COHOMOLOGY GROUPS OF RADICAL EXTENSIONS
Eunmi Choi
Reprinted from the
Journal of the Korean Mathematical Society Vol. 44, No. 1, January 2007
c
°2007 The Korean Mathematical Society
J. Korean Math. Soc. 44 (2007), No. 1, pp. 151–167
COHOMOLOGY GROUPS OF RADICAL EXTENSIONS
Eunmi Choi
Abstract. If k is a subfield of Q(ε
m) then the cohomology group H
2(k (ε
n)/k) is isomorphic to H
2(k(ε
n0)/k) with gcd(m, n
0) = 1. This enables us to reduce a cyclotomic k-algebra over k(ε
n) to the one over k(ε
n0).
A radical extension in projective Schur algebra theory is regarded as an analog of cyclotomic extension in Schur algebra theory. We will study a reduction of cohomology group of radical extension and show that a Galois cohomology group of a radical extension is isomorphic to that of a certain subextension of radical extension. We then draw a cohomological characterization of radical group.
1. Introduction
Let k be a field, k ∗ be the multiplicative subgroup of k and µ(k) be the group of roots of unity in k. For a Galois extension L of k with Galois group G = G(L/k) and for a 2-cocycle α ∈ Z 2 (G, L ∗ ) = Z 2 (L/k, L ∗ ), a crossed product algebra (L/k, α) = P
σ∈G Lu σ with u σ u τ = α(σ, τ )u στ and u σ x = σ(x)u σ
(x ∈ L, σ, τ ∈ G) is called a cyclotomic algebra if L is a cyclotomic extension of k and α has values in µ(L) (i.e., α ∈ Z 2 (L/k, µ(L))). Let H ι 2 (L/k) be the image of a canonical homomorphism ι of H 2 (L/k, µ(L)) into H 2 (L/k, L ∗ ) induced by the inclusion µ(L) ,→ L ∗ . Since µ(L) is a subgroup of the torsion group of L ∗ , ι is injective ([7, p.91]), so we may identify H 2 (L/k, µ(L)) = H ι 2 (L/k) ≤ H 2 (L/k, L ∗ ).
Suppose k is a subfield of the cyclotomic extension Q(ε m ) (Q: the rational number field, ε m : a primitive m-th root of unity). Let L = k(ε n ). Due to [6], m and n are assumed to be either odd or divisible by 4. Then the Galois cohomology group H ι 2 (L/k) is isomorphic to H ι 2 (K/k) where K = k(ε n0) is a subextension of L such that n 0 is a certain divisor of n which is prime to m [13, (7.12)]. Employing this result, Janusz’s reduction theorem on cyclotomic algebras in [6] ([13, (7.9)]) follows that, a cyclotomic algebra (L/k, α) with α ∈ Z 2 (L/k, µ(L)) can be reduced to the case gcd(m, n) = 1, i.e., to (K/k, β) where β is a 2-cocycle in Z 2 (K/k, µ(K)) defined over the smaller group G(K/k).
Received August 9, 2005.
2000 Mathematics Subject Classification. Primary 16K50, 20G10.
Key words and phrases. cohomology group, radical extension.
This study was supported by Hannam University Research Fund 2005.
c
°2007 The Korean Mathematical Society
151
It is well known that every Schur k-algebra (a central simple k-algebra which is a homomorphic image of a group algebra kG for a finite group G) is similar to a cyclotomic k-algebra [13, (3.10)]. The idea of Schur algebra has been generalized to a projective Schur algebra in [9] by replacing group algebra by twisted group algebra; a projective Schur k-algebra is a central simple algebra that is a homomorphic image of a twisted group algebra k α G for a finite group G and α ∈ Z 2 (G, k ∗ ).
The analogue of cyclotomic algebra in the theory of projective Schur algebra is the radical algebra ([1]). A radical k-algebra is a crossed product algebra (L/k, α) where L = k(Ω) is a finite Galois radical extension of k, Ω is a subgroup of L ∗ which is finite modulo k ∗ (i.e., Ω/k ∗ finite), and α ∈ Z 2 (L/k, L ∗ ) is represented by a 2-cocycle with values in Ω.
In this paper we study radical extensions and radical algebras, and obtain a corresponding result to Janusz’s reduction theorem on radical extensions. We derive a reduction of Galois cohomology groups over radical extension fields, indeed prove that for a radical extension L of k, there exists a Galois radi- cal extension K of k in L such that the cohomology group of G(L/k) is iso- morphic to that of G(K/k) (in Theorem 10). We then verify a cohomologi- cal characterization of radical groups that a homomorphism of radical groups R(K/k) → R(L/k) commutes with certain homomorphisms of cohomology groups (in Theorem 14).
All notations are standard. H 2 (L/k, M ) is the 2-dimensional cohomology group H 2 (G, M ) where G = G(L/k) and M is a G-module, while Z 2 (L/k, M ) is the 2-cocycle group. If M = L ∗ , we write H 2 (L/k, L ∗ ) = H 2 (L/k). Let ε d
(d > 0) denote a primitive d-th root of unity, a|b denote the division of b by a, while a t ||b denote the highest power t of a to be a t |b.
2. Preliminaries
Lemma 1. ([13, 7.10]) Let H be a cyclic normal subgroup of G and M be a fi- nite G-module. Let N H = Q
h∈H h. If N H (M ) = M H then inf : H 2 (G/H, M H )
→ H 2 (G, M ) is an isomorphism, where inf is the inflation map from G/H to G and M H is the subset of M consisting of elements fixed by H.
For finite Galois extensions K and L of k with K < L, the norm N L/K : L → K, x 7→ ( Q
σ∈G(L/K) σ)(x) is a homomorphism for x ∈ L. If H = G(L/K) is normal in G = G(L/k), then N L/K corresponds to N H in Lemma 1. In particular it is clear that N Q(ε
pi+1)/Q(ε
pi) hε p
i+1i = hε pii = hε pi+1i H for a prime p and i > 0, thus the following theorem is due to Lemma 1.
i H for a prime p and i > 0, thus the following theorem is due to Lemma 1.
Theorem 2. ([13, (7.12)]) Let k ≤ Q(ε m ) and L = Q(ε m , ε n ). Let n 0 = 4 δ p 1 · · · p s where p i are distinct odd prime divisors of n not dividing m, and δ = 1 if 4|n, 4 6 |m; δ = 0 otherwise. Let K = Q(ε m , ε n0). Then H ι 2 (K/k) ∼ = H ι 2 (L/k).
As a consequence of Theorem 2, Janusz proved the next theorem on algebras.
Theorem 3. ([6], [13, (7.9)]) If k ≤ Q(ε m ) then any cyclotomic algebra over k is similar to the cyclotomic algebra (Q(ε m , ε t )/k, α) with t = 4 δ p 1 · · · p s ; δ = 0 if 4|m and δ = 1 otherwise, where all p i are distinct odd primes not dividing m.
Theorem 2 and 3 can be generalized to any cyclotomic extension field L containing finitely many roots of unity in [4].
Theorem 4. Let k ≤ Q(ε m ) and L = Q(ε m , ε n1, . . . , ε nw). Let
). Let
n 0 1 = 4 δ1p 1 · · · p s with distinct odd primes p j |n 1 , p j - m (1 ≤ j ≤ s) and δ 1 = 1 if 4|n 1 , 4 - m; δ 1 = 0 otherwise. And for 1 < i ≤ w, let
n 0 i = 4 δip i1 · · · p iui
with p ij |n i , p ij - m(1 ≤ j ≤ u i ), and p ij - n v (1 ≤ v < i) where p ij are distinct odd primes, and δ i = 1 if 4|n i , 4 - m, 4 - n v (1 ≤ v < i); δ i = 0 otherwise.
Then H ι 2 (L/k) ∼ = H ι 2 (K/k) where K = Q(ε m , ε n01, . . . , ε n0w). Furthermore, a crossed product algebra (L/k, α) with α ∈ Z 2 (L/k, µ(L)) is similar to (K/k, β) where β is a 2-cocycle having values in µ(K).
). Furthermore, a crossed product algebra (L/k, α) with α ∈ Z 2 (L/k, µ(L)) is similar to (K/k, β) where β is a 2-cocycle having values in µ(K).
Proof. We will prove this when L = Q(ε m , ε n1, ε n2). Write n 0 1 = 4 δ1p 1 · · · p s
). Write n 0 1 = 4 δ1p 1 · · · p s
and n 0 2 = 4 δ2q 1 · · · q u where p i [resp. q j ] are distinct odd prime divisors of n 1
[resp. n 2 ] with p i - m, q j - m and q j - n 1 for 1 ≤ i ≤ s, 1 ≤ j ≤ u. And δ 1 = 1 if 4|n 1 , 4 - m; δ 2 = 1 if 4|n 2 , 4 - m, 4 - n 1 ; and δ 1 = δ 2 = 0 otherwise.
Let K = Q(ε m , ε n01, ε n02) and E = Q(ε m , ε n01, ε n2). Then k < K = Q(ε mn01, ε n02) ≤ E = Q(ε mn01, ε n2)
) and E = Q(ε m , ε n01, ε n2). Then k < K = Q(ε mn01, ε n02) ≤ E = Q(ε mn01, ε n2)
). Then k < K = Q(ε mn01, ε n02) ≤ E = Q(ε mn01, ε n2)
) ≤ E = Q(ε mn01, ε n2)
)
for, gcd(m, n 0 i ) = 1(i = 1, 2). Since each q j in the factorization of n 0 2 satisfies q j - mn 0 1 ; δ 2 = 1 if 4|n 2 and 4 - mn 0 1 ; and δ 2 = 0 otherwise, we are able to use Theorem 2 on K and E to get H ι 2 (K/k) ∼ = H ι 2 (E/k).
Moreover with l = lcm(m, n 2 ), we have
E = Q(ε m , ε n2)(ε n01) = Q(ε l , ε n01) ≤ L = Q(ε m , ε n2)(ε n1) = Q(ε l , ε n1) and each p i in the factorization of n 0 1 satisfies
) = Q(ε l , ε n01) ≤ L = Q(ε m , ε n2)(ε n1) = Q(ε l , ε n1) and each p i in the factorization of n 0 1 satisfies
)(ε n1) = Q(ε l , ε n1) and each p i in the factorization of n 0 1 satisfies
) and each p i in the factorization of n 0 1 satisfies
p i - l; δ 1 = 1 if 4|n 1 and 4 - l; and δ 1 = 0 otherwise.
Thus by applying Theorem 2 to L and E, we have H ι 2 (E/k) ∼ = H ι 2 (L/k).
Now the isomorphism (L/k, α) ∼ = (K/k, β) of crossed product algebra with
α ∈ Z 2 (L/k, µ(L)), β ∈ Z 2 (K/k, µ(K)) follows immediately by Theorem 3. ¤
3. Norm on radical extension fields
A field L is a (finite) radical extension of k if there is a subgroup Ω of L ∗ such that L = k(Ω) and Ω/k ∗ is a (finite) torsion group. We may exhibit L by k(
n1√
a 1 , . . . ,
nw√
a w ) with a i ∈ k ∗ and n i > 0 (1 ≤ i ≤ w). Moreover if L = k(Ω) is a Galois extension over k, then Ω is G(L/k)-invariant, so k(Ω) contains enough roots of unity, i.e., L = k({ε ni,
ni√
a i |1 ≤ i ≤ w}) with primitive n i -th roots of unity ε ni. Clearly Ω is not determined uniquely, while Ω/k ∗ is unique and finite.
The most interesting case of radical extension is L = k(λ) with λ ∈ L ∗ and λ n ∈ k. If L/k is Galois radical then we may regard L = k(ε n , λ). In particular if ε n ∈ k then L/k is a cyclic extension of degree dividing n(see [8, Theorem 14.4]). The radical extension L = k(λ) is said to be irreducible if degree [L : k] = n. Thus L/k is irreducible radical if and only if L = k(λ) and λ ∈ L ∗ is a root of an irreducible polynomial X n − a ∈ k[X], and, if and only if the order of λk ∗ in L ∗ /k ∗ is equal to the degree of λ over k.
Remark that, in case of a cyclotomic extension L = Q(ε m , ε n1, . . . , ε nw), the reduction in Theorem 4 would follow immediately by taking n = lcm(n 1 , . . . , n w ) and applying Theorem 2 to m and n. However when L = k(
n1√
), the reduction in Theorem 4 would follow immediately by taking n = lcm(n 1 , . . . , n w ) and applying Theorem 2 to m and n. However when L = k(
n1√
a 1 , . . . ,
nw√ a w ) is a Galois radical extension, to get a kind of reduction we need to know how to choose n 0 i from n i each other explicitly, not just taking n 0 from n =lcm of n i ’s.
Lemma 5. Let L = k(Ω) be a finite Galois radical extension of k. Then Ω/k ∗ is a finite G-module where G = G(L/k).
Proof. The finite group Ω/k ∗ has a G-module structure by defining an action σλ = σ(λ 0 )k ∗ where σ ∈ G and λ ∈ Ω/k ∗ such that λ = λ 0 k ∗ for λ 0 ∈ Ω. ¤
We begin with a radical extension containing an n-th root of 1 6= a ∈ k ∗ . Theorem 6. Let k = Q(ε m ) and K = k( √
na) be a Galois radical extension of k, where m and n are positive odd integers. For a prime p dividing n, let E = k(
pn√
a) and F = k(ε pn , √
na) be Galois radical extensions of k. Then (i) E = F , or E/F is cyclic of degree p and N E/F hε pn ,
pn√
ai = hε n , √
nai.
(ii) Furthermore we assume p z ||m and p b ||n.
a) If 0 ≤ z ≤ b then F/K is cyclic of degree p, and N F/K hε pn , √
nai
= hε n , √
na p i < hε n , √
nai.
b) If b < z then E = K, or F = K so N E/K hε pn ,
pn√
ai = hε n , √
nai.
Proof. The Galois radical extensions K and E form k < K = k(ε n , √
na) < F = k(ε pn , √
na) < E = k(ε pn ,
pn√ a).
Since E = k(
pn√
a) = F ( √
pλ) where λ = √
na ∈ F and √
pλ is a root of X p − √
na ∈
F [X], and since ε pn ∈ F , E/F is a cyclic extension of degree dividing p [8,
Theorem 14.4]. Hence E = F or [E : F ] = p.
If [E : F ] = p, any σ ∈ G(E/F ) maps
pn√
a to a zero of X pn − a, say σ(
pn√
a) = ε l pnpn√
a for 0 ≤ l < pn. But
pn√
a p is a zero of X n −a, so
pn√
a p ∈ F ∗ ,
pn
√
a p = σ(
pn√
a p ) = ε pl pnpn√
a p and ε pl pn = 1. Thus l = tn for some t = 0, . . . , p−1.
Moreover since each σ ∈ G(E/F ) leaves ε pn fixed, we have σ(
pn√
a)ε tn pnpn√
a, σ 2 (
pn√
a) = ε 2tn pn pn√
a, . . . , σ p−1 (
pn√
a) = ε (p−1)tn pn pn√ a, hence it follows that
N E/F (
pn√
a) = Y
σ∈G(E/F )
σ(
pn√
a) = ε tn(1+···+(p−1))
pn
pn√ a p =
pn√ a p .
Thus N E/F h
pn√
ai = h
pn√
a p i = h √
nai and N E/F hε pn i = hε p pn i = hε n i.
Suppose that m = p z m 0 and n = p b n 0 with p - m 0 n 0 . If 0 ≤ z ≤ b then lcm(m, pn) = p z m 0 p b+1 n 0
p z gcd(m 0 , p b+1−z n 0 ) = m 0 p b+1 n 0
gcd(m 0 , n 0 ) = p b+1 lcm(m 0 , n 0 ) and similarly lcm(m, n) = p b lcm(m 0 , n 0 ). Set lcm(m 0 , n 0 ) = l. Then Q(ε m , ε pn )
= Q(ε pb+1l ) and Q(ε m , ε n ) = Q(ε p
bl ) because p - l, hence [Q(ε m , ε pn ) : Q(ε m , ε n )] = φ(p b+1 l)
φ(p b l) = φ(p b+1 ) φ(p b ) = p (where φ is the Euler phi function), which shows that [F : K] = p.
Now each τ ∈ G(F/K) maps ε pn ∈ F to ε l pn for some 0 ≤ l < pn. Because ε p pn ∈ K ∗ , we have ε p pn = τ (ε p pn ) = ε pl pn , hence p(l − 1) ≡ 0 (mod pn) and l = tn + 1 with 0 ≤ t < p. Indeed, the cyclic group G(F/K) of order p consists of automorphisms τ t such that τ t (ε pn ) = ε tn+1 pn for 0 ≤ t < p. Thus we have
N F/K (ε pn ) = ε
P
p−1t=0
(tn+1)
pn = ε p+pn pn
(p−1)2= ε p pn ,
so N F/K hε pn i = hε p pn i ≤ hε n i. Comparing the orders, we get N F/K hε pn i= hε n i.
And N F/K h √
nai = h Q
τ ∈G(F/K) τ ( √n
a)i = h √
na p i < hε n , √
nai, this is (ii-a).
In case of m = p z m 0 , n = p b n 0 with 0 < b < z, gcd(m, pn) = pgcd(m, n), lcm(m, pn) = lcm(m, n), so Q(ε m , ε pn ) = Q(ε m , ε n ) and [F : K] = 1. Hence E = K, or [E : K] = p and N E/K hε pn ,
pn√
ai = N E/F hε pn ,
pn√
ai = hε n , √
na i.
¤ We observe that the assumption in (ii-b), i.e., m = p z m 0 , n = p b n 0 with b < z implies that k = Q(ε m ) already contains enough p-th roots of unity.
Corollary 7. Assume the same context as in (ii-b) Theorem 6 for radical extensions K = k( √
na) = k(Ω K ) and E = k(
pn√
a) = k(Ω E ) with finite Ω K /k ∗ and Ω E /k ∗ . If H = G(E/K) 6= 1 then N H (Ω E /k ∗ ) = Ω K /k ∗ .
Proof. E = k(ε pn ,
pn√
a) is a cyclic extension of K = k(ε n , √
na) of degree p with cyclic Galois group H. The mapping N H = Q
σ∈H σ in Lemma 1 determines N H hε pn , pn√
ai = hε n , √
nai by Theorem 6. By the action in Lemma 5, we have
N H (Ω E /k ∗ ) = N H hε pn k ∗ ,
pn√
ak ∗ i = hε n k ∗ , √
nak ∗ i = Ω K /k ∗ .
¤ We shall denote N H (Ω E /k ∗ ) by the same notation N E/K (Ω E /k ∗ ) which is the norm map. As a generalization of Theorem 6, we have the following theorem.
Theorem 8. Let k = Q(ε m ) and L = k( √
na) = k(Ω) be a Galois radical exten- sion of k. Assume a prime p with p z ||m and p b ||n. Let F = k(
n/p√
a) = k(Ω F ) be a Galois radical extension with L 6= F . If 0 < b ≤ z then N L/F (Ω/k ∗ ) = Ω F /k ∗ .
Proof. Let E = k(ε n ,
n/p√
a) be a Galois radical extension. Then k < F = k(ε n/p ,
n/p√
a) < E = k(ε n ,
n/p√
a) < L = k(ε n , √
na), and L = E( √
pλ) where λ =
n/p√
a ∈ E and √
pλ is a root of a polynomial X p −
n/p√
a in E[X]. Since ε p ∈ E, L/E is a cyclic radical extension of degree dividing p. So L = E or [L : E] = p. Because b ≤ z, k contains enough p-th roots of unity so F = E. But since L 6= F , we have [L : F ] = p and
N L/F hε n , √
nai = hε p n , √
na p i = hε n/p ,
n/p√ ai.
Similar to Corollary 7, we consequently have N L/F (Ω/k ∗ ) = Ω F /k ∗ . ¤ 4. Cohomology group on radical extension fields
We shall discuss cohomology groups over Galois radical extension fields, and have a reduction of cohomology group that is an analog of Theorem 2. Let H be a normal subgroup of G and M be a G-module. The inflation-restriction sequence on cohomology group
(1) 1 → H r (G/H, M H ) inf
→ H r (G, M ) res
→ H r (H, M )
is exact if r = 1. When r > 1, the sequence is exact if H i (H, M ) = 1 for all 1 ≤ i ≤ r − 1 (refer to [12, (3.4.2), (3.2.3)]).
Theorem 9. Let n = p b n 0 and m = p z m 0 with an odd prime p - n 0 m 0 . Let k = Q(ε m ), and L = k( √
na) = k(Ω) and L 0 = k(
n0√
a) = k(Ω 0 ) be Galois radical extensions of k with L 6= L 0 . If 0 < b ≤ z then the inflation map is an isomorphism on cohomology groups
H 2 (L 0 /k, Ω 0 /k ∗ ) inf
∼ = H 2 (L/k, Ω 0 /k ∗ ).
Proof. Obviously L = k(ε n , √
na) and L 0 = k(ε n0,
n0√
a). Since b ≤ z, ε pb is contained in k < L 0 so ε n ∈ L 0 . Thus L = L 0 (
pb√
λ) where λ =
n0√
a and
pb√ λ is a root of X pb−
n0√
a in L 0 [X]. So L/L 0 is cyclic of degree p c for some c ≤ b.
Due to Lemma 5, Ω/k ∗ is a G(L/k)-module with module action that, for any τ ∈ G(L/k) and √
nak ∗ ∈ Ω/k ∗ , τ ( √
nak ∗ ) = τ ( √
na)k ∗ = ε l n √
nak ∗ ∈ Ω/k ∗
for some l > 0. Similarly Ω 0 /k ∗ is a G(L 0 /k)-module. Moreover it is easy to see that Ω 0 /k ∗ is also a G(L/k)-module by regarding
n0√
a as ( √
na) pb. Write H = G(L/L 0 ). From the sequence of groups
1 → G(L/L 0 ) → G(L/k) → G(L 0 /k) → 1, we consider the inflation-restriction sequence
H 2 (L 0 /k, (Ω 0 /k ∗ ) H ) inf → H 2 (L/k, Ω 0 /k ∗ ) res → H 2 (L/L 0 , Ω 0 /k ∗ ).
Since H is cyclic, it follows from [12, (1.5.6)] and [12, (3.2.1)] that H 2 (L/L 0 , Ω 0 /k ∗ ) = H 0 (H, Ω 0 /k ∗ ) = (Ω 0 /k ∗ ) H
N L/L0(Ω 0 /k ∗ ) .
Every σ ∈ H leaves all elements in Ω 0 /k ∗ fixed, so (Ω 0 /k ∗ ) H = Ω 0 /k ∗ . More- over due to Theorem 8, we compute the norm N L/L0 directly to get
N L/L0(Ω 0 /k ∗ ) = hε p nc0k ∗ ,
n0√
k ∗ ,
n0√
a pck ∗ i = hε n0k ∗ ,
n0√
k ∗ ,
n0√
ak ∗ i = Ω 0 /k ∗ for gcd(p c , n 0 ) = 1. Hence H 2 (L/L 0 , Ω 0 /k ∗ ) = 1.
Again since H is finite cyclic and Ω 0 /k ∗ is a finite H-module, we use the Herbrand’s quotient of Ω 0 /k ∗ (refer to [3, (23.2)]) that
1 = h 2/1 (Ω 0 /k ∗ ) = |H 2 (L/L 0 , Ω 0 /k ∗ )|
|H 1 (L/L 0 , Ω 0 /k ∗ )|
hence it follows that H 1 (L/L 0 , Ω 0 /k ∗ ) = H 2 (L/L 0 , Ω 0 /k ∗ ) = 1. Therefore we can conclude from (1) that the sequence
1 → H 2 (L 0 /k, (Ω 0 /k ∗ ) H ) inf → H 2 (L/k, Ω 0 /k ∗ ) res → H 2 (L/L 0 , Ω 0 /k ∗ ) = 1 is exact, so have an isomorphism H 2 (L 0 /k, Ω 0 /k ∗ ) ∼ = H 2 (L/k, Ω 0 /k ∗ ). ¤
For a given finite radical extension L over k, we observed in Theorem 9 that the cohomology group over G(L/k) can be decreased down to that over G(L 0 /k), where L 0 is the Galois radical extension of k smaller than L by ‘one’
prime factor power p b . We can strengthen this observation with the following theorem which will go to reduction of Galois cohomology groups.
Theorem 10. Let k = Q(ε m ) and L = k( √
na) be the same fields as in Theorem 9. Assume n = p b 11· · · p b uu with distinct primes p i and b i > 0. For each p i
with distinct primes p i and b i > 0. For each p i
(1 ≤ i ≤ u), write m = p z iim i (with p i - m i and z i ≥ 0). Suppose b i ≤ z i for some 1 ≤ i ≤ u, and after appropriate renumbering, we assume that b i ≤ z i for all s + 1 ≤ i ≤ u. Let n 0 = p b 11· · · p b ss, and let the Galois extensions be
· · · p b ss, and let the Galois extensions be
L 0 = k(
n0√
a) = k(Ω 0 ), L j = k(
n0pbs+1s+1 ···pbs+js+j√
a) = k(Ω j ) for 1 ≤ j ≤ u − s.
Then H 2 (L j /k, Ω j−1 /k ∗ ) ∼ = H 2 (L j−1 /k, Ω j−1 /k ∗ ) for all 1 ≤ j ≤ u − s, so
H 2 (L/k, Ω 0 /k ∗ ) and H 2 (L 0 /k, Ω 0 /k ∗ ) are isomorphic.
Proof. Since L 0 , L j (1 ≤ j ≤ u − s) are all Galois radical extensions of k, and since n = p b 11· · · p b ss· p b s+1s+1· · · p b uu = n 0 · p b s+1s+1· · · p b uu, we have L u−s = L and
· p b s+1s+1· · · p b uu = n 0 · p b s+1s+1· · · p b uu, we have L u−s = L and
= n 0 · p b s+1s+1· · · p b uu, we have L u−s = L and
, we have L u−s = L and
k < L 0 = k(ε n0,
n0√
a) < L 1 = k(ε n
0
p
bs+1s+1,
n0pbs+1s+1√
a) < · · ·
< L j = k(ε n
0
p
bs+1s+1···p
bs+js+j,
n0pbs+1s+1 ···pbs+js+j√
a) < · · · < L u−s = L.
Since m = p z iim i and 0 < b i ≤ z i for s + 1 ≤ i ≤ u, each ε pbi
i
∈ k. Together with ε n0 ∈ L 0 , ε n
0
p
bs+1s+1···p
bs+js+jbelong to L 0 for all 1 ≤ j ≤ u − s. Hence L j = L j−1 (λ j ), where λ j ∈ L j and λ p
bs+j s+j
j ∈ L j−1 (1 ≤ j ≤ u − s), and L j /L j−1 is cyclic of degree p c s+j
jfor some c j ≤ b s+j . Thus we have the isomorphism on cohomology groups
H 2 (L 1 /k, Ω 0 /k ∗ ) ∼ = H 2 (L 0 /k, Ω 0 /k ∗ ) due to Theorem 9, and proceeding inductively we get
(2) H 2 (L j /k, Ω j−1 /k ∗ ) ∼ = H 2 (L j−1 /k, Ω j−1 /k ∗ ) for each 1 ≤ j ≤ u − s.
Let H = G(L 2 /L 0 ), and consider the inflation-restriction sequence (3) H 2 (L 0 /k, (Ω 0 /k ∗ ) H ) inf
→ H 2 (L 2 /k, Ω 0 /k ∗ ) res
→ H 2 (L 2 /L 0 , Ω 0 /k ∗ ).
Since L 2 = L 0 (ξ) with ξ pbs+1s+1 p
bs+2s+2 ∈ L 0 , L 2 /L 0 is a cyclic radical extension of degree dividing p b s+1s+1p b s+2s+2, and in fact [L 2 : L 0 ] = p c s+11 p c s+22 . Thus
p b s+2s+2, and in fact [L 2 : L 0 ] = p c s+11 p c s+22 . Thus
p c s+22 . Thus
H = G(L 2 /L 0 ) ∼ = Z pc1
s+1
p
c2s+2∼ = Z pc1
s+1
× Z pc2
s+2
∼ = G(L 2 /L 1 ) × G(L 1 /L 0 ), and by [7, (2.3.14)], we have
H 2 (L 2 /L 0 , Ω 0 /k ∗ ) ∼ = H 2 (L 2 /L 1 , Ω 0 /k ∗ ) × H 2 (L 1 /L 0 , Ω 0 /k ∗ ).
But since p i - n 0 for s + 1 ≤ i ≤ u, we obtain N L1/L
0(Ω 0 /k ∗ ) = hε n0k ∗ ,
n0√
k ∗ ,
n0√
ak ∗ i pc1s+1= Ω 0 /k ∗ = (Ω 0 /k ∗ ) L1/L
0
and similarly, N L2/L
1(Ω 0 /k ∗ ) = (Ω 0 /k ∗ ) pc2s+2= Ω 0 /k ∗ = (Ω 0 /k ∗ ) L2/L
1.
/L
0and similarly, N L2/L
1(Ω 0 /k ∗ ) = (Ω 0 /k ∗ ) pc2s+2= Ω 0 /k ∗ = (Ω 0 /k ∗ ) L2/L
1.
= Ω 0 /k ∗ = (Ω 0 /k ∗ ) L2/L
1.
By the proof of Theorem 9, both H 2 (L 1 /L 0 , Ω 0 /k ∗ ) and H 2 (L 2 /L 1 , Ω 0 /k ∗ ) are trivial groups, so that H 2 (L 2 /L 0 , Ω 0 /k ∗ ) = 1. Then again the Herbrand quotient of Ω 0 /k ∗ is equal to 1, i.e.,
1 = |H 2 (L 2 /L 0 , Ω 0 /k ∗ )|/|H 1 (L 2 /L 0 , Ω 0 /k ∗ )|.
So H 1 (L 2 /L 0 , Ω 0 /k ∗ ) is trivial, and the inflation-restriction sequence (3) is exact. We thus obtain the isomorphism on cohomology groups
H 2 (L 0 /k, Ω 0 /k ∗ ) inf
∼ = H 2 (L 2 /k, Ω 0 /k ∗ ).
And together with (2), it follows that
H 2 (L 2 /k, Ω 0 /k ∗ ) ∼ = H 2 (L 1 /k, Ω 0 /k ∗ ) ∼ = H 2 (L 0 /k, Ω 0 /k ∗ ).
Applying this process to the cyclic group H = G(L 3 /L 0 ) of order p c s+11 p c s+22
p c s+33 and to the sequence
p c s+33 and to the sequence
H 2 (L 0 /k, (Ω 0 /k ∗ ) H ) inf
→ H 2 (L 3 /k, Ω 0 /k ∗ ) res
→ H 2 (L 3 /L 0 , Ω 0 /k ∗ ), we also get H 2 (L 3 /L 0 , Ω 0 /k ∗ ) = 1, for G(L 3 /L 0 ) ∼ = G(L 3 /L 2 ) × G(L 2 /L 0 ) and H 2 (L 3 /L 2 , Ω 0 /k ∗ ) = 1 = H 2 (L 2 /L 0 , Ω 0 /k ∗ ).
The exactness of the inflation-restriction sequence guarantees the isomor- phism H 2 (L 3 /k, Ω 0 /k ∗ ) ∼ = H 2 (L 0 /k, Ω 0 /k ∗ ). Continuing, we eventually get
H 2 (L/k, Ω 0 /k ∗ ) = H 2 (L u−s /k, Ω 0 /k ∗ ) ∼ = H 2 (L 0 /k, Ω 0 /k ∗ ).
¤ Remark 1. In Theorem 10, we assume a = 1, i.e., √
na = ε n for n = p b 11· · · p b uu
(odd primes). Suppose m = p z iim i with z i < b i for 1 ≤ i ≤ s, and b i ≤ z i
(odd primes). Suppose m = p z iim i with z i < b i for 1 ≤ i ≤ s, and b i ≤ z i
for s + 1 ≤ i ≤ u. Let n 0 = p b 11· · · p b ss, L = Q(ε m , ε n ) and L 0 = Q(ε m , ε n0).
, L = Q(ε m , ε n ) and L 0 = Q(ε m , ε n0).
Then L = L 0 and H 2 (L/k, Ω L /k ∗ ) ∼ = H 2 (L 0 /k, Ω L0/k ∗ ) (this is Theorem 10).
Moreover, by rearrangement if necessary, we assume z 1 = · · · = z t = 0 for t ≤ s, i.e., p i 6 |m for 1 ≤ i ≤ t. By letting n 0 0 = p 1 · · · p t , we can further reduce the cohomology group by Janusz theorem that
H 2 (L/k, µ(L)) = H 2 (L 0 /k, µ(L 0 )) ∼ = H 2 (k(ε n00)/k, µ(k(ε n00))).
))).
Remark 2. It is natural to ask whether the Remark 1 is true for a 6= 1 with the same n and n 0 0 , i.e., is H 2 (k( √
na)/k, h √
nai/k ∗ ) ∼ = H 2 (k(
n00√
a)/k, h
n00√ ai/k ∗ )?
The following proposition provides a partial answer.
Proposition 11. Let L = k( √
na) be a Galois radical extension of k = Q(ε m ).
Let n = p b 11· · · p b uu (b i > 0, odd primes p i ) and m = p z iim i (z i ≥ 0, p i 6 |m i ) for all i. By rearrangement, assume z i ≤ b i for 1 ≤ i ≤ s(≤ u), z i > b i for s + 1 < i ≤ u, and moreover z j = 0 for 1 ≤ j ≤ t(≤ s). Set n 0 = p b 11· · · p b ss
and n 0 0 = p 1 · · · p t . Let K = k(
n00√
(b i > 0, odd primes p i ) and m = p z iim i (z i ≥ 0, p i 6 |m i ) for all i. By rearrangement, assume z i ≤ b i for 1 ≤ i ≤ s(≤ u), z i > b i for s + 1 < i ≤ u, and moreover z j = 0 for 1 ≤ j ≤ t(≤ s). Set n 0 = p b 11· · · p b ss
and n 0 0 = p 1 · · · p t . Let K = k(
n00√
· · · p b ss
and n 0 0 = p 1 · · · p t . Let K = k(
n00√
a) = k(Ω), and F v = k(
n0/pbv −zvv√ a) and B v = k(ε n
0
/p
bv −zvv,
n00√
a) be Galois radical extensions of k for t + 1 ≤ v ≤ s.
Then H 2 (L/k, Ω/k ∗ ) ∼ = H 2 (F v /k, Ω/k ∗ ), H 2 (K/k, Ω/k ∗ ) ∼ = H 2 (B v /k, Ω/k ∗ ), but H 2 (F v /k, Ω/k ∗ ) 6∼ = H 2 (B v /k, Ω/k ∗ ).
Proof. Let L 0 = k(
n0√
a) = k(Ω L0) be a Galois radical extension of k. With the integers n = p b 11· · · p b uu, n 0 = p b 11· · · p b ss and n 0 0 = p 1 · · · p t for t ≤ s ≤ u, it is clear that k < K < L 0 < L and H 2 (L/k, Ω L0/k ∗ ) ∼ = H 2 (L 0 /k, Ω L0/k ∗ ) due to Theorem 10. Hence it is follows that
· · · p b uu, n 0 = p b 11· · · p b ss and n 0 0 = p 1 · · · p t for t ≤ s ≤ u, it is clear that k < K < L 0 < L and H 2 (L/k, Ω L0/k ∗ ) ∼ = H 2 (L 0 /k, Ω L0/k ∗ ) due to Theorem 10. Hence it is follows that
· · · p b ss and n 0 0 = p 1 · · · p t for t ≤ s ≤ u, it is clear that k < K < L 0 < L and H 2 (L/k, Ω L0/k ∗ ) ∼ = H 2 (L 0 /k, Ω L0/k ∗ ) due to Theorem 10. Hence it is follows that
/k ∗ ) ∼ = H 2 (L 0 /k, Ω L0/k ∗ ) due to Theorem 10. Hence it is follows that
H 2 (L/k, Ω/k ∗ ) ∼ = H 2 (L 0 /k, Ω/k ∗ ).
Let p v be one of p t+1 , . . . , p s . Since m = p z vvm v with 0 < z v ≤ b v , the Galois radical extensions of k are
K = k(ε n00,
n00√
a) < F v = k(ε n
0
/p
bv −zvv,
n0/pbv −zvv√
a) < L 0 = k(ε n0,
n00√
a).
Now let K, F v and L 0 denote the cyclotomic extensions K = Q(ε m , ε n0
0
) < F v = Q(ε m , ε n
0
/p
bv −zvv) < L 0 = Q(ε m , ε n0).
Owing to p z vv|| pbv −zvn
0
n
0v
and ε pzvv ∈ K, we have p v - [F v : K] and H 2 (F v /K, µ(F v )) = H 2 (L 0 /F v , µ(L 0 )) = 1. Thus due to Janusz theorem we have
H 2 (K/k, µ(K)) ∼ = H 2 (F v /k, µ(F v )) ∼ = H 2 (L 0 /k, µ(L 0 )).
On the other hand, from the tower of fields k < K = K(
n00√
a) < F v = F v (
n0/pbv −zvv√
a) < L 0 = L 0 (
n0√ a), consider a field A v = k(ε n0,
n0/pbv −zvv √
a). We now observe the followings.
(i) K < B v < F v < A v < L 0
(ii) L 0 = A v (
pzvv√
λ) where λ =
n0/pbv −zvb√
a and
pzvv√
λ is a root of X pzvv − λ ∈ A v [X]. Since ε pzvv ∈ A v , L 0 /A v is a cyclic extension of order p w vv with w v ≤ z v .
∈ A v , L 0 /A v is a cyclic extension of order p w vv with w v ≤ z v .
(iii) Similar to (ii), it can be seen that F v = B v (
n0/n00pbv −zvv√
θ) where θ =
n00
√
a ∈ B v and
n0/n00pbv −zvv√
θ is a root of X n0/n
00p
bv −zvv − θ ∈ B v [X]. Since ε n
0
/n
00p
bv −zvvbelongs to B v = Q(ε m , ε n
0
/p
bv −zvv,
n00√
a), F v /B v is cyclic of degree dividing p b 11−1 · · · p b t
t−1 · p b t+1
t+1· · · p z vv· · · p b ss.
· · · p b ss.
(iv) A v = F v (ε n0) = F v (ε pbv −zv
v
), so G(A v /F v ) ∼ = G(L 0 /F v ).
(v) B v = K(ε n
0
/p
bv −zvv), so G(B v /K) ∼ = G(F v /K).
L 0 = k(ε n0) L 0 = k(ε n0,
n00√
,
n00√
a) = L 0 (
n0√ a) A v = k(ε n0,
n0/pbv −zvv √
a) F v = k(ε n
0
/p
bv −zvv) F v = k(
n0/pbv −zvv√
a) = F v (
n0/pbv −zvb√ a) B v = k(ε n
0
/p
bv −zvv,
n00√ a) K = k(ε n00) K = k(
n00√
a) = k(Ω) = K(
n00√ a) Thus from (ii), L 0 /A v is cyclic of order p w vv with w v ≤ z v , so
N L0/A
v(Ω/k ∗ ) = N L0/A
vhε m , ε n00,
n00√
/A
vhε m , ε n00,
n00√
aik ∗ = hε m k ∗ , ε p nwvv0
0 k ∗ ,
n00√
a pwvv k ∗ i, and this is equal to Ω/k ∗ because gcd(n 0 0 , p v ) = 1 for t + 1 ≤ v ≤ s. Hence
H 2 (L 0 /A v , Ω/k ∗ ) = (Ω/k ∗ ) G(L0/A
v) N L
0/A
v(Ω/k ∗ ) = 1
= H 0 (L 0 /A v , Ω/k ∗ ) = H 1 (L 0 /A v , Ω/k ∗ ).
So the exact sequence
H 2 (A v /k, Ω/k ∗ ) → H 2 (L 0 /k, Ω/k ∗ ) → H 2 (L 0 /A v , Ω/k ∗ )
yields the isomorphism H 2 (A v /k, Ω/k ∗ ) ∼ = H 2 (L 0 /k, Ω/k ∗ ).
From (iv), A v = F v (ε pbv −zv
v
) and G(A v /F v ) ∼ = G(L 0 /F v ) cyclic, so the in- variant set Ω/k ∗ by G(A v /F v ) corresponds to µ(K)/k ∗ by G(L 0 /F v ). Thus
H 2 (A v /F v , Ω/k ∗ ) ∼ = H 2 (L 0 /F v , µ(K)/k ∗ ).
Since H 2 (L 0 /F v , µ(K)) ,→ H 2 (L 0 /F v , µ(L 0 )) = 1 by Janusz theorem, we have H 2 (L 0 /F v , µ(K)) = 1. Moreover since H 2 (L 0 /F v , µ(K)) → H 2 (L 0 /F v , µ(K)/
k ∗ ), we have
H 2 (L 0 /F v , µ(K)/k ∗ ) = 1 = H 2 (A v /F v , Ω/k ∗ ).
So we obtain the isomorphism H 2 (F v /k, Ω/k ∗ ) ∼ = H 2 (A/k, Ω/k ∗ ) from the exact sequence H 2 (F v /k, Ω/k ∗ ) → H 2 (A v /k, Ω/k ∗ ) → H 2 (A v /F v , Ω/k ∗ ) = 1.
We therefore have the isomorphisms
H 2 (L/k, Ω/k ∗ ) ∼ = H 2 (L 0 /k, Ω/k ∗ ) ∼ = H 2 (A v /k, Ω/k ∗ )
∼ = H 2 (F v /k, Ω/k ∗ ).
Now from (v), B v = K(ε n
0
/p
bv −zvv) and G(B v /K) ∼ = G(F v /K) cyclic. As above,
H 2 (B v /K, Ω/k ∗ ) ∼ = H 2 (F v /K, µ(K)/k ∗ ) = 1.
Thus H 2 (K/k, Ω/k ∗ ) → H 2 (B v /k, Ω/k ∗ ) → H 2 (B v /K, Ω/k ∗ ) = 1 is exact, so the isomorphism H 2 (K/k, Ω/k ∗ ) ∼ = H 2 (B v /k, Ω/k ∗ ) follows.
However we observe that H 2 (B v /k, Ω/k ∗ ) is not isomorphic to H 2 ( F v /k, Ω/
k ∗ ). In fact, F v /B v is cyclic of degree d dividing p b 11−1 · · · p b t
t−1 · p b t+1
t+1· · · p z vv· · · p b ss by (iii). Thus the N Fv/B
v(Ω/k ∗ ) = N Fv/B
vhε m k ∗ , ε n00k ∗ ,
n00√
· · · p b ss by (iii). Thus the N Fv/B
v(Ω/k ∗ ) = N Fv/B
vhε m k ∗ , ε n00k ∗ ,
n00√
/B
v(Ω/k ∗ ) = N Fv/B
vhε m k ∗ , ε n00k ∗ ,
n00√
k ∗ ,
n00√
ak ∗ i = hε m k ∗ , ε d n0
0