LXXIV.4 (1996)
Congruence of Ankeny–Artin–Chowla type modulo p
2for cyclic fields of prime degree l
by
Stanislav Jakubec (Bratislava)
Introduction. Let p ≡ 1 (mod 4), let T + U √
p > 1 be the fundamental unit, and let h be the class number of Q( √
p). The following congruence (Ankeny–Artin–Chowla congruence) holds:
h U
T ≡ B
(p−1)/2(mod p).
For a cubic field K ⊂ Q(ζ
p+ ζ
p−1), p ≡ 1 (mod 3) the analogous congruence was proved by Feng Ke Qin in [1]. In general, for fields of the lth degree, the analogous congruence is proved in [5].
The aim of this paper is to prove a congruence of Ankeny–Artin–Chowla type modulo p
2for real Abelian fields of a prime degree l and a prime conductor p. The importance of such a congruence can be demonstrated by the following example. Let K be a cubic field. In [5], the following congruence is proved:
(1) h
KS
1S
2≡ −
34B
(p−1)/3B
2(p−1)/3(mod p).
As is well known, h
K< p for a cubic field. If B
(p−1)/3B
2(p−1)/3≡ 0 (mod p), then S
1S
2≡ 0 (mod p), hence the congruence (1) does not provide any information about h
K. Note that such a prime exists, e.g. p = 5479. We have
B
(p−1)/3= B
1826≡ 0 (mod 5479).
In this paper two applications of the main theorem (Theorem 1), for a quadratic and for a cubic field, will be given.
Let l and p be primes such that p ≡ 1 (mod l) and let K ⊂ Q(ζ
p+ ζ
p−1) with [K : Q] = l. Let a be a primitive root modulo p. As is well known, the conjugates of the unit
1991 Mathematics Subject Classification: Primary 11R29.
[293]
η
a= N
Q(ζp+ζ−1p )/K
ζ
p(1−a)/21 − ζ
pa1 − ζ
p,
generate the group of cyclotomic units C(K) of K. Consider the unit η
2= N
Q(ζp+ζp−1)/K
(ζ
p+ ζ
p−1).
It is easy to prove that if 2 is not an lth power modulo p, then the conjugates of the unit η
2generate the group C(K). Let hεi be the group generated by all conjugates of the unit ε.
According to ([3], Lemma 1, p. 69) for a cyclic field K with [K : Q] = l, there is a unit δ such that [U
K: hδi] = f , where (p, f ) = 1.
The following is taken from [5]. According to [8] and [9] (see also [10], p. 284), we have h
K= [U
K: C(K)], where C(K) = hη
2i is the group of cyclotomic units of K. From [U
K: hδi] = f we have η
2f∈ hδi. Let [hδi : hη
2fi] = e. Clearly [hη
2i : hη
2fi] = f
l−1.
Consider two towers of groups
hη
f2i ⊂ hδi ⊂ U
K, hη
f2i ⊂ hη
2i ⊂ U
K. This implies ef = h
Kf
l−1and hence e = h
Kf
l−2. Let
η
2f= δ
c0σ(δ)
c1. . . σ
l−2(δ)
cl−2. It is easy to prove that
e = N
Q(ζl)/Q(c
0+ c
1ζ
l+ . . . + c
l−2ζ
ll−2).
Note that a unit δ for which f = 1 is called a strong Minkowski unit. As is well known, a strong Minkowski unit exists for a cyclic fields K with l < 23.
The problem of existence of a strong Minkowski unit for cyclic fields of small non-prime degree is solved in [7].
Let a be a fixed primitive root modulo p, let χ be the Dirichlet character of order n, n | p − 1, χ(x) = ζ
nindax. Let g be such that g ≡ a
(p−1)/n(mod p) and g
n≡ 1 (mod p
p). Denote by p a prime divisor of Q(ζ
n) such that p | p and 1/g ≡ ζ
n(mod p
p).
Define the rational numbers A
0(n), A
1(n), . . . , A
n−1(n) in the following way:
(2)
A
0(n) = −1/n,
τ (χ
i)
n≡ n
nA
i(n)
n(−p)
i(mod p
2+i), A
i(n) ≡ (p − 1)/n
(i(p − 1)/n)! (mod p), where τ (χ) is the Gauss sum.
Put m = (p − 1)/2, and
G
j(X) = A
0(m)X
j+ A
1(m)X
j−1+ . . . + A
j(m), F
j(X) = 1
(p − 1)! X
j+ 1
(p + 1)! X
j−1+ 1
(p + 3)! X
j−2+ . . . + 1
(p + 2j − 1)! .
Define
E
n∗= E
2n(2n)! for n = 1, 2, 3, . . . ,
where E
2nare the Euler numbers, i.e. E
0= 1, E
2= −1, E
4= 5, E
6= −61, E
8= 1385, E
10= −50521, E
12= 2702765, E
14= −199360981, . . .
Consider the formal expressions G
j(E
∗) and F
j(E
∗), where (E
∗)
k= E
k∗.
Let β
0, β
1, . . . , β
l−1be the integral basis of the field K formed by the Gauss periods. Let
δ = x
0β
0+ x
1β
1+ . . . + x
l−1β
l−1. Associate with the unit δ the polynomial f (X) as follows:
f (X) = X
l−1+ d
1X
l−2+ d
2X
l−3+ . . . + d
l−1, where
d
i= −lA
i(l) x
0+ x
1g
i+ x
2g
2i+ . . . + x
l−1g
i(l−1)x
0+ x
1+ . . . + x
l−1for i = 1, . . . , l − 1. Put
S
j= S
j(d
1, . . . , d
l−1) = sum of the jth powers of the roots of f (X) for j = 1, . . . , 2l − 1. Hence
S
1= −d
1, S
2= d
21− 2d
2, S
3= −d
31+ 3d
1d
2− 3d
3, . . . Define the numbers T
1, . . . , T
2l−1as follows:
T
i= − 1
(i(p − 1)/l)! 2
i(p−1)/l−1(2
i(p−1)/l−1)B
i(p−1)/l−i p − 1
4l G
i(p−1)/(2l)(E
∗) for i = 1, . . . , l − 1, and
T
l= 1 − q
22 , where q
2= 2
p−1− 1
p ,
T
l+i= − 1
(p − 1 + i(p − 1)/l)! 2
p−1+i(p−1)/l−1(2
p−1+i(p−1)/l− 1)
× B
(p−1+i(p−1)/l)+
p − 1
2 + i p − 1 2l
F
i(p−1)/(2l)(E
∗) for i = 1, . . . , l − 1.
Define
α
i= c
0+ c
1g
i+ c
2g
2i+ . . . + c
l−2g
(l−2)ifor i = 1, . . . , 2l − 1.
Let X
1, . . . , X
2l−1∈ Q and let
g(X) = X
2l−1+ Y
1X
2l−2+ . . . + Y
2l−1be a polynomial such that
X
j= sum of the jth powers of the roots of g(X).
Define the mapping Φ : Q
2l−1→ Q
las follows:
Φ(X
1, . . . , X
2l−1) = (1 − pY
l, Y
1− pY
l+1, . . . , Y
l−1− pY
2l−1).
Now the main theorem of this paper be formulated.
Theorem 1. Let l and p be primes with p ≡ 1 (mod l) and let K ⊂ Q(ζ
p+ ζ
p−1) with [K : Q] = l. Suppose that 2 is not an l-th power mod- ulo p. Let δ be a unit of K such that [U
K: hδi] = f , (f, p) = 1. Let η
2f= δ
c0σ(δ)
c1. . . σ
l−2(δ)
cl−2and α
i= c
0+ c
1g
i+ . . . + c
l−2g
(l−2)ifor i = 1, . . . , 2l − 1. The following congruence holds:
(3) ε
x
0+ x
1+ . . . + x
l−1−l
αlΦ(α
1S
1, . . . , α
2l−1S
2l−1)
≡ (2 + 2p)
f (p−1)/(2l)Φ(f T
1, . . . , f T
2l−1) (mod p
2), where ε = ±1.
R e m a r k. The class number h
Kappears in the preceding congruence implicitly, via the congruence
h
Kf
l−2≡ α
1. . . α
l−1(mod p
2).
This congruence can be proved in the following way:
We have the congruence 1/g ≡ ζ
l(mod p
p) and hence σ
−1(c
0+ c
1ζ
li+ . . . + c
l−2ζ
l(l−2)i) ≡ α
i(mod p
p);
this yields
h
Kf
l−2= e = N
Q(ζl)/Q(c
0+ c
1ζ
l+ . . . + c
l−2ζ
ll−2) ≡ α
1. . . α
l−1(mod p
2).
The congruence (3) gives l congruences (one in each component). If B
(p−1)/lB
2(p−1)/l. . . B
(l−1)(p−1)/l6≡ 0 (mod p),
then from the congruence (3), the numbers α
1, . . . , α
l−1modulo p
2can be calculated. Using the congruence
h
Kf
l−2≡ α
1. . . α
l−1(mod p
2), also h
Kcan be calculated modulo p
2. If
B
(p−1)/lB
2(p−1)/l. . . B
(l−1)(p−1)/l≡ 0 (mod p),
then the numbers α
1, . . . , α
l−1and hence also h
Kcan be calculated at most
modulo p.
Before proving Theorem 1, we show its applications to quadratic and cubic fields.
The quadratic case K = Q( √
p), p ≡ 5 (mod 8). Let δ = x
0β
0+ x
1β
1= x
0−1 + √ p 2 + x
1−1 − √ p 2 > 1 be a fundamental unit of Q( √
p). Then
d
1= −2A
1(2) x
0− x
1x
0+ x
1. Hence
S
1= 2A
1(2) x
0− x
1x
0+ x
1, S
2= 4A
1(2)
2(x
0− x
1)
2(x
0+ x
1)
2, S
3= 8A
1(2)
3(x
0− x
1)
3(x
0+ x
1)
3. For A
1(2) we have
τ (χ)
2≡ 4A
1(2)
2(−p) (mod p
3), A
1(2) ≡ (p − 1)/2
((p − 1)/2)! (mod p).
Hence
A
1(2)
2≡ − 1
4 (mod p
2), A
1(2) ≡ (p − 1)/2
((p − 1)/2)! (mod p).
Therefore
S
1= 2A
1(2) x
0− x
1x
0+ x
1, S
2= − (x
0− x
1)
2(x
0+ x
1)
2, S
3= −2A
1(2) (x
0− x
1)
3(x
0+ x
1)
3. Let
x
0+ x
1−2 + x
0− x
12
√ p = T + U √ p > 1.
It can be proved that |η
2| = |N
Q(ζp+ζp−1)/Q(√
p)
(ζ
p+ ζ
p−1)| < 1, hence it is necessary to start from the unit T − U √
p. We have S
1= 2A
1(2) U
T , S
2= − U
2T
2, S
3= −2A
1(2) U
3T
3. For the numbers T
1, T
2, T
3we have
T
1= − 1
((p − 1)/2)! 2
(p−1)/2−1(2
(p−1)/2− 1)B
(p−1)/2− p − 1
8 G
(p−1)/4(E
∗), T
2=
12(1 − q
2),
T
3= − 1
(3(p − 1)/2)! 2
3(p−1)/2−1(2
3(p−1)/2− 1)B
3(p−1)/2+ (3(p − 1)/4)F
(p−1)/4(E
∗).
It is easy to see that Φ(X
1, X
2, X
3) =
1 − p X
12− X
22 , −X
1− p
− 1
6 X
13+ 1
2 X
1X
2− 1 3 X
3. Hence
ε
x
0+ x
1−2
α2Φ(α
1S
1, α
2S
2, α
3S
3) ≡ (2+2p)
(p−1)/4Φ(T
1, T
2, T
3) (mod p
2).
Since (x
0+ x
1)/(−2) = T and α
1= α
2= α
3= h, we get
εT
hΦ(hS
1, hS
2, hS
3) ≡ (2 + 2p)
(p−1)/4Φ(T
1, T
2, T
3) (mod p
2), where ε = ±1. It can be proved that ε = (−1)
1+r, where r is the number of quadratic residues modulo p in the interval (p/4, p/2).
The cubic case. Let p be a prime such that p ≡ 1 (mod 3), p 6= a
2+ 27b
2and let
δ = x
0β
0+ x
1β
1+ x
2β
2. Then
d
1= −3A
1(3) x
0+ x
1g + x
2g
2x
0+ x
1+ x
2, d
2= −3A
2(3) x
0+ x
1g
2+ x
2g x
0+ x
1+ x
2. For the numbers A
1(3), A
2(3) we have
τ (χ)
3≡ 27A
1(3)
3(−p) (mod p
3), A
1(3) ≡ (p − 1)/3
((p − 1)/3)! (mod p), τ (χ
2)
3≡ 27A
2(3)
3(−p)
2(mod p
4), A
2(3) ≡ (p − 1)/3
(2(p − 1)/3)! (mod p).
As is well known, τ (χ)
3= pJ(χ, χ), where J(χ, χ) is the Jacobi sum.
Let J(χ, χ) = a + bζ
3, a ≡ −1, b ≡ 0 (mod 3) and p = a
2− ab + b
2. Hence
−
a + b 1
g
≡ 27A
1(3)
3(mod p
2), A
1(3) ≡ (p − 1)/3
((p − 1)/3)! (mod p).
The number A
2(3) is determined by the congruence
−1 ≡ 27
2A
1(3)
3A
2(3)
3(mod p
2), A
2(3) ≡ (p − 1)/3
(2(p − 1)/3)! (mod p).
For the numbers T
1, . . . , T
5we have T
i= − 1
(i(p − 1)/3)! 2
i(p−1)/3−1(2
i(p−1)/3−1)B
i(p−1)/3−i p − 1
12 G
i(p−1)/6(E
∗) for i = 1, 2, and
T
3=
12(1 − q
2),
T
3+i= − 1
(p − 1 + i(p − 1)/3)! 2
p−1+i(p−1)/3−1(2
p−1+i(p−1)/3− 1)
× B
(p−1+i(p−1)/3)+
p − 1
2 + i p − 1 6
F
i(p−1)/6(E
∗) for i = 1, 2.
It is easy to prove that for Φ(X
1, . . . , X
5) we have Φ(X
1, . . . , X
5)
=
1 − p
− 1
6 X
13+ 1
2 X
1X
2− 1 3 X
3,
− X
1− p
1
24 X
14+ 1
3 X
1X
3+ 1
8 X
22− 1
4 X
12X
2− 1 4 X
4, 1
2 (X
12− X
2) − p
− 1
120 X
15+ 1
4 X
1X
4+ 1 6 X
2X
3+ 1
12 X
13X
2− 1
6 X
12X
3− 1
8 X
1X
22− 1 5 X
5. Hence
±
−1 3
α3(x
0+ x
1+ x
2)
α3Φ(α
1S
1, . . . , α
5S
5)
≡ (2 + 2p)
(p−1)/6Φ(T
1, . . . , T
5) (mod p
2).
P r o o f o f T h e o r e m 1. Let g
1(X), . . . , g
r(X) be polynomials such that g
i(X) 6≡ 0 (mod X). Let
g(X) ≡ g
1(X)
b1. . . g
r(X)
br(mod X
M).
Let s
jbe the homomorphism defined in [4]. We have
s
j(g(X)) = b
1s
j(g
1(X)) + b
2s
j(g
2(X)) + . . . + b
rs
j(g
r(X)), for j = 1, . . . , M − 1. Define
X
j= b
1s
j(g
1(X)) + b
2s
j(g
2(X)) + . . . + b
rs
j(g
r(X)) for j = 1, . . . , M − 1. Let
g(X) ≡ C
0+ C
1X + C
2X
2+ . . . + C
M −1X
M −1(mod X
M).
Consider the reciprocal polynomial F (X) = X
M −1+ C
1C
0X
M −2+ . . . + C
M −1C
0.
By the definition of the homomorphism s
j(see [4]) we have X
j= sum of the jth powers of the roots of F (X).
The numbers C
1/C
0, C
2/C
0, . . . , C
M −1/C
0can be calculated by the Newton recurrence formula.
According to [2] and [4] we have:
Proposition 1. There is a number π ∈ Q(ζ
p+ ζ
p−1), π | p such that (i) N
Q(ζp+ζ−1p )/Q
(π) = (−1)
mp,
(ii) σ(π) ≡ gπ (mod π
2m+1), g ≡ a
2p(mod p
2), (iii) ζ
p+ ζ
p−1≡ P
2mi=0
a
iπ
i(mod π
2m+1), where 0 ≤ a
i< p and a
i≡ (2/(2i)!) (mod p) for i = 1, . . . , m.
The numbers a
m+ifor i = 1, . . . , m, are defined by a
m+1= 2 p − 1 − p(p + 1)B
p−1p .
If 2 is a primitive root modulo p then the coefficients a
m+2, a
m+3, . . . , a
2mare given by the recurrence formula a
m+1+s≡ 1
4
s+1− 4
4
p(s+1)a
s+1− b
s+1p + b
m+s+1(mod p), where b
s+1and b
m+1+sare the coefficients of X
s+1and X
m+1+s, respec- tively, in the polynomial
2 + 2
2! X + . . . + 2
(p − 1)! X
m+ a
m+1X
m+1+ . . . + a
m+sX
m+s 2. Proposition 2. Let K ⊂ Q(ζ
p+ ζ
p−1) with [K : Q] = n. There is a number π ∈ K with π | p such that
(i) N
K/Q(π) = (−1)
np,
(ii) σ(π) ≡ gπ (mod π
n+1), g ≡ a
p(p−1)/n(mod p
2), (iii) β
0≡ P
2ni=0
a
∗iπ
i(mod π
2n+1), where 0 ≤ a
∗i< p and where a
∗i≡ ((p − 1)/n)/(i(p − 1)/n)! (mod p) for i = 1, . . . , n.
From now on we will use the following transformation. Let γ ≡ c
0+ c
1π + c
2π
2+ . . . + c
2l−1π
2l−1(mod π
2l).
From π
l≡ −p (mod π
2l) we have
γ ≡ (c
0− pc
l) + (c
1− pc
l+1)π + . . . + (c
l−1− pc
2l−1)π
l−1(mod π
2l).
Lemma 1. Let α, β ∈ K with αβ 6≡ 0 (mod π). Let
α ≡ e
0+ e
1π + . . . + e
l−1π
l−1(mod π
2l),
β ≡ b
0+ b
1π + . . . + b
l−1π
l−1(mod π
2l).
Relate to the conjugation σ
i(α) the polynomial
f
i(X) = e
0+ e
1g
iX + e
2g
2iX
2+ . . . + e
l−1g
i(l−1)X
l−1. Let
F (X) ≡ f
0(X)
c0f
1(X)
c1. . . f
l−2(X)
cl−2≡ d
0+ d
1X + . . . + d
2l−1X
2l−1(mod X
2l).
Define X
i= s
i(F (X)). Then
α
c0σ(α)
c1. . . σ
l−2(α)
cl−2≡ β (mod p
2) if and only if
d
0Φ(X
1, . . . , X
2l−2) ≡ (b
0, b
1, . . . , b
l−1) (mod p
2).
P r o o f. Clearly
α
c0σ(α)
c1. . . σ
l−2(α)
cl−2≡ d
0+ d
1π + . . . + d
2l−1π
2l−1(mod π
2l).
The coefficients d
1/d
0, d
2/d
0, . . . , d
2l−1/d
0can be expressed using the num- bers X
1, . . . , X
2l−1by the Newton recurrence formula. Clearly
d
0+ d
1π + . . . + d
2l−1π
2l−1≡ (d
0− pd
l) + (d
1− pd
l+1)π + . . . + (d
l−1− pd
2l−1)π
l−1(mod π
2l).
From the definition of the mapping Φ it follows that
(d
0−pd
l, d
1−pd
l+1, . . . , d
l−1−pd
2l−1) ≡ d
0Φ(X
1, X
2, . . . , X
2l−1) (mod p
2).
Lemma 1 is proved.
Proposition 2 gives
β
0≡ −1/l + (a
∗1− pa
∗l+1)π + . . . + (a
∗l−1− pa
∗2l−1)π
l−1(mod π
2l), where 0 ≤ a
∗i< p and a
∗i≡ ((p − 1)/n)/(i(p − 1)/n)! (mod p) for i = 1, . . . , l − 1. According to [4] we have
a
∗i− pa
∗l+i≡ A
i(l) (mod p
2).
Let δ = x
0β
0+ x
1β
1+ . . . + x
l−1β
l−1. Then δ ≡ − 1
l (x
0+ x
1+ . . . + x
l−1) +
X
l−1 i=1A
i(l)(x
0+ x
1g
i+ x
2g
2i+ . . . + x
l−1g
i(l−1))π
i(mod π
2l).
Let
δ
c0σ(δ)
c1. . . σ
l−2(δ)
cl−2= η
2f.
Then X
jcorresponding to the product on the left-hand side is equal to
c
0s
j(δ) + c
1g
js
j(δ) + . . . + c
l−2g
j(l−2)s
j(δ) = α
jS
jfor j = 1, . . . , 2l − 1. Hence the number corresponding to the left-hand side is
x
0+ x
1+ . . . + x
l−1−l
αlΦ(α
1S
1, . . . , α
2l−1S
2l−1).
Now we prove that the right-hand side is equal to (2 + 2p)
f (p−1)/lΦ(f T
1, . . . , f T
2l−1).
By Proposition 1 we have ζ
p+ ζ
p−1≡
X
2m i=0a
iπ
i(mod π
2m+1),
where 0 ≤ a
i< p and a
i≡ (2/(2i)!) (mod p) for i = 1, . . . , m, and hence ζ
p+ ζ
p−1≡ A
0(m) + A
1(m)π + . . . + A
m−1(m)π
m−1(mod π
2m).
Consider the polynomial
g(X) = A
0(m)X
m−1+ A
1(m)X
m−2+ . . . + A
m−1(m).
Now we shall calculate the numbers
s
i= sum of the ith powers of the roots of g(X)
for i = 1, . . . , 2m − 1 modulo p
2. It is easy to see that for i > m − 1 it is enough to determine s
imodulo p. Let W
1, W
2, . . . be a linearly recurrent sequence modulo p of order m − 1 defined by
W
i= −1
(2i)! 2
2i−1(2
2i− 1)B
2ifor i = 1, . . . , m − 1.
For i > m − 1 we have W
i= −
1
2! W
i−1+ 1
4! W
i−2+ . . . + 1
(2m − 2)! W
i−m+1. Lemma 2. The following congruence holds:
s
n+1≡ −1
(2n + 2)! 2
2n+1(2
2n+2− 1)B
2n+2− n + 1
2 G
n+1(E
∗) (mod p
2) for n + 1 < m.
P r o o f. It is easy to see that A
0(m) ≡ 2 + 2p (mod p
2). Clearly (2 + 2p) 1 − p
2 ≡ 1 (mod p
2).
Consider the polynomial
g
∗(X) = X
m−1+ C
1X
m−2+ . . . + C
m−1, where
C
i≡ 1 − p
2 A
i(m) (mod p
2).
Obviously s
nis equal to the sum of the nth powers of the roots of g
∗(X).
Write
C
1= c
1+ b
1p, C
2= c
2+ b
2p, . . . , C
m−1= c
m−1+ b
m−1p.
It is known that there exists a polynomial f
n(x
1, . . . , x
n) such that s
n= f
n(c
1+ b
1p, . . . , c
n+ b
np).
Write s
n= r
n+ t
np. Clearly r
1+ t
1p = −(c
1+ b
1p). Put t
1= −b
1. According to the Newton recurrent formula we have
r
2+ t
2p + (c
1+ b
1p)(r
1+ t
1p) + 2(c
2+ pb
2) = 0.
The last equation can be rewritten in the form
r
2+ c
1r
1+ 2c
2+ p(t
2+ c
1t
1+ r
1b
1+ 2b
2) = 0.
Define F
2= r
1b
1+ 2b
2. Hence r
2+ c
1r
1+ 2c
2+ p(t
2+ c
1t
1+ F
2) = 0. Put t
2= −(c
1t
1+ F
2).
Further we have
r
3+ t
3p + (c
1+ b
1p)(r
2+ t
2p) + (c
2+ b
2p)(r
1+ t
1p) + 3(c
3+ b
3p) = 0, hence
r
3+ c
1r
2+ c
2r
1+ 3c
3+ p(t
3+ c
1t
2+ c
2t
1+ b
1r
2+ b
2r
1+ 3b
3) = 0.
Define F
3= b
1r
2+ b
2r
1+ 3b
3. Then we put t
3= t
1(c
21− c
2) + c
1F
2− F
3,
t
4= t
1(−c
31+ 2c
1c
2− c
3) − (c
21− c
2)F
2+ c
1F
3− F
4,
t
5= t
1(c
41− 3c
21c
2+ 2c
1c
3+ c
22− c
4) − (−c
31+ 2c
1c
2− c
3)F
2− (c
21− c
2)F
3+ c
1F
4− F
5. Consider the numbers
K
2= −c
1, K
4= c
21− c
2, K
6= −c
31+ 2c
1c
2− c
3, . . .
The numbers c
1, c
2, c
3, . . . are equal to 1/2!, 1/4!, 1/6!, . . . modulo p. Define K
2n= K
2n∗/(2n)!.
Then
K
2n∗(2n)! + 1
(2)! · K
2n−2∗(2n − 2)! + . . . + 1
(2n − 2)! · K
2∗2! + 1
(2n)! = 0, hence
1 (2n)!
K
2n∗+
2n 2
K
2n−2∗+ . . . + 1
= 0,
and therefore K
2n∗= E
2n, the Euler number. Using induction by n we put (4) t
n+1= t
1E
2n(2n)! − F
2E
2n−2(2n − 2)! − . . . − F
n+1,
where
F
2= b
1r
1+ 2b
2, F
3= b
1r
2+ b
2r
1+ 3b
3, F
4= b
1r
3+ b
2r
2+ b
3r
1+ 4b
4, . . . Since t
1= −b
1, we have
t
n+1= − b
1E
2n(2n)! − E
2n−2(2n − 2)! (b
1r
1+ 2b
2)
− . . . − (b
1r
n+ b
2r
n−1+ . . . + (n + 1)b
n+1).
In the last equation we first cancel the brackets by multiplication and then make new brackets by factoring out b
1, . . . , b
n+1. The summand obtained by factoring out b
1is
(5) b
1E
2n(2n)! + r
1E
2n−2(2n − 2)! + . . . + r
n. According to [4, Lemma 3],
r
i≡ − 1
(2i)! 2
2i−1(2
2i− 1)B
2i(mod p).
By substitution into (5) we get the sum E
2n(2n)! − E
2n−2(2n − 2)! · 1
2! 2(2
2− 1)B
2− E
2n−4(2n − 4)! · 1
4! 2
3(2
4− 1)B
4− . . . − 1
(2n)! 2
2n−1(2
2n− 1)B
2n. The following identity holds:
(6) E
2n(2n)! − E
2n−2(2n − 2)! · 1
2! 2(2
2− 1)B
2− E
2n−4(2n − 4)! · 1
4! 2
3(2
4− 1)B
4− . . . − 1
(2n)! 2
2n−1(2
2n− 1)B
2n= (n + 1) E
2n(2n)! . This identity can be proved in the following way. For the functions sec x and tan x, we have
sec x = 1 − E
22! x
2+ E
44! x
4− E
66! x
6+ . . . , tan x = 2
2(2
2− 1)B
2x
2! − 2
4(2
4− 1)B
4x
34! + . . . Then
tan x · sec x = 2
2(2
2− 1) 2! B
2x −
E
2B
22
2(2
2− 1)
2!2! + 2
4(2
4− 1)B
44!
x
3+ . . . The identity (6) follows from the equation
d(sec x)
dx = tan x · sec x.
Using the identity (6), by induction (n + 1 → 1) we have (7) t
n+1= −(n + 1)
b
1E
2n(2n)! + b
2E
2n−2(2n − 2)! + . . . + b
n+1. The numbers a
1, . . . , a
m−1from Proposition 1 are
a
i≡ 2
(2i)! (mod p), 0 < a
i< p.
Define the numbers D
ias follows:
1 − p
2 a
i+ pD
i≡ 1
(2i)! (mod p
2), 0 ≤ D
i< p, hence
D
i≡ 1 p
1
(2i)! − 1 − p 2 a
i(mod p).
Let v
n+1be the sum of the (n+1)th powers of the roots of the polynomial X
m−1+ 1 − p
2 a
1X
m−2+ . . . + 1 − p 2 a
m−1.
According to [4, Lemma 3], the sum of the (n + 1)th powers of the roots of the polynomial
X
m−1+ 1
2! X
m−2+ 1
4! X
m−3+ . . . + 1 (2m − 2)! , is equal to
−1
(2n + 2)! 2
2n+1(2
2n+2− 1)B
2n+2. Hence
v
n+1= −1
(2n + 2)! 2
2n+1(2
2n+2− 1)B
2n+2+ p(n + 1)
D
1E
2n(2n)! + D
2E
2n−2(2n − 2)! + . . . + D
n+1. Therefore
s
n+1= −1
(2n + 2)! 2
2n+1(2
2n+2− 1)B
2n+2(8)
+ p(n + 1)
D
1E
2n(2n)! + D
2E
2n−2(2n − 2)! + . . . + D
n+1− p(n + 1)
b
1E
2n(2n)! + b
2E
2n−2(2n − 2)! + . . . + b
n+1. Since
D
i≡ 1 p
1
(2i)! − 1 − p 2 a
i,
from (8) we get s
n+1= −1
(2n + 2)! 2
2n+1(2
2n+2− 1)B
2n+2− (n + 1)E
2n+2(2n + 2)!
(9)
+ (n + 1) p − 1 2
a
1E
2n(2n)! + a
2E
2n−2(2n − 2)! + . . . + a
n+1+ p n + 1 2
a
m+1E
2n(2n)! + a
m+2E
2n−2(2n − 2)! + . . . + a
m+n+1, where a
1, . . . , a
m, a
m+1, a
m+2, . . . are the numbers from Proposition 1. Since A
i(m) = a
i− pa
m+i, from the equality (9) we get
s
n+1= −1
(2n + 2)! 2
2n+1(2
2n+2− 1)B
2n+2− n + 1
2 G
n+1(E
∗).
Lemma 3. The following congruence holds:
W
m+j≡ − 1
(p − 1 + 2j)! 2
p−1+2j−1(2
p−1+2j− 1)B
p−1+2j+ (m + j)F
j(E
∗) (mod p) for j = 1, . . . , m − 1.
P r o o f. Let n > 2p − 3. Consider the polynomial h(X) = X
n+ 1
2! X
n−1+ 1
4! X
n−2+ . . . + 1 (2n)! .
Let W
i∗be the sum of the ith powers of the roots of h(X). By [4, Lemma 3], W
i∗= − 1
(2i)! 2
2i−1(2
2i− 1)B
2i. The characteristic polynomial of the sequence W
iis
X
m−1+ 1
2! X
m−2+ 1
4! X
m−3+ . . . + 1 (2m − 2)! .
Hence W
iis the sum of the ith powers of the roots of this polynomial.
Let
c
1= 1
2! , c
2= 1 4! , . . . Obviously
W
m= W
m∗+ mc
m,
W
m+1= W
m+1∗+ c
mW
1+ (m + 1)c
m+1− c
1mc
m. We can write
G
m= mc
m, G
m+1= (m + 1)c
m+1− c
1G
m.
By induction we can prove that
W
m+j= W
m+j∗+ c
mW
j+ (c
m+1+ c
mE
1∗)W
j−1(10)
+ (c
m+2+ c
m+1E
1∗+ c
mE
2∗)W
j−2+ . . .
+ (c
m+j−1+ c
m+j−2E
1∗+ . . . + c
mE
j−1∗)W
1+ G
m+j. Now we rewrite the right side of (10). We cancel the brackets by multi- plication and make new brackets by factoring out c
m, c
m+1, . . .
E.g., the summand obtained by factoring out c
mis c
m(W
j+ E
1∗W
j−1+ E
2∗W
j−2+ . . . + E
j−1∗W
1).
From (6) we get
c
m(W
j+ E
1∗W
j−1+ E
2∗W
j−2+ . . . + E
∗j−1W
1) = c
mjE
j∗. By repeating this procedure with each summand we have
(11) W
m+j= W
m+j∗+ jc
mE
j∗+ (j − 1)c
m+1E
j−1∗+ . . . + c
m+j−1E
1∗+ G
m+j. By induction it is easy to prove
(12) G
m+j= (m + 1)c
m+j+ (m + j − 1)c
m+j−1E
1∗+ . . . + mc
mE
j∗. Substituting (12) into (11) we get
W
m+j= W
m+j∗+ (m + j)(c
m+j+ c
m+j−1E
∗1+ . . . + c
mE
j∗).
Since η
2= N
Q(ζp+ζ−1p )/K
(ζ
p+ ζ
p−1) we have η
f2≡ r
0+
X
l−1 i=1r
i(p−1)/(2l)π
i(p−1)/(2l)(mod π
2m),
where r
0= (2 + 2p)
f (p−1)/(2l). From Lemmas 1–3 we deduce that the right- hand side is equal to
(2 + 2p)
f (p−1)/(2l)Φ(f T
1, . . . , f T
2l−1).
It remains to prove that T
l= (1 − q
2)/2. From the proof of Lemma 3 we have
T
l≡ −1
(p − 1)! 2
p−2(2
p−1− 1)B
p−1+ p − 1
2 · 1
(p − 1)! , hence
T
l≡ −1
(p − 1)! 2
p−22
p−1− 1
p pB
p−1+ p − 1
2 · 1
(p − 1)! .
From pB
p−1≡ −1 (mod p) the required congruence follows. Theorem 1 is
proved.
Example 1 (p = 13, l = 3, K ⊂ Q(ζ
13+ ζ
13−1), [K : Q] = 3). By Proposition 1 we have
ζ
13+ ζ
13−1≡ 2 + π + 12π
2+ 3π
3+ 4π
4+ 9π
5+ 11π
6+ 2π
7+ 12π
8+ 8π
9+ 10π
10+ 4π
11(mod π
12), hence
ζ
13+ ζ
13−1≡ A
0(m) + A
1(m)π + . . . + A
5(m)π
5≡ 28 + 144π + 25π
2+ 68π
3+ 43π
4+ 126π
5(mod π
12).
It is easy to see that η
2= N
Q(ζp+ζp−1)/K