doi: 10.

2478/v10062-012-0027-8
ANNALES
UNIVERSITATIS MARIAE CURIE-SKŁODOWSKA
LUBLIN – POLONIA
VOL. LXVII, NO. 2, 2013 SECTIO A 17–32

AKRAM LBEKKOURI

Some results on local fields

Abstract. Let K be a local field with finite residue field of characteristic p.

This paper is devoted to the study of the maximal abelian extension of K of
exponent p − 1 and its maximal p-abelian extension, especially the description
of their Galois groups in solvable case. Then some properties of local fields in
general case are studied too.

1. Presentation. Let us make a description of some standard over-exten-

sions. But, first let us specify some important, although well-known remarks
concerning the characteristics of a local field and its residue field.
Let K be a complete field with respect to a discrete valuation.
1.1. Characteristics.
1.1.1. The case of mixed characteristics. In this case, necessarily
char(K) = 0 and the residue field is necessarily of characteristic p > 0.
Therefore, K contains necessarily the (p − 1)-th roots of unity.
It is worthy to note that in this case the residue field may be finite or
infinite. Indeed, for the infinite case consider the following example.
Let p be an arbitrary but fixed prime number, let K be the completion
of Qnr , the field generated over Qp by all roots of unity of order prime to p
(i.e., the maximal unramified extension of Qp ) which is a complete discrete
valued field of characteristic 0 with infinite residue field of characteristic
p > 0. Then the residue field is equal to the algebraic closure of Fp (namely

2000 Mathematics Subject Classification. 11S15.

Key words and phrases. Local fields, local number fields, Wild ramification, interme-
diate extension, standard p-over-extensions, semi-direct product, inertia group.
18 A. Lbekkouri

the field generated by all Fpln with l prime and n any strictly positive
integer), which is infinite.
1.1.2. The case of equal characteristics. In this case we may have
char(K) = 0 with a residue field necessarily infinite. In such case K may
contain the (p−1)-th roots of unity or not, as well as the p-th roots of unity.
Or we may have char(K) = p > 0 with a residue field that may be finite
or infinite. In such case K contains necessarily the (p − 1)-th roots of unity,
but does not contain the p-th roots of unity.
1.2. Presentation of the standard p-over-extensions. By “standard
p-over-extensions” of a local field K with finite residue field of characteristic
p, we mean the maximal abelian extension M of K of exponent p − 1, and
the maximal p-abelian extension of M .
Some needed theorems on groups. Schur–Zassenhaus theorem, 1937
(first version):
Theorem 1.1. If G is a finite group, N is an invariant (not necessar-
ily abelian) subgroup of G, and if (#N, #G/N ) = 1, then the following
sequence:
1 7→ N 7→ G 7→ G/N 7→ 1
is exact, also N has a complement M in G, that is G is a semi-direct product
of N by M .
Schur–Zassenhaus theorem (second version), see [6], Chap. 7, Th. 7.24.,
page 151:
Theorem 1.2. If N and M are finite groups of relatively prime orders,
then every extension of N by M is a semi-direct product.
Generalized Schur–Zassenhaus theorem, see [5], § 2.3, page 41:
Theorem 1.3. Let K be a closed normal Hall subgroup of a profinite group
G. Then K has a complement L in G (i.e., L is a closed subgroup of G
such that G = KL and K ∩ L = {1}). Moreover, any two complements of
K are conjugate.
Note. A closed subgroup K of a profinite group G is a π-Hall subgroup if
#K is a π-number and |G : K| is a π 0 -number. When π = {p}, a π-Hall
subgroup is simply called p-Sylow subgroup.
In particular, if the exponent and the index of K are relatively, prime
then K is a π-Hall subgroup.
1.2.1. Case of finite residue field. Let p be a prime number and K be
a local field with finite residue field of characteristic p, K = Fpf . Therefore,
the maximal abelian extension of exponent p−1 of K is M = K((K ? )1/p−1 ),
regardless of the characteristic of K.
 Some results on local fields 19

Write Γ = gal(M/K) for the Galois group. From Kummer Theory for
abelian extensions we have that the group Γ is dual to the group K ? /K ?(p−1) ,
under the pairing:
ϕ : Γ × K ? /K ?(p−1) 7−→ F?p

(σ, x) 7−→ σ(y)/y

with (y p−1 = x) where F?p ⊂ K ? has been identified with the group of the
(p − 1)-th roots of unity.
Write N for the maximal abelian extension of exponent p of M (i.e., the
maximal p-abelian extension of M ). The expression of N depends essentially
on the characteristic of K:
1. If char(K) = 0, K is a finite extension
√ of Qp of residual degree f and
p ?
ramification index e, then N = M ( M ). Furthermore we have the tower:
√
√
K ——– M = K p−1 K ? ——– N = M ( p M ? ).
The group ∆ = gal(N/M ) is isomorphic to the filtered Γ-module M ? /M ?p
of Fp -dimension 2 + (p − 1)2 ef .
2. If char(K) = p > 0, K = k((T )), we can not take the same extension
as above. Indeed extracting p-th roots in characteristic p gives rise to purely
inseparable extensions, in nontrivial case. In such case, the field N is ob-
tained by adjoining the zeroes of polynomials of the form X p − X − a with
a ∈ M , (Artin–Schreier polynomials). Then we have N = M (℘−1 (M )),
where ℘ is the endomorphism of M defined by ℘ : x → xp − x. So, we have
the tower:
√

K ——– M = K p−1 K ? ——– N = M (℘−1 (M )).
∆ = gal(N/M ) is isomorphic to the filtered Γ-module M/(℘(M )) of Fp -
dimension +∞.
Clearly all extensions of the two towers above are normal, so we write
the different Galois groups as follows:
• Γ = gal(M/K), which is abelian of degree (p − 1)2 isomorphic to
(Z/(p − 1)Z)2 .
• ∆ = gal(N/M ), which is abelian too of exponent p, isomorphic to a
product of a nonnecessarily countable number of copies of Z/pZ, in general
see §2.4., (but in mixed characteristics case this number is finite).
• G = gal (N/K), which is not necessarily abelian. Furthermore, it is
a semi-direct product G = ∆ o Γ. Indeed, according to Schur–Zassenhaus
Theorems 1.1, 1.2, we get the result for finite case. For general case, from
Krull topology, ∆ is a closed normal subgroup of G and the exponents
are relatively prime, then according to the generalized Schur–Zassenhaus
Theorem 1.3, we get the result too.
1.2.2. On the case of equal and prime characteristics. Let K be a
local field having the same characteristic p > 0 as its residue field which is
neither assumed to be finite nor perfect. In this case we still have the tower:
20 A. Lbekkouri

√
p−1
K ? ——– N = M (℘−1 (M )).


K ——– M = K
The group Γ = gal(M/K) is not necessarily finite but abelian of exponent
p − 1. The group ∆ = gal(N/M ) is isomorphic to an infinite but not
necessarily countable number of copies of Z/pZ.

Proposition 1.4. If K = F ((T )), where F is a field of characteristic p

which is not perfect, then K/℘ (K) is infinite. Here ℘ is the endomorphism
defined by ℘ : x → xp − x.

Proof. Consider T1n , where n > 0 and p does not divide n. If T1n − T1n0 ∈
℘ (K), with n 6= n0 and p does not divide nn0 , then T1n − T1n0 = f p − f , for
some f ∈ K = k ((T )). But f ∈ / K = k [[T ]] necessarily (since n, n0 > 0 and
distinct). Thus f has a leading polar term with degree −r < 0, so f p has
a pole with degree −rp < −r, that is f p − f has a pole of order rp that is
divisible by p yet the difference T1n − T1n0 does not have this property, since
n and n0 are distinct and not divisible by p, which ends the proof. 

1.2.3. The field M = K((K ? )1/p−1 )/K in the general case.

• In the local case with finite residue field of characteristic p we have seen
that M = K((K ? )1/p−1 )/K is an abelian extension of degree (p − 1)2 , the
Galois group of which is isomorphic to (Z/(p − 1)Z)2 .
• Meanwhile, if K is a complete field with respect to a discrete valuation
having a residue field of characteristic p not necessarily finite, then M =
K((K ? )1/p−1 )/K is not necessarily finite, but it is abelian of exponent p−1,
since K contains the (p − 1)-th roots of unity.
• Otherwise, for such M/K, this extension need not be finite; if it is finite
it need not be Galois; and if it is finite and Galois it need not have that
Galois group. Some examples follow.
Examples.
1. Let K = k((t)), where k = Q(ξ3 ) and ξ3 is a primitive cube root of unity.
So K is a complete discretely valued field.
Let p = 3. Then k((k ? )1/p−1 )/k is infinite. Hence so is K((K ? )1/p−1 )/K.
2. Let k be an algebraically closed field of characteristic 0, and let K =
k((t)). Then K((K ? )1/p−1 )/K is Galois with group Z/(p − 1)Z, not
(Z/(p − 1)Z)2 .
3. Let k be the field of 3 elements, and let K = k((t)).
Let p = 11. Then K((K ? )1/p−1 )/K is Galois with group Z/2Z × Z/10Z.
4. Let k be the field of 3 elements, and let K = k((t)).
Let p = 7. Then K((K ? )1/p−1 )/K has degree 12 (not 36), but it is not
Galois because it is not separable, since t1/3 is in this field.
5. K = Q(ξ3 ) where ξ3 is a 3-rd root of unity. Therefore, M/K =
K((K ? )1/p−1 )/K = Q(ξ3 )((Q(ξ3 )? )1/2 )/Q(ξ3 ) is infinite, since adjoining to
K the square roots of different prime elements of Z[ξ3 ] will lead to disjoint
 Some results on local fields 21

quadratic extensions whose composite has the degree equal to a large power
of 2 (the power being the number of primes).
More generally we have the following results:
6. Consider K = Q(ξp ) where ξp is a p-th root of unity, p being an odd
prime number. Then
√

1/p−1
 q 
K K ? /K = Q(ξp ) 1/p−1 Q(ξp )? /Q(ξp )
is infinite.
Indeed, from the well-known result: For relatively prime integers a1 , . . . ,
√
an , the 2n algebraic numbers ai1 , . . . , aik with i1 < · · · < ik and 0 ≤ k ≤ n
√ √
are linearly independent over Q, so are a Q-basis for Q( ai1 , . . . , aik ).
In particular, the degree of that field over Q is the maximum possible 2n ,
we can deduce that Q((Q? )1/2 )/Q is infinite. Since Q(ξp )/Q is finite, then
Q(ξp )((Q(ξp )? )1/2 )/Q(ξp ) is infinite, therefore Q(ξp )((Q(ξp )? )1/p−1 )/Q(ξp )
is infinite too. The result is proved.
√ √
Note that the degree of Q(ξp )( ai1 , . . . , aik ) over Q(ξp ) is 2n or 2n−1 ;
it depends on whether √ the set√the numbers ai union +p or −p is still inde-
pendent or not and +p or −p belongs to Q(ξp ) depending on whether
p ≡ 1 mod 4 or p ≡ 3 mod 4.
Furthermore, we have the following generalization:
7. For any fractions field K, with characteristic not equal to 2, of a Dedkind
ring A having infinite many prime ideals and any odd prime number p, we
have that M = K((K ? )1/p−1 )/K is infinite.
Indeed, it suffices to notice that when adjoining to K the square roots of
two different prime elements of A will lead to disjoint quadratic extensions.
√ √
In fact, let L = K( p) and L0 = K( q). They are both quadratic. Ne-
√ √
cessarily L ∩ L0 = K otherwise L = L0 , this means that q = a + b p for
√
a, b ∈ K, thus q = a2 +2ab p+b2 p. Clearly b has to be non-zero. If a is also
√
non-zero, then this formula shows that p ∈ K, so a has to be zero. Then
q = b2 p, but once we localize at the ideal generated by the prime number
q; the prime number p is then a unit and q is a uniformizer so this cannot
happen.
8. In contrast, in characteristic 2 we have the counter-example
√
√

F2 (T ) T = F2 (T ) T + 1 .
Note. Concerning items 7 and 8, the different result for characteristic 2
is really just an artifact. More generally, if p is any prime and a positive
integer n is not a power of p, then M = K((K ?1/n )/K is infinite for rings
as in item 7. Of course if p is prime and n = p − 1, then n cannot be a
power of a prime q unless q = 2, which leads to the item 8. But if we take
a different n (e.g. take n = p − 2), then characteristic 2 need not be the
exception.
22 A. Lbekkouri

2. Description of the over-extensions (case of finite residue field).

2.1. Sundries on groups. First let us prove some necessary results:
Proposition 2.1. Let G0 be a subgroup of G = (Z/pZ)n oϕ (Z/ (p − 1) Z)2
of index p, then G0 ∩ (Z/pZ)n is normal in G.
Proof. First note that (Z/pZ)n is the p-Sylow subgroup of G and is nor-
mal in it. Since G0 contains a copy of (Z/ (p − 1) Z)2 , then (Z/ (p − 1) Z)2
normalizes G0 and therefore normalizes G0 ∩ (Z/pZ)n . On the other hand
(Z/pZ)n normalizes G0 ∩ (Z/pZ)n , since (Z/pZ)n is abelian. In consequence
G0 ∩ (Z/pZ)n is normal in G. 
Remark 2.2. The result above does not mean that any subgroup of index
p of (Z/pZ)n is normal in G = (Z/pZ)n oϕ (Z/ (p − 1) Z)2 . See the following
counter-example.
√

Example 2.3 (Counter-example). Let K = Q3 , consider M = K K ? =
√
p3
√

Q3 i, 3 , and consider E = M 1 + 3 , that is a normal 3-extension
√
of M . The Galois closure of E/K is N = M 3 M ? , i.e.,

√ 3 √
q q 
3
N =M 1 + 3, 1 − 3

and gal (N/M ) = (Z/3Z)2 . But E/K is not normal, otherwise there should
0
be an intermediate subextension
√ √E /K of degree 3 of E/K and0 an automor-
phism σ of E that maps 3 to − 3, which is the identity on E , furthermore
p3
√
√ √
σ 1 + 3 must be a cubic root of σ(1+ 3) = 1− 3, but E contains no
such root, since E is strictly contained in N . Hence the subgroup gal (N/E)
is not normal in gal (N/K).
2.2. Case of mixed characteristic.
Remark 2.4. From the expression of Γ above, we know that Γ is of ex-
ponent p − 1, that is M/K is abelian Kummer extension √ relatively to the
number p − 1. Furthermore, the multiplicative group p−1 K ? /K ? contains a
finite number of coclasses ci K ? likewise the multiplicative group K ? /K ?p−1 ,
with the coclasses ci K ?p−1 . ∆ being seen as Γ-module, from the action of
Γ on it, in fact Γ acts on M , on M ? , on M ?p too, and on M ? /M ?p , as well
as on µp which is included in M ; µp being the group of p-th roots of unity.
So, ∆ ' Hom (M ? /M ?p , hζp i). M ? /M ?p is known to be a vector space on
Fp of dimension 2 + [M : K][K : Qp ] then it is a finite p-elementary abelian
group that is a vector space over Fp having the same dimension, thus #∆ =
p2+[M :K][K:Qp ] = p2+n where n = [M : K][K : Qp ] = (p − 1)2 ef and ∆ is p-
elementary abelian having 2 + n generators that is ∆ = hα1 , α2 , . . . , αn+2 i.
Since M ? /M ?p is a Fp [Γ]-module of dimension n + 2, we can assert that
N can be generated over M by n + 2 elements bi such that bpi ∈ M , that
 Some results on local fields 23

is N = M (b1 , b2 , . . . , bn+2 ), hence we can consider ∆ = hα1 , α2 , . . . , αn+2 i

such that αi (bi ) = ζpi bi , and αi (bj ) = bj if i 6= j, where ζp is a primitive p-th
root of unity. To sum up we have the result:
Proposition 2.5. When writing N = M (b1 , b2 , . . . , bn+2 ), with bpi ∈ M , the
group ∆ = gal(N/M ) = hα1 , α2 , . . . , αn+2 i can be defined by αi (bi ) = εi bi
and αi (bj ) = bj if i 6= j, where ε is a primitive p-th root of unity.
Here we need to construct the semi-direct product
(Z/pZ)n oφ (Z/(p − 1)Z)2 ,
that is to write down a non-trivial homomorphism φ : (Z/(p − 1)Z)2 →
Aut((Z/pZ)n ). (The trivial homomorphism φ(g) = Id, where Id is the
identity automorphism of (Z/pZ)n , corresponds to the direct product). The
group Aut((Z/pZ)n ) can be identified with the group GLn (Z/pZ), the group
of invertible n-dimensional matrices over Z/pZ. To write down a group
homomorphism:
φ : (Z/(p − 1)Z)2 → GLn (Z/pZ)
we just need to decide where the two standard generators, say g1 , g2 , of
(Z/(p − 1)Z)2 are sent, and the only constraints are that their images com-
mute and they have order dividing p − 1. Thus to describe all possible ho-
momorphism ϕ we need to find all pairs of matrices A and B in GLn (Z/pZ)
such that Ap−1 = In , B p−1 = In , where In is the identity matrix and
AB = BA.
Note that two commuting matrices are simultaneously diagonalizable.
It comes to pick up two elements of GLn (Z/pZ) of order dividing p − 1.
Indeed since p − 1 is co-prime to p = char(Z/pZ), then every element of
GLn (Z/pZ) of order p − 1 is diagonalizable. Furthermore, a diagonal matrix
is of order dividing p−1 if the diagonal entries belong to (Z/pZ)? . It suffices
to consider the element M of GLn in the form M = diag(d1 , d2 , . . . , dn );
a diagonal matrix such that di ∈ (Z/pZ)? . Such M exists and verifies the
needed properties of order and commutativity. Then for such M consider
the element fM of Aut((Z/pZ)2 ) the matrix of which is exactly M . Then
write
(Z/(p − 1)Z)2 = hg1 , g2 i
with g1 , g2 of order p − 1 and (Z/pZ)n = hα1 , α2 , . . . , αn i with the αi of
order p. Therefore, we have a group homomorphism:
φ : (Z/(p − 1)Z)2 7→ Aut((Z/pZ)n )
gi 7→ φ(gi ) = f (gi ) = fMi
for i = 1 or 2 with Mi = diag(ζi1 , ζi2 , . . . , ζin ); a diagonal matrix. ζij being
elements of (Z/pZ)? , M1 and M2 being the matrices associated to φ. Thus
we can define the semi-direct product as
(Z/pZ)n oφ (Z/(p − 1)Z)2 = hα1 , α2 , . . . , αn i oφ hg1 , g2 i
24 A. Lbekkouri

with 2n relations gi αj gi−1 = ζij αj . Clearly we have several different actions

φ and then several different semi-direct products.
Note that by considering the dual Γ̂ of Γ; Γ̂ = Hom(Γ, F?p ); the matrices
M1 and M2 can be written as Mi = diag(χ1 (gi ), χ2 (gi ), . . . , χn (gi )) where
χj ∈ Γ̂, and then the 2n relations can be written as gi αj gi−1 = χj (gi )αj . So,
we have proved the following result:
Proposition 2.6. (Z/pZ)n oφ (Z/(p − 1)Z)2 = hα1 , α2 , . . . , αn i oφ hg1 , g2 i
with 2n relations gi αj gi−1 = ζij αj for i = 1, 2 and j = 1, . . . , n. That is
gi αj gi−1 = χj (gi )αj with χj ∈ Γ̂.
Therefore, by the use of the Proposition 2.6, we get the result:
Proposition 2.7. Let G = ∆ oφ Γ = (Z/pZ)n+2 oφ (Z/(p − 1)Z)2 . Write
G = hα1 , α2 , . . . , αn+2 i oφ hσ, τ i. Then the semi-direct product is defined
by the 2(n + 2) relations: σαi σ −1 = ζi αi , and τ αi τ −1 = ξi αi , for i =
1, . . . , n + 2; ζ, ξ being elements of (Z/pZ)? .
That is by considering the dual Γ̂ = Hom(Γ, F?p ) of Γ; and by writing
the matrices M1 and M2 , images of σ and τ by the action defining the
semi-direct product, as follows: M1 = diag(χ1 (σ), χ2 (σ), . . . , χn+2 (σ)), and
M2 = diag(χ1 (τ ), χ2 (τ ), . . . , χn+2 (τ )), where χi ∈ Γ̂, the 2(n + 2) relations
can be written as: σαi σ −1 = χi (σ)αi , and τ αi τ −1 = χi (τ )αi .
2.3. Uncountability of the product ∆ for local functional fields.
Proposition 2.8. Let K = F((T )), where F is a finite
√
field. Then ∆ =
−1
gal(N/M ) (where N = M (℘ (M )) and M = K p−1
K ? ) is a sum of an
uncountable number of copies of Z/pZ.
Proof. F is a finite field. The group M/℘ (M ) is a direct sum of a non-
necessarily countable number of copies of Z/pZ (with ℘ : x → xp − x).
Furthermore, M/℘ (M ) is isomorphic to ∆. M/K is Kummer and abelian
of degree (p − 1)2 . We can write M p = V ((X)); V ((X)) being the field
of Laurent series, where V = F( p−1 √ (ε)) where ε is a generator of the
?
multiplicative group F and X = p−1
T.
Let MM be the maximal ideal of M . Using Hensel’s Lemma we can
easily see that MM ⊆ ℘(M ), (indeed by considering b = − ∞ pi
P
i=0 a , b ∈ M
and bp − b = a). Then M/℘ (M ) is a quotient of M/MM which is a vector
space of a countable dimension on Z/pZ. A base is given by all elements
ζX −t , where t are positive integers and ζ, ranging over a finite base of F over
Z/pZ. Note that every element of M/MM is a finite linear combination
of these elements. A countable product of copies of Z/pZ will give infinite
sums. Then ∆ is necessarily an uncountable product of copies of Z/pZ,
which ends the proof. 
 Some results on local fields 25

3. The intermediate extension.

3.1. Some theorems on groups. P. Hall theorem (1928), see [6] Ch. 5,
Th 5.23, page 85:
Theorem 3.1. Let G be a solvable group of order nm where (n, m) = 1
then G contains at least one subgroup of order n and any such subgroups
are conjugate.
Lemma 3.2 (Galois, see for example [2], Ch. 3, Th. 7). A transitive sub-
group of Sp , the group of permutations of p elements, is solvable if and only
if it contains a unique Sylow p-subgroup of order p.
3.2. Existence of the intermediate extension.
Proposition 3.3. Let K be any commutative field, for every separable
extension L/K of degree p, where p is an odd prime number, such that
G = gal(LC )/K the Galois group of the Galois closure of L/K is solvable.
Then there exists a cyclic extension F/K of degree m dividing p − 1 such
that LF/F is cyclic of degree p and LF/K is Galois (i.e., LC = LF ).
Furthermore, if L/K is not cyclic (LF/K is hence not abelian), then L
has exactly p conjugates over K in LF .
Proof. Write G for the Galois group of its Galois closure. G is solvable, its
order is divisible by p but not by p2 . Furthermore, it can be considered as
transitive subgroup of the symmetric group Sp . According to Lemma 3.2
G contains a unique subgroup P of order p hence it is normal in G. P is
contained in its normalizer N (P ) in Sp . Also N (P ) can be considered as the
affine linear group GA1 (Fp ), thus we have the isomorphism F?p → Aut(P ),
so we get a split short exact sequence:
1 → P → N (P ) → F?p → 1.
Furthermore, N (P )  is isomorphic
 to the group of all 2 × 2 matrices over
x 0
GF (p) of the form . In consequence G/P is cyclic of order m
y 1
dividing p − 1. Therefore, and since G ⊂ N (P ), it is also a semi-direct
product G = P o M with M cyclic of order m. If the semi-direct product
is a direct product, then it is cyclic since m and p are co-prime.
Otherwise G is not abelian. In such case since Q M is cyclic, all its con-
jugates are cyclic too. Write m in the form m = ri=1 mαi i , where mi are
different prime numbers, and N for the number of the conjugates of M
(note that according to Hall’s theorem (Theorem 3.1) all the subgroups of
G of order m are conjugate). Since M is cyclic, it contains one and only
one subgroup Mi of order mαi i (Sylow mi -subgroup of G) which is cyclic
too. Conversely every Sylow mi -subgroup of G can be embedded in some
conjugate of M . Therefore, the number N must divide mp, being N ≡ 1
modulo mi for every i, thus (N, m) = 1. In consequence the number of
26 A. Lbekkouri

conjugates of M is exactly p if G is not cyclic. Set F the field fixed by P ,

then the Galois closure of L/K is LC = LF . This ends the proof. 
3.3. Intermediate extension, explicit determination. We assume
that K is a local field with a finite residue field of characteristic p.
As seen before, the compact group Γ is Γ ' Hom(K ? /K ?p−1 , µp−1 ), then
by duality Γ ' K ? /K ?p−1 . Hence the exponent of Γ is just equal to p − 1,
and M/K is Kummer abelian extension relatively to the number p − 1, and
then the intermediate subextension F/K announced in Proposition 3.3 is
cyclic Kummer.
3.4.√Description of the Galois groups. Since F/K is Kummer, F =
K( m c), with c ∈ K ? , K ? being the√ multiplicative group of nonzero elements
m
√ m


of K. Furthermore, K( c) = K d if and only if there exists an integer
k ≥ 1 with (k, m) = 1 such that d ∈ ck K ?m . By considering the quotient
group K ? /K ?m , the order of the class cK ?m in it is m.
Since m divides p − 1, K ? /K ?m ≈ (Z/mZ) × (Z/mZ); therefore K ? /K ?m
is of order m2 . The number of the distinct Kummer cyclic extensions of
K of degree m is exactly the number of cyclic subgroups of order m in
(K ? /K ?m ).
So it is easy to deduce that the number of the cyclic distinct Kummer
extensions of K of degree m equals the number of the cyclic subgroups of
order m included in (Z/mZ) × (Z/mZ), so by writing m = pα1 1 . . . pαr r , we
get that this number equals
pα1 1 + pα1 1 −1 . . . pαr r + pαr r −1 .

Furthermore, gal (F/K) ≈ H, a cyclic group of order m dividing p−1 that

can be embedded in µp−1 , the group of the (p − 1)-th roots of unity, and by
considering the Galois closure LC of L/K, gal (LC /K) ≈ gal (LC /F ) o H,
(a semi-direct product). Then from local class field theory, see for example
[3], we have the isomorphism between the three groups gal (F/K) ≈ H ≈
K ? /NF/K (F ? ) of order m, and the surjective homomorphism
s : K ? /K ?m ≈ (Z/mZ) × (Z/mZ) 7→ K ? /NF/K (F ? ) .
The group gal (LF/K). Since gal√ (F/K) is cyclic
√ of order m dividing
p − 1, write gal(F/K) =<  > with ( m c) = ξm ( m c), where ξm a primitive
m-th root of unity and name the extension of  to F (π) by  too. Since
gal(F (π)/F ) is cyclic of order p write gal(F (π)/F ) =< σ >. LF/K is
Galois, consider any element τ of gal(LF/K), thus τ = σ i j , with 1 ≤ i ≤ p
and 1 ≤ j ≤ m, then from the normality of < σ > in gal(LF/K), we have
the identity τ στ −1 = σ t with 1 ≤ t ≤ p − 1.
Consider the affine group AGL (1, p), that is the set of all maps from
Fp to itself of the form x 7→ ax + b where a 6= 0 in Fp . gal (LF/K) has
order mp and is isomorphic to a subgroup of AGL (1, p), which is isomor-
phic to the subgroup GL2 (Z/pZ), consisting of the matrices of the form
 Some results on local fields 27

   
a b a b
, where an automorphism δ corresponds to the matrix
0 1 0 1
when δ (ξp ) = ξpa , and δ(x) = ξpb x where ξp , is a primitive p-th root of unity.
That is by picking a generator g of (Z/pZ)? , we can use  for a generator
 of
g 0
gal (F/K),  : x 7→ gx that corresponds to the matrix and for a
0 1
generator
  σ of gal(LF/F ), σ : x 7→ x + 1 that corresponds to the matrix
1 1
then σ−1 = σ g .
1 0
For any element τ of gal(LF/K) we have τ = σ i j , with 1 ≤ i ≤ p and
j
1 ≤ j ≤ m, τ στ −1 = σ g , also g must verify g m = 1 in Fp .
The group (Z/pZ)? has ϕ (m) elements of order m, where ϕ (·) is the
Euler’s totient (indicator). Meanwhile the equation xm = 1 modulo p has
exactly m solutions in (Z/pZ)? , since m divides p − 1 which is the order of
(Z/pZ)? , these solutions are the elements of the cyclic subgroup of order m
of the cyclic group (Z/pZ)? , and is obviously isomorphic to the group of the
m-th roots of unity. √
τ ( m c)
On the other hand τ (c)c = 1 so m√
c
β
= ξm β
. Note that ξm does not depend
√ m√
m
τ ( c) τ ( d)
on c but on the coclass cK ?m only. Indeed m√c = m√d if and only if
τ m dc = m dc , that is dc ∈ K ?m . Now consider θ = σ (π) − π so θ ≡ 0
p
p

modulo π v+1 . Set π1 = τ (π) which is uniformizer too, since LF/F is normal
and totally ramified, then vLF/F (σ (π) − π) = vLF/F (σ (π1 ) − π1 ) > 0 and
integer v is the ramification number (= break=jump) of LF/F .
We can write σ (π1 ) − π1 = u(σ (π) − π) = uθ where u is the unit of LF .
Note that u ≡ 1 modulo π, as σ(τ (π) − π) ≡ τ (π) − π modulo π, therefore
it is worthy to note that the class of τ (θ) θ modulo π is independent of π and
depends on τ and σ only.
Then write θ = στ −1 (π1 ) − τ −1 (π1 ), that is τ (θ) = τ στ −1 (π1 ) − π1 .
Now, since gal(LF/F ) =< σ > is a normal subgroup of gal(LF/K) which
is not abelian we have τ στ −1 = σ a , with 1 ≤ a ≤ p − 1,
τ (θ) =


therefore
(σ a (π1 ) − π1 ). Since the equality between ideals σ π t = π t holds, by
successive substitutions we get σ a (π1 )−π1 ≡ a (σ (π1 ) − π1 ) ≡ a (σ (π) − π)
modulo π v+2 , that is τ (θ) ≡ aθ modulo π v+2 for 1 ≤ a ≤ p − 1, finally we
get τ (θ)
θ ≡ a modulo π
v+1 , that is modulo p for 1 ≤ a ≤ p − 1.

Now, we can take c = NL/K (γ) = NLF/F (γ) where γ = −kak π k−1 and ak
is the coefficient of f (the minimal polynomial of π over K that is over F
too) such that k is the index t that achieves the minimum in the expression
inf 1≤t≤p (vL (tat ) + t) (for more details on this method see [4] §.3.2, the
√
proof of Proposition 3.1). Furthermore, we have p−1 γ ≡ θ modulo π ⇒
28 A. Lbekkouri

√
NLF/F ( p−1 γ) ≡ NLF/F (θ) ≡ θp ≡ θ modulo π, then we immediately get:
√
τ (NLF/F ( p−1 γ)) v+1 , that is modulo p for 1 ≤ a ≤ p − 1.
NLF/F ( p−1 γ) ≡ a modulo π
√

Therefore, we have proved the following result:

Theorem 3.4. Let p be an odd prime number and K be a local field with
char(K) ≡ 0 modulo p having a finite residue field. Consider any separable
extension L/K of degree p.
Then there exists c ∈ K ? , unique up to K ?(p−1) , such that the following
holds: √
• L ( p−1 c) is the Galois closure
√ of L/K. √ √
• For every τ ∈ gal (L ( p−1 c)) /K,√
and σ ∈ gal (L ( p−1 c) /K ( p−1 c)), we
τ ( p−1 c)
have τ στ −1 = σ a , with a = p−1√c modulo p.

In other words we have an explicit determination of the intermediate

√
subextension announced in Proposition 3.3, namely F = K ( p−1 c).
Note. Even if the decomposition of the Galois closure and the justification
of the existence of the intermediate extension do not require the comple-
tion of the base field, the following description of the Galois groups and
the intermediate extension require the completion of the base √ field and the
p−1
finitude of the residue field obligatory. Indeed, M = K( K ? )/K is not
2
necessarily of degree (p − 1) , much more it can be infinite as seen before.
Furthermore the splitting field can be solvable much more cyclic over a local
field and not solvable over a non-local field, see the following Example.
Example (Counter-example). Consider f (X) = X 3 + 3X 2 + 24 as a poly-
nomial of Q3 [X] with f (α) = 0. It is Eisenstein polynomial, and it defines
a normal extension Q3 (α) of Q3 , see [4].
Meanwhile, when considering f as polynomial of Q [X], in spite of the
fact that Q contains the (p − 1)-th roots of unity for p = 3 and the Galois
closure of Q(α)/Q is certainly solvable, the extension Q(α)/Q is not normal
since the equation possesses complex roots.

4. On solvability in the general case. This section adjusts the focus

on some general properties concerning local fields, especially the solvability
of local finite extensions.
In case of local fields with finite residue fields the solvability can be easily
proved. Meanwhile, in the general case, that is when the residue field is infi-
nite, the problem is much more difficult, especially when it is not necessarily
perfect, indeed we can have nonsolvable finite local extensions.
4.1. Sketch of the classical ramification. Let L/K be a finite Galois
extension of local fields. In this paragraph especially, the residue fields are
assumed to be perfect. In this case we can easily define two canonical filtra-
tions (Gi ) (the lower ramification groups) and (Gi ) (the upper ramification
 Some results on local fields 29

groups) in G that are decreasing sequences of subgroups of G. The i-th

lower ramification group is defined to be

Gi = {σ ∈ G; w(σ(α) − α) ≥ i + 1 for all α ∈ RL },

where RL is the ring of the discrete valuation w.

The two filtrations of ramification subgroups are related by the formula
Gi = Gψ (i), where ψ is the Hasse–Herbrand function used to be calcu-
lated in terms of orders of Gi . The said filtrations are compatible with
the subextensions in L/K. In the sense that, if H is a normal subgroup
of G, we can get the filtrations relative to H by taking Hi = Gi ∩ H and
(G/H)i = (Gi H)/H (see [7], Ch. IV). Furthermore, if L/K is abelian when
the residue fields is quasi-finite, the “upper” filtration of G is compatible
with the reciprocity map θ : K ? → G in the sense that θ(Ui ) = Gi for
all i = 0, 1, . . . , n, and (Ui ) is the filtration of the units group in K ? . The
compatibility of the “upper” filtration of G with the class field theory is
studied in [7], Ch. XV.
Consequently, in case of local fields with perfect residue fields, the solv-
ability of G0 , the inertia subgroup of G is ensured, since the quotient of
any two successive lower ramification groups is abelian. In fact it is either
cyclic or a direct sum of groups of order p. Meanwhile, the solvability of G
itself requires the finiteness of the residue field (see for example [7], Corol-
lary 5 of §.2, Ch. IV page 68). Note that in this reference the residue field
is not assumed to be perfect but it is assumed that the residue extension is
separable only.
From now on, let K be a local field (that is a complete field with respect
to a discrete valuation) having a non-necessarily perfect residue field k.
4.2. General case (imperfect residue field). First note that in the
general case, a theory of upper numbering of ramification subgroups corre-
sponding to a theory of the “lower” ramification subgroups by use of some
means like the Hasse–Herbrand ϕ, ψ functions does not exist.
All that has been done till now is a “lower” ramification theory which
is obtained from the double-filtration given in [8], see paragraph §.4.4 that
comes. However, the ramification filtration in the group G does not deter-
mine that in H, for a normal subgroup H of G. In fact we do not have in this
case a formula similar as in the separable case, (see [7], Ch. 4 §.1 Proposi-
tion 2). Indeed, for a finite Galois extension L/K of local fields, the discrete
valuation ring RL of L may not be generated by only one element over RK ,
when we are in the inseparable case in contrary to the separable case.
Moreover, recently for a theory of the upper numbering filtration, Abbes
and Saito made a different interpretation that can be generalized in the gen-
eral situation, giving rise to a quite well-behaved upper-number filtration.
To sum up, they define two decreasing filtrations by ramification groups
on the absolute Galois group (the Galois group of the separable closure)
30 A. Lbekkouri

such that in the classical case where the residue field is perfect, we recover
the classical upper numbering filtration. The definition uses rigid geometry
and log-structures, see [1].
4.3. Description of the extension. Let K be any local field (i.e., a
complete field with respect to a discrete valuation, the residue class field is
imperfect), and let L/K be a finite Galois extension. Write G = gal(L/K)
for its Galois group. The discrete valuation ring of L is denoted by RL and
its maximal ideal by ML . In such case the degree of L/K splits in such a
way [L : K] = ef = etame ewild fsep finsep , (ewild and finsep must be a power
of p, etame is prime to p, meanwhile fsep need not be necessarily prime to p).
Since L/K is a Galois extension, we get that the residue extension l/k
is normal (see for example [7], Ch. 1, §.7, Proposition 20), but it need not
be separable. Consider the set D of all automorphisms of l unvarying all
elements of k, there is a natural surjective homomorphism from ϕ : G → D.
Indeed, let g ∈ G, g preserves RL as well as ML . Therefore, g induces an
automorphism of l = RL /ML . Since g fixes each element of K, it fixes
each element of k as well, the surjectivity of ϕ is proved in the proposition
referred above.
4.4. A filtration of the inertia group. The inertia group G0 is given
by G0 = {σ ∈ G; ∀x ∈ RL , (x − σ(x)) ∈ ML }.
Let L0 be the fixed field of G0 that is the inertia field of L/K and the
maximal subfield of L that is unramified over K. The residue class field
l0 of L0 is the separable closure of k in l (the residue fields) hence l0 /k is
Galois and its degree is fsep , which is the order of G/G0 ' gal(l0 /k).
As it is usually done in this case when the residue class field is imperfect,
for any positive integer i ≥ −1 we define a filtration of the inertia group G0
obtained from the double-filtration given in [8], where the authors consider a
doubly indexed filtration (Gn,i )n,i∈N of the Galois group, from which a lower
ramification filtration that generalizes the last one in the classical case (see
for example [7]) can be deduced.
i+n
Namely Gn,i = {σ ∈ G; ∀x ∈ MiL , (x − σ(x)) ∈ ML }, that is the
subgroup of G consisting of the K-automorphisms of L that induce the
identity on MiL /Mi+n L . Whereas the classical lower ramification filtration
is obtained by considering the sequence of the groups Gn+1,0 .
So, we define Gi = {σ ∈ G; (π − σ(π)) ∈ Mi+1 L } for π an arbitrary but
fixed prime element of L (in the definition of Gi , G can be replaced by G0
since clearly Gi ⊆ G0 for i ≥ 0), that is the subgroup of G consisting of the
K-automorphisms of L that act trivially on RL /Mi+1 L .
By the way, G = G−1 since σ satisfies σ(RL ) = RL , σ(ML ) = ML
and particularly σ(π) is a prime element for any σ ∈ G. Then likewise for
the classical case we have the filtration G = G−1 ⊇ G0 ⊇ G1 · · · ⊇ Gi . . . ,
with the existence of an integer r such that Gi = {1} for i ≥ r, since
RL ' lim i
←−RL /ML .
 Some results on local fields 31

4.5. Results. Then we have the result:

Proposition 4.1. Let L/K be a finite Galois extension of local fields with
imperfect residue class fields. Denote by G0 the inertia group of gal(L/K).
Then G0 is solvable, furthermore it is cyclic in characteristic zero case.
Proof. First for any positive integer i ≥ 1, let σ ∈ Gi . An uniformizer π of
L being fixed we have (σ(π) − π) ∈ Mi+1 L hence π − σ −1 (π) ∈ σ −1 (Mi+1L ),
−1
that is σ ∈ Gi so Gi is a subgroup of G. Furthermore, let σ, τ ∈ Gi . Then
(στ (π) − π) = σ(τ (π) − π) + σ(π) − π ∈ Mi+1 L , i.e., στ ∈ Gi , that is Gi is
normal G. So, the above definition is well justified.
Now, let us fix a set of generators of the residue field extension and their
lifts u1 , . . . , un to RL . Put it another way, RL is generated by π, u1 , . . . , un
as an RK -algebra, where π has valuation 1, and ui are units.
Consider the map:
G → k? ,
g 7→ g(π)/π.
It is clear that this is a homomorphism, and G1 is the kernel of this map.
Then again consider the map:
G1 → k ⊕ k ⊕ · · · ⊕ k; (n + 1 of them),
g 7→ ((g(π) − π)/π 2 , (g(u1 ) − u1 )/π, . . . , (g(un ) − un )/π),
where (g(α) − α)/π i is the residue class of (g(α) − α)/π i mod π, that is also
a homomorphism, and G2 is the kernel. Then we continue to consider:
G2 → k ⊕ k ⊕ · · · ⊕ k; (n + 1 of them),
g 7→ ((g(π) − π)/π 3 , (g(u1 ) − u1 )/π 2 , . . . , (g(un ) − un )/π 2 ),
and so on and so forth, till we get a trivial Gr .
From this, in case of the residue fields having characteristic p > 0, it is
clear that G1 has a filtration by normal subgroups Gi where the subquotients
Gi /Gi+1 are p-elementary abelian groups as Gi /Gi+1 injectively maps to
(1 + MiL )/(1 + Mi+1 L ) which is canonically isomorphic to (l, +) for i ≥ 1.
Furthermore, G0 /G1 is cyclic of order prime to p as it injectively maps
to R?L /(1 + ML ) ' (l? , ×), as well as to Autl (ML /M2L ) ' (l? , ×), and the
field l is of characteristic p. It is worthy to note that the maximal tamely
ramified subfield T of L corresponds to the group GT = G0 ∩ G1 = G1 .
Finally, G1 is a p-group of order ewild finsep , therefore its solvability implies
the last one of G0 .
In case the residue fields are of characteristic zero, since for i ≥ 1 the
subquotients Gi /Gi+1 are isomorphic to a subgroup of (l, +), considered as
an additive group. (l, +) has no finite subgroup except {0}, therefore Gi are
trivial for all i ≥ 1, in consequence G0 is cyclic, ewild = 1 and necessarily p
does not divide e which ends the proof. 
32 A. Lbekkouri

Note. The result of the solvability of the inertia group can be deduced
from the work of I. B. Zhukov in [9] but not for the general case. Indeed, he
added the strong hypothesis that [k : k p ] = p, where k is the residue field,
that this K is particularly assumed to be a two-dimensional local field.
Proposition 4.2. Let K be a local field, and let L/K be a finite Galois
extension. Then L/K is solvable if and only if the maximal separable subex-
tension of l/k is solvable.
Proof. Consider G = gal(L/K), and its inertia group G0 . By use of the
current notations the isomorphism induced by the surjective homomorphism
ϕ defined above makes G/G0 isomorphic to the Galois group of the maximal
separable subextension of l/k which equals D. From Proposition 4.1, G0 is
solvable, so the well-known result of classical group theory: G is solvable if
and only if both G/G0 and G0 are solvable, ends the proof.
Thus L/K is solvable if and only if the Galois group of the maximal
separable subextension of l/k is solvable. 
Acknowledgment. I heartily want to thank Professor I. Fesenko from
Nottingham University (UK) for having read and approved the proof of
Proposition 4.1.

References
[1] Abbes, A., Saito, T., Ramification of local fields with imperfect residue fields, Amer.
J. Math. 124 (5) (2002), 879–920.
[2] Artin, E., Galois Theory, Univ. of Notre Dame Press, Notre Dame, 1942.
[3] Hazewinkel, M., Local class field theory is easy, Adv. Math. 18 (1975), 148–181.
[4] Lbekkouri, A., On the construction of normal wildly ramified over Qp , (p 6= 2), Arch.
Math. (Basel) 93 (2009), 331–344.
[5] Ribes, L., Zalesskii, P., Profinite Groups, Springer-Verlag, Berlin, 2000.
[6] Rotman, J. J., An Introduction to the Theory of Group, Springer-Verlag, New York,
1995.
[7] Serre, J.-P., Local Fields, Springer-Verlag, New York–Berlin, 1979.
[8] Zariski, O., Samuel, P., Commutative Algebra. Volume II, Springer-Verlag, New York–
Heidelberg, 1975.
[9] Zhukov, I. B., On ramification theory in the imperfect residue field case,
Preprint No. 98-02, Nottingham Univ., 1998. Proceedings of the conference:
Ramification Theory of Arithmetic Schemes (Luminy, 1999) (ed. B. Erez),
http://family239.narod.ru/math/publ.htm.

Akram Lbekkouri
BP: 10507
Casa-Bandoeng 20002
Casablanca
Morocco
e-mail: lbeka11@gmail.com

Received August 11, 2012