyclotomic field.
[335]
1. Notations and results. Let ζ
n= exp(2πi/n) for n ∈ N, and let m ∈ N be odd, square free, 3 ≤ m and prime to p. We let K = Q(ζ
mp) and let E
K(p
n), F(K) and ψ be as in the introduction.
Since O
K= Z[ζ
mp] = Z[ζ
m][1 − ζ
p] and m is square free, the set {ζ
mr(1 − ζ
p)
ν: 1 ≤ r ≤ m, (r, m) = 1, ν = 0, 1, . . . , p − 2}
forms a Z-basis of O
K. As is well known,
dim
Fp(Z[ζ
mp]/(p)) = [Q(ζ
mp) : Q] = ϕ(mp), where ϕ denotes the Euler function. Therefore, the set
{ζ
mr(1 − ζ
p)
νmod p : 1 ≤ r ≤ m, (r, m) = 1, ν = 0, 1, . . . , p − 2}
forms an F
p-basis of O
K/(p).
Representing each x mod p ∈ O
K/(p) in this basis, we have x ≡ c
0+ c
1(1 − ζ
p) + c
2(1 − ζ
p)
2+ . . . + c
p−2(1 − ζ
p)
p−2mod p with c
i∈ Z[ζ
m] (i = 0, 1, . . . , p − 2), determined uniquely modulo p.
Let π denote 1 − ζ
pand let
E
K(pπ
i) = {u ∈ E
K: u ≡ 1 mod pπ
i} (i = 1, . . . , p − 2).
Since
E
K(p) ⊃ E
K(pπ) ⊃ . . . ⊃ E
K(pπ
p−2) ⊃ E
K(p
2), and u
p∈ E
K(p
2) for all u ∈ E
K(p), we have
F(K) = (E
K(p)/E
K(pπ)) ⊕ . . . ⊕ (E
K(pπ
k+1)/E
K(pπ
k+2)) ⊕ . . .
. . . ⊕ (E
K(pπ
p−2)/E
K(p
2)).
Thus
dim
FpF(K) = dim
Fp(E
K(p)/E
K(pπ)) + . . .
. . . + dim
Fp(E
K(pπ
k+1)/E
K(pπ
k+2)) + . . .
. . . + dim
Fp(E
K(pπ
p−2)/E
K(p
2)), −1 ≤ k ≤ p − 3.
We define subsets V
k(k = −1, 0, 1, . . . , p − 3) of F(K) by V
k= {x
umod p ∈ F(K) : u ∈ E
K(p),
c
0≡ c
1≡ . . . ≡ c
k≡ 0 mod p, c
k+16≡ 0 mod p}, where u = 1 + px
uand x
u≡ c
0+ c
1(1 − ζ
p) + . . . + c
p−2(1 − ζ
p)
p−2mod p, and e V
kbe the subspace generated by all elements in V
kover F
p. Of course,
F(K) = V
−1∪ V
0∪ V
1∪ . . . ∪ V
p−3∪ {0 mod p} (disjoint union).
For each k (−1 ≤ k ≤ p − 3), we define a mapping π
k: e V
k→ Z[ζ
m]/(p)
by x
umod p 7→ c
k+1mod p and let V
k= π
k( e V
k). Then, since V
k∼ =
E
K(pπ
k+1)/E
K(pπ
k+2) for −1 ≤ k ≤ p − 3, we have dim
FpF(K) =
p−3
X
k=−1
dim
FpV
k.
We now take polynomials S
t(X) ∈ Q[X], Z 3 t ≥ 0, such that S
t(n) = 1
t+ 2
t+ . . . + n
tfor all n ∈ N. For example,
S
0(X) = X, S
1(X) =
12X(X + 1), S
2(X) =
16X(X + 1)(2X + 1), etc.
As is well known, (k + 1)!S
k(X) ∈ Z[X], deg S
k(X) = k + 1, and S
k(−1) = 0 for k ≥ 1.
We define I
k(n) =
X
n 0 l=1S
k− l n
ζ
nl, 0 ≤ k ≤ p − 2, n ∈ N, where P
n0l=1
denotes the sum taken over all l = 1, . . . , n that are prime to n.
Let G be the Galois group of Q(ζ
m) over Q and b G its character group. As is well known, G is isomorphic to (Z/mZ)
×(the multiplicative group of all residue classes prime to m) by assigning σ
µ: ζ
m7→ ζ
mµto µ mod m.
Note that ψ is a Gal(K/Q)-homomorphism. We now state our results, Theorems 1–4.
Theorem 1. (1) There exist units α
k∈ E
K(p), 1 ≤ k ≤ p − 3, such that π
k(ψ(α
k)) = I
k(m) mod p.
(2) For k = 0, there exists α
0∈ E
K(p) such that π
0(ψ(α
0)) = (1 − σ
p)I
0(m) mod p.
(3) For k = −1, there exist β
ν∈ E
K(p), 2 ≤ ν ≤ m/2 and (ν, m) = 1, such that
π
−1(ψ(β
ν)) = (1 − σ
ν)I
p−2(m) mod p.
(4) If B
k+16≡ 0 mod p for some k = 1, 3, . . . , p − 4, then there exists u
k∈ E
K(p) such that
π
k(ψ(u
k)) = 1 mod p,
where B
ndenote the Bernoulli numbers (the definition will be given in Section 4).
For any a ∈ Z, let M (a) ∈ Z denote the non-negative minimal residue
of a mod m, that is, a ≡ M (a) mod m and 0 ≤ M (a) ≤ m − 1. When
b ∈ Z is prime to m, we let M (1/b) ∈ Z be the integer such that b ×
M (1/b) ≡ 1 mod m and 0 ≤ M (1/b) ≤ m − 1. Also, we take M (a/b) for
M (M (a) × M (1/b)).
For any σ
r∈ G and k (0 ≤ k ≤ p − 2), I
k(m)
σr=
X
m 0 l=1S
k− l m
ζ
mlr=
X
m 0 l=1S
k− M (l/r) m
ζ
ml.
We then define for each k, 0 ≤ k ≤ p − 2, the matrix A
k(m) =
S
k− M (l/r) m
1≤r,l≤m (r,m)=(l,m)=1
,
where we index the rows by r, and the columns by l. Since I
k(m)
σ−1= (−1)
k+1I
k(m) for 1 ≤ k ≤ p − 2 (see Lemma 3.2), we have rank A
k(m) ≤
1
2
ϕ(m) for such k.
In addition, we define B
k(m) =
S
k− M (l/r) m
1≤r,l≤m/2 (r,m)=(l,m)=1
, 0 ≤ k ≤ p − 2.
Theorem 2. For each k (1 ≤ k ≤ p − 3), we have det B
k(m) =
−1 2m
k ϕ(m)/2ζ
Q(ζm)+(−k) Y
χ∈G
b
id.6=χ:even
Y
q|m
(1 − χ
1(q)q
k)
× Y
q|m
(1 − q
k) − m
kϕ(m)
,
if k is odd, and det B
k(m) =
1 2m
k ϕ(m)/2ζ
Q(ζm)ζ
Q(ζm)+(−k) Y b
G3χ:odd
Y
q|m
(1 − χ
1(q)q
k),
if k is even. Moreover , if det B
k(m) 6≡ 0 mod p, then dim
FpV
k≥
12ϕ(m).
Here ζ
Q(ζm)and ζ
Q(ζm)+denote the Dedekind zeta functions of Q(ζ
m) and Q(ζ
m)
+(the maximal real subfield of Q(ζ
m)), respectively, and χ
1the primitive Dirichlet character associated with χ. Also, Q
q|m
denotes the product taken over all distinct primes q which divide m.
B
k(m) is, in some sense, a generalization of matrices defined by Carlitz [1] and Tateyama [5].
Let f be the order of the element p mod m in (Z/mZ)
×, and g = ϕ(m)/f .
Let g(m) and g
+(m) denote the number of distinct prime ideals which divide
m in Q(ζ
m) and Q(ζ
m)
+, respectively. We write h
−(Q(ζ
m)) for the relative
class number of Q(ζ
m). By the theorem of Tateyama [5], we can prove
Theorem 3. Assume g(m) = g
+(m) and p does not divide h
−(Q(ζ
m)).
Then
dim
FpV
0≥ (
12
ϕ(m) if p does not decompose in Q(ζ
m)/Q(ζ
m)
+,
1
2
(ϕ(m) − g) if p decomposes in Q(ζ
m)/Q(ζ
m)
+.
Note that when the right-hand side in the inequality above is not positive, or p ≡ 1 mod m, our theorem says nothing. Finally, by the same calculation as in the proof of Theorem 2 we obtain:
Theorem 4. We have det B
p−2(m) =
−1 2m
p−2 ϕ(m)/2ζ
Q(ζm)+(2 − p) Y
χ∈G
b
id.6=χ:even
Y
q|m
(1 − χ
1(q)q
p−2)
× Y
q|m
(1 − q
p−2) − m
p−2ϕ(m)
and if det B
p−2(m) 6≡ 0 mod p, then
dim
FpV
−1≥
12ϕ(m) − 1.
The rest of the article will be devoted to the proofs of the theorems stated above. In Section 2, we discuss some elementary properties of the Fermat quotient of cyclotomic units. The main result here is Lemma 2.5. In Section 3, introducing I
k(m), we prove Theorem 1(1). In Section 4, we prove Theorem 2 and the first part of Theorem 4. In Section 5, discussing I
0(m) and the rank of A
0(m), we prove Theorem 1(2) and Theorem 3. In Section 6, by the argument in Q(ζ
m), we prove Theorem 1(3) and the second part of Theorem 4. Finally, the proof of Theorem 1(4), which is obtained essentially in Washington [6], will be given in Section 7.
Before concluding this section, we classify typical generators of cyclo- tomic units (in the sense of Sinnot) into three types:
I. 1 − ζ
dζ
p(1 6= d | m),
II. 1 − ζ
d(1 6= d | m, d is composite), 1 − ζ
qν1 − ζ
q(q | m, q is a prime, 2 ≤ ν ≤ q − 1), III. 1 − ζ
pν1 − ζ
p(2 ≤ ν ≤ p − 1).
We shall use cyclotomic units of type I to prove Theorem 1(1), (2), type II (units in Q(ζ
m)) to prove Theorem 1(3), and type III (units in Q(ζ
p)) to prove Theorem 1(4).
Also, see Leopoldt [2] for units of type II.
2. Fermat quotient of units of type I. We first prove a preliminary lemma:
Lemma 2.1. For each ν ∈ N, we have (1 − X)
pν≡ 1 − X
pν− p
X
pν−1+ X
2pν−12 + . . . + X
(p−1)pν−1p − 1
mod p
2. P r o o f. We have
(1 − X)
pν= 1 − X
pν+
p
X
ν−1 i=1p
νi
(−X)
i. Since
(p
ν− 1) . . . (p
ν− i + 1)
(i − 1)! ≡ (−1)
i−1mod p
ν, we have
p
νi
≡ (−1)
i−1i p
νmod p
2ν−ordpifor all i = 1, . . . , p
ν− 1,
where ord
pi denotes the exact exponent of the power of p dividing i. Now, we have 2ν − ord
pi ≥ 2ν − (ν − 1) ≥ 2 and ord
p(p
ν/i) ≥ ν − (ν − 1) = 1.
So,
piν6≡ 0 mod p
2if and only if ord
pi = ν − 1, i.e. i = jp
ν−1for some j (1 ≤ j ≤ p − 1). If i = jp
ν−1(1 ≤ j ≤ p − 1), then
p
νi
(−X)
i≡ (−1)
i−1j p(−X)
i= − p
j X
jpν−1mod p
2. This completes the proof.
We define
f (X) = 1 1 − X
pX + X
22 + . . . + X
p−1p − 1
∈ Q(X).
Using Lemma 2.1, we have
(1 − ζ
mζ
p)
pf +1≡ 1 − ζ
mpf +1− p
ζ
mpf+ ζ
m2pf2 + . . . + ζ
m(p−1)pfp − 1
= 1 − ζ
mp− p
ζ
m+ ζ
m22 + . . . + ζ
mp−1p − 1
= (1 − ζ
mp)(1 − pf (ζ
m)) mod p
2, (1 − ζ
mζ
p)
p≡ 1 − ζ
mp− p
ζ
mζ
p+ ζ
m2ζ
p22 + . . . + ζ
mp−1ζ
pp−1p − 1
= (1 − ζ
mp)(1 − pf (ζ
mζ
p)) mod p
2,
and so
(1 − ζ
mζ
p)
pf +1−p≡ 1 − p(f (ζ
m) − f (ζ
mζ
p)) mod p
2.
Consequently, f (ζ
m)−f (ζ
mζ
p) mod p belongs to F(K). Note that (O
K/(p))
×(the multiplicative group of O
K/(p)) has the exponent p
f +1− p. We first prove the existence of a “canonical” element in F(K), a linear combination over F
pof conjugates of f (ζ
m) − f (ζ
mζ
p) mod p (see Lemma 2.3), and next determine the coefficients modp of its image by π
k(see Lemma 2.5). From ζ
pj= ((ζ
p− 1) + 1)
j, it can be easily seen that
1 − ζ
pj= X
j i=1j i
(−1)
i+1(1 − ζ
p)
i. Hence
f (ζ
m) − f (ζ
mζ
p) = 1 1 − ζ
mpp−1
X
j=1
ζ
mjj (1 − ζ
pj) (1)
= 1
1 − ζ
mp p−1X
j=1
X
j i=1ζ
mjj
j i
(−1)
i+1(1 − ζ
p)
i= 1
1 − ζ
mp p−1X
i=1 p−1
X
j=i
ζ
mjj
j i
(−1)
i+1(1 − ζ
p)
i, that is,
f (ζ
m) − f (ζ
mζ
p) = 1 1 − ζ
mpp−1
X
i=1
p−1X
j=1 j i
j ζ
mj(−1)
i+1(1 − ζ
p)
i,
where
ji= 0 if j < i.
Lemma 2.2. For each ν = 1, . . . , p − 1 we have f (ζ
m) − f (ζ
mζ
pν) ≡ 1
1 − ζ
mp p−2X
i=1
p−1X
j=1 νj
i
j ζ
mj(−1)
i+1(1 − ζ
p)
imod p.
P r o o f. Taking ζ
pνfor ζ
pin both sides of (1), we get
f (ζ
m) − f (ζ
mζ
pν) = 1 1 − ζ
mpp−1
X
j=1
ζ
mjj (1 − ζ
pνj).
We denote by P (a) (a ∈ Z) the non-negative minimal residue of a mod p
and use it similarly to M (a) introduced in the previous section. Then
p−1
X
j=1
ζ
mjj (1 − ζ
pνj)
≡
p−1
X
j=1
ζ
mP (j/ν)j/ν (1 − ζ
pj)
= ν
p−1
X
j=1
ζ
mP (j/ν)j (1 − ζ
pj) = ν
p−1
X
j=1
ζ
mP (j/ν)j
X
j i=1j i
(−1)
i+1(1 − ζ
p)
i= ν
p−1
X
i=1
p−1X
j=1 j i
j ζ
mP (j/ν)(−1)
i+1(1 − ζ
p)
imod p.
Now
ji≡
νP (j/ν)imod p, as νP (j/ν) ≡ j mod p. It follows that
p−1
X
j=1 j i
j ζ
mP (j/ν)≡
p−1
X
j=1 1
ν νP (j/ν) i
j/ν ζ
mP (j/ν)≡
p−1
X
j=1 1
ν νP (j/ν) i
P (j/ν) ζ
mP (j/ν)=
p−1
X
j=1 1 ν νj
i
j ζ
mjmod p.
We then have
p−1
X
j=1
ζ
mjj (1 − ζ
pνj) ≡
p−1
X
i=1
p−1X
j=1 νj
i
j ζ
mj(−1)
i+1(1 − ζ
p)
imod p.
This completes the proof.
Let g
i(X) (1 ≤ i ≤ p − 1) be a polynomial with coefficients in F
psuch that g
i(0) = 0 and of degree ≤ i. We define a map e τ
ν(1 ≤ ν ≤ p − 1) from F
p[X] into itself by e τ
ν(g(X)) = g(νX) for all g(X) ∈ F
p[X]. Let N
i(1 ≤ i ≤ p − 1) be the i × i matrix
N
i=
1 1 . . . 1 2
i2
i−1. . . 2 .. . .. . . .. ...
i
ii
i−1. . . i
,
where each component represents a corresponding residue class mod p. Then it is clear that
N
i=
1 0
2 . ..
0 i
1 1 . . . 1 2
i−12
i−2. . . 1 .. . .. . . .. ...
i
i−1i
i−2. . . 1
,
therefore det N
i6= 0, and
N
i
a
iX
ia
i−1X
i−1.. . a
1X
=
e
τ
1(g
i(X)) e
τ
2(g
i(X)) .. . e
τ
i(g
i(X))
,
where g
i(X) = a
iX
i+ a
i−1X
i−1+ . . . + a
1X.
Let (c
i,1c
i,2. . . c
i,i) be the first row of N
−1i. Then
a
iX
i= c
i,1τ e
1(g
i(X)) + c
i,2τ e
2(g
i(X)) + . . . + c
i,iτ e
i(g
i(X)).
Letting e C
i= c
i,1τ e
1+ . . . + c
i,ie τ
i, we write a
iX
i= e C
i(g
i(X)), that is C e
i(g
i(X)) =
0 if deg g
i(X) < i, a
iX
iif deg g
i(X) = i.
Now we take
ji, 1 ≤ i ≤ p − 2, for a polynomial in an indeterminate j with coefficients in F
p.
Since
ji= (1/i!)j
i+ (polynomial with degree ≤ i − 1), for k (0 ≤ k ≤ p − 3) we have
C e
k+1j i
=
0 if i < k + 1, 1
(k + 1)! j
k+1if i = k + 1,
∗ if i > k + 1,
where ∗ means some polynomial in F
p[j], irrelevant for our purpose.
Let τ
ν(1 ≤ ν ≤ p − 1) be an automorphism of Q(ζ
mp) over Q such that τ
ν: ζ
p7→ ζ
pνand τ
ν: ζ
m7→ ζ
m, and C
i(1 ≤ i ≤ p − 1) be the element c
i,1τ
1+ c
i,2τ
2+ . . . + c
i,iτ
iof the group ring of Gal(Q(ζ
mp)/Q) over F
p(we write its action on F(K) additively). Then we obtain the following:
Lemma 2.3. For each k, 0 ≤ k ≤ p − 3, we have C
k+1(f (ζ
m) − f (ζ
mζ
p))
≡ 1
(k + 1)!
1
1 − ζ
mp p−1X
j=1
j
kζ
mj(−1)
k(1 − ζ
p)
k+1+ 1
1 − ζ
mp p−2X
i=k+2
p−1X
j=1
C e
k+1 j ij ζ
mj(−1)
i+1(1 − ζ
p)
imod p.
P r o o f. By Lemma 2.2,
τ
ν(f (ζ
m) − f (ζ
mζ
p)) = f (ζ
m) − f (ζ
mζ
pν)
≡ 1
1 − ζ
mp p−2X
i=1
p−1X
j=1
e τ
ν ji
j ζ
mj(−1)
i+1(1 − ζ
p)
imod p.
Consequently,
C
k+1(f (ζ
m) − f (ζ
mζ
p))
≡ 1
1 − ζ
mp p−2X
i=1
p−1X
j=1
C e
k+1 jij ζ
mj(−1)
i+1(1 − ζ
p)
i≡ 1
1 − ζ
mp p−1X
j=1
C e
k+1 j k+1j ζ
mj(−1)
k(1 − ζ
p)
k+1+ 1
1 − ζ
mp p−2X
i=k+2
p−1X
j=1
C e
k+1 j ij ζ
mj(−1)
i+1(1 − ζ
p)
i≡ 1
1 − ζ
mp p−1X
j=1
1
(k + 1)! j
kζ
mj(−1)
k(1 − ζ
p)
k+1+ 1
1 − ζ
mp p−2X
i=k+2
p−1X
j=1
C e
k+1 jij ζ
mj(−1)
i+1(1 − ζ
p)
imod p, as desired.
We define, for k = 0, 1, . . . , p − 2 and n (∈ N) prime to p, J
k(n) = 1
1 − ζ
np p−1X
j=1
j
kζ
njwhen n ≥ 2 and J
k(1) = 0.
We can deduce that
J
0(n) = 1 1 − ζ
n− 1
1 − ζ
np= (1 − σ
p) 1 1 − ζ
nand
J
k(n)
σ−1≡ (−1)
k+1J
k(n) mod p
for n ≥ 2 by easy calculation, where σ
−1denotes the automorphism which sends ζ
nto ζ
n−1.
Lemma 2.4. For each k (1 ≤ k ≤ p − 2) we have J
k(m) ≡
m−1
X
l=1
p−1X
j=1, M (l/p)≥M (j/p)6=0
j
kζ
mlmod p.
P r o o f. From the equality −m/(1 − ζ
m) = P
m−1l=1
lζ
ml, we obtain
−m 1 − ζ
mp=
m−1
X
l=1
lζ
mlpand
−mζ
mj1 − ζ
mp=
m−1
X
l=1
lζ
mlp+j=
m−1
X
l=1
lζ
mM (lp+j), 1 ≤ j ≤ p − 1,
where M (∗) is the same as in Section 1. Now, since M ((M (lp + j) − j)/p)
= l and
{M (lp + j) : l = 1, . . . , m − 1} = {0, 1, . . . , m − 1}\{M (j)}, we obtain
−mζ
mj1 − ζ
mp=
m−1
X
l=1
M
M (lp + j) − j p
ζ
mM (lp+j)=
m−1
X
l=0
M
l − j p
ζ
ml=
m−1
X
l=1
M
l − j p
ζ
ml+ M
−j p
(−ζ
m− ζ
m2− . . . − ζ
mm−1)
=
m−1
X
l=1
M
l − j p
− M
−j p
ζ
ml.
Therefore
J
k(m) = − 1 m
p−1
X
j=1
−mζ
mj1 − ζ
mpj
k= − 1 m
p−1
X
j=1
j
k m−1X
l=1
M
l − j p
− M
−j p
ζ
ml,
that is,
(2) J
k(m) = − 1 m
m−1
X
l=1
p−1X
j=1
j
kM
l − j p
− M
−j p
ζ
ml.
It is clear that M
l − j p
=
M (l/p) − M (j/p) if M (l/p) ≥ M (j/p), M (l/p) − M (j/p) + m if M (l/p) < M (j/p), and
M
l − j p
−M
−j p
=
M (l/p) if M (j) = 0,
M ((l − j)/p) − {m − M (j/p)}
= M ((l − j)/p) + M (j/p) − m if M (j) 6= 0.
For these reasons, M
l − j p
−M
−j p
=
M (l/p) − m if M (l/p) ≥ M (j/p) 6= 0,
M (l/p) if M (l/p) < M (j/p) or M (j) = 0.
Substituting this in (2), we have J
k(m) = − 1
m
m−1
X
l=1
n
p−1X
j=1, M (l/p)≥M (j/p)6=0
j
k(M (l/p) − m)
+
p−1
X
j=1, M (l/p)<M (j/p) or M (j)=0
j
kM (l/p) o
ζ
ml= − 1 m
m−1
X
l=1
n
M (l/p)
p−1
X
j=1
j
k− m
p−1
X
j=1, M (l/p)≥M (j/p)6=0
j
ko
ζ
ml. The result follows from a well known fact that P
p−1j=1
j
k≡ 0 mod p.
Lemma 2.5. We have J
k(m) ≡ (−m)
km−1
X
l=1
S
k− l m
ζ
mlmod p, 1 ≤ k ≤ p − 2.
P r o o f. It is sufficient to prove (3)
p−1
X
j=1, M (l/p)≥M (j/p)6=0
j
k≡ (−m)
kS
k− l m
mod p.
We deal first with the case where m < p. Let x
i= P (−i/m) and y
i= (i + x
im)/p ∈ Z for each i (1 ≤ i ≤ p − 1), where P (∗) is the same as in the proof of Lemma 2.2. Then we have 1 ≤ y
i≤ m.
Suppose that x
i< x
jfor i 6= j, 1 ≤ i, j ≤ p − 1. Then, as j − i = (y
j− y
i)p − (x
j− x
i)m and −(p − 2) ≤ j − i, we have
y
j− y
i≥ −(p − 2) + m
p = −1 + m + 2 p > −1.
Therefore, y
j≥ y
iif x
j> x
i. For each l (1 ≤ l ≤ m − 1), let
A
l= {j : 1 ≤ j ≤ p − 1, M (l/p) ≥ M (j/p) 6= 0}.
Since y
l6= m, or 1 ≤ y
l≤ m − 1, and M (j/p) = M (y
j), we have A
l= {j : 1 ≤ j ≤ p − 1, M (y
l) ≥ M (y
j) 6= 0}
= {j : 1 ≤ j ≤ p − 1, M (y
l) ≥ M (y
j), y
j6= m}
= {j : 1 ≤ j ≤ p − 1, y
l≥ y
j}.
If m < i ≤ p − 1, we have i − m = y
ip − (x
i+ 1)m; therefore x
i−m= x
i+ 1
and y
i−m= y
i. For that reason, any index i, with common value y
i, can be
represented in the form i = j + mν (1 ≤ j ≤ m, Z 3 ν ≥ 0) and, for such
i’s, the maximal value of x
iequals x
j.
Consequently,
A
l= {j : 1 ≤ j ≤ p − 1, x
j≤ x
l} and
X
p−1j=1, M (l/p)≥M (j/p)6=0
j
k= (−m)
kX
j∈Al
− j m
k≡ (−m)
kX
j∈Al
x
kj= (−m)
k(1
k+ 2
k+ . . . + x
kl) = (−m)
kS
k(x
l)
≡ (−m)
kS
k− l m
mod p, as desired.
Next we consider the case where p < m. Let again x
i= P (−i/m) and y
i= (i + x
im)/p ∈ Z for i (0 ≤ i ≤ m − 1). Since 0 ≤ y
ip = i + x
im ≤ m − 1 + (p − 1)m = pm − 1, we have 0 ≤ y
i≤ m − 1/p, that is, 0 ≤ y
i≤ m − 1. It is clear that {x
i: 0 ≤ i ≤ p − 1} = {0, 1, . . . , p − 1}. Since i − j = (y
i− y
j)p − (x
i− x
j)m for 0 ≤ i, j ≤ m − 1, we see that x
i= x
jif and only if i ≡ j mod p, and that y
i< y
jif x
i< x
j. There exist i
0and ν
i(0 ≤ i
0≤ p − 1, 1 ≤ ν
i) such that i = i
0+ ν
ip for each i (p ≤ i ≤ m − 1).
Since i = y
i0p − x
i0m + ν
ip = (y
i0+ ν
i)p − x
i0m, we have y
i= y
i0+ ν
iand x
i= x
i0.
Consequently, for l (1 ≤ l ≤ m − 1), we have
A
l= {j : 1 ≤ j ≤ p − 1, M (l/p) ≥ M (j/p)} = {j : 1 ≤ j ≤ p − 1, y
l≥ y
j}
= {j : 1 ≤ j ≤ p − 1, y
l0≥ y
j} = {j : 1 ≤ j ≤ p − 1, x
l0≥ x
j}.
Now, clearly, −l/m ≡ −l
0/m ≡ x
l0mod p. Hence X
j∈Al
j
k=
p−1
X
j=1,xj≤xl0
j
k= (−m)
kp−1
X
j=1,xj≤xl0
− j m
k≡ (−m)
kp−1
X
j=1,xj≤xl0
x
kj= (−m)
k(1
k+ 2
k+ . . . + x
kl0) = (−m)
kS
k(x
l0)
≡ (−m)
kS
k− l m
mod p.
We have just completed the proof of Lemma 2.5.
3. I
k(m) as a partial sum of J
k(m). For any divisor d of m we define M
d=
l : l = m
d ν, 1 ≤ ν ≤ d, (ν, d) = 1
.
It is clear that ϕ(d) = ]M
d(the cardinality of M
d), M
1= {m}, {1, 2, . . . , m}
= S
d|m
M
d(disjoint union), and M
d= {l : 1 ≤ l ≤ m, (l, m) = m/d}. Now we have, from Lemma 2.5, for k = 1, . . . , p − 2,
J
k(m) ≡ (−m)
kX
ml=1
S
k− l m
ζ
ml= (−m)
kX
d|m
X
d 0 ν=1S
k−
m d
ν m
ζ
m(m/d)ν= (−m)
kX
d|m
X
d 0 ν=1S
k− ν d
ζ
dνmod p,
where P
d0ν=1
is the same as in Section 1. So, we have (−m)
−kJ
k(m) ≡ X
d|m
I
k(d) mod p.
By the M¨obius inversion formula, it follows that I
k(m) ≡ X
d|m
{µ(m/d)(−d)
−kJ
k(d)} mod p
for k (1 ≤ k ≤ p − 2), where µ is the M¨obius function. Each J
k(d) mod p (d | m) belongs to V
kby the definition, so is I
k(m) mod p.
We have thus proved Theorem 1(1).
Lemma 3.1. For all k ∈ N, we have
S
k(−X) = (−1)
k+1S
k(X − 1).
P r o o f. From the definition of S
k(X), we have S
k(n + 1) = (n + 1)
k+ S
k(n) for all n ∈ N. Hence
S
k(X + 1) = (X + 1)
k+ S
k(X).
Using this formula, we see that
S
k(−n) = −(−n + 1)
k+ S
k(−n + 1)
= −(−n + 1)
k− (−n + 2)
k+ S
k(−n + 2)
= . . . = −(−n + 1)
k− (−n + 2)
k− . . . − (−1)
k+ S
k(−1)
= (−1)
k+1{1
k+ 2
k+ . . . + (n − 1)
k} = (−1)
k+1S
k(n − 1), for all n ∈ N. This proves our lemma.
The next two lemmas, on properties of I
k(m), will be used in the follow- ing section.
Lemma 3.2. For all k (1 ≤ k ≤ p − 2), we have
I
k(m)
σ−1= (−1)
k+1I
k(m).
P r o o f.
I
k(m)
σ−1= X
m 0 l=1S
k− l m
ζ
mm−l= X
m0 l=1S
k− m − l m
ζ
ml= X
m 0 l=1S
kl m − 1
ζ
ml= (−1)
k+1X
m 0 l=1S
k− l m
ζ
ml= (−1)
k+1I
k(m).
Lemma 3.3. For k = 0, we have
I
0(m)
σ−1+ I
0(m) = −µ(m).
P r o o f.
I
0(m)
σ−1= X
m 0 l=1− l m
ζ
m−l=
X
m0 l=1− m − l m
ζ
ml= − X
m 0l=1
ζ
ml− X
m 0 l=1− l m
ζ
ml= −µ(m) − I
0(m).
This completes the proof of the lemma.
4. Determinant of B
k(m). Let A
k(m) and B
k(m) (0 ≤ k ≤ p − 2) be as in Section 1. Recall that
G = Gal(Q(ζ
m)/Q) = (Z/mZ)
×= {σ
r: ζ
m7→ ζ
mr: 1 ≤ r < m, (r, m) = 1};
we shall identify G with {l : 1 ≤ l < m, (l, m) = 1}.
Let H = Gal(Q(ζ
m)
+/Q) = (Z/mZ)
×/{±1}; we shall identify H with {l : 1 ≤ l < m/2, (l, m) = 1}. By Lemma 3.1,
(4) S
k− m − l m
= (−1)
k+1S
k− l m
, k ∈ N.
Now, let us prove Theorem 2 and the first part of Theorem 4. First we assume k (1 ≤ k ≤ p − 2) is odd. Then, by means of (4), we can take S
k(−l/m) for a function on H. Then it is easily verified, by a result on the group determinant, that
det B
k(m) = Y
χ∈H
b X
l∈H
S
k− l m
χ(l).
Moreover, by (4), X
l∈H
S
k− l m
χ(l) = 1 2
X
l∈G
S
k− l m
χ(l)
for all χ ∈ b H (the set of all even characters in b G). Therefore, (5) det B
k(m) =
1 2
ϕ(m)/2Y b
G3χ:even
X
l∈G
S
k− l m
χ(l).
Next we assume k is even. We fix an arbitrary odd character ξ in b G.
Then, from (4), we have S
k− m − l m
= −S
k− l m
,
so that S
k(−l/m)ξ(l) can be regarded as a function on H. Hence, similarly to the case of k odd, we have
det
S
k− M (l/r) m
ξ
M
l r
l,r∈H
= Y
χ∈H
b X
l∈H
S
k− l m
ξ(l)χ(l)
= Y
χ∈H
b 1 2
X
l∈G
S
k− l m
ξ(l)χ(l)
=
1 2
ϕ(m)/2Y b
G3χ:even
X
l∈G
S
k− l m
ξ(l)χ(l)
=
1 2
ϕ(m)/2Y b
G3χ:odd
X
l∈G
S
k− l m
χ(l).
On the other hand, as ξ(M (l/r)) = ξ(l)ξ(r
−1),
.. . . . . S
k− M (l/r) m
ξ
M
l r
. . . .. .
r,l∈H
=
. .. 0
ξ(r
−1)
0 . ..
r∈H
.. . . . . S
k− M (l/r) m
. . . .. .
r,l∈H
. .. 0 ξ(l) 0 . ..
l∈H
.
Consequently,
(6) det B
k(m) =
1 2
ϕ(m)/2Y b
G3χ:odd
X
l∈G
S
k− l m
χ(l).
We recall the Bernoulli numbers and polynomials defined by t
e
t− 1 = X
∞ n=0B
nt
n/n! and B
n(X) = X
n i=0n i
B
iX
n−i,
respectively. It is well known that S
k(X − 1) = 1
k + 1 (B
k+1(X) − B
k+1) for any k ∈ N.
Since S
k− l m
= (−1)
k+1S
kl m − 1
= (−1)
k+1k + 1
B
k+1l m
− B
k+1, we have
X
l∈G
S
k− l m
χ(l)
= (−1)
k+1k + 1
X
m l=1B
k+1l m
χ(l) − (−1)
k+1k + 1 B
k+1X
m l=1χ(l), for any χ ∈ b G, where χ is understood to be a Dirichlet character defined modulo m. Clearly, B
k+1P
ml=1
χ(l) 6= 0 if and only if k is odd and χ is the principal character 1
m∈ b G, where 1
m(l) = 1 if (l, m) = 1, and 1
m(l) = 0 if (l, m) > 1. Moreover, in that case, B
k+1P
ml=1
χ(l) = B
k+1ϕ(m).
We also recall the generalized Bernoulli numbers defined by B
n,χ= m
n−1X
m a=1χ(a)B
na m
, n ∈ N, χ ∈ b G,
satisfying
L(1 − n, χ) = − B
n,χn ,
where L(s, χ) denotes the L-function attached to χ. By elementary results on L-functions, we have, for k (1 ≤ k ≤ p − 2),
L(−k, χ) = L(−k, χ
1) Y
q|m
(1 − χ
1(q)q
k), where Q
q|m