LXXXII.1 (1997)
A note on Sinnott’s index formula
by
Kazuhiro Dohmae (Tokyo)
0. Introduction. Let k be an imaginary abelian number field with exactly two ramified primes. The letters E and C denote the group of units and the group of circular units in k respectively. Sinnott’s index formula for this case is the following (see Proposition 4.1, Theorem 4.1 and Theorem 5.1 in [6]).
Theorem A (Sinnott). Let k be an imaginary abelian number field with conductor m = p
e11p
e22, where p
1and p
2are distinct prime numbers and both e
1and e
2are positive integers. Denote by k
i(i = 1, 2) the maximal subfield of k which is unramified outside p
i∞. Let G be the Galois group of k over Q. Further , T
piand D
pidenote the inertia group and the decompo- sition group of p
iin G (i = 1, 2). Then the group C has finite index in E, and
(1) [E : C] = [k
1: Q][k
2: Q]
[k : Q] · 2
−g0· 2
ε1[G:Dp1]+ε2[G:Dp2]+δ1+δ2−1· Qh
+, where Q is the unit index of k, h
+the class number of the maximal real subfield of k and g
0some rational integer. Moreover , ε
iand δ
iare defined by
ε
i= 0 if k
3−iis imaginary, 1 otherwise,
δ
i= 0 if k
3−iis real and [D
pi: T
pi] is odd , 1 otherwise,
for i = 1, 2. Finally, the rational integer g
0satisfies µ ≤ g
0≤ ν, where µ = ]{1 ≤ i ≤ 2 : k
iis imaginary},
ν = ]{1 ≤ i ≤ 2 : [k
i: Q] is even}.
In general, the formula (1) contains the unknown factor 2
−g0. But if both k
1and k
2are imaginary, then we have µ = ν = 2 and g
0= 2. Hence,
1991 Mathematics Subject Classification: Primary 11R27.
[57]
in this case, (1) reads
(2) [E : C] = [k
1: Q][k
2: Q]
[k : Q] · 2
−1· Qh
+.
In a previous paper [1], we gave another proof of (2) by constructing a system of fundamental circular units (i.e., a basis of the free part of C) of k.
It is the main purpose of this note to prove the following completely explicit version of Theorem A.
Theorem B. Let the notation be as in Theorem A. Then (3) [E : C] = [k
1: Q][k
2: Q]
[k : Q] · 2
ε1[G:Dp1]+ε2[G:Dp2]−1· Qh
+.
The proof includes the explicit construction of a system of fundamental circular units of k. By the way, comparing (1) with (3), we obtain
(4) g
0= δ
1+ δ
2.
Kuˇ cera kindly wrote me a direct proof of the inequality g
0≥ δ
1+ δ
2. But I have never found any direct proof of (4) .
In the last section, we also mention our result on a real abelian number field with exactly two ramified primes.
1. Notation. Let k be an imaginary abelian number field of conductor m = p
e11p
e22. We note that k is a subfield of the mth cyclotomic field K = Q(ζ
m), where ζ
m= e
2π√−1/m
. Let N be the subgroup of T = (Z/mZ)
×which corresponds to Gal(K/k) under the natural isomorphism
(Z/mZ)
×3 t mod m 7→ (t, K) ∈ Gal(K/Q),
where the automorphism (t, K) maps ζ
mto ζ
mt. Throughout this paper, we use the following symbols:
• q
1= p
e11, q
2= p
e22;
• ζ = ζ
m;
• K
1= Q(ζ
q1), K
2= Q(ζ
q2);
• k
1= k ∩ K
1, k
2= k ∩ K
2;
• W is the group of roots of unity in k;
• D(q
1) = h−1, N
K1/k1(1 − ζ
aq2) : 1 ≤ a < q
1∧ (a, p
1) = 1i;
• D(q
2) = h−1, N
K2/k2(1 − ζ
q1b) : 1 ≤ b < q
2∧ (b, p
2) = 1i;
• D(m) = h−1, N
K/k(1 − ζ
x) : 1 ≤ x < m ∧ (x, p
1) = (x, p
2) = 1i;
• D = D(q
1)D(q
2)D(m);
• C = D ∩ E is the group of circular units in k;
• T
1= {a (= a mod m) ∈ T : a ≡ 1 mod q
2};
• T
2= {b (= b mod m) ∈ T : b ≡ 1 mod q
1};
• T
10= {a ∈ T
1: there exists b ∈ T
2such that ab ∈ N };
• T
20= {b ∈ T
2: there exists a ∈ T
1such that ab ∈ N };
• J = −1 mod m ∈ T ;
• J
1≡ −1 mod q
1∧ J
1≡ 1 mod q
2;
• J
2≡ 1 mod q
1∧ J
2≡ −1 mod q
2.
Lemma 1. With the above notation we have:
(1) Gal(K
1/k
1) = T
10, (2) Gal(K
2/k
2) = T
20, (3) Gal(k/k
1) = T
10T
2/N , (4) Gal(k/k
2) = T
1T
20/N . P r o o f. Easy.
R e m a r k. By the statements (1) and (2) of Lemma 1, we can see that k
iis imaginary if and only if J
i6∈ T
i0(i = 1, 2).
In order to prove the theorem, we have to consider the following four cases separately:
I. Both k
1and k
2are imaginary (⇔ J
16∈ T
10∧ J
26∈ T
20);
II. k
1is imaginary and k
2is real (⇔ J
16∈ T
10∧ J
2∈ T
20);
II
0. k
1is real and k
2is imaginary (⇔ J
1∈ T
10∧ J
26∈ T
20);
III. Both k
1and k
2are real (⇔ J
1∈ T
10∧ J
2∈ T
20).
But case I was treated in [1], and cases II and II
0are similar. So it is sufficient to consider cases II and III. In case II, we use the following symbols:
• a
1(= 1), a
2, . . . , a
r1a system of representatives for T
1/hJ
1iT
10;
• b
1(= 1), b
2, . . . , b
r2a system of representatives for T
2/T
20;
• d
1(= 1), d
2, . . . , d
sa system of representatives for T
10T
20/N ;
• Y = {a
i1b
i2d
j: 1 ≤ i
1≤ r
1∧ 1 ≤ i
2≤ r
2∧ 1 ≤ j ≤ s};
• Y
0= Y − {1, a
2, . . . , a
r1, b
2, . . . , b
r2};
• M = {a
i1q
2: 2 ≤ i
1≤ r
1} ∪ {q
1b
i2: 2 ≤ i
2≤ r
2} ∪ Y
0;
• M
0= M ∪ {0}.
And, in case III, we use the following ones:
• a
1(= 1), a
2, . . . , a
r1a system of representatives for T
1/T
10;
• b
1(= 1), b
2, . . . , b
r2a system of representatives for T
2/T
20;
• d
1(= 1), d
2, . . . , d
sa system of representatives for T
10T
20/hJ iN ;
• Y = {a
i1b
i2d
j: 1 ≤ i
1≤ r
1∧ 1 ≤ i
2≤ r
2∧ 1 ≤ j ≤ s};
• Y
0= Y − {1, a
2, . . . , a
r1, b
2, . . . , b
r2};
• M = {a
i1q
2: 2 ≤ i
1≤ r
1} ∪ {q
1b
i2: 2 ≤ i
2≤ r
2} ∪ Y
0;
• M
0= M ∪ {0}.
R e m a r k. We note that Y is a system of representatives for T /hJ iN .
Further, the cardinality of M is equal to
12[k : Q]−1, which is the unit rank
of k.
Now let l : k
×→ R[G] be the G-module homomorphism defined by l : k
×3 α 7→ X
σ∈G
log |α
σ| · σ
−1∈ R[G].
It is easy to see that Ker l ∩ E = Ker l ∩ C = W . Hence [E : C] = [l(E) : l(C)].
Furthermore, any finite subset {v
1, . . . , v
r} of C is a system of fundamental circular units of k if and only if {l(v
1), . . . , l(v
r)} is a Z-basis of l(C).
For any a, b, y ∈ T , we define circular units v
aq2, v
q1b, v
yby v
aq2= N
K1/k11 − ζ
aq21 − ζ
q2, v
q1b= N
K2/k21 − ζ
q1b1 − ζ
q1, v
y= N
K/k(1 − ζ
y).
Then we notice the following facts:
(1) If a ≡ a
0mod T
2N , then v
aq2= v
a0q2. (2) If b ≡ b
0mod T
1N , then v
q1b= v
q1b0. (3) If y ≡ y
0mod N , then v
y= v
y0.
Let C
0be the subgroup of C generated by {v
x: x ∈ M }. Later, we shall see that l(C
0) has finite index in l(C) and l(E). Hence we have
[E : C] = [l(E) : l(C)] = [l(E) : l(C
0)]
[l(C) : l(C
0)] .
2. Computation of [l(C) : l(C
0)]. First, from the definition of C
0, we can easily see the following lemma.
Lemma 2. l(C
0) is generated by {l(v
x) : x ∈ M }.
We choose two integers l
1and l
2such that
l
1≡ 1 mod q
1, p
1l
1≡ 1 mod q
2, p
2l
2≡ 1 mod q
1, l
2≡ 1 mod q
2. Then l
1∈ T
2and l
2∈ T
1. We define f
1, g
1, f
2and g
2by
f
1= [hl
1iT
20: T
20], g
1= [T
2: hl
1iT
20], f
2= [hl
2iT
10: T
10], g
2= [T
1: hl
2iT
10].
Let {1, s
2, . . . , s
g1} and {1, t
2, . . . , t
g2} be systems of representatives for T
2/hl
1iT
20and T
1/hl
2iT
10respectively.
Proposition 3. In case II, l(C) is generated by l(v
ai1q2) (2 ≤ i
1≤ r
1), l(v
y) (y ∈ Y
0) and
1
2
l(v
q1l1), . . . ,
12l(v
q1lf1−11
),
l(v
q1s),
12(l(v
q1s) − l(v
q1sl1)), . . . ,
12(l(v
q1sl1f1−2
) − l(v
q1slf1−11
))
(s = s
2, . . . , s
g1).
Moreover , [l(C) : l(C
0)] = 2
r2−g1.
Proposition 4. In case III, l(C) is generated by l(v
y) (y ∈ Y
0)
and
1
2
l(v
q1l1), . . . ,
12l(v
q1lf1−11
), l(v
q1s),
12(l(v
q1s) − l(v
q1sl1)), . . . ,
12(l(v
q1sl1f1−2
) − l(v
q1slf1−11
))
(s = s
2, . . . , s
g1) and
1
2
l(v
l2q2), . . . ,
12l(v
lf2−1 2 q2), l(v
tq2),
12(l(v
tq2) − l(v
tl2q2)), . . . ,
12(l(v
tlf2−22 q2
) − l(v
tlf2−1 2 q2))
(t = t
2, . . . , t
g2).
Moreover , [l(C) : l(C
0)] = 2
r1+r2−g1−g2.
R e m a r k. In the next section, we shall see that l(C
0) has finite index in l(E). Hence
rank
Zl(C
0) = rank
Zl(C) = rank
Zl(E).
On the other hand, we notice that the cardinality of every system of gener- ators for l(C
0) or l(C) stated in the above propositions is equal to the unit rank of k. This implies that these systems of generators are bases.
We can prove Propositions 3 and 4 in a similar fashion. So we only prove Proposition 3. Let L be the subgroup of R[G] generated by the elements stated in the proposition.
First we prove L ⊂ l(C). Fix a b
i2(1 ≤ i
2≤ r
2). Then X
1≤i1≤r1
1≤j≤s
l(v
ai1bi2dj)
= X
1≤i1≤r1
1≤j≤s
l(N
K/k(1 − ζ
ai1bi2dj))
= 1 2
X
1≤i1≤r1
1≤j≤s
l(N
K/k(1 − ζ
ai1bi2dj)) + 1 2
X
1≤i1≤r1
1≤j≤s
l(N
K/k(1 − ζ
J1ai1bi2dj))
= 1 2
X
0≤h1≤1 1≤i1≤r1
1≤j≤s
l(N
K/k(1 − ζ
J1h1ai1bi2dj))
= 1 2 l
Y
0≤h1≤1 1≤i1≤r1
1≤j≤s
N
K/k(1 − ζ
J1h1ai1bi2dj)
.
Now, by Lemma 1(4), we have Y
0≤h1≤1 1≤i1≤r1
1≤j≤s
N
K/k(1 − ζ
J1h1ai1bi2dj) = N
k/k2(N
K/k(1 − ζ
bi2)).
So we get X
1≤i1≤r1
1≤j≤s
l(v
ai1bi2dj)
= 1
2 l(N
k/k2(N
K/k(1 − ζ
bi2)))
= 1
2 l(N
K2/k2(N
K/K2(1 − ζ
bi2)))
= 1 2 l
N
K2/k21 − ζ
q1bi21 − ζ
q1l1bi2= 1 2 l
N
K2/k21 − ζ
q1bi21 − ζ
q1− 1 2 l
N
K2/k21 − ζ
q1l1bi21 − ζ
q1= 1
2 (l(v
q1bi2) − l(v
q1l1bi2)).
Hence
1
2 (l(v
q1bi2) − l(v
q1l1bi2)) ∈ l(C).
From this, we can easily deduce that L ⊂ l(C).
Next we prove l(C) ⊂ L. For this purpose, it is sufficient to prove that l(v
y) ∈ L for y ∈ Y − Y
0, because it is obvious that l(v
q1bi2) ∈ L for 2 ≤ i
2≤ r
2. Fix an a
i1(2 ≤ i
1≤ r
1). Then
X
1≤i2≤r2
1≤j≤s
l(v
ai1bi2dj) = l(N
k/k1(N
K/k(1 − ζ
ai1)))
= l(N
K1/k1(N
K/K1(1 − ζ
ai1)))
= l
N
K1/k11 − ζ
ai1q21 − ζ
ai1q2l2= l
N
K1/k11 − ζ
ai1q21 − ζ
q2− l
N
K1/k11 − ζ
ai1q2l21 − ζ
q2= l(v
ai1q2) − l(v
ai1q2l2) ∈ L.
If a
i1b
i2d
j6= a
i1, then a
i1b
i2d
j∈ Y
0by the definition of Y
0. Hence l(v
ai1) ∈ L. Now we fix a b
i2(1 ≤ i
2≤ r
2). Then, as we have seen above,
X
1≤i1≤r1
1≤j≤s
l(v
ai1bi2dj) = 1
2 (l(v
q1bi2) − l(v
q1l1bi2)).
If a
i1b
i2d
j6= b
i2, then l(v
ai1bi2dj) ∈ L. Hence l(v
bi2) ∈ L. We have thus proved that l(C) ⊂ L.
Finally, by computing the determinant of the transition matrix, we can see that
[l(C) : l(C
0)] = 1
1 2
f1−1·
12f1−1g1−1= 2
f1g1−g1= 2
r2−g1. This completes the proof of Proposition 3.
3. Computation of [l(E) : l(C
0)]. For each t ∈ T , we let c
0= 1,
c
tq2= 1 if t ∈ T
2N, 0 otherwise, c
tq1= 1 if t ∈ T
1N,
0 otherwise, c
t= n 1 if t ∈ N ,
0 otherwise.
Further, we define b
0,t= 2,
b
aq2,t= (c
(aq2)t+ c
J (aq2)t) − (c
tq2+ c
J tq2) (a ∈ T ), b
q1b,t= (c
(q1b)t+ c
J (q1b)t) − (c
tq1+ c
J tq1) (b ∈ T ),
b
y,t= (c
yt+ c
J yt) − 1
[k : k
2] (c
ytl1q1+ c
J ytl1q1)
− 1
[k : k
1] (c
ytl2q2+ c
J ytl2q2) + 2
[k : Q] (y ∈ T ).
We denote by σ
t(t ∈ T ) the automorphism of k over Q which is the image of (t, K) under the canonical surjection
Gal(K/Q) → G = Gal(k/Q).
We omit the proof of the next lemma because it is just the same as those of Lemmas 3.5 and 3.6 in [1].
Lemma 5. Let Q be the unit index of k and h
+the class number of k
+. Then
[l(E) : l(C
0)] = 1
[k : Q] · |det(b
x,t)
x∈M0,t∈Y| · Qh
+.
In what follows, we compute the determinant of the matrix (b
x,t)
x∈M0,t∈Y. We define
A = {J
1h1a
i1q
2: 0 ≤ h
1≤ 1 ∧ 1 ≤ i
1≤ r
1}, B = {q
1b
i2: 1 ≤ i
2≤ r
2} in case II, and
A = {a
i1q
2: 1 ≤ i
1≤ r
1}, B = {q
1b
i2: 1 ≤ i
2≤ r
2} in case III. Then it is easy to see that
X
y∈A
c
yt= X
y∈B
c
yt= 1 for any t ∈ T .
Lemma 6. (1) For any x ∈ A, there exist rational numbers α
i1(1 ≤ i
1≤ r
1) such that
c
xt+ c
J xt= α
1(c
0+ c
0) +
r1
X
i1=2
α
i1(c
(ai1q2)t+ c
J (ai1q2)t) for any t ∈ T .
(2) For any x ∈ B, there exist rational numbers β
i2(1 ≤ i
2≤ r
2) such that
c
xt+ c
J xt= β
1(c
0+ c
0) +
r2
X
i2=2
β
i2(c
(q1bi2)t+ c
J (q1bi2)t) for any t ∈ T .
P r o o f. As cases II and III are similar, we prove the lemma for case II.
If x ∈ M , there is nothing to prove. So we suppose that x 6∈ M . (1) Since
c
(J1ai1q2)t
+ c
J (J1ai1q2)t
= c
(J1ai1q2)t
+ c
(ai1q2)t
= c
(ai1q2)t
+ c
J (ai1q2)t
, it suffices to consider the case x = q
2. Let us define
A
0= {a
i1q
2: 1 ≤ i
1≤ r
1}, A
1= {J
1a
i1q
2: 1 ≤ i
1≤ r
1}.
Then
X
y∈A0
(c
yt+ c
J yt) = X
y∈A
(c
yt+ c
J yt) − X
y∈A1
(c
yt+ c
J yt).
On the other hand, we can see that X
y∈A1
(c
yt+ c
J yt) = X
y∈A1
(c
(J y)t+ c
J (J y)t) = X
y∈A0
(c
yt+ c
J yt).
Hence X
y∈A0
(c
yt+ c
J yt) = 1 2
X
y∈A
(c
yt+ c
J yt) = 1
2 (1 + 1) = 1
2 (c
0+ c
0).
Consequently, we obtain c
q2t+ c
J q2t= 1
2 (c
0+ c
0) −
r1
X
i1=2
(c
(ai1q2)t+ c
J (ai1q2)t).
(2) From x 6∈ M , we have x = q
1. Since
r2
X
i2=1
(c
(q1bi2)t+ c
J (q1bi2)t) = 1 + 1 = c
0+ c
0, we get
c
q1t+ c
J q1t= (c
0+ c
0) −
r2
X
i2=2
(c
(q1bi2)t+ c
J (q1bi2)t).
Using the above lemma, we can deduce that
|det(b
x,t)
x∈M0,t∈Y| = r
1r
2· |det(c
xt+ c
J xt)
x∈M0,t∈Y|
by the same argument as in the proof of Lemma 3.8 in [1]. So, in order to accomplish our purpose, we have to compute the determinant of the matrix (c
xt+ c
J xt)
x∈M0,t∈Y.
For each t ∈ Y , there exists exactly one u ∈ Y such that tu ∈ hJ iN , because Y is a system of representatives for T /hJ iN . We denote this u by t
0. Then the map Y 3 t 7→ t
0∈ Y is bijective.
Lemma 7. Suppose x ∈ M , t ∈ Y and c
xt0+ c
J xt06= 0. Then:
(1) if x ∈ Y
0, then t = x;
(2) if x = a
i1q
2(2 ≤ i
1≤ r
1), then t ∈ Y
0or t = a
i1; (3) if x = q
1b
i2(2 ≤ i
2≤ r
2), then t ∈ Y
0or t = b
i2. P r o o f. (1) Straightforward.
(2) If c
(ai1q2)t0
+ c
J (ai1q2)t0
6= 0, then a
i1t
0∈ hJ iT
2N . Hence there is a y = b
i2J
hd
jsuch that a
i1yt
0= a
i1b
i2J
hd
jt
0∈ hJ iN , and so we have a
i1b
i2d
jt
0∈ hJ iN . Therefore t = a
i1b
i2d
j, and we obtain t = a
i1or t ∈ Y
0.
(3) In case III, the proof is similar to (2). So we consider case II. If c
(q1bi2)t0+c
J (q1bi2)t06= 0, then b
i2t
0∈ hJ iT
1N . Hence there is a w = J
1h1a
i1d
jsuch that wb
i2t
0= J
1h1a
i1b
i2d
jt
0∈ hJ iN , and so a
i1b
i2J
2h1d
jt
0∈ hJ iN . As
J
2∈ T
20, there exists a d
j0such that J
2h1d
j≡ d
j0mod hJ iN , and then a
i1b
i2d
j0t
0∈ hJ iN . Therefore t = b
i2or t ∈ Y
0.
From this, we can obtain the following conclusion.
Proposition 8. Let Q be the unit index of k and h
+the class number of k
+.
(1) In case II, we have
[l(E) : l(C
0)] = [k
1: Q][k
2: Q]
[k : Q] · 2
r2−1· Qh
+. (2) In case III, we have
[l(E) : l(C
0)] = [k
1: Q][k
2: Q]
[k : Q] · 2
r1+r2−1· Qh
+. In particular , l(C
0) and l(C) have finite indices in l(E).
4. Proof of Theorem B. First we consider case II. By Propositions 3 and 8(1), we have
[E : C] = [l(E) : l(C
0)]
[l(C) : l(C
0)] = [k
1: Q][k
2: Q]
[k : Q] · 2
r2−12
r2−g1· Qh
+= [k
1: Q][k
2: Q]
[k : Q] · 2
g1−1· Qh
+.
Now, since σ
l1is the inverse of the Frobenius automorphism for p
1in k, we have
g
1= [T : hl
1iT
1N ] = [G : D
p1].
Therefore
[E : C] = [k
1: Q][k
2: Q]
[k : Q] · 2
[G:Dp1]−1· Qh
+.
Next we consider case III. By Propositions 4 and 8(2), we have [E : C] = [k
1: Q][k
2: Q]
[k : Q] · 2
g1+g2−1· Qh
+. As also g
1= [G : D
p1] and g
2= [G : D
p2], we obtain
[E : C] = [k
1: Q][k
2: Q]
[k : Q] · 2
[G:Dp1]+[G:Dp2]−1· Qh
+. This completes the proof of the theorem.
5. The real case. Let k be a real abelian number field with con-
ductor p
e11p
e22. Let T
1, T
2, T
10, T
20, N be the same as in Section 1, and
{a
i1}, {b
i2}, {d
j} be systems of representatives for T
1/T
10, T
2/T
20, T
10T
20/N re- spectively. Then
{v
ai1q2
: 2 ≤ i
1≤ r
1} ∪ {v
q1bi2