Chapter 1

Field Extensions

Throughout this chapter k denotes a field and K an extension field of k.

1.1 Splitting Fields

Definition 1.1 A polynomial splits over k if it is a product of linear polynomials
in k[x]. ♦

Let ψ : k → K be a homomorphism between two fields. There is a unique

extension of ψ to a ring homomorphism k[x] → K[x] that we also denote by ψ;
explicitly,
n
X  Xn
i
ψ λi x = ψ(λi )xi .
i=0 i=0

Hence it makes sense to ask if a polynomial in k[x] has a zero in K. Similarly,

it makes sense to ask if a polynomial in k[x] splits in K[x].

Definition 1.2 Let f ∈ k[x] be a polynomial of degree ≥ 1. An extension K/k is

called a splitting field for f over k if f splits over K and if L is an intermediate
field, say k ⊂ L ⊂ K, and f splits in L[x], then L = K. ♦

The second condition in the definition could be replaced by the requirement

that K = k(α1 , . . . , αn ) where α1 , . . . , αn are the zeroes of f in K.
The main result in this section is the existence and uniqueness up to isomor-
phism of splitting fields.
Remarks. 1. If k ⊂ L ⊂ K, and K is a splitting field for f ∈ k[x], then K
is also a splitting field for f over L. The converse is false as one sees by taking
f = x2 + 1 and k = Q ⊂ L = R ⊂ K = C.
2. Let K be a splitting field for f over k. If F is an extension of K and α is
a zero of f in F , then α ∈ K. To see this, write f = β(x − α1 ) . . . (x − αn ) with
β ∈ k and α1 , . . . , αn ∈ K, and observe that 0 = f (α) = β(α − α1 ) . . . (α − αn ),
so α = αi for some i.

1
2 CHAPTER 1. FIELD EXTENSIONS

3. Let K be a splitting field for f over k, and let α be a zero of f in K.

Then f = (x − α)g for some g ∈ K[x]. Because f splits in K, so does g. Hence
K is a splitting field for g over k(α).

Theorem 1.3 Let k be a field and f ∈ k[x]. Then f has a splitting field, say
K/k, and [K : k] ≤ (deg f )!.
Proof. Induction on n = deg f . If deg f = 1, then f = αx + β with α, β ∈ k,
so f already splits in k, so we can take K = k.
Suppose that n > 1. If f is already split we may take K = k, so we may
assume that f has an ireducible factor, say g, of degree ≥ 2. By Proposition
??.??, g has a zero in the extension field k(α) = k[x]/(g); the degree of this
extension is deg g ≤ n. Now write f = (x − α)h where h ∈ k(α)[x]. Since
deg h = n − 1, the induction hypothesis says there is an extension L/k(α) over
which h splits, and [L : k(α)] ≤ (n − 1)!. Certainly f also splits over L, and
[L : k] = [L : k(α)][k(α) : k] ≤ n!. If α1 , . . . , αn are the zeroes of f in L, then
k(α1 , . . . , αn ) is a splitting field for f over k. 
The proof of Theorem 1.3 involves a choice of an irreducible factor of f . It is
conceivable that choosing a different factor might produce a different splitting
field. Before showing that is not the case, and hence that a splitting field is
unique up to isomorphism, we need the following lemma.

Lemma 1.4 Let ϕ : k → k 0 be an isomorphism of fields. Let f ∈ k[x] be

irreducible. If α is zero of f in some extension of k and β is an extension of
ϕ(f ) in some extension of k 0 , then there is an isomorphism ψ : k(α) → k 0 (β)
such that ψ|k = idk and ψ(α) = β.
Proof. The map ϕ extends to an isomorphism k[x] → k 0 [x] and sends (f ) to
(ϕ(f )), so induces an isomorphism between the quotient rings by these ideals.
The composition of the obvious isomorphisms

k(α) → k[x]/(f ) → k 0 [x]/(ϕ(f )) → k 0 (β)

is the desired isomorphism. 

Theorem 1.5 Let k be a field and f ∈ k[x]. Let ϕ : k → k 0 be an isomorphism

of fields. Let K/k be a splitting field for f , and let K 0 /k 0 be an extension such
that ϕ(f ) splits in K 0 . Then
1. there is a homomorphism θ : K → K 0 such that θ|k = ϕ;
2. if K 0 is a splitting field for ϕ(f ) over k 0 , then K 0 ∼
= K.
Proof. (1) We argue by induction on [K : k]. Write f = p1 . . . pn as a product of
irreducibles pi ∈ k[x]. Then ϕ(f ) = ϕ(p1 ) . . . ϕ(pn ), and each ϕ(pi ) is irreducible
in k 0 [x].
If [K : k] = 1, then K = k, and we can take θ = ϕ.
1.2. NORMAL EXTENSIONS 3

Suppose that [K : k] > 1. Then some pi , say p1 , is not linear. Let α ∈ K

be a zero of p1 , and let β ∈ K 0 be a zero of ϕ(p1 ). By Lemma 1.4, there is an
isomorphism ψ : k(α) → k 0 (β) such that ψ|k = ϕ.
Now K is a splitting field for f over k(α), and ϕ(f ) splits over k 0 (β). Since
[K : k(α)] < [K : k] we can apply the induction hypothesis to obtain θ : K → K 0
such that θ|k(α) = ψ. Hence
(2) Certainly θ is injective, so it remains to show it is surjective. However,
if f = ¸(x − α1 ) . . . (x − αn ) then ϕ(f ) = ¸(x − ϕ(α1 )) . . . (x − ϕ(αn )); since K
and K 0 are splitting fields K = k(α1 , . . . , αn ) and K 0 = k 0 (ϕ(α1 ), . . . , ϕ(αn )), θ
is also surjective, and hence an isomorphism. 

Theorem 1.6 A polynomial of positive degree has a unique splitting field up to

isomorphism.

1.2 Normal extensions

Definition 2.1 A finite extension K/k is normal if every irreducible polynomial
in k[x] that has a zero in K actually splits over K. ♦

Theorem 2.2 An extension K/k is normal if and only if it is the splitting field
of a polynomial.

Proof. (⇒) Write K = k(α1 , . . . , αn ). The minimal polynomial pi of αi has a

zero in K so splits in K. Hence f = p1 · · · pn splits in K. But K is generated
by the zeroes of f , so K is the splitting field of f over k.
(⇐) Suppose K = k(α1 , . . . , αn ) is the splitting field of a degree n polynomial
g ∈ k where α1 , . . . , αn are the zeroes of g. Let f ∈ k[x] be irreducible and
suppose that α ∈ K is a zero of f . We must show that f splits in K.
Think of f g ∈ K[x], and let L be the splitting field for f g over K. Write
f = (x − α)(x − β) . . . (x − ω) where α, β . . . , ω ∈ L. Since α and β are zeroes of
f , there is an isomorphism φ : k(α) → k(β) such that φ(α) = β and φ|k = idk
(Lemma 1.4).
Now think of g ∈ k(α)[x] and g = φ(g) ∈ k(β)[x]. Now K is a splitting
field for g over k(α) and φ(g) splits in K(β) so, by Theorem 1.5, there is a map
θ : K → K(β) such that θ|k(α) = φ. Hence θ|k = φ|k = idk and θ(g) = g. Now
0 = θ(g(αi )) = g(θ(αi )) so θ(αi ) belongs to K. Hence θ(K) ⊂ K. In particular,
β = θ(α) ∈ K; the same argument shows that all the zeroes of f belong to K.

The next result shows that a finite extension K/k can be embedded in a
unique smallest normal extension L/k. The extension L/k is called the normal
closure of K/k.

Theorem 2.3 Let K/k be a finite extension. Then there is a finite extension
L/K such that
4 CHAPTER 1. FIELD EXTENSIONS

1. L is normal over k, and

2. if K ⊂ F ⊂ L and F is normal over k, then F = L, and

3. if L0 /K is a finite extension such that L0 satisfies (1) and (2), then there
is a K-isomorphism θ : L → L0 .

Proof. (1) Write K = k(α1 , . . . , αn ), let pi ∈ k[x] be the minimal polynomial

of αi , set p = p1 p2 · · · pn , and let L be the splitting field of p over K. Then
k ⊂ K ⊂ L, L is the splitting field for p over k, and [L : k] < ∞. By Theorem
2.2, L is normal over k.
(2) If K ⊂ F ⊂ L and F is normal over k, then each pi splits in F because
it is irreducible and has a zero in F , whence p splits in F , so F = L.
(3) If L0 /K is a finite extension such that L0 satisfies (1) and (2), then
each pi splits in L0 , so by Theorem 1.5 there is a map θ : L → L0 such that
θ|K = idK . By (2) applied to L0 , and the inclusions K ⊂ θ(L) ⊂ L0 , it follows
that θ(L) = L0 , whence θ is an isomorphism as claimed. 

Example 2.4 Let p be an odd prime. What is the splitting field of xp − 2 over
Q and what is its degree?
By Eisenstein’s criterion f = xp − 2 is irreducible. Let α be the real pth root
of 2. Then f is the minimal polynomial of α over Q, so [Q(α) : Q] = p.
If β ∈ C is a zero of f , then (βα−1 )p = 1, so βα−1 is a zero of xp − 1 =
(x−1)(xp−1 +· · ·+x+1). By ??.??, the cyclotomic polynomial xp−1 +· · ·+x+1
is irreducible. Let ζ = e2πi/p . The zeroes of f are α, ζα, ζ 2 α, . . . , ζ p−1 α, so the
splitting field of f over Q is L := Q(α, ζ).
Since L contains both Q(α) and Q(ζ) its degree over Q is divisible by both
p and p − 1. Hence p(p − 1) divides [L : Q]. Let g be the minimal polynomial
of α over Q(ζ). Then g divides xp − 2 so has degree ≤ p. Therefore

[L : Q] = [Q(ζ)(α) : Q(ζ)][Q(ζ) : Q] = (p − 1) deg g ≤ (p − 1)p,

whence [L : Q] = p(p − 1). ♦

1.3 Finite fields

We already saw parts of the next result in Section ??.

Theorem 3.1 Let K be a finite field. Then

1. |K| = pn for some positive integer n, where p = char K;

n
2. K is the splitting field for xp − x over Fp ;

3. any field of order pn is isomorphic to K.

1.3. FINITE FIELDS 5

Proof. Since char K = p, Fp is a subfield of K. Hence K is a finite dimensional

vector space over Fp , and so has pn elements where n = dimFp K.
It follows that K ∗ := K\{0} is an abelian group of order pn − 1. Hence
pn −1
λ = 1 for every non-zero λ ∈ K. It follows that every element of K is a
n n
zero of xp − x. In other words, xp − x has pn distinct zeroes in K. Hence K
n
is the splitting field for xp − x over Fp . It now follows from Theorem 1.5 that
any field of order pn must be isomorphic to K. 

Theorem 3.1 does not show that a field of order pn exists. It just shows what
it has to be if it exists.
We will write Fpn for the field of pn elements (if it exists!). To prove its
n
existence we will show that xp − x has pn distinct zeroes, and that the set of
n
these zeroes is equal to the splitting field of xp − x over Fp .

To see whether a polynomial has repeated zeroes we look at its derivative.

Definition 3.2 The formal derivative of a polynomial f = a0 + a1 x + · · · + an xn

in k[x] is
f 0 = D(f ) := a1 + 2a2 x + · · · + nan xn−1 .

Notice that D(f g) = f 0 g + f g 0 and D(λf ) = λf 0 if λ ∈ k.

Lemma 3.3 Let f be a non-zero polynomial in k[x]. Then f has a multiple

zero in some extension field of k if and only if deg(gcd(f, f 0 )) ≥ 1.

Proof. (⇒) Let K be an extension of k, and suppose that α ∈ K is a multiple

zero of f . Then f = (x − α)2 g for some g ∈ K[x]. Hence x − α divides both f
and f 0 in K[x]. As remarked on page ??, the gcd of two polynomials, in this
case f and f 0 , is the same in K[x] as in k[x], so that gcd has degree ≥ 1.
(⇐) If f and f 0 have a common factor of degree ≥ 1 in some K[x], then they
have a common factor of the form x − α in K[x] with K/k the splitting field of
f f 0 . If we write f = (x − α)g, then f 0 = (x − α)g 0 + g, whence x − α divides g,
so α is a multiple zero of f . 

n
Proposition 3.4 The polynomial xp − x ∈ Fp [x] has pn distinct zeroes in its
splitting field.
n
Proof. Since the derivative of xp − x is 1, the result follows from the previous
lemma. 

Lemma 3.5 Let p be a prime and R a commutative ring in which p = 0. Then

the map φ : R → R defined by φ(a) = ap is a ring homomorphism.
6 CHAPTER 1. FIELD EXTENSIONS

Proof. Certainly φ(1) = 1. It is clear that φ(ab) = φ(a)φ(b), and

p  
X p
φ(a + b) = ai bp−i .
i=0
i

p


The integers i are divisible by p whenever 1 ≤ i ≤ p − 1, so are zero in R.
Hence φ(a + b) = φ(a) + φ(b). 
We call the map φ in Lemma 3.5 the Frobenius map.
If K is a field of characteristic p, then φ is injective. Hence if K is a finite
field of characteristic p, then φ is also surjective and hence an isomorphism of
K with itself. In particular, every element of K is a pth power.

Theorem 3.6 For each prime p and positive integer n, there is a unique field
n
with pn elements, namely the splitting field of xp − x.
n
Proof. Let K be the splitting field of xp − x over Fp . Let φ : K → K be the
n
Frobenius map. Notice that α ∈ K is a zero of xp − x if and only if φn (α) = α.
pn
Hence, if α and β are zeroes of x − x, so are α ± β, αβ and α−1 . Hence the
n
zeroes of xp − x are a subfield of K. Since K is generated over Fp by the zeroes
n n
of xp − x, we conclude that K is exactly the set of zeroes of xp − x. 

Proposition 3.7 The multiplicative group of non-zero elements in a finite field

is cyclic.

Proof. Suppose that |K| = pn . Write e = pn − 1 = q1n1 · · · qtnt as a product of

powers of distinct primes. We will show there is an element in K\{0} of order
e. Define
ei = eqi−1 and di = eqi−ni .
Since ei < e, there is some αi ∈ K that is not a zero of xei −1. Define βi = (αi )di
and β = β1 · · · βt . The order of βi divides qini , but if it were smaller then αi
would be a zero of xei − 1; hence the order of βi is qini . It follows that the order
of β is e. 

1.4 Separability
Let’s begin with a warning: an irreducible polynomial can have multiple zeroes
in its splitting field. For example, let k = Fp (t) be the rational function field over
Fp , and let f = xp − t ∈ k[x]. By Eisenstein’s criterion applied to f ∈ k[t][x], f
is irreducible but over the extension field K = k(t1/p ) we have f = (x − t1/p )p .
This behavior causes problems.

Definition 4.1 A polynomial f ∈ k[x] is separable if none of its irreducible factors

has a multiple zero in its splitting field.
If every f ∈ k[x] is separable, we say that k is a perfect field.
1.4. SEPARABILITY 7

Let K be an extension of k An element α ∈ K is separable over k if its

minimal polynomial is separable. We say that K is a separable extension of k if
every element in it is separable over k. ♦
We have just seen that Fp (t) is not perfect: t1/p is not separable over Fp (t),
and Fp (t1/p ) is not a separable extension of Fp (t).

Lemma 4.2 If k ⊂ F ⊂ K are fields and K/k is separable, so are K/F and
F/k.
Proof. It follows at once form the definition that F/k is separable if K/k is.
On the other hand, if α ∈ K its minimal polynomial over F divides its minimal
polynomial over k, so has distinct zeroes. Hence α is separable over F . 
It is quite a bit harder to prove the converse of this lemma.

Proposition 4.3 Fields of characteristic zero are perfect.

Proof. Let f be an irreducible polynomial with coefficients in a field of charac-
teristic zero. Since the characteristic is zero, the derivative f 0 is not zero. Since
f is irreducible and deg f 0 < deg f , it follows that gcd(f, f 0 ) = 1, whence f has
no multiple zeroes by Lemma 3.3. 

Theorem 4.4 A field of characteristic p > 0 is perfect if and only if every

element of it is a pth power (if and only if the Frobenius map is surjective).
Proof. Let k be the field in question and write k p = {αp | α ∈ k}.
(⇒) Let α ∈ k and set f = xp −α. Since f 0 = 0, gcd(f, f 0 ) = f so Lemma 3.3
implies that f has a multiple zero. But k is perfect so irreducible polynomials
in k[x] do not have multiple zeroes. Hence f is reducible. Write f = gh with
1 ≤ deg g ≤ p − 1. Let K be the splitting field for f over k. If λ ∈ K is a zero
of f , then λp = α so

f = xp − α = xp − λp = (x − λ)p = gh.

Hence g = (x − λ)d with 1 ≤ d ≤ p − 1. But the coefficients of g belong to k, so

λd ∈ k. Also λp = α ∈ k. Since (p, d) = 1, there are integers u and v such that
du + pv = 1. Hence
λ = (λd )u (λp )v ∈ k.
In particular, α = λp is the pth -power of an element in k.
αj xj be an
P
(⇐) Suppose to the contrary that k is not perfect. Let f =
irreducible polynomial in k[x] having a repeated zero in some extension field.
Then deg(gcd(f, f 0 )) ≥ 1 but f is irreducible so gcd(f, f 0 ) = f . Hence f 0 = 0.
It follows that αj = 0 if p does not divide j. Hence f = β0 + β1 xp + β2 x2p + · · · .
By hypothesis there are elements γi ∈ k such that γip = βi so f = (γ0 + γ1 x +
γ2 x2 + · · · )p . This contradicts the irreducibility of f . We conclude that k must
be perfect. 
8 CHAPTER 1. FIELD EXTENSIONS

Corollary 4.5 A finite field is perfect.

Every algebraic extension of a finite field, and every extension of a charac-

teristic zero field is a separable extension.

Definition 4.6 If K = k(α) we say that K is a simple extension of k and that α

is a primitive element of K over k. ♦

For example, if (n, d) = 1 then α = e2πid/n is a primitive element for the

extension of Q obtained by adjoining all nth roots of one.

Theorem 4.7 (The primitive element theorem) A finite separable exten-

sion is simple. In particular, if k is finite or of characteristic zero every finite
extension of k is of the form k(α).

Proof. Let K be a finite separable extension of k.

Suppose k is finite. Then K is also finite so its multiplicative group K\{0}
is cyclic by Proposition 1.3.7. If α is a generator of this group, then K = k(α).
Suppose k is infinite. By induction it suffices to show that a finite separable
extension of the form k(α, β) is equal to k(γ) for a suitable γ.
Let f and g be the minimal polynomials of α and β over k. Write m = deg f
and n = deg g, and let {α = α1 , . . . , αm } and {β = β1 , . . . , βn } be the zeroes of
f and g respectively. Since the extension is separable, these zeroes are distinct.
Consider the mn equations

(α − αi ) + (β − βj )X = 0, 1 ≤ i ≤ m, 1 ≤ j ≤ n.

Since k is infinite there is some λ ∈ k that is not a solution to any of these

equations. Define γ := α + βλ.
Notice that β is a common zero of the polynomials g(x) and f (γ − λx) in
k(γ)[x]. If ξ 6= β were another common zero lying in some extension field of
K, then g(ξ) = f (γ − λξ) = 0, so ξ = βj for some j > 1 and γ − λξ = αi for
some i, whence α + βλ − λβj = αi thus contradicting the choice of λ. Hence
gcd(g(x), f (γ − λx)) = x − β. In particular, x − β ∈ k(γ)[x] by the remark on
page ??, so β ∈ k(γ). Hence α = γ − λβ is also in k(γ), and we conclude that
k(α, β) = k(γ). 

Corollary 4.8 If K/k is a finite, normal, separable extension, then K is the

splitting field of an irreducible separable polynomial over k.

Proof. By the Primitive Element Theorem, K = k(α). Let f be the minimal

polynomial of α. Then f is separable and irreducible of degree [K : k]. Since f
has one zero in K and K/k is normal, f splits in K. Hence K is the splitting
field for f . 
1.5. AUTOMORPHISMS OF SEPARABLE EXTENSIONS 9

1.5 Automorphisms of separable extensions

The notion of separabilty was defined in terms of individual elements. However,
it eventually proves more useful to be able to characterize whether or not an
extension K/k is separable in terms of automorphisms of K. We shall see that
non-separable extensions are more rigid than separable ones in the sense that
they possess far fewer automorphisms.
The next example illustrates the matter.

Example 5.1 The extension K/k = Fp (t1/p )/Fp (t) is not separable. If φ is an
automorphism of this extension then φ(t1/p ) has the same minimal polynomial
as t1/p , namely xp − t = (x − t1/p )p , so φ(t1/p ) = t1/p , whence φ = idK . Hence
Aut(K/k) = {1}. ♦
√
Compare this to√a separable √extension √ like Q( 2) where p there is a Q-
automorphism of Q( 2) sending 2 → − 2. Of course, Q( 3 ) does not have
any automorphisms
√ but for a rather different reason: although the √ minimal
polynomial of 3 2 has three distinct zeroes, only one of them is in Q( 3 2. Hence
to really understand the difference between separable and non-separable ex-
tensions from the perspective of automorphisms we should focus on separable
extensions that are normal.

Proposition 5.2 Let k(α) be a degree d extension of k and let f ∈ k[x] be the
minimal polynomial of α. Let φ : k → F be a homomorphism, and suppose that
φ(f ) splits over F .

1. If α is separable over k there are exactly d distinct extensions of φ to maps

φi : k(α) → F such that φi |k = φ.

2. If α is not separable there are < d such extensions.

Proof. Let β1 , . . . , βn be the distinct zeroes of f in F . Then n ≤ d = deg f

and n = d if and only if α is separable over k.
Any homomorphism ψ : k(α) → F that extends φ is completely determined
by ψ(α). But ψ(α) = βi for some i, so there are at most n different ψs. By
Lemma 1.4, there is such an extension for all i. Hence there are exactly n
extensions of φ, and the result follows. 

Theorem 5.3 Let K/k be a degree d extension and let φ : k → F be a homo-

morphism. Suppose the minimal polynomials of all the elements in K split in
F.

1. If K/k is separable there are exactly d distinct extensions of φ to maps

φi : K → F such that φi |k = φ.

2. If K/k is not separable there are < d such extensions.

10 CHAPTER 1. FIELD EXTENSIONS

Proof. (1) By the Primitive Element theorem K = k(α), so the result follows
from part (1) of Proposition 5.2.
(2) We argue by induction on d. Since K/k is not separable, d > 1. Let
α ∈ K be non-separable over k. Set s = [k(α) : k] and t = [K : k(α)], so st = d
and t < d.
Every extension of φ to K can be obtained in two steps: first extend φ to
k(α), then extend the extension from k(α) to K. By part (2) of Proposition
5.2, there are < s extensions of φ to k(α), and by the induction hypothesis and
(1), each of those extensions has ≤ t extensions to K, giving a total of < d
extensions of φ to K. 

Theorem 5.4 The following conditions on a finite extension K/k are equiva-
lent:

1. K/k is a normal and separable;

2. | Aut(K/k)| = [K : k];

3. K Aut(K/k) = k.

If K/k is normal but not separable, then | Aut(K/k)| < [K : k].

Proof. The final sentence in the statement of this theorem follows from Theo-
rem 5.3(2).
(1) ⇒ (2) If we set F = K, then the hypotheses of Theorem 5.3(1) hold and
the conclusion of that result gives (2).
(2) ⇔ (3) Artin’s Theorem says that

[K : K Aut(K/k) ] = | Aut(K/k)|,

so (2) and (3) are equivalent.

(2) ⇒ (1) Write d = [K : k] and let φ1 , . . . , φd be the distinct elements
in Aut(K/k). If F denotes the normal closure of K, then the hypotheses of
Theorem 5.3 hold, and each ψi may be considered as a map φi : K → F that
extends the inclusion φ : k → F . The conclusion of Theorem 5.3 then shows
that K/k is separable.
By the Primitive Element Theorem, K = k(α) for some α ∈ K. Let p ∈
k[x] be the minimal polynomial of α. Since the φi s are distinct, the elements
φ1 (α), . . . , φd (α) are distinct. Hence p has d = deg p distinct zeroes in K,
whence K is a splitting field for p. Hence K/k is normal. 

Theorem 5.5 An extension generated by separable elements is separable.

Proof. Let K = k(α1 , . . . , αn ) and suppose that each αi is separable over k.

We argue by induction on n. If n = 0, there is nothing to do, so suppose that
k 0 = k(α1 , . . . , αn−1 ) is separable over k, set α = αn so that K = k 0 (α) and
consider the extensions k 0 /k and K/k 0 .
1.5. AUTOMORPHISMS OF SEPARABLE EXTENSIONS 11

Let F be the normal closure of k 0 over k (see page 3). Let φ : k → F be the
inclusion; the hypotheses of Theorem 5.3 now hold, so there are exactly [k 0 : k]
distinct homomorphisms φi : k 0 → F such that φi |k = idk .
By hypothesis α is separable over k and hence over k 0 . By Proposition
5.2 applied to k 0 (α), there are exactly [k 0 (α) : k 0 ] extensions of each φi to
homomorphisms K = k 0 (α) → F . This gives a total of [k 0 : k].[k 0 (α) : k 0 ] = [K :
k] homomorphisms ψ : K → F such that ψ|k = idk . By part (2) of Theorem
5.3, K/k is separable. 

Corollary 5.6 The splitting field of a separable polynomial is a separable ex-

tension.
Proof. Let f ∈ k[x] be a separable polynomial. Its splitting field is generated
by the zeroes of the irreducible factors of f , and by hypothesis these zeroes are
separable. Theorem 5.5 implies that this splitting field is separable over k. 
12 CHAPTER 1. FIELD EXTENSIONS
Chapter 2

Galois Theory

The fundamental idea of Galois theory is to study an extension K/k through

analysis of its automorphism group Aut(K/k).

2.1 The Galois correspondence

Definition 1.1 A finite, normal, separable extension K/k is called a Galois ex-
tension. We then write
Gal(K/k) = Aut(K/k),
and call this the Galois group of the extension. ♦

We restate Theorem 1.5.4.

Theorem 1.2 The following are equivalent:

1. K/k is Galois;

2. | Aut(K/k)| = [K : k];

3. K Aut(K/k) = k.

Lemma 1.3 Let K/k be a Galois extension and F an intermediate field. Then

1. K/F is a Galois extension, and

2. {ψ ∈ Gal(K/k) | ψ|F = idF } = Gal(K/F ).

Proof. (1) Because K/k is normal it is a splitting field for some f ∈ k[x] ⊂ F .
Then K/F is a splitting field for f over F , and hence normal. By Lemma 1.4.2,
K/F is separable.
(2) Because K/F is a Galois extension we can speak of its Galois group.
There is a natural injective group homomorphism Gal(K/F ) → Gal(K/k) and
the image is as claimed. 

13
14 CHAPTER 2. GALOIS THEORY

Theorem 1.4 (The Galois correspondence) Suppose K/k is a Galois ex-

tension. There is an order-reversing bijection

{intermediate fields F , k ⊂ F ⊂ K} ←→ {subgroups of Gal(K/k)}

implemented by
F 7→ Gal(K/F )
and
K H ← H.
Proof. To prove these maps are mutual inverses we must show that

F = K Gal(K/F ) and H = Gal(K/K H ).

The first equality is given by Theorem 5.4. If H is a subgroup of Gal(K/k),

then K/K H is a Galois extension and H ⊂ Gal(K/K H ), so

|H| = [K : K H ] = | Gal(K/K H )|.

Hence H = Gal(K/K H ).
0
“Order-reversing” means that if H ⊂ H 0 , then K H ⊃ K H . Actually, the
lattices involved in the bijection are anti-isomorphic.

Remarks. Consider a Galois extension K/k and an intermediate extension
k ⊂ F ⊂ K.
1. We will always think of Gal(K/F ) as the subgroup of Gal(K/k) consisting
of automorphisms of K/k that are the identity on F .
2. Notice that [F : k] = | Gal(K/k)|/| Gal(K/F )|.
3. Since Gal(K/k) is finite it has only a finite number of subgroups. Hence
there are only a finite number of intermediate extensions F . This is not apriori
obvious because one has intermediate extensions k(α) for all α ∈ K and there
are infinitely many choices for α if k is infinite. For example, it is not √
at all
5
obvious
√ why√ only finitely many different extensions of Q appear as Q(a 2+
7 3
b −11 + c 17) as a, b, c ∈ Z vary!

Theorem 1.5 Let K/k be a Galois extension. Then

1. F/k is normal if and only if Gal(K/F ) is a normal subgroup of Gal(K/k).
2. If F/k is normal, then Gal(F/k) ∼
= Gal(K/k)/ Gal(K/F ).
Proof. (1) (⇐) Let H be a normal subgroup of Gal(K/k). To show that K H /k
is normal, it suffices to show that every irreducible polynomial f ∈ k[x] having
one zero α ∈ K H actually splits in K H .
Certainly f splits in K so we must show all its zeroes are in K H . Let β ∈ K
be a zero of f . Then there is σ ∈ Gal(K/k) such that σ(α) = β. If η ∈ H, then
σ −1 ησ ∈ H, so α = σ −1 ησ(α) = σ −1 η(β), whence η(β) = σ(α) = β. Therefore
β ∈ KH .
2.2. ELEMENTARY EXAMPLES 15

(⇒) Suppose that F/k is normal. Because K/k is separable, F/k is separa-
ble. Hence F/k is a Galois extension.
Let α ∈ F and let p ∈ k[x] be its minimal polynomial. Then p splits in
F . Hence, if σ ∈ Gal(K/k), then σ(α) ∈ F because it is a zero of p. Hence
σ(F ) ⊂ F for all σ ∈ Gal(K/k). Now, if η ∈ H, then σ −1 ησ(α) = σ −1 σ(α) = α,
so σ −1 ησ|F = idF . In other words, σ −1 ησ ∈ Gal(K/F ). Hence Gal(K/F ) is a
normal subgroup of Gal(K/k).
(2) Using the equality between the order of a Galois group and the degree
of the corresponding extension this is just a restatement of the fact that [K :
k] = [K : F ][F : k]. 
Remark. Consider k ⊂ F ⊂ K and suppose that K/k and F/k are Galois.
By the proof of the previous theorem, if σ ∈ Gal(K/k), σ(F ) ⊂ F , so there is
a group homomorphism Φ : Gal(K/k) → Gal(F/k), Φ(σ) = σ|F . Obviously,
ker Φ = Gal(K/F ). By part (2) of Theorem 1.5, Φ is surjective. This says, as
we already know, that any η ∈ Gal(F/k) can be extended to and automorphism
of K/k in [K : F ] different ways.

2.2 Elementary examples

Example 2.1 The Galois group of Fpn /Fp is the cyclic group of order n gen-
erated by the Frobenius automorphism σ(a) = ap .
n
Since Fpn is the splitting field for xp − x over Fp , it is a normal extension.
It is a separable extension also because Fp is finite. Hence Fpn /Fp is a Galois
extension of degree n.
The Frobenius automorphism is an automorphism of any field of character-
istic p, and it fixes the elements of Fp because Fp − {0} is a group of order p − 1
(so ap−1 = a for all 0 6= a ∈ Fp ). Hence σ ∈ Gal(Fpn /Fp ).
r r
Now, σ r (a) = ap , so σ n = 1. However, if σ r = 1, then ap = a for all
r
a ∈ Fpn , so every element in Fpn is a zero of xp − x; the degree of a polynomial
is at least as big as its number of zeroes, so pr ≥ |Fpn | = pn , whence r ≥ n. It
follows that the order of σ is n, so the subgroup of Gal(Fpn /Fp ) has order ≥ n.
n
However, | Gal(Fp /Fp )| = [Fpn : Fp ] = n, so
Gal(Fpn /Fp ) = hσi.
where σ is the Frobenius automorphism. ♦

Example 2.2 If n = mr, the Galois group of Fpn /Fpm is the cyclic group of
order r generated by the mth power of the Frobenius automorphism. In other
words,
Gal(Fpn /Fpm ) = hσ m i
m
where σ m (a) = ap . ♦
2 2
√ √ 2.3 The splitting field of f = (x − 2)(x − 3) over Q is K =
Example
Q( 2, 3). This is a Galois extension and | Gal(K/Q)| = [K : Q] = 4. There
are only two groups with four elements, Z4 and Z2 × Z2 .
16 CHAPTER 2. GALOIS THEORY

There are elements σ, τ in Gal(K/Q) defined by

√ √ √ √
σ( 2) = − 2 σ( 3) = 3
√ √ √ √
τ ( 2) = 2 τ ( 3) = − 3,

so Gal(K/Q) ∼= Z2 × Z2 . Its subgroup lattice

√ is This is easy to compute. For
example, if H = {1, στ }, then K H = Q( 6). ♦

Example 2.4 Let K be a splitting field for f = x4 − 2 ∈ Q[x].

Set α = 21/4 . The zeroes of f are ±α and ±iα so K = Q(α, i) = Q(α)(i),
whence [K : Q] = 8. Hence G = Gal(K/Q) has eight elements. Elements of G
must permute the zeroes of f , so G is a subgroup of the symmetric group S4 .
Since α and iα have the same minimal polynomial over Q(i), namely x4 − 2,
there is a σ ∈ G such that σ(i) = i and σ(α) = iα. Similarly, there is a τ ∈ G
such that τ (α) = α and τ (i) = −i.
We will now show that G is isomorphic to the dihedral group D4 , the sym-
metry group of the square. A straighforward calculation shows that σ 4 = τ 2 = 1
and τ στ = σ −1 , so there is a homomorphism D4 → G. Notice that σ 2 6= 1;
since a non-trivial normal subgroup of D4 contains σ 2 , the map D4 → G is
injective and hence surjective. Thus G = hσ, τ i ∼= D4 .
Before determining the intermediate fields we make some calculations. Since
σ 2 (α) = −α, σ 2 (α2 ) = α2 . Also στ (i) = −i, στ (α) = iα, and στ (iα) = α, so
στ (α + iα) = α + iα.
The subgroup lattice of G is
D4
  HH
  HH
2
hτ, σ i hσi hστ, σ 2 i
@ @
@ @
@ @
hτ i hσ 2 τ i hσ 2 i hστ i hσ 3 τ i
HH @ 
HH @  
H @ 

{1}

As we said above, every non-trivial normal subgroup of D4 contains σ 2 .

Since D4 /hσ 2 i ∼
= Z4 , the non-trivial normal subgroups of G are hσ 2 i and those
of index two, namely

{1, σ 2 }, {1, σ, σ 2 , σ 3 }, {1, σ 2 , τ, τ σ 2 }, {1, σ 2 , τ σ, τ σ 3 }.

2.3. POLYNOMIALS OF DEGREE ≤ 4 17

The corresponding intermediate fields are the normal extensions of Q, namely

2 √
K hσ i = Q(i, α2 ) = Q(i, 2),
K hσi = Q(i),
2 2 √
K hτ,σ i
= K hσ i ∩ K hτ i = Q( 2),
2 √
K hστ,σ i
= Q(i 2).

The other intermediate fields are

K hτ i = Q(α),
K hστ i = Q((1 + i)α),
2
K hσ τi
= Q(iα),
3
hσ τ i
K = Q((1 − i)α),

all of which are degree four extensions of Q. ♦

Exercises.
Let β be an element of F4 that is not in F2 .
1. Find the minimal polynomial of β over F2 .
2. Show that x2 + (β + 1)x + 1 is irreducible in F4 [x].
3. Is the cubic x3 + x2 + β ∈ F4 [x] irreducible? If not, find its factors.
4. Show that F16 contains an element α that is a primitive fifth root of one
over F2 , and that F16 = F2 (α). Find the minimal polynomial of α over
F4 , and show that α4 is the other zero of this polynomial.
5. Show that α is a zero of x3 + x2 + β ∈ F4 [x].
6. Show that the Galois group of x5 − 1 over F4 is Z2 .
7. Factor x4 + x3 + x2 + x + 1 over F4 .
8. You have shown above that F16 = F4 (α) where α is a primitive fifth root
of 1. Does there exist an element α ∈ Gal(x5 −1/F4 ) such that σ(α) = α2 ?

2.3 Polynomials of degree ≤ 4

In this section we determines the Galois groups of quadratics, cubics, and quar-
tics.
Notation. The Galois group of a separable polynomial f ∈ k[x] is Gal(f /k) :=
Gal(K/k) where K is a splitting field of f .

Definition 3.1 A subgroup H of the symmetric group Sn is transitive if, given

any i, j ∈ {1, . . . , n}, there is an η ∈ H such that η(i) = j. ♦
18 CHAPTER 2. GALOIS THEORY

Lemma 3.2 Let f ∈ k[x] be a separable, irreducible, polyomial of degree n.

Then Gal(f /k) is a transitive subgroup of the symmetric group Sn , and n divides
| Gal(f /k)|.

Proof. Let K be the splitting field of f and write K = k(α1 , . . . , αn ) where

the αi are the distinct zeroes of f . Because f is irreducible it is the minimal
polynomial of each αi , so [k(α) : k] = deg f = n. Hence n divides [K : k] =
| Gal(K/k)|.
If σ ∈ Gal(K/k), then the minimal polynomial of σ(αi ) is the same as the
minimal polynomial of αi so is f . Hence σ(αi ) = αj for some j. Hence Gal(K/k)
permutes the zeroes of f and this gives a homomorphism Gal(K/k) → Sn . It is
injective because if σ is the identity on the αi s it is the identity on k(α1 , . . . , an ).
Because αi and αj have the same minimal polynomial there is an automorphism
of K sending αi to αj . Hence Gal(K/k) is a transitive subgroup of the symmetric
group. 
The discriminant. Let f ∈ k[x] be a monic, irreducible, separable poly-
nomial. The element Y
δ(f ) := (αi − αj )
1≤i<j≤n

depends (up to a sign) on the order in which we label the zeroes of f , so is not
an invariant of f . However, the discriminant of f , which is defined to be
Y
D(f ) := δ(f )2 = (αi − αj )
1≤i6=j≤n

is independent of the labelling of the zeroes of f .

If τ ∈ Sn is a transposition, then τ (δ) = −δ and τ (D) = D.
It follows that D is invariant under Gal(f /k), so belongs to K Gal(K/k) = k.

Lemma 3.3 Let f ∈ k[x] be monic, irreducible, and separable. Then Gal(f /k)
is contained in the alternating group if and only if the discriminant D(f ) is a
square in k.

Proof. Clearly D(f ) is a square in k if and only if δ(f ) belongs to k; if and

only if δ(f ) is fixed by every element of Gal(f /k). But δ(f ) is fixed by even
permutations and sent to −δ(f ) by odd permutations, so δ(f ) ∈ k if and only
if Gal(f /k) consists of even permutations. 
Quadratic Polynomials. Let f = (x − α)(x − β) = x2 + bx + c ∈ k[x].
Then b = −(α + β) and c = αβ. The discriminant is

(α − β)2 = α2 − 2αβ + β 2 = (α + β)2 − 4αβ = b2 − 4c.

Since S2 ∼ = Z2 , A2 = {1}. The lemma says that Gal(f /k) is trivial if and only
if D(f ) is a square, i.e., if and only if b2 − 4c is a square in k. In other words f
splits in k if and only if b2 − 4c is a square, a result known to the ancients.
2.3. POLYNOMIALS OF DEGREE ≤ 4 19

Cubic Polynomials. If f = x3 + ax2 + bx + c, a tedious computation gives

D(f ) = a2 b2 − 4b3 − 4a3 c + 18abc − 27c2 .

You should do this tedious computation at least once in your life: write f =
(x − α1 )(x − α2 )(x − α3 ), express each of a, b, and c in terms of α1 , α2 , α3 , then
multiply out D(f ) = (α1 − α2 )2 (α1 − α3 )2 (α2 − α3 )2 and rewrite it in terms of
a, b, c.
We can always make a linear change of variable to bring a cubic polynomial
into the form f = x3 + px + q. Doing so, the discriminant takes the simpler
form
−4p3 − 27q 2 .

The only transitive subgroups of S3 are S3 itself and A3 , so the Galois group of
an irreducible, separable cubic is either S3 or A3 ∼
= Z3 , with the two possibilities
being determined by whether D(f ) is or is not a square in k.

Quartic Polynomials. The transitive subgroups of S4 are:

S4
the six conjugates of h(1234)i ∼
= Z4
the three conjugates of H = h(1234), (13)(24)i ∼= D4
A4
V = {1, (12)(34), (13)(24), (14)(23)} ∼
= Z2 × Z2 .

The last two are the only ones contained in A4 .

Example 3.4 Let

f = x4 + ax2 + b ∈ Q[x]

be irreducible and suppose that k/Q is an extension that does not contain a
zero of f . Because f is a quadratic in x2 its zeroes are {±α, ±β} in its splitting
field. We will show that

Z4
 if αβ −1 − βα−1 ∈ k
∼
Gal(f /k) = Z2 × Z2 if αβ ∈ k

D4 otherwise.


Let’s write G = Gal(f /k).

Notice that G can not contain a 3-cycle because such an element would fix
one of the zeroes, say α, but would not fix −α. That is absurd. Hence G can
not equal to S4 or A4 .
To compute the discriminant notice that

x4 + ax2 + b = (x2 − α2 )(x2 − β 2 ) = x4 − (α2 + β 2 )x2 + α2 β 2

20 CHAPTER 2. GALOIS THEORY

so −a = α2 + β 2 and b = α2 β 2 . Therefore

δ = (α + α)(α − β)(α + β)(−α − β)(−α + β)(−β − β)

= −4αβ(α2 − β 2 )2
= −4αβ b2 − 4a .

Hence aβ ∈ k ⇐⇒ δ ∈ k ⇐⇒ G ⊂ A4 ⇐⇒ G ∼
= Z2 × Z2 .
Claim: if φ := αβ −1 −α−1 β belongs to k, G does not contain a product of two
disjoint 2-cycles. Proof: Suppose to the contrary G contained such an element.
Looking at the list of possible G above, G would therefore have to contain all
products of disjoint 2-cycles. Hence G would contain the automorphism σ such
that σ(α) = β. But σ(φ) = −φ, so φ 6∈ k. This is a contradiction, so the claim
must hold. ♦ It follows from the claim that G ∼ = Z4 if φ ∈ k.
The following 4-cycles can not belong to G:

α 7→ −α 7→ −β 7→ β,
α 7→ β 7→ −β 7→ −α,
α 7→ −β 7→ β 7→ −α,

The other three 4-cycles all fix φ, so if G ∼

= Z4 , then φ ∈ k. Thus, φ ∈ k ⇐⇒
G∼= Z4 . ♦

2.4 Generic Polynomials

The symmetric group Sn acts on the polynomial ring k[t1 , . . . , tn ] in the obvious
way:

σ(ti ) := tσ(i) .

We extend the action by requiring σ to act as the identity on k and to be an

automorphism of the polynomial ring. Thus, if σ = (123), then the action of σ
on a polynomial f is given by replacing each t1 by t2 , each t2 by t3 , and each
t3 by t1 , and leaving the other tj s untouched.
The action extends to the function field k(t1 , . . . , tn ).
A polynomial f which is invariant under Sn , that is, σ(f ) = f for all σ ∈ Sn ,
is called a symmetric polynomial. The invariants k[t1 , . . . , tn ]Sn form a subring
of k[t1 , . . . , tn ]. Among the invariants are the elementary symmetric polynomials
2.4. GENERIC POLYNOMIALS 21

defined as follows:
ε0 = 1
ε 1 = t1 + · · · + tn
X
ε2 = tp tq = t1 t2 + t1 t3 + · · · + t2 t3 + · · · + tn−1 tn
p<q
X
ε3 = tp tq t r
p<q<r
.. ..
. .
ε n = t 1 t 2 · · · tn .

Theorem 4.1 1. k[t1 , · · · , tn ]Sn = k[ε1 , · · · , εn ];

2. k[ε1 , · · · , εn ] is a polynomial ring in n variables;
3. k(t1 , · · · , tn ) is a Galois extension of k(ε1 , · · · , εn ) with Galois group Sn .
Part (1) of Theorem 4.1 says that every polynomial that is invariant un-
der the action of the symmetric group can be written as a polynomial in the
elementary symmetric polynomials ε1 , · · · , εn . For example, the polynomials
td1 + · · · + tdn are obviously invariant. We have
t21 + · · · + t2n = ε21 − 2ε2 .
Try finding analogous expressions for d = 3, 4, . . ..

Lemma 4.2 Let f = (x − t1 ) · · · (x − tn ) ∈ k(t1 , . . . , tn )[x]. Then

f = xn − ε1 xn−1 + ε2 xn−2 − · · · + (−1)n−1 εn−1 x + (−1)n εn .
Proof. Straightforward. 
We can view these previous two results in another way. Let s1 , . . . , sn be in-
determinates over k, and consider the field k(s1 , . . . , sn ). The general polynomial
of degree n is
f = xn − s1 xn−1 + s2 xn−2 − · · · + (−1)n−1 sn−1 x + (−1)n sn .
It belongs to k(s1 , . . . , sn )[x].

Theorem 4.3 Gal(f /k(s1 , . . . , sn )) ∼

= Sn .
Proof. Let t1 , . . . , tn be indeterminates and ε1 , · · · , εn the elementary symmet-
ric polynomials in t1 , . . . , tn . By Theorem 4.1, k(s1 , . . . , sn ) is isomorphic to
the subfield k(ε1 , · · · , εn ) of K = k(t1 , · · · , tn ). We identify k(s1 , . . . , sn ) with
k(ε1 , · · · , εn ). By Lemma 4.2, f splits over K. In fact f = (x − t1 ) · · · (x − tn ),
so K/k(s1 , . . . , sn ) is the splitting field of f . By Theorem 4.1(3), the Galois
group of f is isomorphic to Sn . 
22 CHAPTER 2. GALOIS THEORY

Corollary 4.4 (Abel, Galois) Let n ≥ 5. The general polynomial of degree

n is not solvable by radicals.
Proof. We will prove this in the next chapter—a polynomial is solvable by
radicals if and only if its Galois group is solvable. However, the symmetric
group Sn is not solvable if n ≥ 5. 
Chapter 3

Solvability by radicals

A polynomial f ∈ k[x] is solvable by radicals if the zeroes of f are given by a

“formula” that involves only +, −, ×, ÷, nth roots, and elements of k and the
coefficients of f . The paradigmatic example is that the zeroes of ax2 + bx + c
are given by √
−b ± b2 − 4ac
2a
provided that a 6= 0 and char k 6= 2. (What happens when char k = 2?)
The formula for the zeroes of a quadratic polynomial was known to several
ancient civilizations. A formula for the cubics was not found until the 16th
century, and is generally credited to Scipio Ferro and Niccolo Tartaglia, although
the first published solution appeared in a book by Cardano. The zeroes of
x3 + px + q are
A − B, ωA − ω 2 B, ω 2 A − ωB,
√
where ω is a primitive cube root of unity, say 21 (−1 + −3), and
s 
r 2  2
q q p
A= 3 − + +
2 2 3
s 
r 2  2
3
q q p
B= + + +
2 2 3

and the cube roots are chosen so that AB = −3p. That is quite some formula!
Notice though that if p and q belong
√ to some field k, the three zeroes belong to
the field k(ω, α, A, B) where α = −3D and A3 , B 3 ∈ k(ω, α). In other words
the zeroes belong to a field K that is obtained by successively adjoining roots
of elements.
One very interesting aspect of this formula, an aspect that was a great
√ puzzle
at the time, is that the three roots could all be real numbers even if −3D is
not a real number.
This was soon followed by a general solution to the quartic, and attention
soon shifted to the quintic. Despite intense efforts during the 17th century no
general solution was found and the suspicion then arose that there might not be

23
24 CHAPTER 3. SOLVABILITY BY RADICALS

a formula for the solution to the general quintic. This was confirmed by Ruffini
and Abel.

3.1 Roots of unity

The polynomials xn − 1 and their Galois groups are of fundamental importance.
Understanding them is a necessary preparation for Galois’s theorem that (with
a proviso on the characteristic of the field) a polynomial is solvable by radicals
if and only if its Galois group is solvable.

Definition 1.1 An element ζ in a field is called a primitive nth root of unity if

ζ n = 1 but ζ i 6= 1 for every 1 ≤ i ≤ n − 1. ♦

Proposition 1.2 Let k be a field of characteristic p > 0. If p divides n, then

k does not contain a primitive nth root of unity.

Proof. Write n = pd. Then xn − 1 = (xd − 1)p , so every nth root of unity is a
dth root of unity. 
In particular, if char k divides n, no extension of k can contain a primitive
nth root of unity. The next proposition shows that if char k does not divide n
then there is an extension of k containing a primitive nth root of unity.

Lemma 1.3 The nth roots of unity form a cyclic subgroup of the multiplicative
group of a field.

Proof. The set Γ of nth roots of unity in k is a subgroup of the multiplicative

group k × = k−{0}. Let m > 0 be minimal such that am = 1 for all a ∈ Γ. Then
|Γ| ≤ m because xm − 1 has at most m zeroes in k. However, Γ ∼ = Za1 ⊕ · · · ⊕ Zat
for some integers ai , so m is the least common multiple of the ai s, whence
m ≤ a1 · · · at = |Γ|. Hence |Γ| = m, so the ai s are relatively prime and Γ ∼ = Zm .


Proposition 1.4 If the characteristic of k does not divide n, then

1. xn − 1 is separable over k;

2. the splitting field of xn − 1 contains a primitive nth root of unity;

3. Gal(xn − 1/k) is abelian;

4. Gal(xn − 1/k) is isomorphic to a subgroup of the group of units in Z/(n).

Proof. (1) and (2). Since char k does not divide n, the derivative of xn − 1 is
non-zero, hence relatively prime to xn − 1. Thus xn − 1 has n distinct zeroes
and these form a cyclic group {1, ω, . . . , ω n−1 } by Lemma 1.3. In particular, ω
is a primitive nth root of unity.
3.2. SOLVABILITY BY RADICALS 25

(3) and (4) Let K denote the splitting field of xn − 1. If σ ∈ Gal(K/k),

then σ(ω) is also a zero of xn − 1, so σ(ω) = ω i(σ) for a unique i(σ) ∈ Zn . A
simple calculation shows that the map i:Gal(K/k) → Zn , σ 7→ i(σ), satisfies
i(στ ) = i(σ)i(τ ) and i(σ −1 ) = i(σ)−1 , so is a homomorphism to the group of
units in Zn . 

Corollary 1.5 If char k does not divide n, then there is a Galois extension of
k generated by a primitive nth root of unity, say ζ, and Gal(k(ζ)/k) is cyclic.

Proof. 

Corollary 1.6 Let p be a prime and k a field of characteristic not p. Then

Gal(xp − 1/k) is isomorphic to a subgroup of Zp−1 .

Proof. The group of units in the ring Zp is isomorphic to Zp−1 by Proposition

1.3.7. 

Example 1.7 Gal(x5 − 1/F4 ) = ∼ Z2 .

It is easy to see that f = x4 + x3 + x2 + x + 1 is irreducible over F2 . Hence if
α is a zero of f , [F2 (α) : F2 ] = 4. The four distinct zeroes of f are α, α2 , α3 , α4 ,
so F2 (α) is the splitting field for f over F2 and F2 (α) is a Galois extension of
F2 . Since F2 (α) ∼= F16 , it has a subfield isomorphic to F4 , and [F2 (α) : F4 ] = 2.
Hence F2 (α) is the splitting field of the separable polynomial x5 − 1 over F4 ,
and Gal(x5 − 1/F4 ) = Gal(F2 (α)/F4 ) ∼ = Z2 .
Over F4 , f is no longer irreducible: we can write F4 = {0, 1, β, β + 1} with
β 2 = β + 1; then f = (x2 + βx + 1)(x2 + (β + 1)x + 1). ♦

3.2 Solvability by radicals

First the formality.

Definition 2.1 A polynomial f ∈ k[x] is solvable by radicals if there is a chain of

field extensions
k = K0 ⊂ K1 ⊂ · · · ⊂ K r (2-1)
of the form
√

Ki+1 = Ki ai
ni

for various positive integers ni and elements ai ∈ Ki , 0 ≤ i ≤ r − 1, such that

f splits in Kr .
In this situation we call each Ki+1 /Ki a simple radical extension, Kr /k a
radical extension, and (2-1) a radical sequence. ♦

Definition 2.2 We call K/k a cyclic extension if it is Galois with cyclic Galois
group. ♦
26 CHAPTER 3. SOLVABILITY BY RADICALS

Theorem 2.3 Let k be a field containing a primitive nth √ root of unity. Then
K/k is a cyclic extension of degree n if and only if K = k( n a) for some a ∈ k.

Proof. (⇐)√ Let ζ ∈ k be a primitive nth root of 1. Suppose that K = k(α)

where α = n a. Then K/k is the splitting field of the polynomial xn − a ∈ k[x]
because it has n distinct zeroes α, ζα, . . . , ζ n−1 α in K.
If σ ∈ Gal(K/k), then σ(α) = ζ i(σ) α for some i(σ) ∈ Zn . The map σ 7→ i(σ)
is a homomorphism Gal(K/k) → Zn . But it is also injective, and hence an
isomorphism because | Gal(K/k)| = [K : k] = n = |Zn |.
(⇒) Suppose Gal(K/k) = hσi ∼ = Zn . By Dedekind’s Proposition ??, the
distinct maps idK , σ, σ 2 , . . . , σ n−1 : K → K are linearly independent over K.
Hence there is some β ∈ K such that

α := β + ζ −1 σ(β) + · · · + ζ 1−n σ n−1 (β)

is non-zero. Now σ(α) = αζ so αn is invariant under σ, hence invariant under

Gal(K/k). Thus αn = a ∈ k. Because xn − a ∈ k[x] has n distinct zeroes
α, ζα, . . . , ζ n−1 α in K, it is separable over k and splits in K.
Every element of Gal(K/k) sends k(α) to k(α), so restriction gives a ho-
momorphism Gal(K/k) → Gal(k(α)/k), φ 7→ φ|k(α) , which √ is clearly injective.
Hence [k(α) : k] ≥ n, and it follows that K = k(α) = k( n a). 

Lemma 2.4 Let F/k be an extension. If f ∈ k[x] is separable, then Gal(f /F )

is isomorphic to a subgroup of Gal(f /k).

Proof. First observe that f remains separable over F because its irreducible
factors over F divide its irreducible factors over k, and therefore have no re-
peated zeroes.
Let L = F (α1 , . . . , αn ) be a splitting field for f over F , where α1 , . . . , αn are
the zeroes of f . Then K := k(α1 , . . . , αn ) is a splitting field for f over k.
If σ ∈ Gal(L/F ) = Gal(f /F ), then each σ(αi ) is a zero of f so equal
to some αj . Hence σ(K) ⊂ K. The map σ 7→ σ|K is a group homomorphism
Gal(f /F ) → Gal(K/k) = Gal(f /k). This map is injective because if σ|K = idK ,
then σ( αi ) = αi for all i whence σ = idL . 

Theorem 2.5 Let f ∈ k[x] and suppose that char k does not divide deg f . If
Gal(f /k) is solvable, then f is solvable by radicals.

Proof. Set n = (deg f )!. By Proposition 1.4, there is an extension F/k gener-
ated by a primitive nth root of unity. By Lemma 2.4, Gal(f /F ) is isomorphic
to a subgroup of Gal(f /k) so is also solvable. Since F/k is a radical extension,
it suffices to show that f is solvable by radicals over F . Hence we can, and will,
assume that k contains a primitive nth root of unity.
Let K/k be a splitting field of f , and choose a solvable series

Gal(K/k) = G0 ⊃ G1 ⊃ . . . ⊃ Gr = {1}
3.2. SOLVABILITY BY RADICALS 27

such that each Gi /Gi+1 is cyclic. Define Ki := K Gi to obtain a sequence of

fields
k = K0 ⊂ K 1 ⊂ . . . ⊂ Kr = K
with each Ki+1 /Ki a cyclic extension whose degree, d√say,

divides n. Since Ki
contains a primitive nth root of unity, Ki+1 = Ki d a for some a ∈ Ki by
Theorem 2.3. Hence K is a radical extension of k. 

Lemma 2.6 The normal closure of a radical extension is a radical extension.

Proof. Let K = k(α1 , . . . , αr ) be a radical extension. We argue by induction
on r. Thus, we assume that the normal closure, say L0 , of K 0 = k(α1 , . . . , αr−1 )
is a radical extension of k.
The normal closure of K, say L, must contain L0 , so is equal to the normal
closure of L0 (αr ). Notice that L0 (αr ) is a radical extension of L0 , so is a radical
extension of k. Let f be the minimal polynomial of αn over L0 and let β1 , . . . , βd
be the zeroes of f in L. Then L = L0 (β1 , . . . , βd ). Since L0 (βi ) ∼
= L0 (αr ), L0 (βi )
0
is a radical extension of k. Since L is generated by the fields L (βi ), the second
remark above implies that L is a radical extension of k. 
Remark. We need a more precise version of Lemma 2.6. With the notation
in its proof, suppose that αini ∈ k(α1 , . . . , αi−1 ) for all i. Then L is built from
k by adjoining only nth
i roots too.

Theorem 2.7 Suppose that f ∈ k[x] is solvable by radicals, say by taking var-
ious nth
i roots of unity. If char k does not divide any of the ni s, then Gal(f /k)
is solvable.
Proof. Let n be the least common multiple of the various ni s. Then char k
does not divide n. By Lemma 2.6 and the hypothesis, there is a radical Galois
extension L/k in which f splits, and the remark after that Lemma shows that
L involves adjoining only nth roots. Let K/L be the splitting field for xn − 1
where n = [L : k].
Let F/k be the splitting field of f . Since
Gal(K/k)
Gal(f /k) ∼
=
Gal(K/F )
it suffices to show that Gal(K/k) is solvable.
Now K contains a primitive nth root of unity, say ζ, and we have a chain

k ⊂ k(ζ) = K0 ⊂ K1 ⊂ · · · ⊂ Kr = K
√

in which each Ki+1 = Ki d ai for some ai ∈ Ki and some integer di dividing
n. By Theorem 2.3, Ki+1 /Ki is a cyclic extension.
Now k(ζ)/k is Galois with abelian Galois group, and
Gal(K/k)
Gal(k(ζ)/k) ∼
=
Gal(K/k(ζ))
28 CHAPTER 3. SOLVABILITY BY RADICALS

so Gal(K/k) is solvable if and only if Gal(K/k(ζ)) is solvable.

Let Gi be the subgroup of Gal(K/k(ζ)) corresponding to Ki . Then there is
a chain
Gal(K/k(ζ)) = G0 ⊃ G1 ⊃ · · · ⊃ Gr = {1}

in which Gi+1 is normal in Gi because Ki+1 /Ki is normal, and Gi /Gi+1 is cyclic
because Ki+1 /Ki is cyclic. 

Example 2.8 Let k = Fp (t) be the rational function field over the field of p
elements. If f = xp − x − t ∈ k[x], then Gal(f /k) ∼
= Zp , so is solvable. However,
f is not solvable by radicals.
Let α be a zero of f . Then

(α + 1)p − (α + 1) − t = αp + 1 − α − 1 − t = 0

so α, α + 1, α + 2, . . . , α + p − 1 are the distinct zeros of f . Hence f is separable.

The splitting field of f is k(α) and the minimal polynomial of α is f , so [k(α) :
k] = p. Hence the Galois group is Zp .
♦

The proof of the next result uses Cauchy’s Theorem, which will be proved
later, and a technical result on the symmetric group that will also be proved
later. In particular, we use the fact that the symmetric group Sn is not solvable
if n ≥ 5.

Theorem 2.9 Let f ∈ Q[x] be an irreducible polynomial of prime degree, say

p. If f has exactly two non-real zeroes, then Gal(f /Q) ∼
= Sp . In particular, if
p ≥ 5, then f is not solvable by radicals.

Proof. By the Fundamental Theorem of Algebra f splits in C. Let K be the

splitting field for f over Q, and set G = Gal(K/Q) = Gal(f /Q). By Lemma
2.4, the action of G on the zeroes of f gives an injective group homomorphism
G → Sp .
By Lemma 3.2, p divides [K : Q] = |G| so, by Cauchy’s Theorem, G has
an element σ of order p. This must be a p-cycle, say σ = (12 . . . p). Complex
conjugation is a Q-automorphism of C, so restricts to a Q-automorphism of K.
But conjugation fixes the p − 2 real zeroes of f , and permutes the two non-real
zeroes so is a 2-cycle. Let’s assume this transposition is (1n).
Notice that σ n is again a p-cycle and σ n = (1 n 2n−1 · · · ), so after relabelling
we can assume that G contains (12 . . . p) and (12). But these two elements
generate all of Sp so G = Sp . 

Exercise. In S4 , (13) and (1234) do not generate S4 , so check the “without

loss of generality” claim in the last sentence of the proof—you need to use the
fact that p is prime.
3.3. CYCLOTOMIC POLYNOMIALS 29

Example 2.10 The polynomial f = x5 − 6x + 3 ∈ Q[x] is not solvable by

radicals. By Eisenstein’s criterion f is irreducible. Since f 0 (x) = 5x4 − 6 has
only two real zeroes, say ±α, f has at most three real zeroes by Rolle’s Theorem.
Since f (−α) > 0 > f (α) and f (x)to ± ∞ as x → ±∞, f has three real zeroes.
Hence Gal(f /Q) ∼ = S5 . ♦

3.3 Cyclotomic polynomials

Definition 3.1 For each positive integer n define

ζn := e2πi/n .

We call Q(ζn ) the nth cyclotomic extension of Q.

The minimal polynomial of ζn over Q is called the nth cyclotomic polynomial
and is denoted by Φn (x). ♦
Claim: Q(ξm , ξn ) = Q(ξ` ) where ` = lcm{m, n}. To see this, first write
` = ma = nb where (a, b) = 1. Then ξ`a is a primitive nth root of unity, so
ξn ∈ Q(ξ` ). Similarly, ξm ∈ Q(ξ` ). So it remains to prove that ξ` ∈ Q(ξm , ξn ).
It suffices to show that Q(ξm , ξn ) contains a primitive `th root of unity. Notice
that ξ`a = ξn and ξ`b = ξm . There are integers c and d such that 1 = ac + bd.
Hence ξ` = ξnc ξmd
. ♦

We write
µn := {the group of nth roots of unity} ⊂ C.
There is an isomorphism of groups Zn → µn defined by a 7→ ξna . If 1 ≤ a < n,
ξna is a primitive dth root of unity where d = gcd(a, n).
If we view Zn as a ring and write Un for its group of units, then ξna is a
primitive nth root of unity if and only if a ∈ Un . Hence the set of primitive nth
roots of unity is the image in µn of Un .

Lemma 3.2 The nth cyclotomic polynomial is

Y Y
Φn (x) = (x − ξ) = (x − ξna )
ξ∈µn 1≤a<n
ξ primitive gcd(a,n)=1

Proof. It is clear that these two products are equal.



Definition 3.3 The Euler φ-function φ : N → N is defined by

φ(n) := the number of integers 1 ≤ m ≤ n such that (m, n) = 1.

Corollary 3.4 Let n be a positive integer. Then

30 CHAPTER 3. SOLVABILITY BY RADICALS

1. deg Φn (x) = φ(n);

2. [Q(ζn ) : Q] = φ(n);
3. Gal(Q(ζn )/Q) is isomorphic to the group of units in Zn .
Proof. The set Un := {m | 1 ≤ m ≤ n and (m, n) = 1} has cardinality φ(n).
The set Un has two other descriptions: it is the set of m such that ζnm is a
primitive nth root of unity; its image in Zn is the group of units. The map
i(σ)
Gal(Q(ζn )/Q) → Zn , σ 7→ i(σ) defined by σ(ζn ) = ζn , is an isomorphism to
Un .