You are on page 1of 12

GALOIS THEORY AT WORK

KEITH CONRAD
1. Examples
Example 1.1. The eld extension Q(

2,

3)/Q is Galois, as it is the splitting eld of


(X
2
2)(X
2
3) over Q. The extension has degree 4 (why?), so the Galois group has size
4. Since

2 and

3 each have two Q-conjugates, all 4 choices of assignments of them to
their Q-conjugates must be realized by an element of the Galois group (otherwise its size
couldnt reach 4). The results are collected in Table 1, which shows in the third column that

2 +

3 has 4 dierent values under the action of the Galois group. This shows

2 +

3
must generate the eld Q(

2,

3): otherwise

2 +

3 would have degree at most 2 over


Q, but then it would have at most 2 dierent Q-conjugates, which isnt the case. (The
polynomial whose roots are the four Q-conjugates of

2 +

3 is X
4
10X
2
+ 1, so this
must be an irreducible polynomial in Q[X].)
(

2) (

3) (

2 +

3)

2

3

2 +

3

2

2

3

2 +

3
Table 1
Since Gal(Q(

2,

3)/Q) has order 4, the only proper nontrivial subgroups are of order
2. There are 3 automorphisms of order 2, so there are three subgroups of order 2 and thus
Q(

2,

3)/Q has 3 intermediate elds of degree 2 over Q. We already know two of them:
Q(

2) and Q(

3). Since

6 =

2

3 is also in Q(

2,

3) and doesnt lie in Q(

2) or
Q(

3), the third quadratic eld in Q(

2,

3) is Q(

6). Here are all the subelds:


Q(

2,

3)
L
L
L
L
L
L
L
L
L
L
r
r
r
r
r
r
r
r
r
r
Q(

2)
M
M
M
M
M
M
M
M
M
M
M
Q(

3) Q(

6)
q
q
q
q
q
q
q
q
q
q
q
Q
In Table 1, the subgroup xing Q(

2) is the rst and second row, the subgroup xing


Q(

3) is the rst and third row, and the subgroup xing Q(

6) is the rst and fourth row


(since (

2)(

3) =

2

3).
1
2 KEITH CONRAD
Example 1.2. We will use the Galois correspondence to nd the intermediate elds of
Q(
4

2)/Q. This extension is not Galois but it is a subextension of Q(


4

2, i)/Q, which is
Galois and its Galois group is isomorphic to D
4
= 'r, s` where r and s permute the 4th
roots of 2 according to the standard generating symmetries of the square having the 4th
roots of 2 as vertices. In terms of their eect on
4

2 and i, we have r(
4

2) = i
4

2, r(i) = i
and s(
4

2) =
4

2, s(i) = i. (Viewing elements of Q(


4

2, i) as complex numbers, s acts on


them like complex conjugation.)
We can use Galois theory for Q(
4

2, i)/Q to study the extension Q(


4

2)/Q. A eld
diagram is in (1.1). We dont have to work out all the subelds of Q(
4

2, i) to nd the
subelds of Q(
4

2). We only need to nd the subgroup corresponding to Q(


4

2), the
subgroups containing that, and their xed elds. (Larger subgroups correspond to smaller
subelds.)
(1.1)
Q(
4

2, i) 1

2
Q(
4

2) H

4
Q D
4
The subgroup H D
4
xing Q(
4

2) has size [Q(


4

2, i) : Q(
4

2)] = 2. Since s is a
nontrivial element of D
4
and xes Q(
4

2), H = 1, s = 's`. Therefore elds strictly


between Q and Q(
4

2) are in bijection with subgroups strictly between 's` and D


4
= 'r, s`.
From the known subgroup structure of D
4
, the only subgroup lying strictly between H and
D
4
is 'r
2
, s`. Therefore only one eld lies strictly between Q(
4

2) and Q. Since Q(

2) is
such a eld it is the only such eld.
Remark 1.3. While Galois theory provides the most systematic method to nd the inter-
mediate subelds in Q(
4

2)/Q, it can be done in other ways. Suppose Q F Q(


4

2)
with [F : Q] = 2. Then
4

2 has degree 2 over F. Since


4

2 is a root of X
4
2, its minimal
polynomial over F has to be a quadratic factor of X
4
2. There are three quadratic factors
with
4

2 as a root, but only one of them, X


2

2, has coecients in Q(
4

2). Therefore
this must be the minimal polynomial of
4

2 over F, so

2 F. Therefore Q(

2) = F by
counting degrees.
A Galois extension is said to have a given group-theoretic property (being abelian, non-
abelian, cyclic, etc.) when its Galois group has that property.
Example 1.4. Any quadratic extension of Q is an abelian extension since its Galois group
is cyclic of order 2.
Example 1.5. The extension Q(
3

2, ) is a non-abelian extension since its Galois group


is isomorphic to S
3
.
Theorem 1.6. If L/K is a nite abelian extension then every intermediate eld is an
abelian extension of K. If L/K is cyclic then every intermediate eld is cyclic over K.
Proof. Every subgroup of Gal(L/K) is normal when Gal(L/K) is abelian, so every inter-
mediate eld is Galois over K. The quotient of an abelian group by a subgroup is abelian
and the quotient of a cyclic group by a subgroup is cyclic.
GALOIS THEORY AT WORK 3
Theorem 1.7. If L
1
/K and L
2
/K are nite abelian extensions inside a common extension
of K then L
1
L
2
is a nite abelian extension of K.
Proof. A composite of Galois extensions is Galois, so L
1
L
2
/K is Galois. Any element
Gal(L
1
L
2
/K) can be restricted to L
1
and L
2
, where it is an automorphism since L
1
and
L
2
are Galois over K. So we get a function Gal(L
1
L
2
/K) Gal(L
1
/K) Gal(L
2
/K) by
([
L
1
, [
L
2
), which is a homomorphism (check!). The kernel is trivial, since if is the
identity on L
1
and L
2
then it is the identity on L
1
L
2
. Thus Gal(L
1
L
2
/L) embeds into the
direct product of the Galois groups of L
1
and L
2
ove K. A subgroup of a direct product of
abelian groups is abelian, so L
1
L
2
/K is an abelian extension.
The analogue of Theorem 1.7 for nite cyclic extensions is false: a compositum of two
cyclic Galois extensions need not be a cyclic extension. Look at Example 1.1.
2. Applications to Field Theory
We will prove the complex numbers are algebraically closed (the Fundamental Theorem
of Algebra) using Galois theory and a minimal amount of analysis. We need one property
of the real numbers, one property of the complex numbers, and one property of nite groups:
Every odd degree polynomial in R[X] has a real root. In particular, no polynomial
of odd degree greater than 1 in R[X] is irreducible.
Every complex number has a complex square root.
If a prime power divides the size of a nite group, there is a subgroup with that
prime power size. In particular, a group of size 2
m
for m 1 has a subgroup of size
2
m1
.
The rst fact is a consequence of the intermediate value theorem. The second fact can be
explained by writing a nonzero complex number as re
i
and then using

re
i/2
. (Writing
a complex number in the form a + bi, can you nd a purely algebraic formula for a square
root of a +bi in terms of a and b?) The third fact is an extension of Sylows rst theorem.
Were now ready to prove the theorem.
Theorem 2.1. The complex numbers are algebraically closed.
Proof. To prove a eld is algebraically closed, it suces to show that the only nite algebraic
extension of the eld is the eld itself. So let E/C be a nite extension. Since E is a nite
extension of R, and were in characteristic 0, we can construct a nite Galois extension
K/R containing E.
Let H be a 2-Sylow subgroup of G = Gal(K/R) and F the corresponding xed eld.
Then F/R is an extension of odd degree, so every element of F has odd degree over R.
There are no irreducible polynomials in R[X] of odd degree greater than 1, so F = R.
Therefore G = H is a 2-group (i.e., a group whose size is a power of 2).
If K = C, then Gal(K/C) has size 2
m
where m 1. Therefore this group has a subgroup
of size 2
m1
, whose xed eld is a quadratic extension of C. A quadratic extension of C
has the form C(

d) for some nonsquare d C

. But every nonzero complex number is a


complex square, so we have a contradiction. Thus K = C. Since C E K, we conclude
that E = C.
Our next application concerns constructible numbers. A necessary condition for a com-
plex number to be constructible (using only an unmarked straightedge and compass) is that
4 KEITH CONRAD
the number has 2-power degree over Q. We can show in a strong way that this condition is
not sucient.
Theorem 2.2. For every integer n 2 there is a eld extension E/Q of degree n containing
no proper intermediate subelds.
Proof. It can be shown that that for every n 2 there is an irreducible polynomial f(X) of
degree n in Q[X] whose splitting eld over Q has Galois group G
f

= S
n
. An algorithm for
constructing an example, relying on the Chinese remainder theorem, is in [2, pp. 641642].
Armed with such an irreducible polynomial f(X) of degree n, let L = Q(r
1
, . . . , r
n
) be
a splitting eld of f(X) over Q, where the r
i
s are roots of f(X). The extension Q(r
1
)/Q
has degree n. This extension will turn out to have no proper intermediate elds. We are
going to explain why this follows from a result in group theory.
When we let G
f
act on the r
i
s just as permutations (ignoring the automorphism aspect
on the elds) we get an embedding G
f
S
n
. We are told that G
f

= S
n
, so this embedding
is an isomorphism: every permutation of the roots comes from an element of G
f
. The
subgroup of G
f
corresponding to Q(r
1
) is Gal(L/Q(r
1
)), which are the elements of G
f
that
x r
1
. Since G
f

= S
n
by the permutation action on the roots of f(X), the automorphisms
xing r
1
are naturally a copy of S
n1
in S
n
. (The subgroups of S
n
xing any one number
are all conjugate to each other, and it can be shown that these are all of the subgroups of
size (n 1)! in S
n
.)
L (1)

(n1)!
Q(r
1
) S
n1

n
Q S
n
It is a theorem of group theory that every copy of S
n1
in S
n
is a maximal subgroup: there
is no subgroup of S
n
properly between S
n1
and S
n
.
1
So by the Galois correspondence,
there is no eld properly between Q and Q(r
1
).
Example 2.3. The polynomial X
4
X 1 has Galois group S
4
over Q. Therefore if is
a root of this polynomial, [Q() : Q] = 4 and there is no quadratic intermediate eld.
Corollary 2.4. For any k 2, there are complex numbers which have degree 2
k
over Q
but are not constructible.
Proof. If a complex number is constructible then Q() can be built up by successive
quadratic extensions. In particular, when [Q() : Q] > 2 there are elds intermediate
between Q and Q(). Taking n = 2
k
in Theorem 2.2 gives examples where [Q() : Q] = 2
k
and there are no elds strictly between Q and Q(), so is not constructible.
Example 2.5. Continuing with the previous example, any root of X
4
X1 has degree a
power of 2 over Q but is not constructible since the eld it generates over Q cant be built
up by successive quadratic extensions.
Theorem 2.6. If L/K be Galois with degree p
r
, where p is a prime, there is a chain of
intermediate elds
K = F
0
F
1
F
2
F
r
= K
1
This is because the action of Sn on {1, 2, . . . , n} is doubly transitive when n 2; the stabilizer subgroup
of a point in a doubly transitive action is a maximal subgroup.
GALOIS THEORY AT WORK 5
where [F
i
: F
i1
] = p for i 1 and F
i
/K is Galois.
Proof. The group Gal(L/K) is a nite group of size p
r
. One of the consequences of the
Sylow theorems is the existence of a rising chain of subgroups from the trivial subgroup to
the whole group where each subgroup has index p in the next one and each subgroup is
normal in the whole group. Now apply the Galois correspondence.
The next application, which is an amusing technicality, is taken from [4].
Theorem 2.7. Let p be a prime number. If K is a eld of characteristic 0 such that every
proper nite extension of K has degree divisible by p then every nite extension of K has
p-power degree.
Aside from examples resembling K = R (where p = 2 works), elds which t the condi-
tions of the theorem are not easy to describe at an elementary level. But the technique of
proof which is used for the theorem is a rather pleasant use of group theory.
Proof. Let L/K be a nite extension. We want to show [L : K] is a power of p.
Since K has characteristic 0, L/K is separable, so we can embed L in a nite Galois
extension E/K. Since [L : K][[E : K], it suces to show [E : K] is a power of p. By the
Sylow theorems (just the rst one, actually), Gal(E/K) contains a p-Sylow subgroup, say
H. Let F = E
H
, so [F : K] is the index of H in Gal(E/K). This index is prime to p by the
denition of a Sylow subgroup, so [F : K] is prime to p. But every proper nite extension of
K has degree divisible by p, so it must be that F = K. Therefore [E : K] = [E : F] = #H
is a power of p.
Remark 2.8. Theorem 2.7 is even true when K has positive characteristic, but then one
has to consider the possibility that K has inseparable extensions and additional reasoning
is needed. See [4].
3. Primitive Elements
Galois theory provides a method to prove an element is a primitive element in a Galois
extension, as follows.
Theorem 3.1. When L/K is a nite Galois extension and L, the degree [K() : K]
is the size of the Galois orbit of . In particular, is a primitive element for L/K if and
only if () = for any = id
L
in Gal(L/K).
Proof. Since is separable over K and L/K is normal, [K() : K] is the number of roots
in L of the minimal polynomial of over K, and these roots are the orbit of under
Gal(L/K). To say L = K() requires that Gal(L/K) take through as many elements as
the degree [L : K] = #Gal(L/K), so dierent elements of Gal(L/K) must act in dierent
ways on , which is equivalent to no element xing except id
L
(() = () if and only
if (
1
)() = ).
Example 3.2. In the extension Q(
4

2, i)/Q, the number


4

2 +i has dierent values under


the 8 elements of the Galois group (see Table 2), since i and
4

2 are linearly independent


over Q (if ai +b
4

2 = 0 with rational a and b then a


2
=

2b
2
, so irrationality of

2 forces
a = 0 and b = 0). Thus Q(
4

2, i) = Q(
4

2 +i) by Theorem 3.1.


6 KEITH CONRAD
(
4

2)
4

2 i
4

2
4

2 i
4

2
4

2 i
4

2
4

2 i
4

2
(i) i i i i i i i i
(
4

2 +i)
4

2 +i i
4

2 +i
4

2 +i i
4

2 +i
4

2 i i
4

2 i
4

2 i i
4

2 i
Table 2
Example 3.3. Let F be any eld and T
1
, . . . , T
n
be indeterminates over F. The T
i
s are
roots of the polynomial
(3.1) (X T
1
)(X T
2
) (X T
n
) = X
n
s
1
T
n1
+s
2
T
n2
+ (1)
n
s
n
,
where
s
k
=

1i
1
<<i
k
n
T
i
1
T
i
k
is the sum of the products of the T
i
s taken k at a time. When S
n
acts on the eld L :=
F(T
1
, . . . , T
n
) by permutations of the variables T
i
, dierent permutations in S
n
permute the
variables in dierent ways. The xed eld L
Sn
consists of the symmetric rational functions:
those which are unchanged by any permutations of the variables T
1
, . . . , T
n
. The s
k
s are
symmetric, so F(s
1
, . . . , s
n
) L
Sn
. We will use Galois theory to show equality occurs here.
Let K = F(s
1
, . . . , s
n
), so K L
Sn
L. Thus [L : K] [L : L
Sn
] = #S
n
= n!. At
the same time, the T
i
s are all roots of the same degree n polynomial (3.1) in K[X], so
L/K is a splitting eld of a polynomial of degree n, which means [L : K] n!. Hence
[L : K] = n! = [L : L
Sn
], which forces L
Sn
= K: every symmetric rational function in
T
1
, . . . , T
n
over F is a rational function of the elementary symmetric functions s
1
, . . . , s
n
.
The extension L/K must have a primitive element, and here is an explicit choice:
T
1
T
2
2
T
n
n
. To show this works, we take our cue from Theorem 3.1 and look at the
S
n
-orbit. For any S
n
,
(T
1
T
2
2
T
n
n
) = T
(1)
T
2
(2)
T
n
(n)
.
If xes T
1
T
2
2
T
n
n
then
T
(1)
T
2
(2)
T
n
(n)
= T
1
T
2
2
T
n
n
,
so by the denition of equality in F[T
1
, . . . , T
n
] we have (1) = 1, . . . , (n) = n, so = id
L
.
Therefore if = id
L
, (T
1
T
2
2
T
n
n
) = T
1
T
2
2
T
n
n
, so T
1
T
2
2
T
n
n
is a primitive element of
L/K.
Example 3.4. Suppose K(, )/K is a nite extension such that K()/K and K()/K
are both Galois. Assume K()K() = K. We will show K(, ) = K(+) when K has
characteristic 0. (From the proof of the primitive element theorem, K(, ) = K( + c)
for all but nitely many c K, so the upshot is that we are going to prove that c = 1 is a
good choice when K() and K() are both Galois over K.)
Let H = Gal(K(, )/K( + )). We will show this group is trivial. Pick H, so
( +) = +. Therefore
() = ().
Since K() and K() are Galois over K, () K() and () K(), so () K()
and () K(). This common dierence is therefore in K() K() = K. Write
() = t, so
() = +t, () = t.
GALOIS THEORY AT WORK 7
Applying repeatedly,
j
() = + jt for all integers j. Choose j 1 such that
j
is the
identity. Then = + jt, so jt = 0. Since we are in characteristic 0 and j is a positive
integer, we must have t = 0, so () = and () = . Therefore is the identity on
K(, ).
Where in the proof did we use the condition that K has characteristic 0? Only in one
place: to know that if jt = 0 in K where j a positive integer then t = 0. The choice of
j comes from
j
= id, so j can be chosen as the order of in the Galois group, and this
order divides the size of the Galois group, which is [K(, ) : K]. The proof carries over to
elds of characteristic p as long as j 0 mod p (so j is invertible in characteristic p), and
this can be arranged by requiring that [K(, ) : K] is not a multiple of p.
If we drop the Galois condition, then of course the whole argument in Example 3.4
falls apart, and the result itself need not be true anymore. A simple example over Q is
Q(
3

2)/Q and Q(
2
3

2)/Q. Neither is a Galois extension, and the eld they generate is


Q(
3

2,
2
3

2) = Q(
3

2, ), which has degree 6 over Q, but


3

2+
2
3

2 =
3

2 (because
1 + +
2
= 0), which generates the cubic eld Q(
3

2)/Q. Can you nd an example where


the conclusion of Example 3.4 isnt true and only one of the two extensions is not Galois?
4. Relations among Galois groups
The next few theorems are concerned with Galois groups of composite extensions. We
got a glimpse of this already in Theorem 1.7.
Theorem 4.1. If L
1
and L
2
are Galois over K in a common eld then [L
1
L
2
: K] =
[L
1
: K][L
2
: K] if and only if L
1
L
2
= K. More generally, there is an embedding
Gal(L
1
L
2
/K) Gal(L
1
/K)Gal(L
2
/K) and it is an isomorphism if and only if L
1
L
2
=
K.
Proof. Consider the map Gal(L
1
L
2
/K) Gal(L
1
/K) Gal(L
2
/K) by ([
L
1
, [
L
2
).
This is a group homomorphism (check!) and the kernel is those which x pointwise L
1
and L
2
, so also L
1
L
2
. Thus the kernel is trivial so Gal(L
1
L
2
/K) embeds into the direct
product. This is an isomorphism if and only if [L
1
L
2
: K] = [L
1
: K][L
2
: K], or equivalently
[L
1
L
2
: L
2
] = [L
1
: K]. We will show this equality occurs if and only if L
1
L
2
= K.
Consider the restriction homomorphism Gal(L
1
L
2
/L
2
) Gal(L
1
/K). Any automor-
phism in the kernel xes pointwise L
2
and L
1
, so also L
1
L
2
. Thus the kernel is trivial.
What is the image in Gal(L
1
/K)? We compute this by rst computing the xed eld of the
image: the xed eld is the set of elements of L
1
which are xed pointwise by Gal(L
1
L
2
/L
2
).
The elements xed by Gal(L
1
L
2
/L
2
) are the elements of L
2
(thats the Galois correspon-
dence), so the image of the map Gal(L
1
L
2
/L
2
) Gal(L
1
/K) has xed eld L
1
L
2
, hence
the image is Gal(L
1
/L
1
L
2
). Thus Gal(L
1
L
2
/L
2
)

= Gal(L
1
/L
1
L
2
), so in particular
[L
1
L
2
: L
2
] = [L
1
: L
1
L
2
]. This is [L
1
: K] if and only if L
1
L
2
= K.
Example 4.2. Let K = Q, L
1
= Q(

2) and L
2
= Q(

3). Then L
1
L
2
= Q(

2,

3) has
degree 4 over Q and L
1
L
2
= Q, so Gal(Q(

2,

3)/Q) is isomorphic to Gal(Q(

2)/Q)
Gal(Q(

3)/Q): any four assignments of (

2,

3) to (

2,

3) extends to an automor-
phism of Q(

2,

3).
Theorem 4.1 admits a converse: if L/K is a nite Galois extension and Gal(L/K)

=
H
1
H
2
for groups H
1
and H
2
then there are elds L
1
and L
2
between L and K such that
L
1
L
2
= L, L
1
L
2
= K, and L
i
/K is Galois with Gal(L
i
/K)

= H
i
. The proof is left as an
exercise to the reader in using the Galois correspondence.
8 KEITH CONRAD
Corollary 4.3. For a nite Galois extension L/K and an arbitrary nite extension L

/K,
the extension LL

/L

is nite Galois and Gal(LL

/L

)

= Gal(L/L L

). In particular,
[LL

: L

] = [L : L L

], [LL

: K] =
[L : K][L

: K]
[L L

: K]
.
Proof. The eld diagram looks the following.
LL

4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
L
x
x
x
x
x
x
x
x
x
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
L

L L

x
x
x
x
x
x
x
x
x
K
Since L/K is nite Galois, L is a splitting eld over K for a separable polynomial f(X)
K[X]. Then LL

is a splitting eld over L

for f(X), so LL

/L

is Galois. By the ideas


from the second paragraph of the proof of Theorem 4.1, the restriction homomorphism
Gal(LL

/L

) Gal(L/K) is injective and the xed eld of its image is Gal(L/L L

), so
Gal(LL

/L

)

= Gal(L/L L

). (That L

/K may not be Galois is irrelevant.)


From the isomorphism of Galois groups, [LL

: L

] = [L : L L

]. The formula for


[LL

: K] follows by writing it as [LL

: L

][L

: K].
Example 4.4. If L/K is Galois with Galois group G and F/K is any extension with
L F = K, then LF/F is a Galois extension of F with Gal(LF/F)

= Gal(L/L F) =
Gal(L/K) = G. This provides a way to try to realize Galois groups over K as Galois groups
over nite extensions of K. We will return to this point at the end of Section 5.
Corollary 4.3 can fail if both extensions are not Galois. If K = Q, L = Q(
3

2) and
L

= Q(
3

2) then [LL

: L

] = 2 and [L : L L

] = 3.
In the proof of Corollary 4.3, the assumption that L

/K is nite was only used at the


very end. The rest of the argument goes through even if L

/K is innite, and here is a


worthwhile application of that generality.
Example 4.5. Let L/K be a nite Galois extension and let L

= K(u) be the rational


function eld over K in the indeterminate u. Then LL

= L(u). The intersection L

L =
K(u)L is K because the only elements of K(u) which are algebraic over K are the elements
of K themselves. Therefore Corollary 4.3 says the extension L(u)/K(u) of rational function
elds is Galois with Gal(L(u)/K(u))

= Gal(L/K). Explicitly, the way this isomorphism
works is that elements of Gal(L/K) act on L(u) by simply acting on coecients, keeping u
xed. For instance, C(u) = R(u)(i) is a quadratic extension of R(u) and Gal(C(u)/R(u))
is the identity and complex conjugation acting on coecients.
GALOIS THEORY AT WORK 9
Returning to Theorem 4.1, it is natural to ask how the Galois group of a composite of
Galois extensions lies in the direct product of the Galois groups. What is the image of the
embedding?
Theorem 4.6. When L
1
and L
2
are nite Galois extensions of K, the image of the
embedding Gal(L
1
L
2
/K) Gal(L
1
/K) Gal(L
2
/K) which sends to ([
L
1
, [
L
2
) is
(
1
,
2
) :
1
=
2
on L
1
L
2
.
Proof. Let H = (
1
,
2
) Gal(L
1
/K) Gal(L
2
/K) :
1
=
2
on L
1
L
2
. The image of
the embedding on Gal(L
1
L
2
/K) lies in H. We will check the image has the same size as
H, so the groups coincide.
To count #H, we count for each
1
Gal(L
1
/K) how large the inverse image is of

1
[
L
1
L
2
under the restriction map Gal(L
2
/K) Gal(L
1
L
2
/K). This is a surjective
homomorphism with kernel Gal(L
2
/L
1
L
2
), so
#H = #Gal(L
1
/K) #Gal(L
2
/L
1
L
2
)
= [L
1
: K][L
2
: L
1
L
2
]
=
[L
1
: K][L
2
: K]
[L
1
L
2
: K]
= [L
1
L
2
: K] by Corollary 4.3
= #Gal(L
1
L
2
/K).

When L
1
L
2
= K in Theorem 4.6 we recover part of Theorem 4.1.
5. The inverse Galois problem
The determination of Galois groups goes in two directions. First, if we are given a
separable irreducible polynomial over a eld then we can ask for methods to compute the
Galois group of the splitting eld. For example, the calculation of a Galois group over
a nite eld is very easy: once we know the degree of the extension we know the Galois
group explicitly as a cyclic group. The situation of base eld Q is more dicult. Methods
from algebraic number theory can be brought to bear on the problem of computing Galois
groups of irreducible polynomials over Q, but the situation is really satisfactory insofar as
computations of new individual examples are concerned only for extensions of moderately
small degree [2, '14.8].
The second problem about Galois groups, which is what we will discuss here, is called the
inverse Galois problem. It asks which nite groups arise as Galois groups of a given eld.
The choice of base eld is important, because if we leave that up for grabs then one can
show every nite group is the Galois group of some eld extension. This will be a pretty
use of Cayleys theorem embedding every nite group in some symmetric group.
Theorem 5.1. Every nite group is the Galois group of some nite Galois extension in
any characteristic.
Proof. Let F be a eld. By the symmetric function theorem,
F(T
1
, . . . , T
n
)
Sn
= F(s
1
, . . . , s
n
),
where the s
i
s are the elementary symmetric functions of the T
i
s. Therefore (Artins the-
orem) F(T
1
, . . . , T
n
)/F(s
1
, . . . , s
n
) is Galois with Galois group S
n
. Any nite group G
embeds into some symmetric group S
n
, and thus can be interpreted as a Galois group.
10 KEITH CONRAD
Theorem 5.1 leaves a lot to be desired: to realize G as a Galois group the proof tells
us to embed G S
n
and then G is the Galois group of the extension E/E
G
where E =
F(T
1
, . . . , T
n
) and G acts on E by permuting the variables using the embedding of G in
S
n
. The base eld E
G
= F(T
1
, . . . , T
n
)
G
of this Galois extension is rather mysterious!
A famous open question, and what is usually understood by the phrase inverse Galois
problem, asks if every nite group is a Galois group over Q. Fixing the base eld to be Q
makes the question a real challenge. It is believed the answer is armative, i.e., that every
nite group should arise as a Galois group over Q.
The inverse Galois problem over nite elds is easy: only cyclic groups can occur, and
every cyclic group does occur (over any nite eld). But the inverse Galois problem with
base eld Q is still open. Schur, in the 19th century, used number-theoretic techniques to
realize S
n
and A
n
as Galois groups over Q using splitting elds of classical polynomials
(like the truncated exponential series

n
k=0
X
k
/k!, whose splitting eld over Q is S
n
unless
4[n, when it is A
n
). Towards the end of the 19th century, Hilbert introduced geometric
methods into the subject when he showed that if a nite group could be realized as a Galois
group over Q(T) then it could be realized as a Galois group over Q (in innitely many
ways) by suitable specializations. The buzzword here is Hilberts irreducibility theorem
[3]. For instance, X
n
X T is irreducible over Q(T) and it turns out that its Galois
group over Q(T) is S
n
, so Hilberts work implies ineectively that for many rational
numbers t the polynomial X
n
X t is irreducible over Q and the Galois group is S
n
.
(With more work, the particular choice t = 1 goes through [6], which incidentally provides
a nice concrete example for the proof of Theorem 2.2.)
Its worth emphasizing why the realization of the S
n
s as Galois groups over Q does not
settle the inverse Galois problem over Q even though every nite group is a subgroup of
some symmetric group. The Galois correspondence goes backwards, so realizing a group as
a Galois group over a eld realizes its subgroups as Galois groups over intermediate elds
(same top eld, changing base eld). It is the quotients of the Galois group that are realized
as Galois groups over the base eld. Alas, the symmetric groups have very few quotient
groups: when n 5, the only normal subgroups of S
n
are (1), A
n
, and S
n
, so the only
quotient groups of S
n
are trivial, of size 2, or S
n
itself.
By number-theoretic methods, Shafarevich proved in the 1950s that every solvable group
is a Galois group over Q. (The proof had an error concerning the prime 2, but this was
later repaired.) Most recent work on the inverse Galois problem has taken the geometric
approach inspired by Hilberts ideas. As we said above, Hilbert showed that the inverse
Galois problem over Q would be settled (by specialization) if we can settle the inverse
Galois problem over Q(T). This is good, because the inverse Galois problem over the
related eld C(T) is settled: every nite group occurs as a Galois group over C(T). This is
based on Riemanns existence theorem from the theory of Riemann surfaces (these surfaces
provide a geometric interpretation of nite extension elds of C(T)). See [8] (esp. Chapter
1) for a basic introduction to these ideas and [7] for a general survey.
As a nice use of Corollary 4.3, lets lift the inverse Galois problem from Q to its nite
extensions.
Theorem 5.2. If every nite group can be realized over Q then every nite group can be
realized over any nite extension of Q.
Proof. Pick a nite group G and a nite extension F/Q. We want to realize G as a Galois
group over F. Following Example 4.4, if we can realize G as a Galois group of an extension
GALOIS THEORY AT WORK 11
L/Qwhere LF = Q, then LF/F is Galois and Gal(LF/F)

= Gal(L/LF) = Gal(L/Q)

=
G. Thus G is realized over F. So the problem is to get a realization of G over Q where the
intersection of the top eld and F is Q.
There are only nitely many elds between Qand F (including Qand F). Let the number
of them be n. By hypothesis all nite groups are realizable over Q, so in particular the
n-fold direct product G
n
is a Galois group over Q. Let G
n
= Gal(E/Q). (We should write
G
n
= Gal(E/Q), but we will work with equality rather than isomorphism for simplicity.)
Inside of G
n
are the normal subgroups N
i
(1 i n) where the ith coordinate is 1 and
there is no restriction in other coordinates. So N
i

= G
n1
and G
n
/N
i

= G. Let E
i
be the
subeld of E corresponding to N
i
, so E
i
/Q is Galois with Gal(E
i
/Q)

= G
n
/N
i

= G. So
each of the elds E
1
, . . . , E
n
realizes G as a Galois group over Q. We are going to prove by
counting that at least one of the elds E
1
, . . . , E
n
intersects F in Q. Then the composite
of F with that E
i
will realize G as a Galois group over F.
For i = j, E
i
E
j
= Q. Indeed, E
i
E
j
is the largest subeld of E that lies in E
i
and
E
j
, so by the Galois correspondence Gal(E/E
i
E
j
) is the smallest subgroup of Gal(E/Q)
containing Gal(E/E
i
) = N
i
and Gal(E/E
j
) = N
j
. That means Gal(E/E
i
E
j
) = 'N
i
, N
j
`.
From the denition of N
i
and N
j
, the subgroup they generate in G
n
is G
n
. Thus E
i
E
j
= Q.
Now associate to each E
i
the subeld E
i
F of F. There are n elds E
i
and n subelds
of F. If the correspondence from the subelds E
i
to the intersections E
i
F is a bijection
then E
i
F = Q for some i and were done. If the correspondence is not injective then
we have a repeated intersection E
i
F = E
j
F for some i = j. But any element in both
intersections is in E
i
E
j
= Q, which means E
i
F = E
j
F = Q and again were done.
The only special feature about nite extensions of the rational numbers which was used
in the proof is that there are only nitely many elds between it and Q. So if every nite
group arises as a Galois group over a eld K and F/K is a nite extension with nitely
many intermediate elds (e.g., F/K is separable) then the proof of Theorem 5.2 shows every
nite group arises as a Galois group over F.
6. What Next?
There are two important aspects of eld extensions which are missing by a study of Galois
theory of nite extensions, and we briey address them:
(1) Galois theory for innite extensions
(2) transcendental extensions
A eld extension L/K of innite-degree is called Galois when it is algebraic, separable,
and normal. That means each element of L is the root of a separable irreducible in K[X]
and that every irreducible in K[X] with a root in L splits completely over L. A typical
example of an innite Galois extension of Q is Q(
p
) =

n1
Q(
p
n), the union of all
p-th power cyclotomic extensions of Q, where p is a xed prime. Even if an algebraic
extension L/K is innite, any particular element (or nite set of elements) in L lies in
a nite subextension, so knowledge of nite extensions gets used to understand innite
extensions. In fact, another way of describing an innite Galois extension is that it is a
composite of nite Galois extensions.
In an innite Galois extension L/K, one still denes the Galois group Gal(L/K) as
the group of K-automorphisms of L, and one can associate subgroups of the Galois group
to intermediate elds and an intermediate (xed) eld to each subgroup of the Galois
group. However, this correspondence is no longer a bijection! This was rst discovered
12 KEITH CONRAD
by Dedekind, who saw in particular examples that dierent subgroups of a Galois group
could have the same xed eld. So it looks like Galois theory for innite extensions breaks
down. But it isnt really so. Krull realized that if you put a suitable topology on the
Galois group then a bijection can be given between all intermediate elds and the closed
subgroups in that topology. (See [1] and [5].) This not only rescued Galois theory for
innite extensions, but gave a dynamic new impetus to the study of topological groups.
Anyone wishing to understand innite Galois theory should learn about the p-adic numbers
and their topological and algebraic structure, as this is used in the simplest examples of
innite Galois groups.
Turning away from Galois theory, the next most important class of eld extensions are
transcendental extensions. These are eld extensions in which every element of the top
eld is transcendental (that is, not algebraic) over the bottom eld. The simplest example
of a transcendental extension of a eld F is the eld F(u) of rational functions over F
in an indeterminate u, or more generally the eld F(u
1
, . . . , u
n
) of rational functions in n
independent variables over F. Since these extensions of F have innite degree, the notion
of eld degree is no longer important. In its place is the concept of transcendence degree,
which is a nonlinear analogue of a basis and measures how transcendental the extension
is. The need to understand transcendental eld extensions is not driven for its own sake,
but for other areas of mathematics, such as algebraic geometry.
References
[1] J. Bastida, Field extensions and Galois theory, Addison-Wesley, Reading, MA 1984.
[2] D. Dummit and R. Foote, Abstract Algebra, 3rd ed., Wiley, New York, 2004.
[3] C. Hadlock, Field Theory and its Classical Problems, Math. Assoc. America, Washington, D.C., 1978.
[4] I. Kaplansky, Fields and Rings, 2nd ed., Univ. of Chicago Press, Chicago, 1972.
[5] P. Morandi, Field and Galois theory, Springer-Verlag, New York, 1996.
[6] H. Osada, The Galois groups of the polynomials X
n
+ aX
l
+ b, J. Number Theory 25 (1987), 230238.
[7] J-P. Serre, Topics in Galois theory, Jones and Bartlett, Boston, MA, 1992.
[8] H. Volklein, Groups as Galois Groups an Introduction, Cambridge Univ. Press, Cambridge, 1996.

You might also like