LXXIII.2 (1995)
Norm residue symbol and cyclotomic units
by
Charles Helou (Media, Penn.)
Introduction. Let l be a prime number ≥ 5, ζ a primitive lth root of unity in C, K = Q(ζ), O
K= Z[ζ] the ring of integers of K, λ = 1 − ζ, prime element of O
Kdividing l, and b K the λ-adic completion of K. For α, β in K
∗= K − {0}, denote by (α, β)
λthe norm residue symbol (i.e. Hilbert symbol) as defined in [5] where it is written
α,β(λ), or in [2] where the symbol is the inverse of the one in [5] used here. Let [α, β] denote the element of the finite field F
l= Z/lZ defined by
(α, β)
λ= ζ
[α,β].
The symbol [α, β] is a bilinear skew-symmetric function of α, β, with respect to the multiplicative and additive structures of K
∗and F
lrespectively; and if [α, β] = 0, then α and β are said to be orthogonal.
The cyclotomic units, in O
K, are u
n= 1 − ζ
n1 − ζ ,
for n ∈ Z not divisible by l (or 1 ≤ n ≤ l − 1). They generate a subgroup C of the multiplicative group of all units of O
K.
Let a ∈ Z and α
1= a − ζ. Terjanian conjectured ([9]) that every prime number l ≥ 5 has the property (called LC) that the following two conditions are equivalent:
(i) α
1is orthogonal to C, (ii) a ≡ 0, 1, or −1 (mod l
2).
He showed that (ii) implies (i), and if (i) holds then a is divisible by l
2or a
l−1≡ 1 (mod l
2). But the latter condition, in conjunction with the congruence a ≡ 0, 1, or −1 (mod l), implies a ≡ 0, 1, or −1 (mod l
2). Thus, Terjanian’s conjecture amounts to the assertion that if α
1is orthogonal to C then a ≡ 0, 1, or −1 (mod l).
In this paper, we use the Artin–Hasse explicit reciprocity law to give an
[147]
expression for [α
1, u
n], when a 6≡ 1 (mod l), involving the Fermat quotient q(x) = x
l−1− 1
l (for x ∈ Z, l - x) and the polynomial sums in F
lS
nr(a) = X
1≤k≤l−1 k≡r (mod n)
a
kk (for n, r ∈ Z, l - n).
Some properties of these sums are established, in order to simplify the ex- pression for [α
1, u
n]. This yields necessary and sufficient conditions for α
1to be orthogonal to C, in the form of equations over F
l.
Furthermore, we assume l ≡ 1 (mod 4) and consider the quadratic sub- field E = Q( √
l) of K. When n is not a quadratic residue modulo l, the norm of u
nin K|E is an element of C which, by Dirichlet’s class number formula, is equal to ε
2h, where ε is the fundamental unit and h the class number of E. Also, by a result on real quadratic fields, h < √
l, so that l - h.
Thus, if α
1is orthogonal to C then it is orthogonal to ε. Upon using again the Artin–Hasse law, an expression is obtained for [α
1, ε] in terms of ε and the polynomial
F (X) = X
l−1 k=1k l
2
X
k, where
kl2
is the Legendre symbol. It follows that, assuming the validity of a conjecture by Ankeny, Artin and Chowla ([1]) concerning ε, if α
1is orthogonal to C, then a is a root of F in F
l. This polynomial F has in F
lthe trivial roots 0, 1, −1 (with 1 of multiplicity (l − 1)/2) and takes in K as values for X = ζ
m(1 ≤ m ≤ l − 1) the quadratic Gauss sums; it can also be written in terms of the polynomial
A(X) = Y
1≤k≤(l−1)/2
(X − ζ
k2),
whose coefficients lie in O
E. The polynomials F and A were studied by Jacobi ([6]) and Dirichlet ([4]) respectively.
1. Application of the Artin–Hasse law. Let a ∈ Z, a 6≡ 1 (mod l), α
1= a−ζ and c ∈ Z such that (a−1)c ≡ 1 (mod l). For any n ∈ Z with l - n, let σ
nbe the element of the Galois group of K|Q defined by σ
n(ζ) = ζ
n. Denote by log α the λ-adic logarithm of any unit α in b K; in particular, for α = 1 + x with x ≡ 0 (mod λ),
log(1 + x) = X
∞ k=1(−1)
k−1k x
k.
Denote by Tr the trace from b K to the field Q
lof l-adic numbers (whose restriction to K coincides with the trace from K to Q). Further notations will be introduced as needed.
Proposition 1. For a 6≡ 1 (mod l) and for any 1 ≤ n ≤ l − 1, [α
1, u
n] = n(a
l− a)
l(a − 1) − n + 1
2 q(a − 1) + cq(c) − n l Tr
log(1 + cλ) σ
n(λ)
. P r o o f. We have α
1= a − 1 + λ ≡ (a − 1)(1 + cλ) (mod λ
l), so that ([9], (R10) or [5], p. 54)
[α
1, u
n] = [a − 1, u
n] + [1 + cλ, u
n].
For the first bracket, since u
n=
n−1
X
k=0
ζ
k≡
n−1
X
k=0
(1 − kλ) ≡ n − n(n − 1)
2 λ (mod λ
2), we have ([9], (R22) or [5], p. 110)
[a − 1, u
n] = n − 1
2 q(a − 1).
For the second bracket, since u
n= σ
n(λ)/λ, we have
[1 + cλ, u
n] = [1 + cλ, σ
n(λ)] − [1 + cλ, cλ] + [1 + cλ, c].
Moreover ([9], (R19) or [5], p. 54),
[1 + cλ, σ
n(λ)] = n[1 + cσ
−1n(λ), λ], and ([9], (R18) or [5], p. 55)
[1 + cλ, cλ] = [1 + cλ, 1] + [1, cλ] = 0, and ([9], (R23) or [5], p. 110)
[1 + cλ, c] = cq(c).
Therefore
(1) [α
1, u
n] = n − 1
2 q(a − 1) + n[1 + cσ
n−1(λ), λ] + cq(c).
By the Artin–Hasse reciprocity law ([2], Ch. 12, Th. 10, whose third part is missing a factor 1/λ of ζ log α; cf. [5], p. 94),
(2) [1 + cσ
n−1(λ), λ] = − 1 l Tr
ζ
λ log(1 + cσ
n−1(λ))
.
Let b σ
nbe the Q
l-automorphism of b K extending σ
n. For any α ∈ b K,
Tr(b σ
n(α)) = Tr(α), and if α ≡ 1 (mod λ), then, by continuity of b σ
nin
the λ-adic topology, b σ
n(log α) = log(b σ
n(α)). Therefore, in the right-hand
side of (2), the argument of Tr may be replaced by
σζnn(λ)
log(1 + cλ), which gives, since ζ
n= 1 − σ
n(λ),
(3) Tr
ζ
λ log(1 + cσ
n−1(λ))
= Tr
log(1 + cλ) σ
n(λ)
− Tr(log(1 + cλ)).
Similarly, letting N be the norm in b K|Q
l, and using the λ-adic continuity of the b σ
n’s and the multiplicative-additive homomorphism property of log, we have, for α ≡ 1 (mod λ),
(4) Tr(log(α)) = log(N (α)) ≡ N (α) − 1 (mod l
2).
In particular, for α = 1 + cλ = 1 + c − cζ, the norm is (1 + c)
l− c
l, so that
(5) 1
l Tr(log(1 + cλ)) ≡ (c + 1)
l− c
l− 1
l (mod l).
Moreover, in F
l, we have c = 1/(a − 1) and (a − 1)
l= a − 1, so that (6) (c + 1)
l− c
l− 1
l = 1
a − 1
a
l− a
l − (a − 1)q(a − 1)
, in F
l. The result now follows from (1), (2), (3), (5) and (6).
Corollary. We have [α
1, u
n] = n(a
l− a)
l(a − 1) − n + 1
2 q(a − 1) + cq(c) + n l
X
l k=1(−c)
kk Tr
λ
kσ
n(λ)
. P r o o f. Let D be the different of the extension b K|Q
l([8], Ch. 3). For x in K, if x ∈ (l b
2)D
−1then Tr(x) lies in the ideal (l
2). Since the ring of integers of b K is Z
l[ζ] and the irreducible polynomial of ζ over Q
lis Φ(X) =
XX−1l−1, we have
D = (Φ
0(ζ)) = (lλ
−1) = (λ
l−2).
Thus x ≡ 0 (mod λ
l) implies Tr(x) ≡ 0 (mod l
2). Since log(1 + cλ)
σ
n(λ) ≡ X
l k=1(−1)
k−1k c
kλ
kσ
n(λ) (mod λ
l), it follows that
Tr
log(1 + cλ) σ
n(λ)
≡ X
l k=1(−1)
k−1k c
kTr
λ
kσ
n(λ)
(mod l
2).
Hence the result follows by substitution into Proposition 1.
Notations. For k ∈ Z such that l - k, the element 1/k is well defined in
F
l; by abuse of notation, it will also be used in congruences modulo l. By
contrast, the appearance of l in a denominator means that the factor of 1/l
is divisible by l and that we are considering the quotient. For an integer
1 ≤ n ≤ l − 1, we denote by n
0the specific representative of (1/n) (mod l) contained in the same interval, i.e. 1 ≤ n
0≤ l − 1 and nn
0≡ 1 (mod l).
For a real number x, the greatest integer ≤ x is written [x]. For a positive integer n and any m ∈ Z, we denote by res
n(m) the least residue ≥ 0 of m modulo n, i.e.
res
n(m) = m − n[m/n].
Lemma 1. For any integers 1 ≤ n ≤ l − 1 and 2 ≤ k ≤ l, Tr
λ
kσ
n(λ)
= l X
0≤s<kn0/l
k − 1 [sl/n
0]
(−1)
[sl/n0], while for k = 1,
Tr
λ
σ
n(λ)
= l − n
0. P r o o f. For 1 ≤ n ≤ l − 1 and 1 ≤ k ≤ l,
Tr
λ
kσ
n(λ)
= X
l−1 i=1(1 − ζ
i)
k1 − ζ
in.
The general term in the latter sum can be written in the form 1 − ζ
inn01 − ζ
in(1 − ζ
i)
k−1= X
l−1i=1
(1 − ζ
i)
k−1n
X
0−1 j=0ζ
inj. Moreover, for any m ∈ Z,
(7)
X
l−1 i=1ζ
mi=
l − 1 if l | m,
−1 if l - m.
The case k = 1 now follows easily, and so we assume 2 ≤ k ≤ l. In view of what precedes, and after replacing (1 − ζ
i)
k−1by its binomial expansion, we get
(8) Tr
λ
kσ
n(λ)
=
n
X
0−1 j=0k−1
X
t=0
k − 1 t
(−1)
tX
l−1 i=1ζ
(nj+t)i.
The inner sum in (8) is equal to l − 1 or to −1 according as t ≡ −jn or t 6≡ −jn (mod l). Therefore the right-hand side of (8) splits into two sums obtained by replacing the inner sum respectively by −1 for all terms, and by l for those terms such that t ≡ −jn (mod l). The first sum is
(−1)
n
X
0−1 j=0k−1
X
t=0
k − 1 t
(−1)
t= −
n
X
0−1 j=0(1 − 1)
k−1= 0,
and (8) is thus reduced to the second sum, i.e.
(9) Tr
λ
kσ
n(λ)
= l X
t∈T
k − 1 t
(−1)
t,
where T is the set of integers t satisfying the two conditions: 0 ≤ t ≤ k − 1, and t ≡ −jn (mod l) for some 0 ≤ j ≤ n
0− 1. The latter condition is equivalent to : tn
0+ j = sl for some s ∈ N and 0 ≤ j ≤ n
0− 1, i.e.
j = res
n0(sl) and t = [sl/n
0] for some s ∈ N. The former condition on t is then equivalent to 0 ≤ sl/n
0< k. The result now follows by substitution for t in terms of s in (9).
Lemma 2. (a) For any integer 1 ≤ k ≤ l − 1, the binomial coefficient
klis divisible by l and we have
1 k ≡ 1
l
l k
(−1)
k−1(mod l).
(b) For any algebraic integer α, we have the congruence (in any ring of algebraic integers containing α)
X
l−1 k=1α
kk ≡ (α − 1)
l− α
l+ 1
l (mod l).
(c) For any x, y ∈ Z not divisible by l, we have q(xy) ≡ q(x) + q(y) (mod l).
If x ≡ y (mod l), then
q(x) ≡ q(y) − 1
x · x − y
l (mod l).
If xy ≡ 1 (mod l), then
q(y) ≡ −q(x) − xy − 1
l (mod l).
(d) For any x ∈ Z not divisible by l, we have X
l−1k=1
kx l
1
k ≡ xq(x) (mod l).
(e) For any integer 0 ≤ r ≤ l − 1, we have
l − 1 r
≡ (−1)
r1 − l
X
r j=11 j
(mod l
2).
P r o o f. (a) The divisibility of the binomial coefficient by l is clear, and we have
1 l
l k
= (l − 1)(l − 2) . . . (l − (k − 1))
k! ≡ 1
k! (−1)
k−1(k − 1)! (mod l).
Hence the result.
(b) It follows from (a) and the binomial formula that X
l−1k=1
α
kk ≡ − 1
l X
l−1 k=1l k
(−α)
k≡ − 1
l ((1 − α)
l+ α
l− 1) (mod l).
(c) First,
q(xy) = x
l−1− 1
l y
l−1+ y
l−1− 1
l ≡ q(x) + q(y) (mod l).
Next, if x ≡ y (mod l), let x = y + hl, with h in Z, then, by the binomial formula,
q(x) = (y + hl)
l−1− 1
l ≡ y
l−1+ (l − 1)hly
l−2− 1
l ≡ q(y) − hy
l−2(mod l).
The conclusion follows, since y
l−2≡ (1/y) (mod l).
Now if xy ≡ 1 (mod l), then, by what precedes, q(x) + q(y) ≡ q(xy) ≡ q(1) − 1
xy · xy − 1
l (mod l), and the result follows.
(d) For every 1 ≤ k ≤ l − 1, we have
kx ≡ res
l(kx) (mod l) and kx − res
l(kx)
l =
kx l
, so that, by (c) above,
q(k) + q(x) ≡ q(kx) ≡ q(res
l(kx)) − 1 kx
kx l
(mod l).
Hence, by summation, X
l−1k=1
q(k) + (l − 1)q(x) ≡ X
l−1 k=1q(res
l(kx)) − 1 x
X
l−1 k=1kx l
1
k (mod l).
Since the map k 7→ res
l(kx) induces a permutation of the set {1, . . . , l − 1}, then P
l−1k=1
q(res
l(kx)) = P
l−1k=1
q(k), and the result follows.
(e) The property holds trivially for r = 0, so we assume 1 ≤ r ≤ l − 1.
In Z[X], we have the congruence
(X − 1)(X − 2) . . . (X − r) ≡ (−1)
rr! + (−1)
r−1X
r j=1r!
j
X (mod X
2).
Hence we get the congruence in Z (l − 1)(l − 2) . . . (l − r) ≡ (−1)
rr!
1 −
X
rj=1
1 j
l
(mod l
2).
The result follows upon division of both sides by r!.
Proposition 2. For a 6≡ 1 (mod l) and for any 1 ≤ n ≤ l − 1, [α
1, u
n] = n − 1
2 q(a − 1) − q(n
0) a − 1 + n
X
l−1 k=2(−1)
kk(a − 1)
kX
1≤s<kn0/l
k − 1 [sl/n
0]
(−1)
[sl/n0].
P r o o f. Substituting the expressions for Tr(λ
k/σ
n(λ)), obtained in Lemma 1, into the sum that occurs in the expression for [α
1, u
n], established in the Corollary to Proposition 1, and separating the terms corresponding to k = 1 and to k = l from the others, we get, in F
l,
(10) n l
X
l k=1(−c)
kk Tr
λ
kσ
n(λ)
= − nc + n X
l−1 k=2(−c)
kk
X
0≤s<kn0/l
k − 1 [sl/n
0]
(−1)
[sl/n0]+ n l
cn
0− c
ln
X
0−1 s=0l − 1 [sl/n
0]
(−1)
[sl/n0]. By Lemma 2(e),
n
X
0−1 s=0l − 1 [sl/n
0]
(−1)
[sl/n0]≡
n
X
0−1 s=01 − l
[sl/n
X
0] j=11 j
(11)
≡ n
0− l
n
X
0−1 s=1X
1≤j≤sl/n0
1
j (mod l
2).
In the last double sum, every term 1/j (for 1 ≤ j ≤ l − 1) occurs as many times as there are integers jn
0/l ≤ s ≤ n
0− 1, i.e. n
0− 1 − [jn
0/l] times.
Hence
n
X
0−1 s=1X
1≤j≤sl/n0
1 j =
X
l−1 j=1n
0− 1 −
jn
0l
1 (12) j
= (n
0− 1) X
l−1 j=11 j −
X
l−1 j=1jn
0l
1 j .
Applying Lemma 2(b) and (d) to evaluate the last two sums in (12) modulo l,
then substituting the result into (11), we get
(13)
n
X
0−1 s=0l − 1 [sl/n
0]
(−1)
[sl/n0]≡ n
0(1 + lq(n
0)) (mod l
2).
Substituting (13) into (10), we obtain, in F
l, (14) n
l X
l k=1(−c)
kk Tr
λ
kσ
n(λ)
= − cn − cq(c) − cq(n
0) + n
X
l−1 k=2(−c)
kk
X
0≤s<kn0/l
k − 1 [sl/n
0]
(−1)
[sl/n0]. In the last double sum, we isolate the terms corresponding to s = 0 and we use Lemma 2(b) to evaluate the resulting sum, which is (in F
l)
X
l−1 k=2(−c)
kk = c − (c + 1)
l− c
l− 1
l = c − 1
a − 1
a
l− a
l − (a − 1)q(a − 1)
. Rewriting (14) accordingly and then substituting it into the expression of [α
1, u
n] in the Corollary to Proposition 1 yields the desired result.
2. A formula for [α
1, u
n]
Lemma 3. For any integers 1 ≤ n, n
0≤ l − 1 such that nn
0≡ 1 (mod l) and 2 ≤ k ≤ l − 1, if d = (nn
0− 1)/l, then
X
1≤s<kn0/l
k − 1 [sl/n
0]
(−1)
[sl/n0]= X
d r=1X
1≤j≤k−1 j≡[(rn−1)/d] (mod n)
k − 1 j
(−1)
j.
P r o o f. If n = n
0= 1 then d = 0 and the equality holds trivially. We may thus assume n, n
0> 1, so that d ≥ 1. Let s = hd + r, with h, r ∈ N and 0 ≤ r ≤ d − 1 (euclidean division of s by d). We first prove that, for 1 ≤ s < kn
0/l,
(15)
sl n
0= hn +
rn − 1 d
.
If r = 0 then one can easily see that both sides in (15) are equal to hn − 1, and (15) holds. Assume then 1 ≤ r ≤ d − 1. Since
sl
n
0= (hd + r)ln
dl + 1 < hn + rn d , and rn/d is not an integer (d being prime to n), we have
sl n
0≤ hn +
rn − 1 d
.
The inverse inequality is obtained as follows: since s < kn
0/l, we have (hd + r)n < ld; hence
hn +
rn d
< l ≤
rn − d
rn d
l, so that
hn +
rn d
≤ (hd + r)ln dl + 1 = sl
n
0,
which implies the desired inequality and ends the proof of (15).
Now set j = [sl/n
0], and note that 1 ≤ s < kn
0/l if and only if 1 ≤ j ≤ k − 1. Moreover, since n
0< l and d < n, the maps s 7→ [sl/n
0] and r 7→ [(rn − 1)/d] are strictly increasing and therefore injective. Furthermore, any integer 1 ≤ j ≤ k − 1 satisfying j ≡ [(rn − 1)/d] (mod n), for some 0 ≤ r ≤ d − 1, is of the type j = [sl/n
0] for a convenient 1 ≤ s < kn
0/l (namely, if j = hn + [(rn − 1)/d], then s = hd + r). Therefore, in view of (15), the integers j = [sl/n
0], for 1 ≤ s < kn
0/l, are partitioned into the congruence classes j ≡ [(rn−1)/d] (mod n) (1 ≤ j ≤ k −1), for 0 ≤ r ≤ d−1 or, what is the same, for 1 ≤ r ≤ d. Hence the identity of the statement.
Lemma 4. Let m, n be positive integers, r ∈ Z and ζ
na primitive n-th root of unity in C. Then
X
0≤j≤m j≡r (mod n)
m j
(−1)
j= 1 n
n−1
X
k=1
ζ
n−kr(1 − ζ
nk)
m.
P r o o f. For any k ∈ Z, writing the binomial expansion of (1−ζ
nk)
m, then partitioning the resulting sum according to the congruence classes modulo n of the exponent, we get
(1 − ζ
nk)
m=
n−1
X
t=0
X
0≤j≤m j≡t (mod n)
m j
(−1)
jζ
nkt.
Multiplying this identity by ζ
n−kr, for k = 0, 1, . . . , n − 1 respectively, then adding the resulting equalities, we get
(16)
n−1
X
k=0
ζ
n−kr(1 − ζ
nk)
m=
n−1
X
t=0
X
0≤j≤m j≡t (mod n)
m j
(−1)
j n−1X
k=0
ζ
nk(t−r).
Since P
n−1k=0
ζ
nk(t−r)is equal to n if t ≡ r (mod n) and equal to 0 if t 6≡ r (mod n), the right-hand side of (16) is equal to n P
0≤j≤m j≡r (mod n)
m j
(−1)
j.
Hence the result.
Proposition 3. For a 6≡ 1 (mod l) and 1 ≤ n ≤ l − 1, letting d = (nn
0− 1)/l (with 1 ≤ n
0≤ l − 1 such that nn
0≡ 1 (mod l)), we have
[α
1, u
n] = n − 1
2 q(a − 1) + q(n) a − 1 − 1
a − 1
n−1
X
t=1
f
n,a(ζ
nt),
where f
n,a(X) = g
n(X)h
a(X) is the product of the following two polynomials in Z[X]
g
n(X) = X
d r=1X
n−[(rn−1)/d], h
a(X) = (X − a)
l− (X − 1)
l+ (a − 1)
ll(X − 1) .
Also, ζ
nis a primitive n-th root of unity in C, and P
n−1t=1
f
n,a(ζ
nt) is an element of Z independent of the choice of ζ
n.
P r o o f. It follows from Proposition 2 and Lemmas 3 and 4 that [α
1, u
n] = n − 1
2 q(a − 1) − cq(n
0) (17)
+
n−1
X
t=1
X
d r=1ζ
−t[(rn−1)/d]n
X
l−1 k=2(−c)
kk (1 − ζ
nt)
k−1. By Lemma 2(b), for 1 ≤ t ≤ n − 1, we have in Z[ζ
n] the congruence (18) (1 − ζ
nt)
X
l−1 k=2(−c)
kk (1 − ζ
nt)
k−1≡ (c(ζ
nt− 1) − 1)
l− c
l(ζ
nt− 1)
l+ 1
l − c(ζ
nt− 1) (mod l).
Since (a − 1)c ≡ 1 (mod l) and c(ζ
nt− 1) − 1 ≡ c(ζ
nt− a) (mod l), so that (a − 1)
lc
l≡ 1 (mod l
2) and (c(ζ
nt− 1) − 1)
l≡ c
l(ζ
nt− a)
l(mod l
2), the fractional term in the right-hand side of (18) is
≡ c
l(ζ
nt− a)
l− (ζ
nt− 1)
l+ (a − 1)
ll ≡ c(ζ
nt− 1)h
a(ζ
nt) (mod l).
Also, the right-hand side of (18) is an integral multiple of 1 − ζ
nt, which is relatively prime to l. Therefore, we can divide (18) through by 1 − ζ
ntto get (19)
X
l−1 k=2(−c)
kk (1 − ζ
nt)
k−1≡ −ch
a(ζ
nt) + c (mod l).
Substituting (19) into (17), and taking into account that, by Lemma 2(c),
q(n
0) ≡ −q(n) − d (mod l), we obtain
[α
1, u
n] = n − 1
2 q(a − 1) + cq(n) + cd (20)
− c
n−1
X
t=1
X
dr=1
ζ
−t[(rn−1)/d]n
h
a(ζ
nt)
+ c X
d r=1n−1
X
t=1
ζ
−t[(rn−1)/d]n
.
In the last double sum, P
n−1t=1
ζ
−t[(rn−1)/d]n
= −1, for 1 ≤ r ≤ d, since [(rn − 1)/d] is not divisible by n. Moreover, P
dr=1
ζ
−t[(rn−1)/d]n
= g
n(ζ
nt), for any t. Hence the result by substitution into (20).
Note that the polynomial (X−a)
l−(X−1)
l+(a−1)
lhas all its coefficients divisible by l (by the binomial expansion), and admits 1 as a root, so that h
a∈ Z[X]. Moreover, any element of the Galois group of Q(ζ
n)|Q permutes the nth roots of unity ζ
ntin the sum θ = P
n−1t=1
f
n,a(ζ
nt), and therefore it leaves θ invariant. Thus θ lies in Q, and since it is an integral element, it lies in Z.
Lemma 5. Let n be a positive integer.
(a) If ζ
nis a primitive n-th root of unity in C, and m ∈ Z, then
n−1
X
t=1
ζ
nmtζ
nt− 1 = n − 1
2 − res
n(m − 1).
(b) If d is a positive integer relatively prime to n, then
d−1
X
j=1
jn d
= (d − 1)(n − 1)
2 .
P r o o f. Both statements are trivially true if n = 1, since the sums in- volved, as well as the right-hand sides, are then equal to 0. We thus assume n ≥ 2.
(a) Let r = res
n(m − 1), so that m ≡ r + 1 (mod n) and 0 ≤ r ≤ n − 1.
Then (21)
n−1
X
t=1
ζ
nmtζ
nt− 1 =
n−1
X
t=1
ζ
n(r+1)t− 1 ζ
nt− 1 +
n−1
X
t=1
1 ζ
nt− 1 . For the first sum in the right-hand side of (21), we have
n−1
X
t=1
ζ
n(r+1)t− 1 ζ
nt− 1 =
X
r j=0n−1
X
t=1
ζ
njt= n − 1 − r.
For the second sum, consider the polynomial f (X) =
n−1
Y
t=1
(X − ζ
nt) = X
n− 1 X − 1 =
n−1
X
j=0
X
j. Then
f
0(X) f (X) =
n−1
X
t=1
1 X − ζ
nt, and thus
n−1
X
t=1
1
1 − ζ
nt= f
0(1) f (1) =
P
n−1j=1
j
n = n − 1 2 . Hence the result, by substitution into (21).
(b) Note that, for a real number x 6∈ Z, we have [n − x] = n − [x] − 1.
Since d is prime to n, it follows that, for 1 ≤ j ≤ d − 1, jn/d 6∈ Z. Hence, for 1 ≤ j ≤ d − 1, we have
(d − j)n d
+
jn d
= n − 1.
Therefore 2
d−1
X
j=1
jn d
=
d−1
X
j=1
jn d
+
(d − j)n d
= (d − 1)(n − 1).
Hence the result.
Lemma 6. In the notations and under the conditions of Proposition 3, we have
(a) For any integer 1 ≤ e ≤ n − 1,
n−1
X
t=1
ζ
neth
a(ζ
nt) ≡ (e − 1)
a
l− a
l − (a − 1)q(a − 1)
+
n−1
X
r=1
r X
1≤k≤l−1 k≡l+e−r−1 (mod n)
a
k− 1
k (mod l).
(b) Also,
n−1
X
t=1
f
n,a(ζ
nt) ≡ (d − 1)(n − 1) 2
a
l− a
l − (a − 1)q(a − 1)
+
d−1
X
j=0 n−1
X
r=1
r X
1≤k≤l−1 k≡l+[−jn/d]−r (mod n)
a
k− 1
k (mod l).
P r o o f. (a) For any 1 ≤ t ≤ n − 1, using the binomial expansions of (ζ
nt− a)
land (ζ
nt− 1)
lin the defining expression of h
a, we have
(ζ
nt− 1)h
a(ζ
nt) = (a − 1)
l− a
l+ 1
l −
X
l−1 k=11 l
l k
(−1)
k−1(a
k− 1)ζ
nt(l−k). Using Lemma 2(a), we convert this into a congruence modulo l in Z[ζ
n];
then dividing by ζ
nt− 1, which is relatively prime to l, we get, in the ring obtained by localization of Z[ζ
n] at a prime ideal above l,
h
a(ζ
nt) ≡ (a − 1)
l− a
l+ 1
l · 1
ζ
nt− 1 − X
l−1 k=1a
k− 1
k · ζ
nt(l−k)ζ
nt− 1 (mod l).
Hence, using Lemma 5(a), we deduce
n−1
X
t=1
ζ
neth
a(ζ
nt) ≡ (a − 1)
l− a
l+ 1 l
n − 1
2 − (e − 1)
− X
l−1 k=1a
k− 1 k
n − 1
2 − res
n(l − k + e − 1)
(mod l).
In view of Lemma 2(b), X
l−1 k=1a
k− 1
k ≡ (a − 1)
l− a
l+ 1
l (mod l),
so that (22)
n−1
X
t=1
ζ
neth
a(ζ
nt)
≡ (e − 1) a
l− (a − 1)
l− 1
l +
X
l−1 k=1res
n(l + e − k − 1) a
k− 1
k (mod l).
The sum in the right-hand side of (22) can be rewritten in terms of r = res
n(l + e − k − 1) which takes the values 0 ≤ r ≤ n − 1, each of which corresponds to those 1 ≤ k ≤ l − 1 such that k ≡ l + e − r − 1 (mod n). The result then follows immediately from (22).
(b) For every 1 ≤ j ≤ d, let e(j) = n − [(jn − 1)/d]. From the definitions, we have
n−1
X
t=1
f
n,a(ζ
nt) = X
d j=1n−1
X
t=1
ζ
nte(j)h
a(ζ
nt).
Thus, in view of (a) above,
n−1
X
t=1
f
n,a(ζ
nt) ≡
a
l− a
l − (a − 1)q(a − 1)
X
dj=1
e(j) − d
(23)
+ X
d j=1n−1
X
r=1
r X
1≤k≤l−1 k≡l+e(j)−r−1 (mod n)
a
k− 1
k (mod l).
Moreover, for 1 ≤ j ≤ d − 1, e(j) = n − [jn/d], while for j = d, e(d) = 1, so that, in view of Lemma 5(b), P
dj=1
e(j) = (d − 1)(n − 1)/2 + d. Also, for 1 ≤ j ≤ d, −[(jn − 1)/d] − 1 = [−jn/d], so that in the triple sum in (23), the inner summation is for k ≡ l + [−jn/d] − r (mod n); furthermore, the latter congruence class (mod n) is the same for j = d as for j = 0, so that the summation may take place for 0 ≤ j ≤ d − 1. Hence the result.
Definition. For a, n, r ∈ Z, with n > 0, we introduce the polynomial sums in F
lS
nr(a) = X
1≤k≤l−1 k≡r (mod n)
a
kk .
Proposition 4. For a 6≡ 1 (mod l) and 1 ≤ n ≤ l − 1, let d = (nn
0− 1)/l (with 1 ≤ n
0≤ l − 1 such that nn
0≡ 1 (mod l)). Then
[α
1, u
n] = d(n − 1)
2 q(a − 1) − (d − 1)(n − 1)
2(a − 1) · a
l− a
l + q(n) a − 1
− 1
a − 1
d−1
X
j=0 n−1
X
r=1
r(S
l+[−jn/d]−rn
(a) − S
l+[−jn/d]−rn
(1)).
P r o o f. The result is derived from Proposition 3, by substituting for P
n−1t=1
f
n,a(ζ
nt) its expression from Lemma 6(b), and writing the latter in terms of the above defined sums S
nt(a) and S
nt(1) in F
l.
3. The sums S
nr(a) and the expression for [α
1, u
n]
Lemma 7. The sums S
nr(a) have the following properties, for any a, n, r ∈ Z, with n > 0:
(a) If a 6≡ 0 (mod l), then
S
nl−r(a) = −aS
nr(a
0),
where a
0is an inverse of a modulo l in Z.
(b) For any a ∈ Z,
n−1
X
r=0
S
nr(a) = (a − 1)
l− a
l+ 1 l
(the summation could be, instead, over any residue system modulo n in Z).
(c) For any integer e ≥ 1, S
nr(a) =
e−1
X
h=0
S
enhn+r(a).
(d) If l - n, then
a
rS
n−rl(a) = X
rl/n<j<(r+1)l/n
a
njnj , provided a 6≡ 0 (mod l) if r ≤ 0.
(e) If l - n and a
n≡ 1 (mod l), then
n−1
X
r=1
ra
rS
n−lr(a) = q(n).
(f) If l - n and m ∈ Z, then
n−1
X
r=1
rS
nm−r(a) = X
l−1 k=1res
n(m − k) a
kk ,
n−1
X
r=1
rS
nr(a) = X
l−1 k=1res
n(k) a
kk . (g) For any a ∈ Z,
S
20(a) = 1
2 · (a − 1)
l− (a + 1)
l+ 2
l , S
21(a) = 1
2 · (a − 1)
l− 2a
l+ (a + 1)
ll .
(h) If l - n and a 6≡ 1 (mod l), then
n−1
X
r=1