CHAPTER IX

APPLICATIONS OF GALOIS THEORY

1. Finite Fields
Let F be a finite field. It is necessarily of nonzero characteristic p and its prime field is the field with p
elements Fp . Since F is a vector space over Fp , it must have q = pr elements where r = [F : Fp ]. More
generally, if E F are both finite, then E has q d elements where d = [E : F ].
As we mentioned earlier, the multiplicative group F of F is cyclic (because it is a finite subgroup of the
multiplicative group of a field), and clearly its order is q 1. Hence each non-zero element of F is a root of
the polynomial X q1 1. Since 0 is the only root of the polynomial X, it follows that the q elements of F
are roots of the polynomial X q X = X(X q1 1). Hence, that polynomial is separable and F consists of
the set of its roots. (You can also see that it must be separable by finding its derivative which is 1.) We
may now conclude that the finite field F is the splitting field over Fp of the separable polynomial X q X
where q = |F |. In particular, it is unique up to isomorphism. We have proved the first part of the following
result.
Proposition. Let p be a prime. For each q = pr , there is a unique (up to isomorphism) finite field F
with |F | = q.
Proof. We have already proved the uniqueness. Suppose q = pr , and consider the polynomial X q X
Fp [X]. As mentioned above Df (X) = 1 so f (X) cannot have any repeated roots in any extension, i.e. it
is separable. Let F be its splitting field over Fp . We have F = Fp [x1 , . . . , xq ] where x1 , . . . , xq are the q
distinct roots of f (X). However, it is not hard to see that the set {x1 , . . . , xq } is in fact already a field. For,
x is a root of f (X) = X q X if and only if xq = x, and the fact that raising to the qth power is a ring
homomorphism in characteristic p (i.e., (x + y)q = xq + y q and (xy)q = xq y q ) tells us that the set of roots
is a subring. If x 6= 0, we have (x1 )q = (xq )1 = x1 , so every element in this subring is invertible in the
subring. Hence F is the set of roots of f (X) and has q elements as claimed. 
We want to fit all these finite fields in the same field and show how they are related to one another. To
this end, we shall use a result to be proved later. Namely, if F is any field, then we shall show later that
there is an algebraic extension F which is algebraically closed and which is unique up to F -isomorphism.
Such an extension is called an algebraic closure of F , or with abuse of terminology the algebraic closure of
F . Let p denote the algebraic closure of the prime field Fp . For any q = pr there is a unique subfield of
p isomorphic to every field with q elements, namely the splitting filed of X q X over Fp . We shall denote
this instance of a field with q elements by Fq .
The finite fields Fq are coherently related. Namely, first suppose that Fq Fq0 . The the latter may be
viewed as a vector space over the former of dimension d = [F0q : Fq ]. That is, F0q is isomorphic to a direct
0
sum of d copies of Fq a, whence q 0 = |F0q | = |Fq |d = q d , or pr = prd . It follows that r divides r0 .
0
Conversely, suppose r divides r0 , i.e. r0 = rd, q = pr , and q 0 = pr = q d . We claim that Fq Fq0 . To
d
see this note that the latter is the splitting field over Fp of X q X and the former is the splitting field of
X X. However, dividing yields
1

X q 1 1
d d
Xq X
= = X (q1)(k1) + X (q1)(k2) + + X q1 + 1
X X
q X q1 1
Typeset by AMS-TEX
89
90 IX. APPLICATIONS OF GALOIS THEORY

where k = (q d 1)/(q 1) = q d1 + q d2 + + q + 1. (Note the confusing double use of the formula for
0
a geometric sum!) It follows that X q X divides X q X so any splitting field of the latter contains a
splitting field of the former. Hence, by the uniqueness of splitting fields in p , Fq0 Fq as claimed.
Remark. p is in fact the union of the finite subfields Fq where q = pr . For, it is algebraic over Fp by
definition, so any element is in a finite extension of Fp , hence in a finite subfield of p . However, the above
analysis assures us that the fields Fq are the only such subfields.
We are now in a position to calculate the Galois group G(Fq0 /Fq ) in the case q 0 = q d . First, define
q : p p by q (x) = xq . It is not hard to see that q is a ring homomorphism, so at the very least it is
a monomorphism. It is in fact also an epimorphism and hence an automorphism of p . (See the exercises.)
Theorem. Let q 0 = q d where q = pr for a given prime p. The restriction of q to Fq0 is an automorphism
of Fq0 and it generates the Galois group G(Fq0 /Fq ) which is cyclic of order d.
Proof. First note that q fixes Fq which is the set of elements in p satisfying xq = x. It is also not
hard to see that it carries Fqd into itself. Since it is a monomorphism and that field is finite, its restriction
is an automorphism of Fqd . Moreover, we have

i i
iq (x) = xq so q = Id xq = x for all x Fqd .

If this were to happen for some i < d, Fq would in fact have q i elements instead which is nonsense. Hence,
the restriction of q has order d. On the other hand, [Fq0 : Fq ] = d, so by Galoiss main theorem |G(Fq0 /Fq )|
has order d and hence is generated by q . 

Exercises.
1. Verify explicitly that q : p p preserves sums, products, and the identity element 1 p .
2. Show that q : p p is an epimorphism, hence an automorphism. (You already know from the
discussion in the text that it is a monomorphism.)
3. (a) An automorphism of p is specified if we know its restriction d to every Fq with q = pd . Also,
0
if d0 = cd, q = pd , and q 0 = pd , then the restriction of d0 to Fq must be d . Suppose conversely, that we
are given for each positive integer d an automorphism d of Fpd , and that the family of all d satisfies the
consistency condition just enunciated. Show that there is an automorphism of p which restricts to d on
Fpd for every d.
(b) Is every such a power of ; i.e., is G(p /Fp ) cyclic with generator ? Hint: This is not easy.

2. Extension of the base field

Let K F be a finite, normal separable extension. We often want to know what happens to the Galois
group when we extend F to a larger field L which may not even be algebraic over F . For example, the
extension Q[i] Q is normal and separable with Galois group cyclic of order 2. Similarly, we can say the
same for R[i] R, and in fact there is a natural way to identify the two Galois groups. The first problem in
dealing with this situation in general is that there is no reason even to assume that K and L can be imbedded
in the same field. Assume then that there is a field which contains both K and L. (In the example, all the
fields are contained in C.) We denote by KL the smallest subring of which P contains both K and L. (This
definition is not quite standard!) KL as before is the set of finite sums xi yi with xi K and yi L and
it would ordinarily not be a subfield of ; we would have to form fractions in order to get a field. However,
if K F is finite then K = F [x1 , . . . , xn ] for appropriate elements of K, so KL = L[x1 , . . . , xn ] is in fact
finite over L and is thereby a field. In this case we shall call KL the compositum of K and L. Note that
we could have extensions K 0 F and L0 F which are F -isomorphic to the previous extensions with K 0
and L0 contained in some field 0 but with K 0 L0 not isomorphic to KL. This seems paradoxical, but it can
happen since the construction of KL depends to some extent on the common enclosing field . We shall
analyze this situation in more detail when we discuss ring theory later in this course.
 3. EXTENSION OF THE BASE FIELD 91

Theorem. (Natural Irrationalities). Let K F be a finite, normal, separable extension, and let L
be an extension of F such that K and L are contained in a common field. Then KL L is a finite,
normal, separable extension. Moreover, restricting L-automorphisms of KL to K yields a monomorphism
G(KL/L) G(K/L), and the image is G(K/K L).

LK 1
 
 
 
1 K L G(KL/L)
 
 
 
G(K/K L) K L
| |
G(K/F ) F

Proof. First note that KL is normal over L. For, we can write K = F [x1 , . . . , xk ] as generated by the
roots of a separable polynomial with coefficients in F ; hence KL = L[x1 , . . . , xk ] is also such a splitting field
over L. Let G(KL/L). The restriction of to K certainly fixes F and by normality it must carry K
into itself; so it is an element 0 G(K/F ). is certainly completely determined by its effect on x1 , . . . , xk
and since these are in K, it is determined by its restriction 0 to K. Hence 7 0 is a monomorphism. An
element x K is fixed by the subgroup which is the image of this monomorphism if and only if it is fixed by
all G(KL/L), i.e., if and only if x L. But since x K in any case, this holds if and only if x K L.
By Galoiss Main Theorem, the image subgroup must be G(K/K L). 

Example 1. Let = 3 2 and let = e(2i)/3 as earlier. Let L = Q(). We use the fact that is not
algebraic to conclude that L K = Q in this case. (Can you prove it?) Hence the diagram looks like

L[,
 ] 1
 
 
 
1 Q[,
 ] L S3
 
 
 
S3 Q

Example 2. Let K Q be a finite, normal, separable extension contained in C. Then K R is the

maximal real subfield of K. If K is real then KR = R and nothing interesting occurs. If K is not contained in
R, then we must have KR = C since it its a proper extension of R and C is the only such algebraic extension.
It follows that G(K/Q) has a subgroup of order 2 generated by the restriction of complex conjugation.

Exercises.
1. (a) Let K be a field and let P and Q be groups of automorphisms of K. Show that K P Q = K P K Q .
(b) Let K F be a field extension and let L and M be intermediate fields. Show that G(K/LM ) =
G(K/L) G(K/M ).
(c) Suppose that K F is a finite normal separable extension. Conclude that in the Galois correspondence
between subgroups of G(K/F ) and intermediate subfields, intersections of subgroups correspond to composita
of subfields and products of subgroups correspond to intersections of subfields.
2. Prove that if K and N are finite normal extensions of F both contained in the same field , then the
compositum KN is normal over F .
92 IX. APPLICATIONS OF GALOIS THEORY

3. Cyclotomic extensions
A root of the polynomial X 1 for some n > 0 is called a root of unity. For example, in C, if we put
n

= 2/n, then X n 1 has the n roots eik , k = 0, 1, . . . , n 1. These appear in the complex plane as the
vertices of a regular n-gon inscribed in the unit circle.

Let the base field have characteristic p. If p | n then

D(X n 1) = nX n1 = 0

so X n 1 is not separable. Conversely, if p does not divide n, then X n 1 and nX n1 clearly have no
roots in common, so X n 1 is separable. For this reason, we shall always assume when discussing nth roots
of unity that gcd(p, n) = 1 if p > 0. Of course, if the characteristic is 0, there is no need for any extra
assumption.
Let K be an extension of F . The roots of X n 1 in K are distinct (assuming as above that gcd(p, n) = 1),
and it is clear that they form a subgroup of K under multiplication. Hence, the set of roots forms a cyclic
group of order n. Assume further that X n 1 splits completely in K. Then, the order of this group is
n. In that case, a generator of the group of nth roots of unity in K is called a primitive nth root of unity.
Every root is then a power i with 0 i n 1, and such a power is also primitive (i.e. a generator) if and
only if gcd(i, n) = 1. It follows that the number of primitive roots is (n) where is the Euler function.
Moreover, again since the roots are all powers of , we have in K[X]

Y
n1
Xn 1 = (X i )
i=0

Note that this splitting already takes place in F [] which is a splitting field for X n 1. F [] is called a
cyclotomic extension of F . Notice that the existence of a primitive nth root of unity in some extension
of F implies that X n 1 has distinct roots, so it is separable and necessarily n is relatively prime to the
characteristic of F .
Theorem. Let n be relatively prime to the characteristic of F . Let be a primitive nth root of unity.
Then G(F []/F ) is isomorphic to a subgroup of U (Z/nZ) (the group of units of Z/nZ) and hence is abelian
of order dividing (n).
Proof. G = G(F []/F ) is completely determined by its effect on . Since is an automorphism,
() must also be a primitive root so we have () = i where i = i is relatively prime to n. Define a map
: G U (Z/nZ) by () = i mod n. As just noted, is one-to-one. It is also a homomorphism since if
() = i and () = j , then () = ( i ) = ()i = ( j )i = ji . Hence ( ) = ji mod n = ( )(). 
Remarks. As we discovered last quarter (in an exercise), U (Z/nZ) is often cyclic. In fact that will be
true if n is an odd prime power or twice an odd prime power or if n = 2 or 4. Hence, in those cases the
Galois group of the cyclotomic extension is also cyclic. In all other cases, the group of units is abelian but
non-cyclic, so the Galois group need not be cyclic. (Of course, depending on F , it could turn out to be
cyclic.)
 4. RADICAL EXTENSION 93

Theorem. If is a primitive nth root of unity in some extension of Q, then G(Q()/Q)

= U (Z/nZ).
In particular, it is of order (n).
Proof. By Galoiss Main Theorem, it suffices to show that [Q() : Q] = (n). This turns out to be
rather difficult.
Let (X) denote the minimal polynomial of over Q. Since is a root of X n 1, it follows that
X 1 = (X)(X) where by Gausss Lemma, we may assume that (X) and (X) are monic with
n

coefficients in Z. We shall show that every primitive root i (with gcd(i, n) = 1) is also a root of (X). It
will follow that [Q() : Q] = deg (X) (n). Since we already know [Q() : Q] | (n) from the theorem
just proved, it will follow that [Q() : Q] = (n) as claimed.
First note that it suffices to show that for each q prime and not dividing n, q is a root of (X). For,
let i = qi0 where q is such a prime. If we already know q is a (primitive) root, then we can apply the same
0
reasoning to it and conclude by induction on i that ( q )i = i is such a root.
Suppose then that ( ) 6= 0 (where q is a prime not dividing n). It follows that q is a root of (X)
q

(since it is certainly a root of X n 1) so is a root of (X q ). Since (X) is the minimal polynomial of ,

we have (X q ) = (X)H(X), and as above H(X) is monic with coefficients in Z. Hence,

(X)q (X q ) (X)H(X) mod q.

Consider the images of the polynomials (X) and (X) in Z/qZ[X]. Because of the above congruence, they
must have a common irreducible factor in Z/qZ[X] so the image of the product (X)(X) has a repeated
irreducible factor in Z/qZ[X] so it is not separable. However, this product is X n 1 which is separable in
Z/qZ[X] since q does not divide n. It follows that ( q ) = 0 as claimed. 
Corollary. The minimal polynomial over Q of a primitive nth root of unity is
Y
n (X) = (X i )
0<i<n
(i,n)=1

Note. n (X) is called the nth cyclotomic polynomial.

Proof. The argument in the proof shows that n (X) divides the minimal polynomial (X), but they
have the same degree. 
Corollary. Let n be a primitive nth root of unity. Then if gcd(n, m) = 1, we have Q(nm ) =
Q(n )Q(m ) and Q(n ) Q(m ) = Q.
Proof.
Exercise.

Exercises.
1. Let gcd(n, m) = 1.
(a) Show that Q(nm ) = Q(n )Q(m ). Hint: Show that the compositum contains a primitive nth root of
unity.
(b) Show that Q(n ) Q(n ) = Q. Hint: Use U (Z/nmZ) = U (Z/nZ) U (Z/mZ).

4. Radical Extension
Let F be a field of characteristic relatively prime to n and consider the polynomial X n a where a F
is not zero. Note that with this assumption on n, that polynomial is separable. Let K be a splitting field of
X n a over F . We want to determine the Galois group G(K/F ).
Let K be one root of X n a. If K is any other root, n = n = a / is an nth root of unity.
Since K contains n distinct roots of X n a, it must contain n distinct nth roots of unity, so it contains a
primitive nth root of unity . Clearly, K = F (, ), and the roots of X n a are , , 2 , . . . , n1 .
94 IX. APPLICATIONS OF GALOIS THEORY

F (, )

F () F ()

F
The relation among the subfields may be somewhat different than the diagram suggests. For example, we
could have F (). For example, take F = Q, n = 4, and a = 4. Then

X 4 + 4 = (X 2 2x + 2)(X 2 + 2x + 2)

so we may take = 1 + i which is a root of the first factor. However, in this case = i is a primitive 4th
root of unity. More trivially, we could have F . At the other extreme, we could have F , and this is
the case we now investigate.
Theorem. Let F contain a primitive nth root of unity . If is a root of X n a where a 6= 0 is in
F , then F () is normal and separable over F , [F () : F ] | n, and G(F ()/F ) is cyclic. Conversely, if K
is finite, normal, and separable over F (containing a primitive nth root of unity) and G(K/F ) is cyclic of
order d | n, then K = F () where d F .
Proof. Suppose is a root of X n a as in the statement of the theorem. Let G(F ()/F ); since ()
is another root of X n a, and since we know all its roots by the above analysis, it follows that () = i()
where 0 i() n 1. Define : G(F ()/F ) Z/nZ by () = i() mod n. It follows easily from the
fact that F that is a group homomorphism, and since is completely determined on F () by (),
is a monomorphism. Hence, G(F ()/F ) is cyclic of order dividing n.
Conversely, suppose that G(K/F ) is cyclic of order d | n. Let = n/d . is a primitive dth root of unity.
(Why?) Let generate G(K/F ) and for each x K form

hx, i = x + 1 (x) + 2 2 (x) + + (d1) d1 (x)

(called the Lagrange resolvent of x.) We have

(hx, i) = (x) + 1 2 (x) + 2 3 (x) + + (d1) d (x)

= (x) + 1 2 (x) + 2 3 (x) + + (d1) x
= ( 1 (x) + 2 2 (x) + 2 3 (x) + + (d) x)
= (x + 1 (x) + 2 2 (x) + + (d1) d1 (x))

i.e. (hx, i) = hx, i. Thus if = hx, i then () = so (d ) = ()d = d so since generates

G(K/F ), d F . Hence, is a root of X d a where a = d .
Suppose that 6= 0. Let m(X) be the minimal polynomial of over F . We have i () = i is a root of
m(X) for every i = 0,1, . . . , d 1, and since these are all distinct, deg m(X) d. However, m(X) | X d a
so deg m(X) = d, and K = F () as claimed. (Note also that this shows that X d a = m(X) is irreducible.)
To complete the proof, it suffices to show that = hx, i 6= 0 for at least one x K. But, this follows
from Artins Theorem on independence of characters because if

x + 1 (x) + 2 2 (x) + + (d1) d1 (x) = 0

for all x K, the automorphisms Id, , 2 , . . . , d1 form a dependent set of characters. 

If we start with the polynomial X n a, it might not be irreducible over F so that |G(F ()/F )| could be
a proper divisor of n. However, if n is prime, either X n a splits completely in F or it is irreducible. This
follows from
 5. RADICAL EXTENSION 95

Proposition. Let p be prime. Let a F and suppose a 6= bp for any b F . Then X p a is irreducible
over F .
Proof. First suppose that p is not the characteristic of F . Let K = F (, ) be a splitting field of X p a
where p = a and is a primitive pth root of unity as above. In K[X], we have

Y
p1
Xp a = (X i )
i=0

so the minimal polynomial m(X) of is a product of factors of the form X i . Hence, the product of its
roots (except for sign, its constant term) which is in F is of the form c = r k for some r and k. If r < p
(i.e. m(X) 6= X p a), then we can find s and t such that ps + rt = 1. Hence,

= ps rt = as (c/ k )t = as ct kt

so it follows that kt F , and since ( kt )p = p = a, this contradicts the hypothesis of the Proposition.
Suppose instead that p is the characteristic of F . Then we know that X p a splits in its spitting field into
(X )p . The minimal polynomial of must be of the form (X )i for some i. If i < p, then considering
the term iX i1 in (X )i , we can see that F , contrary to the hypothesis. 
In the case n = p is prime and is not in F , we can extend the analysis a bit further.

F (, )

F () F ()

F
In the above diagram, [F () : F ] = p, and by our previous discussion of cyclotomic extensions, [F () :
F ] | (p) = p 1. It follows that F () F () = F (since its degree must divide both p and p 1.) Let H be
the subgroup of the Galois group with fixed field F (), and let K be the subgroup with fixed field F (). The
first is normal, but the second need not be normal. (See the Exercises.) ) The subgroup H K fixes both
and so it fixes F (, ). Hence H K = {Id}. On the other hand, the fixed field of HK must be contained
both in F () and F () so it is F ; hence HK = G(F (, )/F ). It follows that if is not in F , the Galois
group is the semidirect product of the normal subgroup H with the non-normal subgroup K. Furthermore,
by natural irrationalities, H is isomorphic to G(F ()/F ) which we know to be cyclic of order dividing p 1.
In addition, by the theorem proved at the beginning of this section G(F (, )/F ()) is necessarily cyclic of
order p.

Exercises.
1. Let p be a prime other than the characteristic of of F . Let a 6= 0 be an element of F which is not a pth
power of an element of F . Consider a splitting field L = F (, ) of X p a where p = a and is a primitive
pth root of unity. Show that F () is normal over F if and only if F .
2. Find the Galois groups of each of the splitting field of each of the following polynomials over the indicated
base fields. You need notrestrict your attention just to the results in the preceding section.
(a) X 4 2 over Q, Q[ 2], and F3 .
(b) (X 3 2)(X 3 5) over Q, Q[] (where is a primitive third root of unity), and R.
(c) X 4 + X 2 + 1 over Q.
96 IX. APPLICATIONS OF GALOIS THEORY

5. Solvability by Radicals
Cubic Equations.
If f (X) = X 3 + aX 2 + bX + c is an irreducible cubic polynomial with coefficients in Q, then by using a
transformation of the type X X u with suitable u, we may assume the coefficient of X 2 is 0. (See the
Exercises.) Hence, we shall suppose f (X) = X 3 + pX + q is irreducible in Q[X]. Let K be the splitting field
of f (X) over Q. G = G(K/Q) must be a isomorphic to a subgroup of S3 which is transitive on the roots
x1 , x2 , x3 of f (X). Identify G with that subgroup. Since the only such transitive subgroups of S3 are S3
itself and the alternating group A3 (which is cyclic of order 3), it follows that these are the only possibilities
for G. Let
= (x1 x2 )(x1 x3 )(x2 x3 ).
is fixed by every element of A3 , and its square D = 2 is fixed by every element of S3 . In fact, using

x1 x2 + x1 x3 + x2 x3 = p
x1 x2 x3 = q

one may verify the identity

D = (4p3 + 27q 2 ).
(Simply expand both sides in terms of x1 , x2 , x3 .)
D is called the discriminant of the polynomial f (X).
Since is fixed under A3 , it follows that Q() is the fixed field of A3 G. Hence, we conclude that
Q G = A3 . Otherwise, it is S3 . In any case G(K/Q()) = A3 is cyclic of order 3. Extend Q() by
adjoining , a primitive cube root of unity. [Q(, ) : Q()] is 1 or 2 since satisfies a quadratic equation
over Q. Since [K : Q()] = 3, it follows that [K Q(, ) : Q()] divides both 2 and 3, so it equals 1. Hence,
K Q(, ) = Q(), and by natural irrationalities G(K()/Q(, )) is also cyclic of order 3.

K() 1


y y
1 K Q(, ) A3


y y
A3 Q()


y y degree 1 or 2
G Q

It now follows from our theorem on cyclic extensions that K() can be obtained from Q(, ) by adjoining
an element where 3 Q(, ). Hence, every root of f (X) can be expressed as a polynomial in , , and
.
Such a description of the roots is called a solution by radicals because each constituent in the solution
is either a root of unity or a cube root of something already obtained. But note that we may need to go
beyond the splitting field of f (X) to express it that way.
(Note also that in thiscase is actually a root of a quadratic equation, so it may be expressed in terms
of the quadratic radical 3.)
A method for solving of cubic equations by radicals was obtained by Italian mathematicians in the 16th
century. It is called Cardanos method because he included it in what was probably the first relatively modern
algebra textbook, although it had been discovered earlier by the mathematicians del Ferro and Tartaglia.
Cardano himself never claimed credit for the method, which he attributed to Tartaglia, but he incurred the
latters wrath for making it public. Thus Cardano treated mathematics in the modern fashion, as knowledge
which should be freely available with suitable credit being given to those who made the discoveries, while
 5. SOLVABILITY BY RADICALS 97

Tartaglia treated it a a proprietary product to be used for his commercial benefit. (He ran a school to teach
the children of merchants elementary arithmetic, and being able to solve cubics added to his prestige and
presumably his profits.) We find echoes of such attitudes in modern societys use of science and technology.
Here is the method. First, let
p p
= (p/3)3 + (q/2)2 = D/108 = 3 /18.

(Since there are two square roots of any rational number, choose the positive square root if is real and a
positive multiple of i if is imaginary.) Next, choose and 0 in C such that

3 = (q/2) + ,
(0 )3 = (q/2) .

There are of course three choices for since each of and 2 would do as well, and similarly for 0 . Note
however that
(0 )3 = (q/2)2 2 = (p/3)3
so again given the freedom we have to vary a cube root, we may as well assume 0 = p/3 or 0 = p/3
Then the three roots of X 3 + pX + q = 0 in C are

x1 = + 0
x2 = + 2 0
x3 = 2 + 0

One way to prove this is to show that with the given values of the roots, the coefficients of

(X x1 )(X x2 )(X x3 )

are the same as the coefficients of f (X). The algebra necessary to accomplish this is rather involved and might
better be postponed until we have discussed elementary symmetric functions in a later section. However, a
more direct derivation of these formulas is outlined in the exercises.
It is interesting to note in this connection that since D could be positive, the formulas could easily involve
non-real complex radicals even if ultimately the solutions are real. (See the exercises.) This observation
may have been the first convincing evidence that mathematics really required imaginary numbers for further
progress.
A related more complicated method for solving quartics by radicals was discovered subsequently by
Ferrara, who was a student of Cardano.
The general case.
In general, let F be a field of characteristic 0. (Some of the theory will still work in characteristic p.) We
say f (X) F [X] is solvable by radicals if there is a tower of fields

F = F1 F2 = F1 (2 ) F3 = F2 (3 ) Fk = Fk1 (k )

where at each stage some power ni i = i Fi1 , i = 2, . . . , k, and at the top Fk contains a splitting field for
f (X). In this case, as for cubics and quartics, the roots of f (X) may be expressed in terms of the radicals
i in perhaps a very involved way.
Theorem (Galoiss Solvability Criterion). Let F be a field of characteristic 0, let f (X) F [X], and let
K be a splitting field of f (X). f (X) is solvable by radicals if and only if the Galois group G(K/F ) is a
solvable group.
98 IX. APPLICATIONS OF GALOIS THEORY

Proof. Suppose that f (X) is solvable by radicals. Then there is a tower

F = F1 F2 = F1 (2 ) F3 = F2 (3 ) Fk = Fk1 (k )

where at each stage i is a root of a polynomial of the form X n with = i Fi1 and n = ni is a
positive integer.
Let m be the product of the ni in this tower and let be a primitive mth root of unity (in some extension
of Fk .) First, we reduce to the case F . To do this note first that KF () is a normal extension of
F . (It is not hard to see that in general the compositum of two normal extensions is normal. Use the
fact that a finite extension is normal if it is fixed under automorphisms of some extension normal over
the base field.) Since F () is a normal extension of F , it follows that K F () F is normal, and
G(K F ()/F ) is an epimorphic image of G(F ()/F ) so it is abelian. On the other hand, by natural
irrationalities, G(KF ()/F ())
= G(K/K F ()), so since G(K F ()/F ) = G(K/F )/G(K/K F ()) is
solvable, it suffices to show that G(KF ()/F ()) is solvable. By forming the composita Fi F (), we may
form a tower of radical extensions starting with F () with the last stage containing KF () so the extension
KF () F () inherits the relevant hypothesis.
Suppose then that F . We shall proceed by induction on the number of terms k in the tower. Since
F , F2 F1 = F is a normal extension with cyclic Galois group of order dividing m. Consider the
compositum KF2 . By essentially the same argument as in the previous paragraph, it suffices to show that
G(KF2 /F2 ) is solvable. However, F2 F3 Fk is a tower over F2 with k 1 terms, and the last stage
contains KF2 . Co by induction, G(KF2 /F2 ) is indeed solvable. This completes the first part of the proof.
Suppose conversely that G(K/F ) is solvable. Let m = |G(K/F )| and let be a primitive mth root
of unity in some extension of K. The extension F () F is a radical extension in the trivial sense that
m = 1 F . Also, by natural rationalities G(KF ()/F ()) is isomorphic to a subgroup of G(K/F ) so it
is also solvable. Let G(KF ()/F ()) = G1 G2 Gr = {1} be a tower of subgroups each normal
in the preceding and with cyclic factors. Let F () = K1 K2 Kr = KF () be the corresponding
tower of subfields. Each is normal over the preceding since the corresponding subgroups are. Since at each
stage, the Galois group, Gi /Gi1 , is cyclic and the base field contains the relevant roots of unity, each stage
is a radical extension as required. 
Remark. There is one additional twist which may be added to the above description of solvability by
radicals. Some people, may not consider it valid to call a root of unity a radical. For example, in C, if
n = a and is a primitive
nth root of unity, the roots of X n a are , , . . . , n1 . If a 6= 1, we may
reasonably consider = a to be a true algebraic radical. But we have no way to express except by using
n

numbers of the form ei which involve trigonometric

functions. We noted in the case n = 3 that could in
fact be expressed in terms of a true radical 3, and the same is true in general. Namely, the cyclotomic
extension F () has abelian (hence solvable) Galois group of order dividing (n) < n. Hence, we can apply
the analysis in the above proof by considering true radicals and roots of unity of order less than n. The
details require some thought, but ultimately we can express the roots of the original polynomial in terms
of elements i with each ni i Fi1 , the next lower field in the tower, and ni i 6= 1. The roots of unity do
leave some trace however, since at each stage, we have multiple choices for these true radicals and we must
make one specific choice in some manner.
Theorem (Galois). There exist polynomials in Q[X] of degree 5 which are not solvable by radicals.
Proof. Take f (X) = X 5 4X + 2. We shall show that the Galois group of the splitting field of f (X)
over Q is S5 which we know is not solvable.
First note that f (X) is irreducible. This follows from the following important result.
Lemma (Eisenstein Irreducibility Criterion). Let p be a prime and let

f (X) = an X n + an1 X n1 + + a1 X + a0 Z[X]

where an 6 0 mod p, ai 0 mod p for 0 < i < p, a0 0 mod p, but a0 6 0 mod p2 . Then f (X) is
irreducible.
 5. SOLVABILITY BY RADICALS 99

Proof of the Lemma. Assume f (X) = g(X)h(X) is a proper factorization in Q[X]. By Gausss
Lemma, we may suppose that g(X), h(X) Z[X]. Project this factorization onto Z/pZ[X]. In that ring,
we have
f(X) = g(X)h(X)
where by hypothesis f (X) = an X n and an 6= 0 in Z/pZ. By unique factorization, it follows that g(X) = cX r
and h(X) = dX s where r+s = n. It follows that the constant terms of both g(X) and h(X) must be divisible
by p. From this it follows that the constant term of f (X) must be divisible by p2 which is false.
Note. The Eisenstein Criterion works in A[X] for A any UFD since Gausss Lemma applies, and if p is
an irreducible element of A then A/pA is a field.
We continue with the calculation of the Galois group of f (X) = X 5 4X + 2. Let K be the splitting field
of f (X) and let be a root in K. Since f (X) is irreducible, [Q[] : Q] = 5.
K

Q[]
5

It follows that |G(K/Q)| is divisible by 5. On the other hand, G(K/Q) is isomorphic to some subgroup
of S5 which is of order 5! = 5 4! so the 5-Sylow subgroup of G = G(K/Q) must have order exactly 5
and may be identified with a 5-Sylow subgroup of S5 . Thus, G contains a 5-cycle which by appropriate
renumbering of the roots we can identify with ( 1 2 3 4 5 ). On the other hand, f (X) has exactly p
3 real roots. (For, f 0 (X) = 5Xp
4
4, and f 00 X) = 20X 3 so f (X) has a local maximum at x = 4 4/5
and a local minimum at x = 4 4/5. It is positive at the local maximum, negative at the local minimum,
negative for x very negative, and positive for x very positive. Hence, it crosses the real x-axis exactly 3
times. There are other more algebraic ways to come to the same conclusion.) The other two roots must be
complex conjugates. If we restrict complex conjugation to K, it must interchange the two complex roots
and it fixes the 3 real roots, hence it induces a 2-cycle ( i j ). Thus, G is a subgroup of S5 which contains
a 5-cycle = ( 1 2 3 4 5 ) and a 2-cycle = ( i j ). It follows that G = S5 . For, conjugating by the
powers of yields the 2-cycles ( i + 1 j + 1 ), ( i + 2 j + 2 ), ( i + 3 j + 3 ), and ( i + 4 j + 4 ) (where
the indices should be read mod 5). Hence, G contains a 2-cycle of the form ( 1 k ) = ( k 1 ). By shifting
mod 5 as above, we see that if G contains ( 4 1 ), it contains ( 1 3 ) and if it contains ( 5 1 ), it contains
( 1 2 ). If ( 1 2 ) G (and so also ( 1 5 )), then ( 2 3 ) G and by conjugation ( 1 3 ) G (and so
also ( 1 4 )). Hence ( 1 k ) G for every k, and by shifting, G contains every transposition so it is S5 . A
similar argument works if ( 1 3 ) G. 
Note that the above argument works as long as the polynomial is irreducible of degree 5 and has exactly
3 real roots. In fact, it will work if the polynomial is irreducible of prime degree p and has exactly p 2 real
roots.

Exercises.
1. Show by making an appropriate change of variables Y = X u, that we may replace a general cubic
polynomial by one in which the coefficient of Y 2 is zero.
2. Let f (X) = X 3 + pX + q be an irreducible cubic with rational coefficients.
(a) Substitute X = Y + Z, and show that
f (Y + Z) = Y 3 + Z 3 + (3Y Z + p)(Y + Z) + q
(b) Assume Y, Z are restricted to satisfy the relation 3Y Z + p = 0. Solve for Y in terms of Z, and
substitute in the above expression to obtain
g(Z) = Z 6 + qZ 3 (p/3)3 .
100 IX. APPLICATIONS OF GALOIS THEORY

(c) Use the quadratic formula to find the roots , 0 of the equation

T 2 + qT (p/3)3 = 0.

Let be a cube root of and 0 a cube root of 0 such that 0 = p/3. Tracing the steps backwards, note
that setting Y = , Z = 0 and X = + 0 yields a root of the original equation.
(d) Note that subject to the requirement 0 = p/3 (which is consistent with the restraint Y Z +p/3 = 0),
the only other possible choices are replacing by and 0 by 2 0 or replacing by 2 and 0 by 0 .
Conclude that the three roots of the irreducible cubic are given by the expressions in the text.
3. (a) Show that the roots of a real cubic are real if it has three real roots and imaginary otherwise. Hint:
A real cubic either has three real roots or one real root and two conjugate complex roots. Use the formula
for D in terms of the roots x1 , x2 , x3 .
(b) Let f (X) = X 3 3X + 1. Apply Cardanos method to find its complex roots.
(c) Show the roots of this polynomial are real by calculating its discriminant in terms of its coefficients.
Note that in the formulas you got in part (b), complex numbers enter in an essential way. In other words,
Cardanos method uses of non-real complex numbers to describe the roots although all three are real.
(d) What is the Galois group of the splitting field of this polynomial over Q?
4. (a) Let K = F () where n F for some n > 1. Show that there is a tower of subfields Fi , i = 0,
1, . . . , k, between F and K with F0 = F and Fk = K and such that

Fi = Fi1 (i ) where ipi Fi1 and pi is prime.

(b) Let f (X) = X 3 + pX + q Q[X] be irreducible. Let F be an extension of Q contained in C, and

let N = F (x1 ) where x1 is one of the roots of f (X). Show that if D F (where D is the discriminant
of f (X)) then N is normal over F . (c) Show that if f (X) = X 3 + pX + q Q[X] has 3 real roots then

D R.
(d) Assume f (X) as above is irreducible and has 3 real distinct roots. Show that there is no tower of real
fields starting with Q and ending with a field in which f (X) splits completely and such that each stage is
a radical extension. In short, an irreducible real cubic with 3 real roots cannot be solved by real radicals.
Hint:
By part (a), you can assume the stages are all of prime degree. By forming the composita with
Q[ D], you may assume the tower starts with that field. Somewhere along the way one of the fields would
be a normal extension of degree 3 over the previous stage. (Why?) It would also be obtained by adjoining
a cube root of some element since it would be a radical extension of degree 3. By normality, it would have
to contain a primitive cube root of unity. (Why?)

6. Symmetric polynomials
Before Galois showed that it is not generally true that every equation of degree 5 or higher is solvable by
radicals, Abel derived a related result. He showed that, in a certain sense we shall make clear below, there
are no radical formulas for the roots of a polynomial f (X) of degree n derived from its coefficients.
Let k be a field and let K = k(X1 , . . . , Xn ) be the field of rational functions in the indeterminates
X1 , . . . , Xn i.e. the field of fractions of k[X1 , . . . , Xn ]. The symmetric group Sn may be viewed as a
finite group of automorphisms of K since each permutation of the indeterminates induces an automorphism
of k[X1 , . . . , Xn ] and hence also of K. Let F be the fixed field of S5 . Then by our previous theory,
K = F (X1 , . . . , Xn ) is the splitting field of
Y
f (X) = (X Xi )
= X n 1 X n1 + 2 X n2 + + (1)n n F [X]
 6. SYMMETRIC POLYNOMIALS 101

where
X
1 = Xi
i
X
2 = Xi Xj
i<j
..
.
n = X1 X2 . . . Xn .

(The polynomials are called the elementary symmetric functions of the indeterminates. Any polynomial left
fixed by Sn is called symmetric.) By Galois Theory, K F is a finite, normal separable extension with
Galois group Sn which of course is not solvable for n > 4. Hence, the above polynomial in F [X] is not
solvable by radicals. This would not be very interesting were it not for the following additional facts. First,
K is also a splitting field for the separable polynomial f (X) over the field F 0 = k(1 , . . . , n ), and since
[K : F 0 ] n! = [K : F ], it follows that F = k(1 , . . . , n ). Moreover, we shall show below that if we choose
indeterminates T1 , . . . , Tn then Ti Xi defines an isomorphism k[T1 , . . . , Tn ] k[1 , . . . , n ] and the field
of fractions of the latter is the field F . Thus, we may identify F with the rational function field k(T1 , . . . , Tn ).
If we think of T1 , . . . , Tn as the coefficients of a general polynomial of degree n, then Abels Theorem asserts
that this general equation is not solvable by radicals in the sense discussed previously. That means there are
no general radical formulas for the roots as functions of indeterminate coefficients.
To complete the discussion we now prove the following important result.
Theorem. Let A be any commutative ring and let X1 , . . . , Xn be indeterminates. Every symmetric
polynomial in A[X1 , . . . , Xn ] is uniquely expressible as a polynomial in the elementary symmetric polynomials.
Proof. We shall use the abbreviated notation X = (X1 , . . . , Xn ) and X 0 = (X1 , . . . , Xn1 ).
We proceed by a double induction on n the number of indeterminates and d the degree of the polynomial
f (X). For n = 1 there is nothing to prove so suppose n > 1. Assume f (X) is symmetric of degree d > 0.
If Xn |f (X) then by symmetry, Xi |f (X) for every i, and in this case it is easy to see that X1 . . . Xn = n
divides f (X), i.e.
f (X) = f 0 (X)n .
Since f (X) and n are fixed by Sn it follows that f 0 (X) is also symmetric. (For this, you only need to be able
to cancel factors like Xi ; A need not be a domain.) Now apply induction on d to conclude that f 0 (X) and
hence f (X) are expressible as polynomials in 1 , . . . , n . If on the other hand Xn does not divide f (X), then
we may write f (X) = g(X 0 ) + h(X) where g(X 0 ) is a polynomial in the first n 1 indeterminates and Xn
divides h(X). Since f (X) is fixed by Sn1 , it is easy to see that g(X 0 ) (and also h(X)) is fixed by Sn1 . By
induction, we may suppose that g(X 0 ) = k(01 , . . . , 0n1 ) where 01 , . . . , 0n1 are the elementary symmetric
functions in the first n 1 indeterminates. Consider f1 (X) = f (X) k(1 , . . . , n1 ). (Note that the primes
have been dropped from the s.) Put Xn = 0 in that equation and note that h(X1 , . . . , Xn1 , 0) = 0 since
Xn |h(X) and that i (X1 , . . . , Xn1 , 0) = 0i for i = 1, 2, . . . , n 1. We get that f1 (X1 , . . . , Xn1 , 0) = 0 so
that Xn |f1 (X). We can now apply the first part of the argument to conclude that f1 (X) is expressible as a
polynomial in 1 , . . . , n . Hence, the same is true of f (X) = f1 (X) + k().
We now show that the representation of a symmetric polynomial as a polynomial in the i is unique. To
this end, map the polynomial ring A[T1 , . . . , Tn ] in indeterminates T1 , . . . , Tn to A[X1 , . . . , Xn ] by Ti i .
Suppose f (T ) A[T ] satisfies f () = 0. If Tn |f (T ) then f (T ) = f 0 (T )Tn and f 0 ()n = 0. Hence,
f 0 () = 0 and by induction we may conclude that f 0 (T )and hence f (T )is the zero polynomial. Otherwise,
f (T ) = g(T1 , . . . , Tn1 ) + h(T ) where Tn |h(T ). If we put Xn = 0 in the relation f (1 , . . . , n ) = 0 and use
the fact that n (X1 , . . . , Xn1 , 0) = 0, f () = g(1 , . . . , n1 ) + h(1 , . . . , n1 , n ) = 0 tells us that
g(01 , . . . , 0n1 ) = 0. (Note the primes on the s.) By induction, we may conclude that g(T1 , . . . , Tn1 ) = 0.
Hence, h() = 0 with Tn |h(T ). Applying the first part of the argument, we conclude that h(T ) = 0. 
102 IX. APPLICATIONS OF GALOIS THEORY

Exercises.
1. For n = 2, express the symmetric polynomial (X1 X2 )2 in terms of the elementary symmetric poly-
nomials.
2. (a) For n = 3, express the symmetric polynomial [(X1 X2 )(X1 X3 )(X2 X3 )]2 in terms of the
elementary symmetric polynomials.
(b) Assuming the relation X1 + X2 + X3 = 0 for the first elementary symmetric function, show that the
expression in part (a) gives the discriminant previously calculated for a cubic of the form X 3 + pX + q.
3. The power sum Pk = X1k + X2k + + Xnk is certainly a symmetric polynomial. There are formulas due
to Newton for inductively expressing each power sum in terms of elementary symmetric functions and power
sums of lower degree. These may be used either to express the power sums as polynomials in the elementary
symmetric polynomials with integral coefficients of the elementary symmetric polynomials in terms of the
power sums with rational coefficients. Look these formulas up in an appropriate source, and use them to for
n = 4 to express Pk , k = 1, 2, 3, 4, 5 in terms of the elementary symmetric functions. (You might also study
the proof that the formulas are valid while you are at it.)

7. Division rings
There is a famous theorem of Wedderburn which asserts that there are no finite non-commutative divisions
rings.
Theorem. Every finite division ring is a field.
Proof. Let x D and consider the centralizer of x in D
CD (x) = {y D|xy = yx}.
It is not hard to see that CD (x) is a subring of D, and it is even a sub-division ring since if y 6= 0 commutes
with x, then multiplying on left and right by y 1 shows that y 1 also commutes with x. Clearly, CD (x)
contains the center Z(D) of D (i.e. the set of all elements which commute with every element of D.) Z(D)
is also a sub-division ring and because it is commutative, it is a field.
Let |Z(D)| = q. Then, because everything in sight is a vector space over Z(D), we have |CD (x)| = q d(x)
and |D| = q n . Again, because D is a free module over CD (x), it follows that q n is a power of q d(x) so d(x)|n.
Let G = D be the multiplicative group of non-zero elements of D. |G| = q n 1. The center Z(G)
is clearly just Z(D) so |Z(G)| = q 1. Finally, the centralizer CG (x) (in G of x) is clearly CD (x) so
|CG (x)| = q d(x) 1. Now consider the class equation for G
X
|G| = |Z(G)| + (G : CG (x))
x6Z(G)

which comes down to X

qn 1 = q 1 + (q n 1)/(q d(x) 1).
Consider the nth cyclotomic polynomial n (X)the minimal polynomial over Q of a primitive nth root
of unit n . We know that n (X)|X n 1 but of course n (X) does not divide X d 1 for any d < n. If d|n,
then
X n 1 = H(X)(X d 1)
(where H(X) is just a sum of ascending powers of X d .) It follows that n (X)|H(X). By Gausss Lemma
and associated arguments, we know that n (X) has integral coefficients and in fact the above divisibility
relations take place in Z[X]. It follows that we may put X = q in the above relation to obtain n (q)|H(q) =
(q n 1)/(q d 1) for each d|n. It follows that n (q)|q 1.
To complete the proof, we show that |q 1| < |n (q)| if n > 1. This follows by noting that
Y
n (q) = (q i )
where the product is taken over all primitive nth roots of unity. However, it is easy to see that for any z 6= 1
on the unit circle in the complex plane we have |q 1| < |q z|. (Draw the picture!)