Caltech Algebraic Number Theory 2
Caltech Algebraic Number Theory 2
15
I nd the proofs of the Norm index, reciprocity and existence theorem are best
explained in Langs book which we follow very closely. However, he uses idelic
language which we translate back into the language of congruence groups.
1.6. Proof of the Norm index Theorem. Throughout this section L/K is a
cyclic extension with Galois group G =< >. We recall that the Tate cohomology
of a cyclic group G is periodic with period 2, more precisely there is a class c
2 (G, Z) so that cup-product with c gives an isomorphism
H 2 (G, Z) = H
i (G, M )
i+2 (G, M )
(10)
c : H
=H
for any G-module M and i Z. The Herbrand quotient
0 (G, M )|
|H
q(M ) :=
1 (G, M )|
|H
is dened if both groups are nite and can be viewed as some sort of Euler characteristic of the cohomology of M . It behaves like an Euler characteristic in that for
any (nite length) exact sequence
M 1 M2 M3
one has
0 M G M M MG 0
and
NG
1 (G, M ) MG
0 (G, M ) 0
0H
M G H
and
1 (G, M )
1 (G, M )
H
=H
(11)
by (10).
Let m be a modulus only divisible by the primes ramied in L/K (with exponents
to be determined later in the proof of Lemma 1.6) and let S be the set of primes
dividing m together with the innite primes. Using Lemmas 1.5, 1.6, 1.7 below the
proof of the Norm index theorem is as follows. Consider the map of short exact
sequences
0
NL/K (L )K ,m
NL/K (L )
K
NL/K (L )
m
NL/K (JL
)P m
m)
NL/K (JL
Jm
m)
NL/K (JL
K
NL/K (L )K ,m
Jm
m )P m
NL/K (JL
Jm
QL
(12)
By denition of P
mod m}.
pS
16
M. FLACH
)1
= q(OL,S
| ker()|
(13)
Lemma 1.7 gives
(14)
But we already know the opposite inequality by Theorem 1.3 and this gives the
norm index theorem, Theorem 1.5.
To show (13) we would like to argue as follows. Take the exact sequence of
G-modules
0 OL,S
L JLm Cl(OL,S ) 0
(15)
and conclude
1 =q(OL,S
) q(L )1 q(JLm ) q(Cl(OL,S ))1
0 (G, L )|1 |H
0 (G, JLm )|
=q(OL,S
) |H
) | ker()|1 | coker()|
=q(OL,S
using Lemma 1.5 and the niteness of Cl(OL,S ). Unfortunately, this neat argument
K
Jm
m
0 (G, L ) =
0
does not work since H
m ) are not
NL/K (L ) and H (G, JL ) = NL/K (JL
nite and the Herbrand quotient is not dened. But it is straightforward to work
around this diculty which results in a more clumsy argument as follows. The
sequence (15) induces a commutative diagram
1 ((L )) /
H
/H
0 (O )
L,S
/H
0 (L )
0 (O )
H
L,S
/H
0 (L )
1 (Cl(OL,S ))
H
PPP
PPP
PPP
PPP
(
/H
/H
1 (O )
0 ((L ))
L,S
/H
1 (Cl(OL,S ))
0 ((L ))
H
OOO
OOO
OOO
OO'
)
H (OL,S
/H
0 (J m )
L
/H
0 (Cl(OL,S ))
/H
0 (J m )
L
/H
0 (Cl(OL,S ))
where the second and fourth row are exact and the indicated injections and surjections hold because of Lemma 1.5. An easy diagram chase gives an exact sequence
0 (O ) ker() ker() 0
1 ((L )) H
0H
L,S
and a completely dual argument gives an exact sequence
1 ((L )) 0.
0 (Cl(OL,S )) H
0 coker() coker() H
//H
1 ((L
17
Hence
1 ((L ))|
0 (Cl(OL,S ))| |H
| coker()| | coker()| |H
=
1 ((L ))| |H
0 (O )| | ker()|
| ker()|
|H
L,S
=
0 (Cl(OL,S ))| |H
1 ((L ))|
1 (O )| |H
|H
L,S
1 ((L ))| |H
0 (O )| |H
1 (Cl(OL,S ))|
|H
L,S
=q(OL,S
)1
q(Cl(OL,S )) = q(OL,S
)1 .
Proof. The rst part is of course just Hilberts theorem 90 for which we refer to
Langs book on algebra, for example. Concerning the second, one has an isomorphism of G-modules
Jm
Z
IndG Z
=
=
L
Gp
pmf P|p
pmf
H 1 (Gp , Z)
=
pmf
Hom(Gp , Z) = 0.
pmf
Remark 1.6.1. Note that we have surreptitiously commuted cohomology with an
innite direct sum. What follows immediately from the denition is that cohomology
commutes with arbitrary direct products of coecients (exercise)
H i (G,
M )
H i (G, M ).
=
since there is a natural map (from right to left) which is an isomorphism for P = R
and hence for any retract of Rn . If G is a group of type F P , i.e. so that Z
has a resolution by nitely generated, projective R = ZG-modules, then cohomology
commutes with direct sums (in fact arbitrary direct limits). Now a nite group is
certainly of type F P since the standard resolution consists of nitely generated
ZG-modules.
Lemma 1.6. For p S denote by Gp G the decomposition group. For m
suciently large one has
[K : NL/K L K ,m ] =
|Gp |
pS
where K
,m
18
M. FLACH
Proof. The proof is short if one uses local class eld theory. For each p S and
and this nishes the proof of Lemma 1.6. However, since we wont be using local
class eld theory anywhere else in the entire proof, we would like to indicate how
Lemma 1.6 also has a self-contained proof. The rst thing to notice is
0
q(L
P ) = |H (Gp , LP )| = Kp /NLP /Kp (LP )
|Gp |
of nite index. This follows since it contains NLP /Kp (Kp ) = (Kp )
and (Kp )n
has nite index by explicit computation (see also the section on the proof of the
existence theorem). Since the Gp -action preserves valuations we get
q(L
P ) = q(Z) q(OLP ) = |Gp | q(OLP )
since OKp [Gp ] is cohomologically trivial. These computations already show (17)
but there is a subtle point left to prove, namely the existence of a conductor,
a subgroup U (np ) contained in NLP /Kp (L
P ). It turns out that any nite index
subgroup of E , where E is a local eld of characteristic zero, is open and hence
contains a group U (n) for n large enough. It is enough to show this for the groups
(E )n which one can do by an explicit analysis similar to the one we just gave for
L
P . However the statement is wrong if E is a local eld of characteristic p. Then
E /(E )p is a countable product of copies of Z/p and it turns out that not every
nite index subgroup is open. So here one would also have to show that the norm
subgroup of a nite extension is open.
19
Finally, to make the proof independent of local class eld theory, we also have
to show directly that
surjective. One can now lter OLP by its standard Gp -invariant open subgroups
1 + Pn with subquotients l , l, l, . . . and show that the Norm map is surjective for
all quotients OL
/1+Pn , i.e. NLP /Kp (OL
)(1+pn ) = OK
. Since NLP /Kp (OL
)
p
P
P
P
n
is a nite index subgroup it contains some subgroup 1 + p and we are done.
We summarize the facts about local elds that we have obtained in the proof of
Lemma 1.6.
Corollary 1.6. For a cyclic extension LP /Kp of local elds with group Gp we have
q(L
P ) = |Gp |,
q(OL
) = 1.
P
pS
)=
q(OL,S
|Gp |
|G|
Proof. This is a beautiful application of the S-unit theorem and the Herbrand
index. Recall the S-unit regulator map
log | |p
R
: OL,S
pSL
which is G-equivariant. One way to phrase the unit theorem is to say that it induces
an exact sequence of G-modules
R
0 OL,S
Z R
R
R 0.
pSL
0 XS
Z
Z0
pSL
OL,S
Z R
= XS Z R.
20
M. FLACH
Using elementary representation theory (of nite groups over elds of characteristic
zero) there is then an isomorphism of Q[G]-modules
OL,S
Z Q
= XS Z Q.
) = q(OL,S
/(L)) = q(OL,S
/(L) XS ) = q(XS ).
q(OL,S
Z) q(Z)
pSL
= q(
IndG
Gp
Z) |G|
pS
pS
|Gp |
|G|
.
Since the following two statements have one-line proofs given what we did in this
section, we mention them here but they will probably not be used elsewhere.
Corollary 1.7. In the situation of Theorem 1.5, if an element K ,m generates
an ideal that is a norm, then is the norm of an element.
Proof. Inequality (14) is an equality which implies that ker() = 0 where
:
NL/K (L ) K ,m
NL/K (JLm ) P m
.
NL/K (L )
NL/K (JLm )
pS
Kp /NLP /Kp (L
P ).
21
() 1
The next step is to treat the case of a cyclic extension L/K. We cannot show
directly what we want, that ker() contains some group P m , but we will be able
to show the opposite inclusion ker() P m NL/K (JLm ). Together with the Norm
index theorem this implies equality.
Proposition 1.5. Let L/K be cyclic. Then for the modulus m from Theorem 1.5
we have ker() P m NL/K (JLm ).
Proof. Let : J m Gal(L/K) be the Artin map for the modulus from Theorem
1.5 which we know is divisible only by the ramied primes in L/K. Let a ker()
and
r
pi i
a=
i=1
Ei (mi )
HH
HH
HH
HH
H
LEA
AA
zz
z
AA
z
z
AA
z
z
A
z
LEi D
E
DD
uu
}}
u
}
D
u
}
DD
uu
}}
DD
uu
}}
uu
Ei
L II
II
yy
II
y
II
yy
II
I yyy
K
Gal(LE/E)
= Gal(L/K) =< >
= Gal(LEi /Ei )
22
M. FLACH
are cyclic but, crucially LEi /Ei is cyclotomic for a set of pairwise relatively prime
integers mi , i = 1, ..., r. Moreover, pi splits completely in Ei /K and is prime to all
mj . We can write
(pi i , L/K) = di
and have 1 = (a, L/K) =
di
, hence
di = dn
where n = [L : K] = | < > |. Since the Artin map is always surjective there is a
fractional ideal B in E (which we can take to be prime to m and the mi ) such that
(B, LE/E) = .
By (18) we get for b = NE/K B
(b, L/K) = .
Now we can write
pi i bdi = NEi /K Ai
since b is a Norm (even from E, not only Ei ) and since pi is a norm since it is
totally split in Ei /K. Since (pi i bdi , L/K) = 1 we get (Ai , LEi /Ei ) = 1. But
LEi /Ei is cyclotomic and so we can write
Ai = (i )NLEi /Ei Bi
for some i 1 mod m(mi )mi, (note that we can take any multiple of the modulus (mi )mi, from Lemma 1.8). Applying NEi /K we get
pi i bdi = NEi /K (i )NLEi /K Bi = NEi /K (i )NL/K (NLEi /L Bi )
where NEi /K (i ) 1 mod m. Taking the product over i = 1, ..., r yields
abdn P m NL/K (JLm ).
But n = [L : K] and therefore bdn = NL/K (bd ) is also a norm. Hence
a P m NL/K (JLm ).
Lemma 1.9. Let a, r Z and q be a prime number. Then there exists a prime
p such that a has order q r modulo p.
>1
qr (a) =
r1
r1
aq 1
1)q1 + q(aq
1)q2 + + q.
= (aq
aqr1 1
r1
23
distinct primes p1 , . . . , ps , pi
= pj , so that a has order qi i mod pi with ri > ri .
Then a certainly has order divisible by n modulo m. Let b be a positive integer
such that
ba
mod p1 ps ,
b1
mod m.
r
r
Then a 1 mod mod p1 ps and hence q11 qs s divides . This implies
a 1 mod p1 ps and hence a 1 mod m. Therefore b 1 mod m and
hence a and b are independent modulo m.
Here is a translation of this Lemma in terms of cyclotomic elds.
Lemma 1.11. Let L/K be an extension of degree n, p a prime of K and S a
nite set of prime numbers. Then there exists an integer m, relatively prime to the
primes in S and to p so that
(i) L Q(m ) = Q and hence L K(m ) = K.
(ii) The Artin symbol := (p, K(m )/K) has order divisible by n.
(iii) There exists Gal(K(m )/K) of order divisible by n and independent of
.
Proof. Apply Lemma 1.10 with a = N p. Take m only divisible by primes unramied in L/Q so that L Q(m ) = Q. Then the cyclotomic polynomial m (X) is
irreducible over any subeld of L hence over K := L K(m ) and over K, i.e.
[K(m ) : K] = [K (m ) : K ] = [K(m ) : K ] and K = K. So (i) is satised and
one has isomorphisms
Gal(L(m )/L)
= Gal(K(m )/K)
= Gal(Q(m )/Q)
= (Z/mZ) .
a
so that (ii) holds. Choosing b as in Lemma 1.10 and
By (18) we have (m ) = m
b
setting (m ) = m we get (iii).
24
M. FLACH
Lemma 1.11 results in the following diagram of elds where E is the eld we are
going to construct in the next Lemma.
L(m ) = E(m )
OOO
r
OOO
rrr
r
OOO
r
r
r
OO
r
rrr
K(m )
L LL
E
II
LLL
oo
II
o
LLL
o
II
o
o
o
LLL
II
o
I
oo
LL
o
o
o
Q(m )
K OOO
OOO
u
u
u
OOO
u
OOO
uu
OOO
uu
u
u
Q
Lemma 1.12. (Artins Lemma) Let L/K be a cyclic extension, p a prime of K
unramied in L/K and S a nite set of prime numbers. Then there exists an
integer m, relatively prime to the primes in S and to p, and a nite extension E/K
so that
(0) L K(m ) = K.
(1) L E = K.
(2) L(m ) = E(m ).
(3) p splits completely in E/K.
Proof. Choosing m as in Lemma 1.11 we have
G := Gal(L(m )/K)
= Gal(L/K) Gal(K(m )/K)
and condition (0) is satised. Let be a generator of Gal(L/K) and be as in
Lemma 1.11, let H be the subgroup generated by
{(, ), ((p, L/K), (p, K(m )/K))}
and let E be the xed eld of H. Condition (1) is satised since H and
Gal(L(m )/L)
= 1 Gal(K(m )/K)
generate G. Condition (2) is satised since the intersection of H and
Gal(L(m )/K(m ))
= Gal(L/K) 1
is trivial (uses independence and divisibility by n of and ). Finally, condition
(3) is satised since the decomposition group
< ((p, L/K), (p, K(m )/K)) >
of p in L(m )/K is contained in H.
We can now nally produce our diagram (19). We apply Artins Lemma to
p1 , . . . , pr in succession, making sure that m1 , . . . , mr are pairwise relatively prime.
Then the compositum E = E1 Er satises L E = K as well as (20). This is an
easy exercise in Galois theory.
Now on to the proof of the reciprocity theorem, Theorem 1.6. Given an abelian
extension L/K, let Li /K be the (nite) family of cyclic subextensions. If i denotes
25
mi
the Artin map for Li /K we know by Prop. 1.5 that P
ker(i ) where mi is only
divisible by primes ramied in Li /K. Then for m := i mi we have
Pm
ker(i ) = ker()
i
B = OK,S
(K )n
where S is a large enough set of places containing all innite places and
OK,S = { K|p
/ S ||p 1}
is the ring of S-integers. More precisely we have
Proposition 1.6. Assume K contains a primitive n-th root of unity. Let S be a
nite set of places of K such that
p|pS
p|npS
26
M. FLACH
Then
LS,n := K
n
OK,S
) J mS
P mS (KS,n
where
(22)
Moreover
KS,n
:= K
(Kp )n .
pS
[LS,n : K] = n|S| .
(K )n /(K )n | = |OK,S
/OK,S
(K )n |
[LS,n : K] = |OK,S
of LS,n over K. If an S-unit has an n-th root in K then its n-th root is again an
S-unit, i.e.
(K )n = (OK,S
)n
OK,S
and hence
[LS,n : K] = [OK,S
: (OK,S
)n ].
[LS,n : K] = [OK,S
: (OK,S
)n ] = |(K)/(K)n | ns1 = ns .
So for example, if K = Q we can only take n = 2 and then for S = {p1 , . . . , ps1 , }
we get
LS,2 = Q( p1 , . . . , ps1 , 1)
which has degree 2s .
/ S then OK,S
OK,p
In general LS,n is unramied outside S, since if p
n
n
and satises the monic equation with integer coecients f (X) = X and
f () = nn1 OK,p
. By Theorem 1.6 there is a modulus mS only divisible by
primes in S, and which we can increase to satisfy (21), so that LS,n /K is a class
eld and
S
J mS /P mS NLS,n /K (JLmS,n
)
= Gal(LS,n /K)
is a group of exponent n and order ns . To see the inclusion
(24)
S
(KS,n
) P mS NLS,n /K (JLmS,n
) = ker(LS,n )
[J mS : (KS,n
)] = ns .
We have (KS,n
) = KS,n
/OK,S,n
where
:= OK,S
KS,n
= OK,S
OK,S,n
27
(Kp )n
pS
/OK,S,n
0 KS,n
1
K /OK,S,n
0 OK,S
/OK,S,n
1 J mS
1
n
pS Kp /(Kp )
pS
H 0
J mS H 0
Kp /(Kp )n 1
H 0
where i are the natural inclusions, i are their respective pullbacks and we obtain
isomorphisms on H since
coker(i ) = coker(i ). To see this last claim, note that
K surjects onto both pS Kp /(Kp )n (by weak approximation) and J mS (by our
assumption that the primes in S generate Cl(K)) and that the map coker(i )
coker(i ) has trivial kernel since i was a pullback of i .
The bottom row now consists of nite groups and we get
mS
n
[J
: (KS,n )] = |H| =
Kp /(Kp ) / OK,S /OK,S,n
pS
=
1
n2
1
1
= n2s
s = ns
|n|p ns
|n|
n
p
p
pS
|n|p = 1.
) n = ns .
OK,S /OK,S,n
= OK,S /(OK,S
the extension L = K( n ). Then all p S are split completely in L/K. For any
q
/ S we nd K with () = q in J mS . For p S let P be a place of
L dividing p. By weak approximation we can nd L so that, for all p S,
/ 1 mod Pmp and 1 mod qmp for all P
= q | p. Then NL/K ()/ 1
mod pmp , i.e. () P mS NL/K (JLmS ). We conclude J mS = P mS NL/K (JLmS ), i.e.
28
M. FLACH
Proof. For E = C both sides are 0 since ||x|| = x2 for positive real x. For E = R
both sides are of order 1, resp. 2, according to whether n is odd, resp. even.
Now assume E is non-archimedean. Setting q(A) = | coker(n)|
| ker(n)| for an abelian
group A for which both quantities are nite, q is multiplicative in exact sequences
and q(A) = 1 if A is nite. The exact sequence
E Z 0
0 OE
gives
Q0
0 1 + meE OK
gives
q(OE
) = q(1 + meE ).
[OK : OK ]1 if OE
).
Corollary 1.10. If n K and Q J m is a congruence group such that J m /Q
has exponent n, then there exists a class eld for Q.
Proof. First note that if m is a multiple of m and we set H = J m /Q we have a
commutative diagram analogous to (9)
0 Q J m
0
J m H 0
J m H 0
where now we must use the fact that Q is a congruence subgroup to get surjectivity
of . Indeed, given
Z,
a
p|m , pm
there is K with /p 1 mod pmp and hence n /pn 1 mod pmp . Then
/ n 1 mod mS , i.e. (/ n ) P mS Q. Since J mS /Q has exponent n we have
29
x OE
mod P
for a prime P above p (the choice doesnt matter since Gal(E/L) is abelian). This
implies for any lift
= Gal(E/K) of
x OE
xN p
1 (x)
mod
(P)
and
(P) is a prime above (p). So we get the characterizing property of Frob(p)
1
and this implies that m is G-equivariant. If QE = NL/K
(Q) is a pullback then
1
m
m
QE contains NL/K (P ) which contains PL by the above remark. Hence QE
is a congruence subgroup. Moreover, the G-action on JLm /QE is trivial, since
1
(Q). This proves c).
NL/K (a) = NL/K ((a)) for all G, i.e. (a)/a NL/K
If moreover L/K is cyclic, then Gal(E/K) must be abelian since the lift
of
any generator commutes with all Gal(E/L) and
and Gal(E/L) generate
Gal(E/K). By denition of QL := NL/K (JLm ) P m the composite homomorphism
NL/K
: JLm QL QL /P m
is surjective, hence for any subgroup P m Q QL J m and QE := 1 (Q/P m )
(25)
J m /QE = J m / 1 (Q/P m )
= (QL /P m )/(Q/P m )
= QL /Q.
L
30
M. FLACH
has a class eld E/L. By d) E/K is the class eld for Q QL J m and by Lemma
1.4 there is a class eld for Q J m since Q QL Q.
We can now conclude the proof of Theorem 1.7. Given any congruence subgroup
P m Q J m , let n be its exponent. There is a tower of cyclic extensions
K = K0 K1 K2 Kr = K(n )
giving us a sequence of congruence subgroups
Q0 = Q,
1
1
Qi = N K
(Qi1 ) = NK
(Qi1 QKi ).
i /Ki1
i /Ki1
m
By an easy upward induction and (25) all groups JK
/Qi have exponent n. By
i
Corollary 1.10 Qr has a class eld and by Lemma 1.15 e) and an easy downward
induction all Qi have class elds. In particular Q = Q0 has a class eld.