LECTURE NOTES (PART 3), MATH 500 (FALL 2022)

CHARLES REZK

1. Field extensions
A subfield of a field K is a subring F ⊆ K which is also a field. This implies that the subring M 31 Oct
has 1, which is equal to the identity element of K. subfield

We say that K is an extension field of F if F is a subfield of K. The notation “K/F ” means extension field
“K is an extension field of F ”, so it is the same as “F ⊆ K”. We tend to represent this by the picture
K

F
Every field K contains a prime field F ⊆ K, which is the smallest subfield of K. The prime
field is either isomorphic to Q, which case we say char(K) = 0, or is isomorphic to Fp = Z/p, in
which case we say char(K) = p. Equivalently, (char K) = Ker φ, where φ : Z → K is the unique ring
homomorphism preserving 1.
Degree of an extension. If S is a ring and R ⊆ S is a subring with 1R = 1S , then the ring S also
gets the structure of an R-module, via multiplication in S.
In particular, if F ⊆ K is a field extension, K is naturally an F -vector space. We define the
degree of the extension K/F to be degree

[K : F ] := dimF K.
The extension is finite if [K : F ] < ∞. finite

Example. We have [C : R] = 2, while [R : Q] is uncountably infinite.

The following very important.
Proposition (Tower law). Suppose we have inclusions of field F ⊆ K ⊆ L. Then
[L : F ] = [L : K] [K : F ]
Proof. Note: this formula is correct even for infinite degrees, but is mainly useful when all degrees
are finite.
Let {αi }i∈I be a basis of K as an F -vector space, and let {βj }j∈J be a basis of L as a K-vector
space. I will show that the indexed collection (αi βj )i∈I, j∈J is a basis of L as an F -vector space.
6 αi0 βj 0 if (i, j) 6= (i0 , j 0 ), and thus that the cardinality of this new basis is
(This will also show αi βj =
the product of the cardinalities of the original bases.)
Date: December 2, 2022.

1
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 2

To see this, write x ∈ L as

X
x= xj βj , xj ∈ K,
j
!
X X
= yij αi βj , yij ∈ F,
j i
X
= yij αi βj ,
i,j
where there are only finitely many j ∈ J such that xj = 6 0, for each j ∈ J only finitely many i ∈ I
such that yij 6= 0, and yij = 0 for all i if xj = 0. Then there are only finitely many (i, j) ∈ I × J
such that yij 6= 0. Thus
P we have shown that {αi βj } spans L over F .
Now suppose 0 = i,j yij αi βj . Then
 
X X
0=  yij αi  βj ,
i j
P
so j yij αi = 0 for all j since {βj } is linearly independent. Then yij = 0 for all i and j since {αi }
is linearly independent. 
This gives rise to the general tower law.
Corollary (General tower law). If F = K0 ⊆ K1 ⊆ · · · ⊆ Kr = L, then [L : F ] = [Kr :
Kr−1 ] · · · [K1 : K0 ].

2. Homomorphisms of fields
By a homomorphism of fields, we mean a ring homomorphism φ : K → L between fields such homomorphism
that φ(1) = 1. We have shown that such φ are always injective (since Ker(φ) ( K is a proper ideal,
and 0 is the only proper ideal in a field). Thus I will also call such maps embeddings. embeddings
(This is just convenient language, so we don’t confuse these from other kinds of homomorphisms
of algebric objects. Also, it helps with the fact that DF still think that the constant map 0 should
be a homomorphism.)
Write Emb(K, L) for the set of field embeddings of K into L.
Easy observation: every field embedding gives a field extension. If φ : F → K is a field embedding,
then it induces an isomorphism of fields F ≈ F , where F = φ(F ) ⊆ K is the image of φ, which is a
subfield of K. So K/F is the extension associated to the embedding φ.
We write Aut(K) for the group of automorphisms, i.e., the bijective field homomorphism automorphism of a field
φ : K → K.
Maps of extensions. If K/F and L/F are two extension fields of F , a homomorphism of
extensions is a field homomorphism φ : K → L such that φ|F = idF . homomorphism of ex-
tensions
Given an extension K/F , we write Aut(K/F ) for the group of isomorphisms of K which restrict
to the identity of F . Thus Aut(K/F ) ≤ Aut(K).

3. Construction of field extensions

Let F be a field. I am going to write Irred(F ) ⊆ F [x] for the set of monic and irreducible
polynomials over F .
Given f ∈ Irred(F ), we can form
K := F [x]/(f ).
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 3

Since f is irreducible, (f ) ⊆ F [x] is a maximal ideal, so the quotient is a field. The composite
F  F [x]  K
of the inclusion map with the quotient map is an embedding F → K. We silently identify F with
its image under this embedding, and thus we regard this as a field extension F/K.
This implies silent identification: we regard F [x] as a subring of K[x].
Write α := x ∈ K for the coset x + (f ). Then we have
f (α) = 0 in K.
In other words, by forming K we have “adjoined a root” of the irreducible polynomial f to F . Note
that every element of K can be written in terms of α and elements of F , and in fact uniquely in the
from
cn−1 αn−1 + · · · + c1 α + c0 , c0 , . . . , cn−1 ∈ F, n = deg f.
Thus [K : F ] = n = deg f .
Remark. We can do “arithmetic” in K very easily. If we write elements of K in the above “canonical
form”, i.e., as an expression cn−1 xn−1 + · · · + c1 x + c0 with ck ∈ F , then addition is given by
componentwise addition of coefficients. To multiply we make use of the identity
αn = −(cn−1 αn−1 + · · · + c1 α + c0 ).
Here is how you compute multiplicative inverses in K. Given g ∈ F [x] with deg g < n and
6 0 and image g ∈ F [x]/(f ), to compute g −1 , use polynomial long division to find for each
g =
k = 0, . . . , n − 1,
xk g = −sk f + rk , sk , rk ∈ F [x], deg rk < n.
Thus, each rk is a “canonical form” representing the image of xk g in K = F [x]/(f ). Use linear
algebra to solve for ck ∈ F such that
n−1
X
ck rk = 1.
k=0
Pn−1
Then g −1 = k
k=0 ck x . (If you couldn’t find a solution, then g = 0.)

We can get another extension of F from F [x]. Let

F (x) := Frac F [x],
the field of rational functions in one variable. The composite F  F [x]  F [x] defines a field
embedding, and thus we get a field extension F (α)/F , where α = x. We have that [F (α) : F ] is
infinite. (Note: not necessarily countably infinite, e.g., [F (x) : F ] is uncountable when F is an
uncountable field. To prove this, show that { (x − c)−1 | c ∈ F } is F -linearly independent in F [x].)
Note that α is not the root of any non-zero polynomial f ∈ F [x], basically by construction and
the fact that F [x] embeds in its fraction field.

4. Field extensions generated by a set

Given a field K and a subset X ⊆ K, the subfield generated by X is the intersection of all subfield generated by
X
subfields which contain X.
Proposition. The subfield generated by X is the subset of elements which can be obtained from X
by a finite sequence of the arithmetic operations +, −, ×, ÷ applied to X ∪ {0, 1}. Such a subfield
always contains the prime subfield of K.
Proof. Straightforward: observe that the collection of elements obtain from X ∪ {0, 1} by arithmetic
operations is a subfield, and contains any subfield containing X. 
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 4

Typically, we work relative to a fixed subfield. Thus given a field extension K/F and a subset
S ⊆ K, we write F (S) ⊆ K for the subfield of K generated by F ∪ S. Typically S is finite, so we
write
F (α1 , . . . , αn ) ⊆ K
for the subfield generated by F ∪ {α1 , . . . , αn }.
A simple extension is a K/F such that K = F (α) for a single element α ∈ K. simple extension

Example. Consider
√ D ∈ Q. This has a sqaureroot in C. Pick an element α ∈ C such that α2 = D,
and call it D. √
(If D ≥ 0, then by convention we usually choose D√ > 0, but if D < 0 there is no standard
choice. It doesn’t actually matter which root we use for D in what follows, but a choice needs to
be made for the notation to make √ sense.)
Then we obtain a subfield Q( D) ⊆ C. There are two cases.
√ √ √
(1) D is the square of an element of Q, so D ∈ Q. Then Q( D) = Q, and√[Q( D) : Q] = 1.
(2) D is not the square of an element of Q. Then every element α ∈ Q( D) has a unique
representation as √
α = a + b D, a, b ∈ Q,
√
so [Q( D) : Q] = 2. √
First: the subset K = { a + b D | a, b ∈ Q } is a subfield of C. Clearly it is an abelian
subgroup, is closed under multiplication, and has 1. It remains to show it has multiplicative
inverses. √
Let α = a + b D 6= 0 with a, b ∈ Q, and consider
a −b √
β := 2 + 2 D.
a − b D a − b2 D
2

This makes sense because a2 − b2 D 6= 0, since if not then D = (a/b)2 , which contradicts the
hypothesis on D. √
For uniqueness of the representation, note that if a + b D = √ 0 for some a, b ∈ Q, then
either (i) b = 0, whence a = 0, or (ii) b 6= 0, whence D = ( D)2 = (−a/b)2 , which
contradicts the hypothesis. Thus we must have (a, b) = (0, 0).
√ √
Note: Q( D) = Q( E) iff E = √ Dc2 for some c ∈ Q r {0}. (Exercise: prove this.) Thus WLOG
any of these fields has the form Q( D) √ for a squarefree integer D (i.e., p2 - D for all primes p).
If D 6= 1 is a squarefree integer then D ∈ / Q (by unique factorization in Z, since a2 = Db2 for
some a, b ∈ Z implies that any prime divides D an even number of times).
Example. I claim that
√ √ √ √ √
Q( 2, 3) = { a + b 2 + c 3 + d 6 | a, b, c, d ∈ Q },
√ √
and that [Q( 2, 3) : Q] = 4. √ √
√ The key
√ fact I need√ is that 3
√ ∈
/ Q( 2). This is just a direct calculation. That is, suppose
3 ∈ Q( 2), so that 3 = a + b 2 for some a, b ∈ Q. Then
√
3 = (a2 + 2b2 ) + 2ab 2 =⇒ a2 + 2b2 = 3, 2ab = 0.
√ p
But then 2ab = 0 implies we must have either 3 = a2 or 3/2 = b2 , i.e., that 3 ∈ Q or 3/2 ∈ Q,
but this is not the √ case. √
Let K = { u + v 3 √| u, v ∈ Q( 2) }. √ This is clearly a subring of C (with 1). To see that it is a
subfield,
√ let
√ α = u + v 3 with u, v ∈ Q( 2). Then u2 − 3v 2 6= 0, since otherwise 3 = (u/v)2 and so
3 ∈ Q( 2). Then it is easy to check that
u −v √
α−1 = 2 2
+ 2 2 ∈ K.
u − 3v u − 3v 2
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 5
√ √ √ √ √
√ Since√K = Q( 2)( 3) and 3 ∈ / Q( 2), we have [K : Q( 2)] = 2 and so [K : Q] = [K :
2] [Q( 2), Q] = 4.
√ √ √ √ √ √
Exercise. Q( 2, 3) = Q(α) where α = 2 + 3. Thus Q( 2, 3)/Q is a simple extenison.
√ √ √ √ √
Example. We have that [Q( 8, 32) : Q] = 2, since 32 = 2 8 ∈ Q( 8).
√ √ √ √ √
Exercise. If D, E ∈ Q, then [Q( D, E) : Q] = 4 iff D, E, DE ∈ / Q.
√ √ √ √
Example. Q( 3 2) = { a + b 3 2 + c 3 4 | a, b, c ∈ Q }, and [Q( 3 2) : Q]
√ = 3. √
√ Note that√ to show this, one must prove that the form a + b 2 + c 3 4 is unique (i.e., that
3

3
4 6= a + b 3 2), and that multiplicative inverses of elements of that form have the same form. This
can be done, but we will soon have a better method.

5. Algebraic and transcendental elements

Let K/F be a field extension, and suppose α ∈ K. Consider the subfield F (α) ⊆ K generated F 4 Nov
over F by α. Observe that we can always evaluate a a polynomial f ∈ F [x] at α.
There are two cases:
(1) There exists a non-zero f ∈ F [x] such that f (α) = 0. In this case we say that α is algebraic algebraic
over F .
(2) There does not exist a non-zero f ∈ F [x] such that f (α) = 0. In this case we say that α is
transcendental over F . transcendental

Proposition. Let α ∈ K be algebraic over F . Then there exists a unique irreducible monic
polynomial m ∈ Irred(F ) such that m(α) = 0. Furthermore, a polynomial f ∈ F [x] has α as a root
iff m | f in F [x].
Proof. There exists a unique ring homomorphism ψ : F [x] → K (preserving 1), such that ψ|F = idF ,
and ψ(x) = α. This homomorphism is given by evaluation: ψ(f ) = f (α), and so f (α) = 0 iff
f ∈ Ker(ψ).
Since α is algebraic, Ker(ψ) 6= (0), so there exists a unique monic polynomial m ∈ F [x] such that
(m) = Ker(ψ).
By the isomorphism theorem, φ factors through a ring isomorphism
∼
φ : F [x]/(m) −
→ φ(F [x]) ⊆ K.
Since φ(F [x]) is a subring of a field (with 1), it is an integral domain. Thus f g ∈ (m) implies either
f ∈ (m) or g ∈ (m), and in particular m is irreducible. 
For any α ∈ K algebraic over F , we write m = mα/F ∈ Irred(F ) for the unique monic irreducible
such that m(α) = 0. It is called the minimal polynomial of α over F . minimal polynomial
This proof also gives the following.
Proposition. If α ∈ K is algebraic over F with minimal polynomial m = mα/F , then there is a
unique isomorphism of F -extensions
∼
φ : F [x]/(m) −
→ F (α), such that φ(x) = α.
As a consequence, [F (α) : F ] = deg mα/F .
Proof. We already have a injective ring homomorphism
φ : F [x]/(m)  K, φ(x) = α,
with m irreducible. We have shown that F [x]/(m) is therefore a field, and thus its image L ⊆ K in
K is a subfield. Clearly F [x]/(m) is the generated over F by x, and thus L is generated over F by
α = φ(x).

 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 6
√
Example. The real number α = 3 2 is a root of f = x3 − 2 ∈ Q[x]. This f has no roots in Q, so is
irreducible over Q since deg f = 3. Therefore
Q(α) ≈ Q[x]/(f ).
As a consequence, [Q(α) : Q] = 3, and all elements of Q(α) have a unique expression a + bα + cα2 ,
a, b, c ∈ Q.
What if α is transcendental?
Proposition. If α ∈ K is transcendental over F , then there is a unique isomorphism of F -extensions
∼
φ : F (x) −
→ F (α), such that φ(x) = α,
where F (x) = Frac F [x] is the field of rational functions in one variable over F . As a consequence,
[F (α) : F ] is infinite.
Proof. There is a unique ring homomorphism ψ : F [x] → K such that ψ|F = idF and ψ(x) = α.
Because α is transcendental, Ker(ψ) = (0), so ψ is injective. Therefore ψ extends over the fraction
field F (x) = Frac F [x]:
ψ
F [x] / /K
=


 φ
F (x)

We get a complete classification of simple extensions.
Proposition. Suppose K/F is a field extension with K = F (α) for some α ∈ K. There are two
cases:
(1) [K : F ] < ∞. Then α is algebraic over F , and there is unique isomorphism of F -extensions
of the form
φ : F [x]/(f ) → K, f = mα/F ∈ Irred(F ), φ(x) = α.
(2) [K : F ] = ∞. Then α is transcendental over F , and there is a unique isomorphism of
F -extensions of the form
φ : F (x) → K, φ(x) = α.

6. Finitely generated extensions

An extension K/F is a finite extension if [K : F ] < ∞. finite extension
An extension K/F is a finitely generated extension if K = F (α1 , . . . , αn ) for some finite list finitely generated ex-
tension
of elements α1 , . . . , αn ∈ K.
Note: every finite extension is finitely generated (use a vector space basis as the generating set).
The converse is not true (e.g., a simple transcendental extension).
Proposition. Let L/F be an extension, and α1 , . . . , αn ∈ L a finite list of elements. Let K =
F (α1 , . . . , αn ). Then TFAE.
(1) [K : F ] < ∞.
(2) Every element β ∈ K is algebraic over F .
(3) The element αk is algebraic over F for all k = 1, . . . , n.
Furthermore, if any of these hold, then [K : F ] ≤ d1 · · · dn where dj = deg mαj /F .
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 7

Proof. (1) =⇒ (2). If β ∈ K then [K : F ] = [K : F (β)] [F (β) : F ] < ∞ implies [F (β) : F ] < ∞, and
thus β is algebraic over F .
(2) =⇒ (3). Immeditae.
(3) =⇒ (1). Let Kj = F (α1 , . . . , αj ). By the tower law [K : F ] = [Kn : Kn−1 ] · · · [K1 : K0 ].
Since αj ∈ Kj is algebraic over F , it is algebraic over Kj−1 , since F ⊆ Kj−1 ⊆ Kj . Thus
[Kj : Kj−1 ] = [Kj−1 (αj ) : Kj−1 ] < ∞ for all j = 1, . . . , n, so [K : F ] < ∞.
For the final statement, we use the following lemma to get that
[Kj : Kj−1 ] ≤ [F (αj ) : F ] = dj .

Lemma. Let F ⊆ K ⊆ L and α ∈ L, such that α is algebraic over F and [K : F ] < ∞. Then
[K(α) : K] ≤ [F (α) : F ] and [K(α) : F (α)] ≤ [K : F ].

[K(α):K] K(α)
K [K(α):F (α)]

[K:F ] F (α)
F [F (α):F ]

Proof. We have [F (α) : F ] = d = deg mα/F and [K(α) : K] = e = deg mα/K . Since mα/K is the
minimal polynomial of α over K, it divides any polynomial f ∈ K[x] which has α as a root. In
particular, mα/K | mα/F in F [x], and so e ≤ d. The second claim follows from the tower law:
[K(α) : F (α)] [F (α) : F ] = [K(α) : K] [K : F ]. 
Given subfields F ⊆ K, K 0 ⊆ L, the composite extension is the subfield of L generated over composite extension
F by K ∪ K 0 . It is usually written KK 0 ⊆ L (but be careful: it is not in general a set of linear
combinations of products).
Clearly, if K = F (α1 , . . . , αm ) and K 0 = F (β1 , . . . , βn ), then KK 0 = F (α1 , . . . , αm , β1 , . . . , βn ).
Proposition. If K/F and K 0 /F are subextensions of L/F which are finite over F , then
[KK 0 : K] ≤ [K 0 : F ], [KK 0 : K 0 ] ≤ [K : F ], [KK 0 : F ] ≤ [K : F ][K 0 : F ].
Proof. Factor K 0 /F as a sequence of simple extensions F = K00 ⊆ · · · ⊆ Kn0 = K 0 with Kj0 =
0
Kj−1 (αj ), and use the tower law and the previous lemma. 

7. Algebraic extensions
We say K/F is algebraic if every α ∈ K is algebraic over F . algebraic extension
We have seen that every finite extension is algebraic. There are, however, algebraic extensions
which are not finite.
Example. An algebraic number is an α ∈ C which is algebraic over Q, i.e., is the root of some algebraic number
non-zero f ∈ Q[x].
Let Qalg be the set of algebraic numbers. Then Qalg is a subfield of C, by the following proposition.
The extension Qalg /Q is obviously algebraic, but is infinite.
That Qalg is a subfield follows from the following.
Proposition. If L/F is an extension and α, β ∈ L are algebraic over F , then α + β, αβ, −α, α−1
are algebraic over F .
Proof. Since F (α)/F and F (β)/F are finite extensions, the composite extension F (α, β)/F is also
finite and thus algebraic. Since α + β, αβ, −α, α−1 ∈ F (α, β), these are algebraic elements. 
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 8

√ √
Exercise. Let p1 , . . . , pr be distinct prime numbers. Show that [Q( p1 , . . . , pr ) : Q] = 2r . This
implies that Qalg /Q is an infinite extension.
Note: We can also say that Qalg = L, where the union is over subfields L ⊆ C which are finite
S
over Q.
If we stack algebraic extensions, it is still algebraic.
Proposition. If F ⊆ K ⊆ L such that K/F and L/K are algebraic extensions, then L/F is an
algebraic extension.

P α ∈ L. Since L/K is algebraic, there exists a non-zero f ∈ K[x] with α as a root.

Proof. Suppose
Write f = nk=0 ck xk with ck ∈ K, and let F 0 = F (c0 , . . . , cn ). Since the ck are in K they are
algebraic over F , so [F 0 : F ] < ∞, and thus [F 0 (α) : F ] = [F 0 (α) : F 0 ] [F 0 : F ] ≤ (deg f )[F 0 : F ] is
finite. Thus α is algebraic over F . 

8. Algebraically closed fields

We say that K is algebraically closed if every non-constant f ∈ K[x] has a root in K. algebraically closed
If f has a root c ∈ K then f factors as (x − c)g ∈ K[x] with g ∈ K[x], and we can apply the same
argument to g (unless g is constant). Thus K is algebraically closed iff every non-zero polynomial over
K splits over K, i.e., is a product of degree 1 polynomials. That is, if Irred(K) = { x − c | c ∈ K }. splits
To check that K is algebraically closed, it sufffices to check only that all f ∈ Irred(K) have a
root in K.
Example. The complex numbers C is algebraically closed, by “The Fundamental Theorem of
Algebra”, which is really a theorem of analysis.
Proposition. K is algebraically closed iff for any algebraic extension L/K we have L = K, iff for
any α ∈ L in an extension L of K which is algebraic over K, we have α ∈ K.
Proof. If K is algebraically closed and α ∈ L in some algebraic extension L/K, then α is the root
of some non-zero f ∈ K[x], which splits over F and thus α ∈ K. Therefore L = K.
Suppose K is such that the only algebraic extension L/K has L = K. If f ∈ K[x] is non-constant,
form K(α)/K where f (α) = 0. Then by hypothesis K(α) = K, whence α ∈ K. 
Example. The field Qalg of algebraic numbers is algebraically closed, since if α is algebraic over Qalg ,
then it is algebraic over Q (since Qalg /Q is an algebraic extenison).

9. Algebraic closure of fields

An algebraic closure is an extension F /F which is algebraic, and is such that every non-constant M 7 Nov
algebraic closure
polynomial f ∈ F [x] splits over F , i.e., is a product of degree 1 factors.
Recall that K is algebraically closed if every f ∈ K[x] has a root in K (and thus splits over K).
Proposition. Given an extension K/F , we have that K is an algebraic closure of F iff (i) K/F is
algebraic, and (ii) K is algebraically closed. Algebraic closures are algebraically closed.
Proof. =⇒. (i) is immediate from the definition. For (ii), suppose f ∈ K[x] non-constant. We need
to show f splits over K. Form an extension K(α)/K with α a root of f . Then since both K(α)/K
and K/F are algebraic, so is K(α)/K, so there exists a non-zero g ∈ F [x] with g(α) = 0. But by
definition of algebraic closure, g splits over K, so α ∈ K.
⇐=. Property (i) says K/F is algebraic. It remains to show that every non-constant f ∈ F [x]
splits over K, but this is immediate from (ii), since such an f is also in K[x], and algebraic closure
implies these split over K. 
Example. Any algebraically closed field is its own algebraic closure. Thus F = F for any F .
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 9

Example. The field Qalg of algebraic numbers is an algebraic closure of Q, since Qalg /Q is algebraic
and Qalg is algebraically closed.
Proposition. If K/F is an extension and K is algebraically closed, then K contains a unique
algebraic closure F of F , which is equal to the subset of elements which are algebraic over F .
Proof. A special case of this is F = Q, K = C, and F = Qalg . The general proof is no different.
Let F be the set of all elements of K which are algebraic over F ; this is a subfield (as we saw in
the case of Qalg ). Note that any α ∈ K which is algebraic over F is also algebraic over F (since then
α is algebraic over the finite subextension of F generated by coefficients of its minimal polynomial).
Thus for any f ∈ F [x], since it splits completely over K into linear factors x − αi where the αi
are algebraic over F , hence over F , we have that αi ∈ F . Therefore F is algebraic over F (since any
f ∈ F [x] is also in F [x]), and is algebraically closed. 

10. 2-radical extensions

We basically understand degree 2 extensions (as long as we are not in characteristic 2).
√
Proposition. Let K/F be an extension, with char(F ) 6= 2 and [K : F ] = 2. Then K = F ( d) for
some d ∈ F which is not a square in F .
Proof. Because [K : F ] = 2, we have K = F (α) for any α ∈ K r F . Let f = mα/F = x2 + bx + c
with b, c ∈ F , and let d = b2 − 4c. Then
(2α + b)2 = 4α2 + 4bα + b2 = 4(−bα − c) + 4bα + b2 = b2 − 4c = d,
√ √ √
so we can set √d := 2α+b ∈ K. Clearly d ∈ / F , since otherwise we would have α = (−b+ d)/2 ∈ F .
Thus K = F ( d). 
√
Remark. If char F = 2, then if we try to do this it turns out that d = b ∈ F , so it does not
generate K over F .
Let’s say that a finite extension K/F is 2-radical1 if there exists a finite tower of subfields of the 2-radical extension
form
F = K0 ⊆ K1 ⊆ · · · ⊆ Kr = K, [Kj : Kj−1 ] = 2.
Example. In each of √ √ √ √
4 8 16
Q ⊆ Q( 2) ⊆ Q( 2) ⊆ Q( 2) ⊆ Q( 2) ⊆ · · ·
and
r s r
√ √ √ √
q q q
Q ⊆ Q( 2) ⊆ Q( 1 + 2) ⊆ Q( 1 + 1 + 2) ⊆ Q( 1 + 1 + 1 + 2) ⊆ · · · ,
each intermediate extension has degree 2. (Exercise: prove this.)
Remark. If K/F is 2-radical then [K : F ] = 2r for some r, but the converse is not true. (We will see
an example later.)
Proposition. If K/F and K 0 /F are finite subextensions of L/F which are 2-radical, then the
composite extension is 2-radical.
Proof. Factor K/F as a sequence of degree 2-extensions Kj = Kj−1 (αj ) with K0 = F and Kr = K.
Then we have a chain of extensions
K 0 = K 0 K0 ⊆ K 0 K1 ⊆ K 0 K2 ⊆ · · · ⊆ K 0 Kn = KK 0 .
Each intermediate extension is a simple extension since K 0 Kj = K 0 Kj−1 (αj ), and we know that
[K 0 Kj : K 0 Kj−1 ] ≤ [Kj : Kj−1 ] = 2. 
1I don’t think this is standard terminology.
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 10

Given F ⊆ C, let [
F 2rad = L.
F ⊆L⊆C
L/F is 2-radical
Thus, α ∈ F 2rad iff and only if there exists a finite 2-radical extension L : F with α ∈ L. We see
that F 2rad is a subfield of C, using that if L, L0 are 2-radical extensions over F then so is LL0 .
Say that a field K is squareroot closed if every element of K has a squareroot in K. squareroot closed

Proposition. For F ⊆ C, the subfield F 2rad is the smallest subfield containing F which is squareroot
closed.
Proof. First, F 2rad is clearly squareroot√closed. If α ∈ F 2rad , then there exists a finite 2-radical
extension L/F with α ∈ L, and then L( α)/F is also a 2-radical extension.
Let L ⊆ C be any squareroot closed
√ subfield containing F . Clearly if K ⊆ L, and given K 0 /K
with [K 0 : K] = 2, then K 0 = K( d) for some d ∈ K, and so K 0 ⊆ L. Using this it is easy to see
that any 2-radical extension of F is contained in L, so F 2rad ⊆ L. 

11. Straightedge and compass constructions

Greek geometers developed the notion of “straightedge and compass” constructions in plane
geometry. The idea is that given a set P of points in the plane, you are allowed to construct:
• a line between any two distinct points of P (“straightedge”), and
• a circle with center at a point of P and radius r = distance between two given distinct
points in P (“compass”).
From this you get a bigger set of points P 0 , which contains P as well as all points of intersection
of the lines and circles you drew. You can iterate this to get a bigger set P 00 and so on, until you
obtain the set P of points which are “constructible from P” by ruler and compass.
Note: you need at least 2 points in P to start with.
The problem is to use this to make geometric constructions. For instance, you can:
• Bisect an angle.
• Raise a perpendicular. (I.e., given two points, form a square with those two points as
adjacent vertices.)
• Construct a regular triangle or regular pentagon.
The following problems were unsolved by Greek geometers (by straightedge and compass con-
struction):
• Trisect an angle.
• “Square the circle.” (I.e., given a circle, construct a square with the same area.)
• “Duplicate the cube.” (I.e., given a line segment L construct a line segment L0 so a cube
with edge L0 has twice the volume of a cube with edge L.)
• Construct a regular 7-gon.
In fact, these are all impossible.
Given a set of points P in the plane, designate two of them as 0 and 1, so that we can identify the
plane with C and thus P ⊆ C. Let F ⊆ C be the subfield generated by the P. (You can show that
the subfield F doesn’t depend on which points you choose as 0 and 1.) Then we get the following.
Theorem. A point α is constructible from P iff α ∈ F 2rad .
If P = {0, 1}, then this is F 2rad = Q2rad . We say that α ∈ C is constructible if α ∈ Q2rad . constructible number
The proof is involved. I like the account in Stewart, Galois Theory. (DF §13.3 talks about it too,
but their way of setting things up is a little different.) The basic idea is:
(1) Show how to carry out field operations in C using ruler and compass constructions, and also
how to calculate square roots. Thus all elements of F 2rad are constructible from P.
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 11

(2) Show that all all points constructible from P are in F 2rad . This amounts to showing that
computing intersection points of lines and/or cicles involves solving polynomial equations of
degree at most 2.
Here are some impossibility results which follow from this, using only the fact that α ∈ Q2rad
must have [Q(α) : Q] = 2r .
√ √
• Cannot duplicate the cube. Given r we want to produce r 3 2, i.e., to construct α = 3 2. But
[Q(α) : Q] = 3.
• Cannot trisect every angle. In particular, θ = 2π/3 cannot be trisected. This amounts to
showing that ζ := e2πi/9 is not constructible.
We know that ζ 9 = 1, but ζ 3 6= 1. Since 0 = ζ 9 − 1 = (ζ 3 − 1)(ζ 6 + ζ 3 + 1), we see that
ζ is a root of f = x6 + x3 + 1 ∈ Q[x], so [Q(ζ) : Q] ≤ 3. Let α = ζ + ζ −1 ∈ Q(ζ). Using
f (ζ) = 0, you can show that
α3 = ζ 3 + 3ζ + 3ζ −1 + ζ −3 = 3α − 1.
So α is a root of g = x3 − 3x + 1 ∈ Q[x]. By the rational roots test this has no root in Q so
is irreducbile over Q. Thus [Q(α) : Q] = 3. Since [Q(ζ) : Q] = [Q(ζ) : Q(α)] [Q(α) : Q], we
see that 3 divides [Q(ζ) : Q].
• Cannot
√ square
√ the circle. That is, given a circle with radius r, produce a square with side
π r. But
√ π is not constructible. If it were, then it would be algebraic over Q, and thus
π ∈ Q( π) would be algebraic over Q, but by Lindemann’s theorem it is not.
• Cannot construct the regular heptagon. Show that ζ := e2πi/7 ∈ / Q2rad . Its minimal
6
polynomial over Q is Φ7 = x + · · · + x + 1, so [Q(ζ) : Q] = 6.
Remark. If p is a prime number, then ζp = e2πi/p satisfies
Q(ζp ) = deg Φp = p − 1,
since Φp is irreducible over Q. Thus, it is impossible to construct a regular p-gon for a prime p,
unless it is a Fermat prime, i.e., of the form p = 2m + 1. Examples include Fermat prime
1 2 4 8 16
3 = 2 + 1, 5 = 2 + 1, 17 = 2 + 1, 257 = 2 + 1, 65537 = 2 + 1.
These are the only known examples, and it is unknown whether there are more. (Note that a Fermat
d
prime must always have the form 22 + 1 for some d, because a + 1 | ak + 1 whenever k is odd.)

12. Splitting fields

Let f ∈ F [x] with f 6= 0. A splitting field of f is an extension Σ/F such that splitting field

• f splits over Σ, i.e., f = c(x − α1 ) · · · (x − αn ) for some c, α1 , . . . , αn ∈ Σ with c 6= 0, and

• Σ is generated over F by the roots of f , i.e., Σ = F (α1 , . . . , αn ). (Equivalently: the only
subfield of Σ over which f splits is Σ itself.)
If a polynomial splits over some field then we certainly get a splitting field.
Proposition. If L/F is an extension and f ∈ F [x] splits over L, then the subfield Σ = F (α1 , . . . , αn )
generated by the roots of f in L is a splitting field of f .
Proof. Obvious. 
√
Example. If f = (x2 + 1)(x2 − 5) ∈ Q[x], then Σ = Q(i, 5) is a splitting field.
We always have an “abstract” construction of splitting fields.
Proposition. Every non-zero polynomial f ∈ F [x] admits a splitting field.
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 12

Proof. We use induction on degree n = deg f . If n ≤ 1 take Σ = F .

Suppose n ≥ 2. Let p be any irreducible factor of f , so 1 ≤ deg p ≤ n. Let K := F [x]/(p) = F (α1 )
with α1 = x, and view this as an extension of F .
Over K we have f = (x − α1 )g, g ∈ K[x] with deg g = n − 1. By induction g has a splitting
field Σ/K. I claim that Σ/F is a splitting field for F . Over Σ we have g = c(x − α2 ) · · · (x − αn )
for α2 , . . . , αn ∈ Σ and c ∈ K × , and Σ = K(α2 , . . . , αn ). Thus f = c(x − α1 ) · · · (x − αn ) and
Σ = F (α1 , . . . , αn ). 
This argument also shows the following.
Corollary. If Σ/F is a splitting field of f ∈ F [x], then [Σ : F ] ≤ n! where n = deg f .
Proof. Induction on degree, using that if α ∈ Σ is a root of f , then [F (α) : F ] = deg mα/F ≤ n and
Σ/F (α) is a splitting field of a polynomial of degree n − 1, so [Σ : F ] = [Σ : F (α)] [F (α) : F ] ≤
(n − 1)! · n. 
Notation: I will sometimes write Σ = Σf /F for some choice of splitting field of f ∈ F [x]. (The
splitting field is not unique. However, we will see that it is unique up to isomorphism.)

13. Examples of splitting fields

Example (Cyclotomic extensions). Let ζ ∈ L be a primitive nth root of unity, i.e., an element of W 9 Nov
order n in L× . Then for F ⊆ L, the subfield K = F (ζ) is the splitting field of f = xn − 1.
This is because, since |ζ| = n, the elements 1, ζ, . . . , ζ n−1 are pairwise distinct, and are all clearly
roots of f contained in K. Thus f = (x − 1)(x − ζ) · · · (x − ζ n−1 ) and clearly K is generated over F
by the roots.
The degree of the extension [F (ζ) : F ] will be less than n since f = (x − 1)g (unless n = 1).
Let ζn = e2πi/n ∈ C. The field Q(ζn ) is called a cyclotomic field. We know that if n = p is cyclotomic field
prime, then [Q(ζp ) : Q] = p − 1, since Φp is irreducible over Q.
Example (Splitting field of xp − 2). Let p be a prime number, and let f = xp − 2 ∈ Q[x]. Note that
f is irreducible by Eisenstein’s√criterion.
If α is an root of this (e.g., p 2 ∈ R), so is αζ k where ζ is some fixed primitive pth root of unity.
That is, the roots of f in C are
α, αζ, . . . , αζ p−1 .
As these are distinct (since ζ k =
6 1 if p - k), these are distinct roots of f , so Σ = Q(α, ζ) is a splitting
field of f . (Note that ζ = (αζ)α−1 can be written in terms of roots of f , so it must be in the
splitting field.)
We have the following diagram of subfields.
Σ = Q(α, ζ)

Q(α)
p Q(ζ)
p−1
Q
Since [Σ : Q(ζ)] ≤ [Q(α) : Q] = p, we have [Σ : Q] ≤ (p − 1)p, but it is necessarily divisible by both
p and p − 1, so (since these are relatively prime), [Σ : Q] = p(p − 1).
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 13

14. Separable polynomials

Given a polynomial
f = a0 + a1 x + · · · + an xn ∈ F [x],
define its formal derivative by the formula formal derivative
n−1
Df := a1 + 2a2 x + · · · + nan x ∈ F [x].
Exercise. Check that the formal derivative satisfies: D(f + g) = Df + Dg, D(f g) = (Df )g + f (Dg),
Dc = 0 and D(cf ) = cD(f ) if c ∈ F .
We say that a non-zero polynomial f ∈ F [x] is separable if f and Df are relatively prime in separable polynomial
F [x]. That is, if (f, Df ) is the unit ideal in F [x].
Exercise. Show that if f = gh ∈ F [x] and f is a separable polynomial, so are g and h.
Exercise. Show that any degree 1 polynomial is separable.
Remark. If F ⊆ K and f ∈ F [x], then f is separable as a polynomial over F iff it is separable as a
polynomial over K.
To see this: (i) If f is separable over F , then 1 = uf + vD(f ) for some u, v ∈ F [x], and since
this equation also holds in K[x], f is also separable over K. (ii) If f is separable over K, then any
common divisor d ∈ F [x] of {f, D(f )} is also a common divisor of these in K[x], so d ∈ F × , and
thus f is separable over F .
Thus, we don’t need to say things like “separable over K”, we just say f is separable.
Exercise. Show that if φ : F → K is a homomorphism of fields, then f ∈ F [x] is a separable
polynomial iff φ(f ) ∈ K[x] is a separable polynomial.
You probably know that α ∈ R is a repeated root of some f ∈ R[x] if and only if α is a critical
point of R. This is actually a purely algebraic fact.
Proposition. Let L/F be any extension over which f ∈ F [x] splits. Then f is separable iff f has
no multiple roots in L, iff f and Df have no common roots in L.
Proof. Since f = c(x − α1 ) · · · (x − αn ) over L, it suffices to show for each k that (x − αk )2 | f iff
x − αk | Df . In fact, since f = (x − αk )h with h ∈ L[x], we have
Df = h + (x − αk )(Dh),
so (x − αk )2 | f iff (x − αk ) | h iff (x − αk ) | Df . 
Thus, we can say f is separable if and only if it has simple roots over its splitting field.
Example. The polynomial f = x4 + 2x2 + 1 ∈ Q[x] has Df = 4x3 + 4x. It is not hard to see (e.g.,
using the Euclidean algorithm) they have a common factor x2 + 1. Thus f is not separable. In fact,
f = (x2 + 1)2 over Q, and f = (x − i)2 (x + i)2 over C, so all roots are multiple roots.
Example. The polynomial f = xn − 1 is separable over Q since Df = xn−1 and x - f . Thus f has n
distinct roots over C, as we know.
Here is a generalization, which does not require passing to an extension where f splits.
Proposition. A non-zero polynomial f ∈ F [x] is separable iff for some irreducible factorization
f = g1 · · · gn over F , we have that (i) each gk is separable, and (ii) there are no repeated factors,
i.e., if i 6= j then gi - gj .
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 14

Proof. I’ll show that an irreducible factor g of f divides Df iff either (i) g is not separable or (ii)
g 2 | f . Given this, the claim is immediate from the existence of irreducible factorizations.
So let g ∈ Irred(F ) such that g | f , so f = gh for some h ∈ F [x]. The claim is immediate from
the identity
Df = (Dg)h + g(Dh)
and the fact that g is a prime element of F [x], so that g | Df iff g | (Dg)h iff (g | Dg or g | h) iff
(g | Dg or g 2 | f ). 

15. Irreducible separable polynomials

We would like to be able to say that irreducible polynomials are always separable. This is true
for fields of characteristic 0, e.g., subfields of C, but it not quite true generally.
Proposition. Suppose f ∈ F [x] is irreducible. Then f is separable iff Df 6= 0.
In particular, if char F = 0, all irreducible polynomials over F are separable.
Proof. We have that deg Df < deg f , and if also Df 6= 0 we have f - Df . Since f is irreducible,
this means f and Df are relatively prime. If instead Df = 0, then f and Df are not relatively
prime since f | Df . 
The problem with non-0 characteristic is that it is not generally the case that deg Df = deg f − 1,
and in fact it is possible that Df = 0 even when f is not a constant polynomial.
Example. Consider a field of prime characteristic p and
f = xp − a ∈ F [x], a ∈ F.
Then Df = pxp−1 = 0, so deg Df = −∞ rather than p − 1. So f actually does divide Df in this
case. So f is not separable.
Let b be a root of f in some extension field K = F (b). Then bp = a, and in fact
X p
p
(x − b) = xj (−b)j = xp + (−1)p bp = xp − bp = x − a = f.
j
j=0

Thus f can only have one root in any splitting field, which has multiplicity p. (Note: (−1)p ≡ −1
mod p for every prime p, but the reason is different depending on whether p is odd or even!)
It turns out that there are actually irreducible polynomials of this form. In fact, we’ll show later
that in the above setup, if b ∈/ F then f is irreducible over F . As an example, consider the field
F = Fp (t) of rational functions over Fp and let f = xp − t ∈ F [x]. It turns out that t does not have
a pth root in F , and so f is irreducible over F but not separable. (We will discuss this example
soon.)
You can tell immediately from the form of a polynomial in finite characteristic when its derivative
is 0.
Proposition. If f ∈ F [x] with char F = p 6= 0, then Df = 0 iff f = nk=0 ck xpk , iff f = g(xp ) for
P
some g ∈ F [x].
Example. Let n ≥ 1 and f = xn − 1 ∈ F [x]. For char F = p 6= 0, the polynomial f is separable as
long as p - n.
On the other hand, xp − 1 = (x − 1)p . So in characteristic p, there are no primitive pth roots of
unity.
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 15

16. Homomorphisms from simple finite extensions

Given an embedding λ : F  F 0 of fields, we get an induced homomorphism of polynomial rings,
which by abuse of notation we also call λ:
X X
λ : F [x] → F 0 [x], λ( ck xk ) := λ(ck )xk , ck ∈ F.
Proposition. Let F (α)/F be a finite extension, where α has minimal polynomial m ∈ F [x]. Suppose
given an embedding of fields λ : F  F 0 , and an extension L/F 0 . Let m0 = λ(m) ∈ F 0 [x].
Then for any root root β ∈ L of m0 , there exists a unique embedding µ : F (α) → L such that
µ|F = λ and µ(α) = β.
α7→β
F (α) / µ
/L

F / F0
λ
That is, there is a bijection
   
µ : F (α)  L o / β∈L
such that µ|F = λ such that m0 (β) = 0

µ / µ(α)

In particular, the number of such homomorphisms is equal to the number of distinct roots of
m0 = λ(m) in L.
Proof. Recall that there is an isomorphism F (α) ≈ F [x]/(m) of extensions of F , under which α
corresponds to x. WLOG we can assume F (α) = F [x]/(m).
Now we use various universal properties to describe ring homomorphisms (preserving 1) of the
form µ : F [x]/(m) → L. In particular, given λ : F → F 0 ⊆ L and β ∈ F , there exists a unique ring
homomorphism
e : F [x] → L,
µ e|F = λ0 , µ
µ e(x) = β.
By the homomorphism theorem for quotients, this factors through the quotient map F [x] → F [x]/(m)
iff µ e(m) = λ(m)(β) = 0, such a factorization µ : F [x]/(m) → L exists iff m0 (β) = 0,
e(m) = 0. Since µ
and there is only one such factorization.
F [x]/(m)
:: $
x7→β

x7→β $/
F [x] LO
O µ
e

O O
F / F0
λ


Often we just need the special case when λ = idF .
Corollary. Let F (α)/F be a finite extension, where α has minimal polynomial m ∈ F [x]. Suppose
an extension L/F .
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 16

Then for any root root β ∈ L of m, there exists a unique embedding µ : F (α) → L such that
µ|F = idF and µ(α) = β.
α7→β
F (α) / µ
/L

F
That is, there is a bijection
   
µ : F (α) → L o / β∈L
such that µ|F = idF such that m(β) = 0

µ / µ(α)

In particular, the number of such homomorphisms is equal to the number of roots of m in L.

√
Example. Let f = x3 − 2 ∈ Q[x], and let α = 3 2. Since f = mα/Q , there are three distinct
embeddings µ : Q(α) → C, corresponding to the roots α, αω, αω 2 , where ω = e2πi/3 .

17. Homomorphisms from splitting fields

Given a finite extension K/F , we now have a recipe for constructing homomorphisms of extensions F 11 Nov
K → L over F : write K/F as a composite of simple extensions
F = K0 ⊆ K1 ⊆ K2 ⊆ · · · ⊆ Kn = K, Kj = Kj−1 (αj ),
and inductively construct homomorphisms φj : F (α1 , . . . , αj ) → L extending φj−1 . At each step
there is one choice: φj (αj ) ∈ L can be any root of mαj /Kj−1 .
Proposition. Consider
∼
• an isomorphism of fields λ : F − → F 0,
• a non-zero polynomial f ∈ F [x],
• a splitting field Σ/F of f , and
• an extension L/F 0 over which f 0 = λ(f ) ∈ F 0 [x] splits.
Then there exists a homomorphism φ : Σ → L such that φ|F = λ. The image φ(Σ) of φ is a splitting
field of f 0 over F 0 .
Proof. We use induction on deg f . If f is constant, then Σ = F and we take φ = λ. So suppose
deg f ≥ 1, whence deg f 0 = deg f ≥ 1.
Let α1 be some root of f in Σ, and let m = mα1 /F ∈ Irred(F ) be its minimal polynomial. Then
f = mg for some g ∈ F [x]. Under λ : F [x] → F 0 [x] we get a factorization f 0 = m0 g 0 with m0 = λ(m).
By hypothesis f 0 splits over L, so we choose a root β1 ∈ L of m0 .
By the previous proposition, there exists an isomorphism φ1 fitting in
Σ L

F (α1 )
∼ / F 0 (β1 )
φ1

F
∼ / F0
λ

(It is an isomorphism because F 0 (β1 ) = φ1 (F (α1 )) is the image of φ1 .) We can factor f = (x − α1 )h

over F (α1 ). Now note that we are in the same sitation: we have
• an isomorphism of fields φ1 : F (α1 ) → F 0 (β1 ),
• a nonzero polynomial h ∈ F (α1 )[x],
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 17

• a splitting field Σ/F (α1 ) of h, and

• an extension L/F 0 (α1 ) over which φ1 (h) splits.
Since deg h < deg f , induction applies to produce the desired homomorphism φ. 
This is important even when F = F 0 and ι = idF . In this case, we see that there is an embedding
of extensions Σ → L whenever g splits over L. In particular, we learn that any two splitting fields
of f are isomorphic.
Corollary. Let Σ/F and Σ0 /F be two splitting fields for the same non-zero polynomial f ∈ F [x].
Then Σ and Σ0 are isomorphic as extensions over F .
Proof. By the previous proposition, there exists a map φ : Σ → Σ0 of F -extensions, whose image
φ(Σ) is a splitting field of F , and thus φ(Σ) = Σ0 . 
Important: this isomorphism of splitting fields is not generally unique. In particular, Aut(Σ/F )
is usually not a trivial group.
Notation: I sometimes write Σ = Σf /F for any choice of a splitting field of f ∈ F [x].

18. Automorphisms of field extensions

Proposition. Suppose G ≤ Aut(K) is a group of automorphisms of a field K. Then the set
K G := { α ∈ K | g(α) = α, ∀g ∈ G } is a subfield of K.
We call K G the fixed field of the action of the group G. fixed field
We have seen that if φ ∈ Aut(K/F ) is an automorphism of a field extension, and if α ∈ L is a
root of some f ∈ F [x], then φ(α) is also a root of f . We can apply this observation to spltting fields.
Proposition. Let f ∈ F [x], and let Rf = { α ∈ K | f (α) = 0 } be the set of roots of f in some
extension K/F . Then any φ ∈ Aut(K/F ) restricts to a permutation of the set Rf , and this defines
a group homomorphism
ι : Aut(K/F ) → Sym(Rf ).
Furthermore, if K = F (Rf ) (i.e., if K is a splitting field of f ), then ι is injective, so Aut(K/F ) is
isomorphic to a subgroup of Sym(Rf ).
Proof. It is clear that restricting to Rf ⊆ K defines such a homomorphism, since φ(α) is a root of
f whenever α is, and vice-versa by considering φ−1 .
For injectivity, note that if φ ∈ Aut(K/F ) satisfies φ(α) = α for all roots α of f , then Rf ⊆ K G
and F ⊆ K G , where G = hφi ≤ Aut(Σ/F ) is the cyclic subgroup generated by φ. Since K G is a
subextension of K containing the roots of f we must have F (Rf ) ⊆ K G , so when K = F (Rf ) we
have φ = id. 

19. Examples of automorphisms of field extensions

We can use the techniques we have developed to compute the automorphism groups of some field
extensions.
Example (Important). Consider g = x3 − 2 ∈ Q[x]. This factors
g = (x − α1 )(x − α2 )(x − α3 ) = (x − α)(x − ωα)(x − ω 2 α),
√
where α = 3 2 ∈ R and ω = e2πi/3 . Thus Σ = Q(α1 , α2 , α3 ) ⊆ C is the splitting field. In fact, Σ is
generated over Q by any two of the roots, since any quotient αi /αj with i 6= j is either ω or ω −1 .
Earlier we showed that [Σ : Q] = 6, since it contains the subfield Q(ω) and [Q(ω) : Q] = 2.
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 18

There are six distinct embeddings Σ → C. One of them is the “obvious” inclusion, but there are
others. We construct them in two steps, as indicated in the following diagram.

Q(α, ωα)
ωα7→? /Σ

Q(α)
α7→? / Q(φ(α))

Q Q
with two places to make choices. The possible choices are:
φ(α) α α ωα ωα ω 2 α ω 2 α
φ(ωα) ωα ω α α ω 2 α α
2 ωα
φ(ω 2 α) ω 2 α ωα ω 2 α α ωα α
First construct φ1 : Q(α) → C sending α to some root of g, so φ1 (α) ∈ {α, ωα, ω 2 α}.
Consider the case of φ1 (α) = α. Over Q(α) we have
g = (x − α)g1 , g1 = x2 + αx + α2 ,
so that the roots of g1 in C are {ωα, ω 2 α}. In fact, g1 is irreducible over Q(α). We know this is
true because we know that [Σ : Q] = 6 and [Q(α) : Q] = 3, whence [Σ : Q(α)] = 2 by the tower law.
Since Σ is generated over Q(α) by ωα, we conclude that mωα/Q(α) has degree 2. Since g1 (ωα) = 0
we conclude that g1 = mωα/Q(α) . Thus we can construct φ : Q(α, ωα) → Σ extending φ1 so that
φ(ωα) ∈ {ωα, ω 2 α}.
In general, if φ1 (α) = ω k α, then φ(g1 ) = x2 +ω k αx+ω 2k α2 , whose roots are {α, ωα, ω 2 α}r{ω k α},
and is irreducible over Q(ω k α) because [Σ : Q(ω k α)] = 2. Thus we can construct φ : Q(α, ωα) → C
extending φ1 , so that φ(ωα) is one of the roots of g1 .
The image of φ is Q(α, ωα) again, since the images of the roots of g are still roots of g.
The above argument shows that G = Aut(Σ/Q) is a group of order 6, and examining the possible
formulas for φ ∈ G, we see that G ≈ S3 . In fact, if we label the roots of g as α1 = α, α2 = ωα,
α3 = ω 2 α, then for every σ ∈ S3 there is a unique φ ∈ G such that φ(αk ) = ασ(k) .
√ √
Example.√ Consider
√ g = (x2 − 2)(x2 − √ 3) ∈ Q,√with roots ± 2, ± 3. The splitting field is
Σ = Q( 2, 3). Earlier we showed that 3 ∈ / Q( 2) so [Σ : Q] = 4.
We can construct isomorphisms φ : Σ → Σ according to the diagram:
√
√ √ 37→?
Q( 2, 3) /Σ

√
√ 27→? √
Q( 2) / Q(φ( 2))

Q Q
The choices here are described by
√ √ √ √ √
φ(√2) √2 √2 −√ 2 −√2
φ( 3) 3 − 3 3 − 3
√ √
The first√choice gives φ1 : Q( 2) → C sending 2 to a root of g1 = x2 − 2. Note that the image
of φ1 is Q( 2) under either case.
The remaining factor
√ x2 − 3 ∈ Q[x], so we have that φ1 (g2 ) = g2 . This
of g √is g2 = √ √ √remains
√ over Q( 2) since 3 ∈
irreducible / Q( 2). Thus the second choice extends to φ2 : Q( 2, 3) → C
sending 3 to a root of g2 .
Using this, we see that G = Aut(Σ/Q) ≈ C2 × C2 .
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 19

Example. Consider g = x3 + x2 − 2x − 1 ∈ Q[x]. We see g ∈ Irred(Q) by the rational roots test,

and since g is separable it has three distinct roots α1 , α2 , α3 ∈ C. Picking a root at random, we get
three choices of homomorphism φ : Q(α1 ) → C, determined by φ(α1 ) ∈ {α1 , α2 , α3 }.
What may not be obvious is that g already splits over Q(α1 ). In fact, it turns out that the roots
of this polynomial are
α1 = ζ + ζ −1 , α2 = ζ 2 + ζ −2 , α3 = ζ 3 + ζ −3 , ζ = e2πi/7 .
(These are actually all real numbers. All you need to check this is that ζ 7 = 1 and ζ 6= 1.) Using
this you can check that: α2 = α12 − 2, α3 = α22 − 2. Thus Σ = Q(α1 ) is already a splitting field of g,
and a homomorphism φ : Σ → Σ is determined by where it sends α1 . The possible choices are:
φ(α1 ) α1 α2 α3
φ(α2 ) α2 α3 α1
φ(α3 ) α3 α1 α2
Thus [Σ : Q] = 3 and G = Aut(Σ/Q) ≈ C3 .
Question: what if we didn’t already know the roots of g? How could we have analyzed this in
that case?
√
Example. x4 − 2 ∈ Q[x]. Here the roots are {±α, ±iα}, where α = 4 2. In this case, we can show
that [Σ : Q] = 8 and G = Aut(Σ/Q) ≈ D8 . I’ll leave this as an exercise, except to note that this
relies on the chain of extensions
Q ⊆ Q(α) ⊆ Q(α, i), [Q(α) : Q] = 4, [Q(α, i) : Q(α)] = 2.
(Note that Q(α) ⊆ R so i ∈
/ Q(α).)

20. Normal extensions

We noted that a given finite extension can be the splitting field of many different polynomials.
There is an abstract characterization of when such an extension is a splitting field, which doesn’t
require mentioning a particular polynomial.
An algebraic extension L/F is normal if every f ∈ Irred(F ) which has a root in L splits in L. normal
Note: this definition is different than as given in DF, but is equivalent to their definition as I will
show.
Example. Qalg /F for any subfield F ⊆ Qalg is a normal extension, since it is an algebraic extension
and all polynomials over Qalg (and hence over F ) split in Qalg .
Exercise. Show that every degree 2 extension is normal.
√ √
Example. Q( 3 2)/Q is not normal, since f = x3 − 2 does not split over Q( 3 2).
Theorem. A finite extension L/F is normal if and only if it is a splitting field for some polynomial
f ∈ F [x].
Proof part 1: Finite normal extensions are splitting fields: If L/F is a finite extension then
L = F (α1 , . . . , αm )
for some finite list of elements α1 , . . . , αm ∈ L. Let f = m
Q
k=1 mαk /F ∈ F [x], the product of minimal
polynomials of these elements. Normality of L/F says thet each mαk /F splits over L, and thus f
splits over L. Since L/F is generated by the αk s, it is clear L/F is a splitting field of f . 
We get the second part as a special case of a more general claim.
Lemma. Suppose F ⊆ L ⊆ M , where L = Σf /F is a splitting field of some f ∈ F [x]. If α, β ∈ M
are roots of the same irreducible polynomial g ∈ Irred(F ), then [L(α) : L] = [L(β) : L].
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 20

Proof. Consider the following diagram of subfields of M .

M

L(α) L(β)

F (α) F (β)

F
To prove the result it suffices to show that [F (α) : F ] = [F (β) : F ] and [L(α) : F (α)] = [L(β) : F (β)],
using the tower law. In fact, I claim there exist isomorphisms φ and ψ which are compatible with
the inclusions in
ψ
L(α) / L(β)
O O

O φ O
F (α) / F (β)
a =

a =
F
That is, φ|K is the inclusion of the subfield K ⊆ K(β), and ψ|K(α) is φ.
In fact, there exists φ : F (α) → F (β) sending φ(α) = β, because α, β are both roots of g ∈
Irred(F ).
Note that φ(f ) = f since f ∈ F [x], and that both L(α)/F (α) and L(β)/F (β), being generated
over the ground fields by roots of f , are splitting fields of f relative to their subfields. Thus by
previous theory, φ extends to an isomorphism ψ. 
Proof of part 2: splitting fields are normal extensions. Suppose L/F is a splitting field of f ∈ F [x],
and g ∈ Irred(F ) is some irreducible polynomial with root α ∈ L.
Form a splitting field Σ/L of the polynomial g ∈ F [x] ⊆ L[x]. If β is any root of g in Σ, the
previous lemma says
[L(α) : L] = [L(β) : L].
But α ∈ L so these are 1, so β ∈ L. Thus all roots of g are in L, so g splits over L as desired. 
As a consequence, a splitting field Σ/F contains a splitting field for any f ∈ Irred(F ) which has
a root in Σ.
We can generalize this to infinite extensions.
Theorem. An algebraic extension L/F is normal iff it is a splitting field for a set S ⊆ F [x] r {0}
of polynomials, i.e., if all f ∈ S split over L and L is generated over F by the roots of all f ∈ S.
Proof. =⇒: Let S = { mα/F | α ∈ L }, the set of all minimal polynomials of all elements (which
exist because L/F ) is algebraic. Then L = F (S) and every mα/F splits over L since the extension
is normal.
⇐=: Suppose L/F S is a splitting field of a set of polynomials S ⊆ F [x] r {0}, so L is generated
over F by the set f ∈S Rf , where Rf ⊆ L is the set of roots f in L. Given α ∈ L and g ∈ Irred(F )
such that g(α) = 0, we see that α must be contained in a subfield generated by a finite set of such
roots, so α ∈ F (Rf ) ⊆ L where f = f1 · · · fk for some finite list f1 , . . . , fk ∈ S. Since F (Rf )/F is a
splitting field of f , it is normal so g splits over F (Rf ) and hence over L.

 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 21

We also have the following, which will be important for us.

Proposition. Consider fields F ⊆ K ⊆ L. If L/F is normal then L/K is normal.
Proof. First note that since L/F is normal then it is algebraic by definition, so L/K and K/F are
also algebraic.
Consider f ∈ Irred(K) with a root α ∈ L. Since K/F is algebraic there is a minimal polynomial
g = mα/F ∈ F [x] over F . Since g has α as a root, over K we must have f | g. Since L/F is normal
we have that g splits over Σ, and therefore its factor f splits over Σ. 
Remark (Warning). It is not true √
that L/F normal implies K/F normal. For instance, F = Q,
K = Q(α), L = Q(α, ω), with α = 3 2 and ω = e2πi/3 .
Remark (Warning).
√ It is not true√ that L/K and K/F normal imply L/F normal. For instance,
F = Q, K = Q( 2), and L = Q( 4 2). Both L/K and √ K/F are degree 2 and so normal, but L/F is
4 4
not normal since the minimal polynomial x − 2 of 2 does not split over L ⊆ R.

21. Existence of algebraic closure

Every field F is contained in an algebraically closed field, and thus admits an algebraic closure. M 14 Nov
The general proof is somewhat non-constructive, ultimately relying on the axiom of choice.
Lemma. Given any field F , there exists a field extension F ⊆ K such that every non-constant
f ∈ F [x] has a root in K.
Proof. To do this, we need the notion of a polynomial ring on an infinite set of variables:
R = F [xs | s ∈ S].
Any element in here is just a polynomial involving a finite subset of the variables xs . (That is, R is
set theoretically a union of polynomial rings in finitely many variables.)
Let S = Irred(F ), the set of monic irreducible polynomials. The goal is to produce K/F so that
K has a root for every f ∈ S. Form R := F [xf | f ∈ S] as above. Let I = (f (xf ) | f ∈ S), the
ideal generated by elements f (xf ) ∈ R, where we plug in the variable xf into the polynomial it
corresponds to.
I claim I 6= R. Given that I is a proper ideal, we then know that there exists a maximal ideal
M so that I ⊆ M ( R. Let K = R/M , which is a field, and comes with injective homomorphism
F → K, which we use to identify F as a subfield of K. This field K is thus constructed to have the
property that every irreducible polynomial over F has at least one root in K, as desired.
To show that I 6= R, suppose not and derive a contradiction. Then there exist f1 , . . . , fn ∈ S and
g1 , . . . , gn ∈ R such that
1 = g1 f1 (xf1 ) + · · · + gn fn (xfn ).
Note that there are only finitely variables which appear here (xf1 , . . . , xfn , together with the variables
which appear in the g1 , . . . , gn ). So this is really an identity in a polynomial ring on finitely many
variables, which I’ll write as xf1 , . . . , xfn , xfn+1 , . . . , xfm .
Construct an extension E/F which contains a root αk of fk for all k = 1, . . . , n. Plugging in
xfk = αk when k = 1, . . . , n, and xfk = 0 for the remaining variables into the above identity, gives
1 = 0 in E, which is impossible. This is our contradiction. 
Theorem. Every field F admits an algebraic closure.
Proof. Given F , form
F = K0 ⊆ K1 ⊆ K2 ⊆ · · · ⊆ Kj ⊆ · · · , j ∈ Z>0 ,
where each Kj is chosen
S as in the lemma, so that every non-constant polynomial in Kj−1 [x] has a
root in Kj . Let K = Kj . This set has the structure of a field, compatible with the structures
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 22

on the Kj s. Any polynomial h(x) ∈ K[x] must actually lie in some Kj [x] and so has a root in
Kj+1 ⊆ K. Thus K is algebraically closed, and so contains an algebraic closure F of F . 

22. Uniqueness of algebraic closure

Algebraic closures are also unique, up to isomorphism.
Proposition. If K/F and L/F are algebraic closures of F , then K and L are isomorphic as
F -extensions.
∼
Proof. We will construct an F -embedding φ : K → L, which gives an isomorphism K − → φ(K) ⊆ L.
The image φ(K), being isomorphic to K, is also an algebraic closure of F , and so is algebraically
closed. But since L/F is algebraic then certainly L/φ(K) is algebraic, and therefore L = φ(K).
The map φ is constructed as for maps out of a splitting field, except we need to use Zorn’s lemma.
Let  
F ⊆ E ⊆ K and,
P = (E, ψ) .
φ : E → L such that φ|F = idF .
This can given a partial ordering, so that (E, ψ) ≤ (E 0 , ψ 0 ) iff E ⊆ E 0 and ψ 0 |E = ψ. It is
straightforward that Zorn’s lemma applies to give a maximal element (E0 , ψ0 ). If E0 = K we can
set φ = ψ0 and we are done, so suppose not and derive a contradiction.
Choose α ∈ K r E.b Since α is algebraic over F , the extension E(α)/
b Eb is finite. Let f = m b
α/E
0
be the minimal polynomial. Then f = ψ0 (f ) ∈ L[x] has a root β, so we can extend ψ0 to an
embedding ψ1 : E(α)
b → L. This gives (E(α),
b ψ1 ) ∈ P contradicting maximality of (E, b ψ), so we
have our contradiction. 
Note: the isomorphism this produces is not unique.

23. The Frobenius endomorphism

Let F be a field of characteristic p 6= 0. Then for any a, b ∈ F we have that
(a + b)p = ap + bp ,
p

p!
because p | k = (p−k)! k! when 0 < k < p.
Proposition. If F is a field of characteristic p 6= 0, then the function φ : F → F defined by
φ(a) := ap is a homomorphism of fields.
This map φ : F → F is called the Frobenius endomorphism of F . Frobenius endomor-
phism
Example. In F = Fp , we have that ap = a for every a ∈ Fp .
Example. Let K = F2 [x]/(x2 + x + 1). This is a field because f = x2 + x + 1 clearly has no root in
F2 = {0, 1}. Write γ = x, so K = {0, 1, γ, γ + 1} (a field of order 4) and γ 2 = γ + 1.
Then the Frobenius φ sends φ(γ) = γ 2 = γ + 1. In fact, Aut(K) = {e, φ}, since {γ, γ + 1} are
the two roots of f .
Remark. If f ∈ Fp [x], and K/Fp is some extension field, then the Frobenius automorphism preserves
roots of f . That is, if α ∈ K is a root of f , so is αp .
In particular, α and αp have the same minimal polynomial over Fp .
Example. Let K = F2 /(x3 + x + 1) with γ = x, a field because f = x3 + x + 1 has no root in
F2 = {0, 1}. Write γ = x, and note that K has exactly 8 elements. Then Aut(K) = {e, φ, φ2 }, since
{γ, γ 2 , γ 4 = γ 2 + γ} are the roots of f .
The other three elements of K r F2 are {γ 3 = γ + 1, γ 6 = γ 2 + 1, γ 12 = γ 5 = γ 2 + γ + 1}, which
are roots of g = x3 + x2 + 1, also irreducible over F2 .
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 23

24. Perfect fields

A field F is called perfect if either: perfect

(1) it is characteristic 0, or
(2) it is characteristic p 6= 0, and every element of F is a pth power (i.e., the Frobenius
endomorphism a 7→ ap is an automorphism).
Thus, finite fields are perfect, since any injective map K → K from a finite set to itself must be a
bijection.
Example (A non-perfect field). Let K = F (t), the field of rational functions over any field F of
characteristic p, e.g., F = Fp . Then the Frobenius φ : K → K is not surjective: the element t is not
in the image of φ.
To see this, note that if t = (g/h)p for some g, h ∈ F [t], then g p = t hp , and thus p deg g =
1 + p deg h, which has no solution with g, h non-zero.
Lemma. Suppose char F = p 6= 0. For any a ∈ F , consider f = xp − a ∈ F [x]. The polynomial f
is not separable. Furthermore, either
(1) f ∈ Irred(F ), or
(2) f = (x − b)p for b ∈ F with a = bp .
Proof. To see that f not separable, simply note that Df = pxp−1 = 0, which is certainly not
relatively prime to f .
We have that F splits over an extension field F (b) as f = (x − b)p , where bp = a, since if b is any
root of f we can compute (x − b)p = xp − bp = f . For any 0 < k < p we have
(x − b)k = xk − kb xk−1 + · · · + (−b)k .
In particular (x − b)k ∈ F [x] implies kb ∈ F , and thus b ∈ F since the integer k represents a non-zero
element of the prime field Fp . Thus f is reducible over F iff b ∈ F . 
In other words, in characteristic p, for any a ∈ F we have either a1/p ∈ F or [F (a1/p ) : F ] = p.
Proposition. A field F is perfect iff every f ∈ Irred(F ) is separable.
Proof. When char F = 0 there is nothing to prove, so suppose char F = p 6= 0.
Suppose F is perfect. Then any f ∈ F [x] such that
P Df = 0 itself a pth power of a polynomial in
F [x]. To see this, note that Df = 0 implies f = nk=0 ak xkp , so since F is perfect we can choose
bk ∈ F so that bpk = ak , so
n n X n p n
p kp
X X X
kp k
f= ak x = bk x = bk x = gp, g= bk xk ∈ F [x].
k=0 k=0 k=0 k=0
Thus if f ∈ Irred(F ), we must have Df 6= 0, so (f, Df ) = F [x], so f is separable.
Conversely, if every irreducible over F is separable, then f = xp − a ∈ F [x] is never irreducible
for any a ∈ F (since Df = 0), so a has a pth root in F . 
The Frobenius φ : F → F is always injective. Thus, when char F = p = 6 0, elements a ∈ F can
have at most one pth root in F , or even in any extension of F . That is, ap = bp implies a = b in
such fields.
In particular, there are no primitive nth roots of unity in F whenever p | n.

25. Finite fields

Clearly a field of characteristic 0 is infinite, since it contains Q. Thus any finite field has
char = p 6= 0.
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 24

If K is finite of characteristic p, then [K : Fp ] = n < ∞, whence |K| = pn , since it as a vector

space over Fp it is isomorphic to Fnp . This means that |K| determines char K.
n
Let f = xp − x ∈ Fp [x]. Since Df = −1 6= 0 is a unit in Fp [x], this is separable.
Form a splitting field K = Σf /Fp . Let
n
R := { α ∈ K | f (α) = 0 } = { α ∈ K | αp = α } ⊆ K,
n
the set of roots of f in K. Since f is separable, |R| = pn . Since f = x(xp −1 − 1) = 0, the elements
of R are either 0 or (pn − 1)st roots of unity.
n n n
In fact, R is a subfield of K: it is closed under all field operations, since (α + β)p = αp β p ,
n n n
(αβ)p = αp β p , etc. Obviously f already splits over R, so R = K.
n
We have thus obtained a field of order pn , as a splitting field of f = xp − x over Fp , whose
elements are exactly the roots of f . This field is usually denoted Fpn (by pure mathematicians,
especially number theorists), or sometimes as GF (pn ) (for “Galois field”, especially by people using
finite fields in applications such as cryptography).
Proposition. If K is a field with |K| = pn , then K ≈ Fpn .
Proof. Clearly char K = p. We have shown that K × = K r {0}, being finite, is cyclic of order
n
pn − 1. Thus every element of K is a root of f = xp − x ∈ Fp [x]. Since these are distinct, f splits
over K, which is therefore a splitting for f . The claim follows by uniqueness of splitting fields up to
isomorphism. 
Proposition. The automorphism group Aut(Fpn ) = hφi ≈ Cn , a cyclic group of order n generated
by Frobenius.
Proof. Clearly φ ∈ G = Aut(Fpn ), and φn = id. There is no smaller k > 0 such that φk = id, since
k
in that case every a ∈ Fpn would satisfy ap = a, and we know there are only pk such elements.
Thus |φ| = n, and |G| ≥ n.
Since F× × n
pn is cyclic, there exists ζ ∈ Fpn with |ζ| = p − 1. Clearly Fpn = Fp (ζ). As this is a simple
extension, any automorphism α ∈ G is determined once we know α(ζ). But since [Fpn : Fp ] = n, we
have deg mζ/Fp = n, so there are at most n possibilities for α(ζ). Thus |G| ≤ n.


26. Separable extensions

Given an extension K/F , say that α ∈ K is separable over F it is algebraic over F , and if W 16 Nov
its minimal polynomial m = mα/F is a separable polynomial (i.e., Dm 6= 0 when characteristic is separable
non-zero.).
Therefore, α is separable over F if it is the root of any separable polynomial f ∈ F [x], since
mα/F | f , and any factor of a separable polynomial must also be separable.
We say that K/F is a separable extension if all elements of K are separable over F . separable extension

Example. Every algebraic extension K/F of fields of characteristic 0 is separable.

Example. Let F be a perfect field. We have shown this means exacly that every f ∈ Irred(F ) is
separable. Then any algebraic extension K/F of a perfect field is separable.
Example. The standard example of an inseparable extension is L/K, where F is any field of
characteristic p, L = F (t) is a function field over F , and K = F (tp ). The element t generates L over
K, but is not contained in K, since t = g(tp )/h(tp ) for some g, h ∈ Fp [x] is impossible for degree
reasons.
Furthermore, t is algebraic over K but not separable, since its minimal polynomial mt/K = xp − tp
is not separable.
Proposition. If F ⊆ K ⊆ L such that L/F is separable, then L/K and K/F are separable.
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 25

Proof. By hypothesis, every α ∈ L is a root of a separable polynomial f ∈ F [x]. It is an immediate

consequence that every α ∈ K is separable over F , so K/F is separable. Furthermore, since f
remains separable over the larger field K, we have that α is separable over K, so L/K separable.

We have the following criterion for separability of an element.
6 0. Then α ∈ K is separable over F iff
Lemma. Let K/F be an extension in characteristic p =
p
α ∈ F (α ).
Proof. Note that if α is transcendental over F (and so not separable over F ), then F (α) is isomorphic
/ F (xp ), since x = g(xp )/h(xp )
to a function field F (x) on one variable x. It is clear in this case that x ∈
for polynomials g, h ∈ F [x] is impossible for degree reasons. So we can reduce to the case of α
algebraic over F .
So assume that α is algebraic over F , so we have algebraic extensions
F ⊆ F (αp ) ⊆ F (α).
Let g = mα/F ∈ Irred(F ), so k = deg g = [F (α) : F ]. I will show that α ∈ / F (αp ) iff g is not
separable.
• Suppose g is not separable. Since it is irreducible over F , this means Dg = 0, so g = h(xp )
for some h ∈ F [x] with deg h = k/p. But we know h(αp ) = 0, so [F (αp ) : F ] ≤ deg h =
1
/ F (αp ).
p [F (α) : F ] < [F (α) : F ], and thus α ∈
• Suppose g is separable. Consider f = xp − αp ∈ F (αp ), and recall that f either splits over
F (αp ) or is irreducible over F (αp ). If it is irreducible over F (αp ) then f = mα/F (αp ) , and
therefore f | g, which would imply g has repeated roots, so this does not happen.
So f must split over F (αp ), which means α ∈ F (αp ).

Example (An example of an inseparable extension). Let F be any field with char F = p > 0. Let
L = F (t) be the function field, and let K = F (tp ) ⊆ L. The element t ∈ L is not separable over K
by the lemma, since t ∈/ K = K(tp ).

27. Separably generated extensions

Given a field F in characteristic p > 0, let
F p := { ap ∈ F | a ∈ F },
the subset of elements which are pth powers. This is exactly the image φ(F ) of the Frobenius
endomorphism φ : F → F , so F p is a subfield of F , which is isomorphic to F (but possibly not equal
to F ).
This operation is compatible with the formation of subfields generated by subsets: if K = F (S)
for some subset S ⊆ K, then K p = F p (S p ), where S p = { sp | s ∈ S }. This implies that for any
two subfields F, F 0 ⊆ K, we have (F F 0 )p = F p F 0p .
Finally, note that [K : F ] = [K p : F p ]. This is because Frobenius gives an isomorphism
∼
φ: K − → K p of fields which takes F to F p .
Proposition. Let K/F be a finite extension in characteristic p 6= 0. Then K/F is separable iff
K = F K p , i.e., iff K is generated over F by pth powers in K.
Proof. =⇒. Suppose K/F is separable. Then for any α ∈ K we have α ∈ F (αp ) ⊆ F K p . Therefore
K ⊆ F K p whence K = F K p .
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 26

⇐=. Suppose K = F K p . Let α ∈ K, and consider the diagram of extensions

K
e F Kp
φ
F (α) e0 Kp o ∼ K
p
d F (α ) e e
φ
F d0 F p (αp ) o ∼ F (α)
F d d
φ
Fp o ∼ F
The Frobenius maps gives isomorphisms which tell us that [K p : F p (αp )] = [K : F (α)] and
[F p (αp ) : F p ] = [F (α) : F ]. Note that F K p = F (αp )K p , so e0 ≤ e. But according to the tower law,
e | e0 , whence e = e0 and so F (α) = F (αp ), whence α is separable over F . 
Remark. This criterion doesn’t work if the extension is infinite. For instance, starting with the
function field K = F (t), we can form an infinite chain of non-separable extensions
2
K = F (t) ( K1 = F (t1/p ) ( K2 = F (t1/p ) ( . . . .
is such that L = KLp , since KKnp = Kn−1 , but L/K is certainly not a separable
S
Then L = n Kn
extension.
Corollary. We have the following.
(1) If F ⊆ K ⊆ L are finite extensions such that K/F and L/K are separable, then L/F is
separable.
(2) If F ⊆ L is a field extension, and K, K 0 ⊆ L such that K/F and K 0 /F are finite separable
extensions, then KK 0 /F is separable.
(3) If L = F (α1 , . . . , αn ) such that each αk is separable over F , then L/F is separable.
(4) Given an extension K/F , let K sep ⊆ K be the subset of elements which are separable over
F . Then K sep is a subfield of K.
(5) Every splitting field Σ/F of a separable polynomial f ∈ F [x] is a separable extension.
Proof. These are all trivial if we are in characteristic 0, so assume characteristic is p > 0.
For (1), we have L = KLp = F K p Lp = F Lp , since K p ⊆ Lp .
For (2), we have (KK 0 )p = K p K 0p , whence F (KK 0 )p = F K p K 0p = (F K p )(F K 0p ) = KK 0 .
Statement (3) follows by inductively applying (2) with K = F (α1 , . . . , αj−1 ) and K 0 = F (αj ),
since F K 0p = F (F p (αjp )) = F (αjp ), which equals F (αj ) when αj is separable over F .
Statement (4) and (5) follow immediately from (3).


28. Roots of unity over Q

We want to understand the extension Q(ζn )/Q. To do this we need to understand the minimal
polynomials of nth roots of unity.
An nth root of unity  in C is a root of xn − 1. This is a monic polynomial in Z[x], and so in
particular is a primitive polynomial, i.e., its coefficients are a relatively prime set. By Gauss’s
Lemma, there is an irreducible factorization of xn − 1 over Q into primitive polynomials over Z,
which must be monic (up to a sign). This implies that the minimal polynomial f := m/Q , which is
a factor of xn − 1, is an element of Z[x].
Theorem. Let ζ ∈ C be a primitive nth root of unity. Then every primitive nth root of unity is a
root of f = mζ/Q .
The proof will use the following lemma.
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 27

Lemma. Let  ∈ C be a primitive nth root of unity, and let p be a prime not dividing n. Then for
any f ∈ Irred(Q) which has  as a root, we have f (p ) = 0.
Proof of Theorem using Lemma. Every primitive nth root of unity can be written as ζ k with
gcd(k, n) = 1 and k > 0. Factor k = p1 · · · pr into primes, where no pi divides n. Then using the
lemma and induction we show successively that ζ, ζ p1 , ζ p1 p2 , ζ p1 p2 p3 , . . . are all roots of f , whence
f (ζ k ) = 0. 
Now we prove the lemma. The idea of the proof is this: since f ∈ Z[x], we can reduce mod p
it to f ∈ Fp [x]. This f will have the same degree as f (since f is monic). In a splitting field over
Fp , its roots will still be some nth roots of unity, since f | xn − 1. It will also be separable over Fp
(since xn − 1 is, since p - n).
However f may not be irreducible any more. But if it has irreducible factorization f = g1 · · · gk
over Fp , then we know for any root  of gk , we have that p is also a root of gk . Thus in characteristic
p, the set of roots of f must be “closed under pth powers”. We can hope that this property “lifts”
to f itself.
Proof of Lemma. We can assume f = m/Q . Let g = mp /Q . As noted above these are both factors
of xn − 1, and so are both in Z[x] since xn − 1 is primitive and f, g are monic. We want to show
f = g, so we assume f 6= g and derive a contradiction. If that is the case we must have xn − 1 = f gh
for some h ∈ Z[x].
Let G := g(xp ) ∈ Z[x]. Then G() = g(p ) = 0, so f | G, so G = f k for some monic k ∈ Z[x],
since G is also monic polynomial with integer coefficients.
Now we can reduce everything modulo p, by taking images under the homomorphism π : Z[x] →
Fp [x] which sends integer coefficients to integers modulo p. Write f , g, h, k, G for the images of
f, g, h, k, G under π. Note that since f, g are monic of positive degree, f , g are also monic of positive
degree (but they might not be irreducible over Fp ). Also since elements of Fp satisfy ap = a, we
have g(xp ) = g(x)p . Thus
g p = G = f k, xn − 1 = f gh.
The first identity implies that f , g must have some irreducible factor m ∈ Irred(Fp ) in common,
since neither is a unit, and any irreducible factor of f must also divide g. Since m | f and m | g,
the second identity implies that m2 | xn − 1 in Fp [x]. But xn − 1 ∈ Fp [x] is separable, since
D(xn − 1) = nxn−1 and p - n, and these are relatively prime since x - xn − 1. So xn − 1 cannot have
a repeated irreducible factor, so we have a contradiction. 

29. Cyclotomic polynomials

We write Φn := f ∈ Z[x] for the minimal polynomial of one primitive nth root of unity, and F 18 Nov
hence of all primitive nth roots of unity. Every nth root of unity is a primitive dth root of unity for
some d dividing n, so since xn − 1 is separable we must have
Y
xn − 1 = Φd .
d|n

This means that we can compute the Φn by induction on n, using polynomial long division.
The degree of Φn is the Euler φ-function
φ(n) := |(Z/n)× | = number of d ∈ {1, 2, . . . , n} such that gcd(d, n) = 1.
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 28

Φ1 = x − 1, Φ9 = x6 + x3 + 1,
Φ2 = x + 1, Φ10 = x4 − x3 + x2 − x + 1,
Φ3 = x2 + x + 1, Φ11 = x10 + x9 + · · · + x2 + x + 1,
Φ4 = x2 + 1, Φ12 = x4 − x2 + 1,
Φ5 = x4 + x3 + x2 + x + 1, Φ13 = x12 + x11 + · · · + x2 + x + 1,
Φ6 = x2 − x + 1, Φ14 = x6 − x5 + x4 − x3 + x2 − x + 1,
Φ7 = x6 + x5 + x4 + x3 + x2 + x + 1, Φ15 = x8 − x7 + x5 − x4 + x3 − x + 1,
Φ8 = x4 + 1, Φ16 = x8 + 1.

Exercise. Show that:

(1) if m is odd and m > 1, then Φ2m = Φm (−x), and
(2) if p | m where p is prime, then Φpm = Φ(xp ).
Use this to show that you can reduce the problem of computing cyclotomic polynomials to the case
when n is a product of distinct odd primes.
Remark. It’s not true that every coefficient in Φn must be in {−1, 0, 1}. The first counterexample is:
Φ105 = x48 + x47 + x46 − x43 − x42 − 2x41 − x40 − x39 + x36 + x35 + x34
+ x33 + x32 + x31 − x28 − x26 − x24 − x22 − x20 + x17 + x16 + x15
+ x14 + x13 + x12 − x9 − x8 − 2x7 − x6 − x5 + x2 + x + 1.
In particular, we have that [Q(ζn ) : Q] = deg Φn = φ(n), the value of the Euler φ-function, Euler φ-function
defined by
φ(n) = |(Z/n)× | = |{ k ∈ {1, . . . , n} | gcd(k, n) = 1 }|.

30. Galois extensions

Given a field extension L/F , we say it is a Galois extension iff it is both normal and separable. Galois extension
In other words, L/F is Galois if for every α ∈ L, there exists a separable polynomial f ∈ F [x]
which (i) has α as a root, and (ii) splits over L. (Note: if any f has these properties then so does
the minimal polynomial mα/F .)
Note: this is different from what DF say in two ways. (1) In §14.1 they only define finite Galois
extension (though they do the infinite case in §14.9). (2) Their definition is different than the above
in the finite case, but I will show they are equivalent.
Proposition. A finite extension L/F is Galois iff it is a splitting field of some separable polynomial
f ∈ F [x].
Proof. We have shown that a splitting field of any f is normal, and that it is also separable if f is a
separable polynomial.
Conversely, suppose L/F is finite and Galois. We can write L = F (α1 , . . . , αn ) for some αk ∈ F .
Let mk = mαk /F ∈ F [x]. Each mk is a separable polynomial. Note that there could be repetition in
the list m1 , . . . , mn . Let f = the product of distinct elements from the list of minimal polynomials.
Since f is a product of pairwise distinct up-to-units separable irreducible polynomials, it is
separable. Clearly all αk s are roots of f , so L/F is a splitting field of f . 
Given extensions K/F and L/F over F , write EmbF (K, L) for the set of homomorphisms
φ : K → L with φ|F = idF (“embeddings” of extensions). We will prove the following theorem on
embeddings.
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 29

Theorem (Theorem on Embeddings). Let K/F and L/F be extensions with K/F finite. Then
|EmbF (K, L)| ≤ [K : F ],
with equality iff
(1) K/F is a separable extension, and
(2) every f ∈ Irred(F ) which has a root in K splits over L.
As a consequence, we will get the following (which is DFs definition of finite Galois extension).
Corollary. If L/F is a finite extension, then it is Galois iff |Aut(L/F )| = [L : F ].
Proof. We apply the theorem in the case of K = L. Let G = Aut(L/F ). Note that since K = L, we
have EmbF (L, L) = G. Then the Corollary is an immediate consequence of the theorem: equality
|G| = [L : F ] holds iff L/F is separable and normal. 
Remark. If F is a perfect field, then every finite extension L/F is separable. This applies when F is
characteristic 0 or F is finite. Thus all finite normal extensions over perfect F are Galois.

31. Proof of the theorem on embeddings

I’m going to prove the theorem as a consequence of a slightly more general statement, which will
allow for an inductive argument. Suppose given extensions K/F and L/F 0 , and an isomorphism
λ : F → F 0 . We define
Embλ (K, L) = {embeddings φ : K → L such that φ|F = λ}.
φ
KO /L
O

O O
F
∼ / F0
λ
If F = F0 and λ = idF , then this is just EmbF (K, L).
∼
Proposition. Let K/F and L/F 0 be extensions, with K/F finite. Suppose that λ : F − → F 0 is an
isomorphism of fields. Then
|Embλ (K, L)| ≤ [K : F ],
with equality iff
(1) K/F is separable, and
(2) if f ∈ Irred(F ) has a root in K, then f 0 := λ(f ) ∈ F 0 [x] splits over L.
The theorem is exactly the special case of F = F 0 and λ = idF .
The induction will be by handling one simple extension at a time, so lets do that.
∼
Lemma. Let K/F and L/F 0 be extensions and λ : F − → F 0 an isomorphism. Then for any α ∈ K
we have
|Embλ (F (α), L)| ≤ [F (α) : F ],
with equality iff
(i) α is separable over F , and
(ii) m0 := λ(mα/F ) ∈ F 0 [x] splits over L.
Proof. We have a bijection of sets
Embλ (F (α), L) ←→ { α ∈ L | m0 (α0 ) = 0 }.
Since m0 has at most d := deg m0 = deg m = [F (α) : F ] roots, this gives the inequality.
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 30

Note that α is separable iff m is a separable polynomial, iff m0 is a separable polynomial (since
∼
→ F 0 [x] is an isomorphism, it preserves separability). So (i) just says m0 has no repeated
λ : F [x] −
roots in any extension field.
Thus, (i) and (ii) together are equivalent to the statement that m0 has exactly d roots, or
equivalently that there are exactly d embeddings F (α) → L extending λ. 
Proof of Proposition. We work by induction on n = [K : F ].
When n = 1 so that K = F , then Embλ (K, L) has only one element, and (i) and (ii) hold trivially.
So suppose n ≥ 2.
Pick α ∈ K r F , so that we get a chain of extensions F ( F (α) ⊆ K. Write d := [F (α) : F ] and
e := [K : F (α)], and note that e < ed = n.
To give φ : K → L extending λ amounts two choices:
(a) A choice of µ : F (α) → L extending λ. By the Lemma there are at most d such choices.
(b) Given µ, a choice of φ : K → L extending µ. Since e < n, by induction there are at most e
choices.
Thus X
|Embλ (K, L)| ≤ |Embµ (K, L)| ≤ de = n.
µ∈Embλ (F (α),L)
Now suppose (1) and (2) both hold: K/F is separable, and for every f ∈ Irred(F ) with a root in
K the image f 0 = λ(f ) splits over L.
• Statement (1) implies that α is separable over F , and that m0 = λ(mα/F ) splits over L.
Thus by the Lemma we have that d = |Embλ (F (α), L)|.
• Statement (1) implies that K/F (α) is separable. Furthermore, if f ∈ Irred(F (α)) has a
root β ∈ K, then f 0 = µ(f ) ∈ F 0 [x] splits over L: this is because f | m = mβ/F , and the
hypothesis (2) implies m0 = λ(m) splits over L, and hence so does its factor f 0 .
Therefore since e < n, by induction we have that e = |Embµ (K, L)|.
Therefore, (1) and (2) imply equality.
Conversely, suppose we have equality |Embλ (K, L)| = n. Consider any α ∈ K, and as before let
d = [F (α) : F ] and e = [K : F (α)]. By what we have already proved, we have
0 ≤ |Embλ (F (α), L)| ≤ d, 0 ≤ |Embµ (K, L)| ≤ e for any µ ∈ Embλ (F (α), L).
Thus X X
n= |Embµ (K, L)| ≤ e ≤ de = n.
µ∈Embλ (F (α),L) µ∈Embλ (F (α),L)
This is only possible if d = |Embλ (F (α), L)| and e = |Embµ (K, L)|.
In particular since d = |Embλ (F (α), L)|, the Lemma implies that (i) α separable over F , and (ii)
m0 = λ(mα/F ) splits over L. Since this applies to any α ∈ L, we see that (1) K/F is separable, and
(2) if f ∈ Irred(F ) has a root in K then f 0 = λ(f ) splits over L. 

32. The basic Galois correspondence

Recall that a a finite Galois extension L/F is one which is normal and separable. These are M 28 Nov
exactly the splitting fields of separable polynomials over F .
If L/F is a normal finite extension, then |Aut(L/F )| ≤ [L : F ], and equality holds iff L/F is also
separable.
For a finite Galois extension we write
Gal(L/F ) := Aut(L/F ),
and call it the Galois group. Galois group
We are going to be thinking about intermediate fields of this extension, i.e., F ⊆ K ⊆ L. Note intermediate fields
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 31

that if L/F is finite Galois, then so is L/K. The extension K/F is finite and separable, but might
fail to be normal.
Let G = Gal(L/F ).
• If K is an intermediate field of L/F , then L/K is finite Galois, and Gal(L/K) = Aut(L/K)
is a subgroup of G.
• If H ≤ G is a subgroup of G, then LH is an intermediate field of L/F .
We will need the Emebbing Theorem, which tells us that |Gal(L/K)| = [L : K] whenever L/K
is a finite Galois extension. We are also are going to need the following important but technical
lemma, which we will prove later.
Lemma (Tech Lemma). Let G ≤ Aut(L) be a finite subgroup of automorphisms of a field L. Then
|G| = [L : LG ].
Theorem (Basic Galois correspondence). Let L/F be a finite Galois extension with G = Gal(L/F ).
The operations
H 7−→ LH
and
Gal(L/K) ←−[ K
are inverse one-to-one correspondences
{subgroups of G} ←→ {intermediate fields of L/F }.
Notice that both operations are order reversing, where the ordering is inclusion:
0
H ⊆ H0 =⇒ LH ⊇ LH ,
K ⊆ K0 =⇒ Gal(L/K) ⊇ Gal(L/K 0 ).
That is, if a ∈ L is such that h(a) = a for all h ∈ H, then certainly that is true for all h ∈ H 0 ⊆ H;
and if g ∈ G = Gal(L/F ) is such that g|K 0 = id, then certainly g|K = id since K ⊆ K 0 .
Proof of the basic Galois correspondence, using the Tech Lemma. I show that the two operations
are inverse to each other: doing one and then the other (in either order) gets you back where you
started.
Let H ≤ G be a subgroup, and consider
H =⇒ LH =⇒ Gal(L/LH ).
Observe that H ≤ Gal(L/LH ) by definition of LH = { a ∈ L | h(a) = a ∀h ∈ H }. We have
|Gal(L/LH )| = [L : LH ] = |H|,
Emb Thm Tech Lemma

using that L/LH is finite Galois. Therefore the two groups must be equal: H = Aut(L : LH ).
Let K be an intermediate field of L : F , and consider
K =⇒ Gal(L/K) =⇒ LGal(L/K) .
Observe that K ⊆ LGal(L/K) by definition of Aut(L/K) = { g ∈ G | g|K = id }. We have
[L : LGal(L/K) ] = |Gal(L/K)| = [L : K],
Tech Lemma Emb Thm
again using that L/K is finite Galois. Therefore the two intermediate fields must be equal using the
tower law [L : K] = [L : LGal(L/K) ][LGal(L/K) : K], so: K = LGal(L/K) . 
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 32

33. Lattice Galois correspondence

As an immediate consequence of the proof, we get a relation between degrees of extensions and
order of subgroups.
Theorem (Degrees Galois correspondence). For any intermediate field K of L/F we have
[L : K] = |Gal(L/K)|, [K : F ] = |G : Gal(L/K)|.
For any subgroup H ≤ G we have
[L : LH ] = |H|, [LH : F ] = |G : H|.
Both the set of subgroups of G and the set of intermediate extensions of L/F are posets. The
Galois correspondence gives a “duality” between “joins” and “meets” in these posets.
Theorem (Lattice Galois correspondence). For intermediate fields K1 , K2 of L/F , corresponding
to subgroups H1 = Gal(L/K1 ), H2 = Gal(L/K2 ) of G, the Galois correspondence sends
K1 K2 7→ Gal(L/K1 K2 ) = H1 ∩ H2 ≤ G, K1 ∩ K2 7→ Gal(L/K1 ∩ K2 ) = hH1 ∪ H2 i ≤ G.
Proof. The first claim is immediate. For the second claim, it is easier to prove that LhH1 ∪H2 i =
K1 ∩ K2 , and use the Galois correspondence.


34. Normality in the Galois correspondence

There is an additional part of the correspondence as usually stated. Given an intermediate field
K of a Galois extension L/F , we would like to know what Aut(K/F ) is, and understand when K/F
is a normal extension.
Recall that for H ≤ G, its normalizer is
NG (H) = { g ∈ G | gHg −1 = H } ≤ G.
Theorem (Normality in the Galois correspondence). Let L/F be a finite Galois extension with
0
G = Gal(L/F ). Let H, H 0 ≤ G with K = LH , K 0 = LH .
(1) For any g ∈ G, we have that
K 0 = g(K) ⇐⇒ H 0 = gHg −1 .
(2) There is an isomorphism of groups
Aut(K/F ) ≈ NG (H)/H.
(3) The extension K/F is normal iff H is a normal subgroup of G. If that is so, then
Gal(K/F ) ≈ G/H.
Proof. Proof of (1). First we show that Gal(L/g(K)) = gHg −1 . In fact, for any u ∈ G, we have
that u ∈ Gal(L/g(K)) iff ug(c) = g(c) for any c ∈ K, iff (g −1 ug)(c) = c for all c ∈ K, iff g −1 ug ∈ H.
Thus Gal(L/g(K)) = gHg −1 . The claim follows by the basic Galois correspondence, which says
K 0 = g(K) iff H 0 = gHg −1 .
Proof of (2). By (1) we have that
NG (H) = { g ∈ G | gHg −1 = H } = { g ∈ G | g(K) = K }.
Thus there is a homomorphism of groups
π : NG (H) → Aut(K/F ), π(g) := g|K.
I claim that π is surjective. That is, I claim that any automorphism φ : K → K such that φ|F = idF
extends to an automorphism φe : L → L, which necessarily is an element of NG (H) since φ(K)
e = K.
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 33

Since L/F is normal it is the splitting field of some polynomial f ∈ F [x], and so L/K is also
a splitting field of f . In our proof of the uniqueness of splitting fields, we showed that for any
isomorphism φ : K → K such that φ(f ) = f splits over L, we can extend φ to φe : L → L, which will
also be an isomorphism since φ(L)/F
e is also a splitting field of f .
φ
/L
e
L ∼

φ
K /K
∼

F
Thus we have a surjective homomorphism π : NG (H)  Aut(K/F ) with kernel
Ker(π) = { g ∈ G | g(K) = K, g|K = idK } = H.
∼
Thus this gives an isomorphism NG (H)/H − → Aut(K/F ).
Proof of (3). We know that the finite extension K/F is Galois iff |Aut(K/F )| = [K : F ]. Since
K/F is separable, this implies it is normal iff equality holds in
|NG (H)| [L : F ] |G|
= |NG (H)/H| = |Aut(K/F )| = |EmbF (K, K)| ≤ [K : F ] = = ,
|H| [L : K] |H|
which is true iff NG (H) = H, i.e., iff H E G. 
The list of theorems above (Basic Galois correspondence, Degree Galois correspondence, Lattice
Galois correspondence, and Normality in the Galois correspondence) taken together are “The Galois
Correspondence”.
We have proved everything, except for the “Tech Lemma”.

35. Example: Galois extensions of degree 4

2 − 2)(x2 − 3) ∈ Q[x]. List roots as α =
√ √ √ √
Example. f =√(x√ 1 2, α2 = − 2, α3 = 3, α4 = − 3.
Then L = Q( 2, 3), G ≈ h(1 2), (3 4)i ≤ S4 .
{e} L
2 2 2 2 2 2
√ √ √
h(3 4)i h(1 2)(3 4)i h(1 2)i Q( 2) Q( 6) Q( 3)
2 2 2 2 2 2

G Q

Example. f = Φ5 ∈ Q[x]. List roots as α1 = ζ = e2πi/5 , α2 = ζ 2 , α3 = ζ 3 , α4 = ζ 4 . Then L = Q(ζ),

G ≈ h(1 2 4 3)i ≤ S4 .
{e} L
2 2

h(1 4)(2 3)i Q(ζ + ζ −1 )

2 2

G Q
Note that ζ + ζ −1 = 2 cos 2π/5.
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 34

36. Example: Galois extensions of degree 6

√
Example. f = x3 − 2 ∈ Q[x], F = Q, L = Σf /Q = Q(α, αω, αω 2 ), with α = 3 2 and ω = e2πi/3 . I’ll W 30 Nov
list the roots as α1 = α, α2 = αω, α3 = αω 2 . Then G = Gal(L/Q) ≈ S3 . The following diagram
shows subgroups and corresponding subfields.
{e} L
2 2 2 2 2 2

3 h(2 3)i h(1 3)i h(1 2)i 3 Q(α) Q(αω) Q(αω 2 )

h(1 2 3)i 3 3 3 Q(ω) 3 3 3

2 2
G Q
Example. f = Φ7 = x6 + x5 + x4 + x3 + x2 + x + 1 ∈ Q[x], F = Q, L = Σf /Q = Q(ζ), with
ζ = e2πi/7 . I’ll list the roots as αk = ζ k , k = 1, . . . , 6. Then G = Gal(L/Q) ≈ C6 . We can take as
a generator the element σ ∈ G which sends σ(ζ k ) = ζ 3k , since this has order 6. It corresponds to
the permutation (1 3 2 6 4 5) of roots. The following diagram shows subgroups and corresponding
subfields.
{e} L
2 2

3 hσ i 3
3 Q(ζ + ζ −1 )

hσ 2 i 3 Q(ζ + ζ 2 + ζ 4 ) 3

2 2

G Q
All of the subextensions K/F are normal. Note that σ 3 coincides with complex conjugation, so
3
Lhσ i = L ∩ R. We have β := ζ + ζ −1 = 2 cos 2π/7. You can show directly that this has a minimal
polynomial of degree 3, namely x3 + x2 − 2x − 1. The √ element γ := ζ + ζ 2 + ζ 4 has minimal
polynomial x2 + x + 2 = 0, and that in fact γ = (−1 + i 7)/2.

37. Example: Galois extension of degree 8

Example. f = x4 − 2 ∈ Q[x], F = Q, L = Σf /Q , G = Gal(L/Q) ≤ S4 . The polynomial has roots
√
α1 = α = 2, α2 = iα, α3 = i2 α = −α, α4 = i3 α = −iα.
4

Consider the following partial diagram of subfields.

L = Q(α, i)

Q(α)

Q(i)
Q
Claim: f is irreducible (e.g., by Eisenstein). Also, mi/Q = x2 + 1. Therefore [Q(α) : Q] = 4 and
[Q(i) : Q] = 2.
Also, i ∈/ Q(α) ⊆ R, so [L : Q(α)] ≥ 2. Since [L : Q(α)] ≤ [Q(i) : Q] = 2, we conclude that
[L : Q(α)] = 2 and thus [L : Q] = [L : Q(α)][Q(α) : Q] = 8, and so |G| = 8.
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 35

I claim that G = D4 . To see this, note that the relations α1 = −α3 and α2 = −α4 limit how
elements of G can permute the roots, so that only 8 elements of S4 are possibilities. These include
r = (1 2 3 4) and s = (2 4).
Here is the full diagram of subgroups and subfields.
{e} L

√
hsi hsr2 i hr2 i hsri hsr3 i Q(α) Q(iα) Q( 2, i) Q(α − iα) Q(α + iα)

√ √
hs, r2 i hri hs, sri Q( 2) Q(i) Q(i 2)

G Q
Aside from {e}, the normal subgroups of G are exactly those which contain r2 . √
Thus, the proper
intermediate fields K such that K/Q is normal are precisely the subfields of Q( 2, i). The other
four extensions K/Q of degree 4 are not normal over Q. Instead, there are isomorphisms of fields
Q(α) ≈ Q(iα) and Q(α − iα) ≈ Q(α + iα), because hsi and hsr2 i are conjugate subgroups, and
likewise hsri and hsr3 i.

38. Proof of the lemma on degree of a fixed field

Here is the proof of the “Technical Lemma”, that |L : F | = |G| if F = LG for a finite group of
automorphisms G.
Proof that [L : F ] ≥ |G|. Either [L : F ] is infinite in which case the conclusion is immediate, or
[L : F ] < ∞. In that case, we have G = EmbF (L, L), and we know that
|G| = |EmbF (L, L)| ≤ [L : F ],
using the upper bound on the number of embeddings from the embedding theorem. 
For the rest, we are going to need a bit of linear algebra.
Lemma (Linear algebra lemma). A system of m homegeneous linear equations in n variables always
has a non-trivial solution if n > m.
Equivalently, if a system of m homogeneous linear equations in n variables has only one solution,
then n ≤ m.
Proof. This is a theorem of linear algebra. The space of solutions is the nullspace of the m × n
matrix A = (aij ), consisting of column vectors in x ∈ K n such that Ax = 0. The dimension over K
of the null space is
nullity A = (number of columns of A) − (rank of A) = n − rank A.
If the only solutions are trivial, then nullity A = 0 whence n = rank A. But in general rank A =
min(m, n), so we must have n ≤ m as desired. 
Proof that [L : F ] = |G|. We know that |G| ≤ [L : F ]. Let m = |G|, and write φ1 , . . . , φm for the
distinct elements of G. We suppose there exists a set v1 , . . . , vn of F -linearly independent elements
of L, with n = m + 1, and derive a contradiction. This will show that dimF L ≤ n, and thus
completes the proof.
Consider the homogeneous linear system over L of the form
φi (v1 )x1 + · · · + φi (vn )xn = 0, i = 1, . . . , m.
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 36

with m equations and n = m + 1 variables. This is the matrix equation Ax = 0, where A =

(φi (vj )) ∈ Mm×n (L).
Since n > m there are more variables than equations, so there exists x = (x1 , . . . , xn ) ∈ Ln with
x 6= 0 and Ax = 0. Choose such a non-trivial solutions with the fewest number of non-zero entries,
and relabel the vj ss (and thus the xj s) so that x1 , . . . , xr ∈ L× and xr+1 = · · · = xn = 0. Thus we
have
φi (v1 )x1 + · · · + φi (vr )xr = 0, x1 , . . . , xm ∈ L× , i = 1, . . . , m, r ≥ 1.
Now let ψ ∈ G be any element. Applying it to the above equation gives
(ψφi )(v1 ) ψ(x1 ) + · · · + (ψφi )(vr ) ψ(xr ) = 0, i = 1, . . . , m.
The list ψφ1 , . . . , ψφm just runs through all elements of G in a different order. So we can write this
system as
φi (v1 ) ψ(x1 ) + · · · + φi (vr ) ψ(xr ) = 0, i = 1, . . . , m,
For each i = 1, . . . , m, we form (original equation) ψ(xr ) − (new equation) xr , and get



φi (v1 ) x1 ψ(xr ) − ψ(x1 )xr + · · · + φi (vr ) xr ψ(xr ) − ψ(xr )xr = 0, i = 1, . . . , m.
The last term drops out, so we really have
φi (v1 )y1 + · · · + φi (vr−1 )yr−1 = 0, i = 1, . . . , m,
where yj = xj ψ(xr ) − ψ(xj )xr . By minimality of the solution of Ax = 0, we must have y1 = · · · =
yr−1 = 0, and thus
xj /xr = ψ(xj /xr ), ∀ ψ ∈ G, j = 1, . . . , r − 1.
Thus xj /xr ∈ LG = F , i.e., xj = cj xP ×
r for c1 , . . . , cr ∈ F , with cr = 1.
Now take the original equations j φi (vj )xj = 0 and divide through by xr to get
φi (v1 )c1 + · · · + φi (vr )cr = 0, c1 , . . . , cr ∈ F × , i = 1, . . . , m, r ≥ 1.
In particular, when φi = id, we get c1 v1 + · · · + cr vr = 0 with c1 , . . . , cr ∈ F ×, contradicting linear
independence of the vj s.


39. Minimal polynomials and the Galois correspondence

Any finite extension K/F is contained in a finite normal extension L/F . For instance, take F 2 Dec
L = Σf /K , where f = mα1 /F · · · mαn /F ∈ F [x] with K = F (α1 , . . . , αn ). The extension L/F
constructed this way is called the normal closure of K/F : it is the “smallest” normal extension of normal closure
K with the property that if g ∈ Irred(F ) such that g has a root in K, then g splits over L.
If the original extension K/F is also separable, then the normal closure is a Galois extension.
Suppose we have finite Galois extension L/F with G = Gal(L/F ). Given α ∈ L, what can we
say about: its minimal polynomial g = mα/F , the roots of g in L, and the spliting field Σg/F ⊆ L?
Given α ∈ L, we write
Gα := { g(α) | g ∈ G } ⊆ L.
This is called the subset of Galois conjugates of α in L. Galois conjugates
Note that the g(α) need not be distinct for distinct g, so Gα could have fewer elements than G.
Proposition. Given L/F finite Galois with G = Gal(L/F ), and α ∈ L, we have the following.
(1) For α ∈ L, we have Y
m = mα/F = (x − β) ∈ F [x].
β∈Gα
(2) We have that F (α) = LH where H = Stab(α) = { g ∈ G | g(α) = α }. Thus
deg m = |Gα| = [F (α) : F ] = |G : H|.
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 37

(3) We have that Σm/F = F (g(α), g ∈ G) = LN , where N = g∈G gHg −1 E G is the largest
T

normal subgroup of G which is contained in H. Thus Gal(Σm/F /F ) ≈ G/N .

Q
Proof. For (1), note that any g ∈ G applied to f = β∈Gα (x − β) just permutes the factors, so in
fact f ∈ (L[x])G = LG [x] = F [x]. It must be irreducible over F , since G acts transitively on the set
of roots of f , so it cannot factor over F non-trivially.
For (2), clearly F (α) ⊆ LH , so equality holds since deg m = |Gα| = |G/H| = [LH : F ] using the
orbit/stabilizer theorem.
For (3), clearly the splitting field of m is generated by its roots, which are exactly the the Galois
conjugates of m. We know from the Galois correspondence that
• g(F (α)) = F (g(α)) corresponds to gHg −1 , and
• Σm/F is the composite field of the F (g(α))s, and composite fields correspond to intersection
of subgroups.

We can say this in terms of the action of the Galois group G on the set L of elements in the
Galois extension: there is a bijective correspondence
 
orbits of action of G on L ←→ f ∈ Irred(F ) which have a root in L ,
where the orbit corresponding to f is the set of roots of f . Furthermore, the size of the orbit is the
degree of the corresponding irreducble.
Example. Consider C/R with |G| = 2. Then orbits of G acting on C have the form (i) {c}, c ∈ R,
or (i) {λ, λ}, λ ∈ C r R. These correspond to polynomials in R[x] of the form (i) f = x − c, or (ii)
f = x2 + bx + c with b2 < 4c. In particular, all f ∈ Irred(R) have degree 1 or 2.
Example. Fix a prime p. Then every finite field of characteristic p is isomorphic to Fpn for some
n ≥ 1. Note that each Fpn /Fp is a Galois extension with cyclic Galois group G = hφi ≈ Cn generated
by Frobenius. The intermediate fields of this are exactly the Fpd for all divisors d | n, and note that
since G is abelian there is only one subfield of Fpn which is isomorphic to a given Fpd .
We can use this to count the number of irreducible polynomials of degree n in Fp [x]. In fact, if
we write md := number of monic irreducibles in Fp [x] of degree d, we have the formula
X
pn = dmd , for all n ≥ 1.
d|n

Using this we can inductively compute md :

m1 = (p)/1, m5 = (p5 − p)/5,
m2 = (p2 − p)/2, m6 = (p6 − p3 − p2 + p)/6,
m3 = (p3 − p)/3, m7 = (p7 − p)/7,
m4 = (p4 − p2 )/4, m8 = (p8 − p4 )/8.
To see this, note that if f ∈ Irred(Fp ) with deg f = d, then its splitting field must be isomorphic
Fpd . In fact, f has a root in Fp [x]/(f ) ≈ Fpd , and therefore splits since this is a Galois extension
over Fp . Since Fpn contains a subfield isomorphic to Fpd if and only if d | n, we see that: f has a
root over Fpn iff f splits over Fpn iff d | n.
Now consider the orbits of the action by G ≈ Cn on Fpn , so that
X X X
|Fpn | = |O| = d · (number of orbits of size d) = dmd ,
orbits O d|n d|n

because an orbits of size d is exactly the set of roots of a monic irreducible of degree d, and all of
these split over Fpn .
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 38

40. Primitive element theorem

Lemma. Suppose F is an infinite field. Let V be a finite dimensional F -vector space, and suppose
that W1 , . . . , Wk is a finite collection of subspaces. If V = W1 ∪ · · · ∪ Wk , then V = Wi for some i.
Proof. Let n = dim V ; we argue by induction on n, noting that the cases of n = 0 and n = 1 are
obvious. Consider V 0 ( V any proper subspace, and let Wi0 := V 0 ∩ Wi . Then V 0 = W10 ∪ · · · ∪ Wk0 ,
and so by the inductive argument V 0 = Wi0 for some i, i.e., V 0 ⊆ Wi .
Note that if V 0 has codimension 1 in V , the statement V 0 ⊆ Wi implies either V 0 = Wi or Wi = V .
So there are two possibilities: either
(1) every codimension 1 subspace of V is an element of the set {W1 , . . . , Wk }, or
(2) there exists an i such that Wi = V .
So we can finish the proof by showing that for n ≥ 2 there are infinitely many distinct codimension
1 subspaces of V , which without loss of generality we can take to be F n . Given a ∈ F , let
Va = { (x1 , . . . , xn ) ∈ F n | x1 = ax2 }.
Each Va is the kernel of a surjective linear map F n → F and so has codimension 1 in F n . If a 6= b,
then v = (a, 1, 0, . . . , 0) is in Va but not in Vb , whence Va 6= Vb . Since F is infinite, this gives
infinitely many such subspaces.

The primitive element theorem says that all finite separable field extensions are simple extensions.
Theorem (Primitive element theorem). Let K/F be a finite separable field extension. Then there
exists γ ∈ K such that K = F (γ).
Proof. The case of finite fields is easy: if F is finite then so is K, and if |K| = pk any primitive
(pk − 1)st root of unity in K is certainly a primitive element. Thus we are left with the case of
infinite fields.
Let L be a normal closure of K/F . Then L/F is a finite Galois extension, and hence by the Galois
correspondence there exist only finitely many intermediate extensions (corresponding to the finitely
many subgroups of the Galois group). Therefore K/F also has only finitely many intermediate
extensions.
Let K1 , . . . , Kr be the list of proper intermediate extensions of K/F . Then by the above lemma,
there exists γ ∈ K r (K1 ∪ · · · ∪ Kr ), which necessarily satisfies F (γ) = K. 

41. Cyclotomic extensions

Let K be a field, let n ≥ 1, and suppose there exists a primitive nth root of unity ζ = ζn ∈ K.
This is equivalent to saying there is a cyclic subgroup of order n in K × , with ζn as its generator.
This ζn is a root of f = xn − 1, which is defined over the prime subfield of K. Note that every ζ k
must be a root of f , so f must be a separable polynomial, since the elements in the list 1, ζ, . . . , ζ n−1
are pairwise distinct because ζ is a primitive nth root of unity. Clearly, f and Df = nxn−1 are
relatively prime iff n 6= 0 in the field.
Therefore: a primitive nth root of unity can exist in K only if char K = 0 or char K = p > 0 with
p - n. Conversely, if either char K = 0 or char K = p > 0 with p - n, then K has a finite extension
field which has a primitive nth root of unity.
Now let’s assume char K = 0 or char K = p > 0 and p - n, and consider an extension K = F (ζn )/F .
This is a Galois extension. What can we say about G = Gal(F )?
For any g ∈ G, we have that g(ζn ) = ζnk for some k ∈ Z. This k is well-defined modulo n. Since g
restricts to an automorphism of the group hζn i ⊆ K × , it must take ζn to a another generator of
this group. Thus we must have gcd(k, n) = 1.
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 39

Thus we have a function

π : Gal(K/F ) → (Z/n)× , g(ζn ) = ζnφ(g) .
This is a group homomorphism:
(gh)(ζn ) = g(h(ζn )) = g(ζ φ(h) ) = (ζ φ(g) )φ(h) = ζ φ(g)φ(h) , =⇒ φ(gh) = φ(g)φ(h).
This is injective: g ∈ Ker π iff φ(g) = 1 iff g(ζn ) = ζn .
Thus, for any F (with characteristic not dividing n) we have that G = Gal(F (ζ)/F ) is isomorphic
to a subgroup of (Z/n)× .
Remark. If used a different primitive nth root of unity  = ζna instead of ζn , we actually get the
same formula:
g() = g(ζ a ) = (ζ φ(g) )a = (ζ a )φ(g) = φ(g) .
Thus G is naturally isomorphic to a subgroup of (Z/n)× .
One case when the image of π is surjective is when F = Q.
Proposition. We have that Gal(Q(ζn )/Q) ≈ (Z/n)× .
Proof. Since Φn ∈ Irred(Q), we have [Q(ζn ) : Q] = φ(n) = |(Z/n)× |. 
Remark. The structure of these groups is understood. Suppose n = pk11 · · · pkr r is a prime factorization.
The Chinese Remainder Theorem gives
(Z/n)× ≈ (Z/pk11 )× · · · × (Z/pkr r ).
Then you can show that for p odd,
(Z/pk )× ≈ Cpk−1 × Cp−1 = h1 + pi × hai, k ≥ 1,
k−1
where a = bp and a projects to a generator of (Z/p)× . When p = 2 this is slightly different:
(Z/pk )× ≈ C2k−2 × C2 = h1 + 22 i × h−1i, k ≥ 2.
Example (Cyclotomic extensions of Fp ). Suppose p - n. Then Gal(Fp (ζn )/Fp ) is a cyclic group of
order d, generated by the Frobenius automorphism φ(a) = ap , where d = the order of p in (Z/n)×
Thus, Fp (ζn ) ≈ Fpd , where d = the smallest power of p which is congruent to 1 modulo n.
An extension L/F is abelian if L/F is Galois and Gal(L/F ) is abelian. The cyclotomic extensions abelian extension
provide a large class of abelian extensions, and in fact supply all abelian extensions of Q.
Theorem (Kronecker-Weber). Every finite abelian extension L/Q is contained in some cyclotomic
extension Q(ζn )/Q.
This is a non-trivial theorem, the subject for a course in algebraic number theory.

42. Extensions by nth roots

Let F be a field. Let n ≥ 1, and suppose either char F = 0 or char F = p > 0 with p - n.
Then for any c ∈ F , the polynomial f = xn − c ∈ F [x] is separable. Form its splitting field L/K.
If α, β ∈ L are two roots of f , then
(β/α)n = β n /αn = c/c = 1,
an nth root of unity. Since there are n distinct roots, a primitive nth root of unity must appear as
such a ratio. Thus
L = F (α, ζ), αn = c, ζ ∈ L× , |ζ| = n,
√
where α = n c is a chosen nth root of c, and ζ a chosen nth root of unity. The roots of f in L are
{α, αζ, . . . , αζ n−1 }.
Let K = F (ζ). We have a chain of extensions F ⊆ K ⊆ L, all of which are normal and separable.
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 40

Let G = Gal(L/F ), and let H = Gal(L/K), so Gal(K/F ) ≈ G/H. Since K/F is a cyclotomic
extension, we can identify G/H with a subgroup of (Z/n)× via φ : G/H  (Z/n)× by g(ζ) = ζ φ(g) .
If h ∈ H, then h(ζ) = ζ. Thus h(α) = αζ a for some integer a, which is unique modulo n. We
thus define a function
ψ : H → Z/n, h(α) = αζ ψ(h) .
This is a group homomorphism:
0 0 0
hh0 (α) = h(αζ ψ(h ) ) = αζ ψ(h) ζ ψ(h ) = αζ ψ(h)+ψ(h ) .
Furthermore, it is injective, since L = K(α). Therefore H isomorphic to a subgroup of Z/n.
Recall that a group G is solvable if there exists a chain of subgroups
{e} = G0 E G1 E · · · E Gs = G,
each normal in the next, so that the quotionts Gk /Gk−1 are abelian.
Proposition. Let L/F be a splitting field of xn − c ∈ F [x], where char F does not divide n. Then
G = Gal(L/F ) is a solvable group, of order dividing nφ(n)
Proof. We have H ≤ Z/n and G/H ≤ (Z/n)× , which are both abelian. 
Remark. In general, the action of a particular element g ∈ G on L/F is determined by formulas of
the form
g(α) = αζ a , g(ζ) = ζ b ,
for some a ∈ Z/n and b ∈ (Z/n)× depending on g. Using this, you can see that G will be
isomorphic to some subgroup of the semi-direct product group G0 = (Z/n) oα (Z/n)× , where
∼
α : (Z/n)× −→ Aut(Z/n) is the standard isomorphism, defined by b 7→ (a 7→ ab). Thus the group
structure of G0 is given by
(a, b) · (a0 , b0 ) = (a + ba0 , bb0 ).
Exercise. Show that if p is prime, then the Galois group of the splitting field of f = xp − 2 over Q is
the largest possible, i.e., of order p(p − 1).

43. Solvability by radicals

Assume char F = 0. Recall a radical extension K/F is an extension such that there exists a M 5 Dec
chain of extensions radical extension
√
F = K0 ⊆ K1 ⊆ · · · ⊆ Kr = K, Kj = Kj−1 ( nj cj ), nj ≥ 1, cj ∈ Kj−1 ,
√
where we use the symbol “ n c” to denote any root of the polynomial xn − c.
We say that f ∈ F [x] is solvable by radicals if its splitting field Σ/F is contained in some solvable by radicals
radical extension L/F .
Lemma. If K, K 0 ⊆ L/F such that K/F and K 0 /F are radical extensions, then KK 0 /F is a radical
extension.
√
Proof.
√ A straightforward induction, based on the case of K 0 = F ( n c), in which case KK 0 =
K( n c). 
Lemma. If K/F is a radical extension, then it is contained in some Galois radical extension.
Proof. Given K/F , form a normal closure L/F , which is Galois with Galois group G. Then L is the
composite subfield of the collection of subfields { g(K) | g ∈ G }. Since each g(K)/F is isomorphic
to K/F it is radical, and thus L/F is radical and Galois. 
Proposition. If K/F is contained in a radical extension, then Aut(K/F ) is solvable.
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 41

Proof. Choose a radical extension R/F containing K, and then choose a Galois closure L/F of R/F ,
which is also radical. We have that Aut(K/F ) ≈ NG (H)/H where H = Aut(L/K) ≤ NG (H) ≤ G.
We know that any subgroup and quotient group of a solvable group are solvable. Thus it suffices to
show that G is solvable.
√
Since L/F is radical, there is a chain F = K0 ⊆ · · · ⊆ Kr = L with Kj = Kj−1 ( nj cj ). Inductively
√
define Lj so that L0 = K0 = F , and Lj = the normal closure of Lj−1 ( nj cj ), (i.e., the splitting field
of xnj − cj over Lj−1 ) which is contained in L since L/F is normal. Thus Lr = L.
We thus have a chain of extensions
F = L0 ⊆ L1 ⊆ · · · ⊆ Lr−1 ⊆ Lr = L,
such that each Lj /Lj−1 is normal. Thus the associated chain of Galois groups Gj = Gal(L/Lj ) has
the form
G = G0 D G1 D · · · D Gr−1 D Gr = {e},
with each Gj−1 /Gj = Gal(Lj /Lj−1 ) a solvable group, since Lj /Lj−1 is a splitting field for xnj − cj ∈
Lj−1 [x]. Thus G is solvable. 

44. A polynomial which is not solvable by radicals

Proposition. Let p be a prime number. Let f ∈ Irred(Q) with deg f = p, such that f has exactly
two non-real roots in C. Then the Galois group of the splitting field of f is isomorphic to Sp .
Proof. Let L ⊆ C be the splitting field. Since f is irreducible it is separable, by labelling the roots
we can identify G = Gal(L/Q) with a subgroup of Sp .
We have that [Q(α1 ) : Q] = p, so p divides [L : F ] = |G|. By Cauchy’s theorem there exists an
element of order p in G, which must therefore be a p-cycle σ ∈ Sp .
Let τ ∈ G be the automorphism induced by complex conjugation (defined because Q ⊆ R). Then
τ corresponds to a 2-cycle in Sp , since there are only two non-real roots.
The statement now follows from the following. 
Proposition. For p prime, we have Sp = hσ, τ i, where σ is any p-cycle and τ is any 2-cycle.
Proof. WLOG we can assume τ = (1 2). Replace σ with a power of it which sends 1 to 2, which will
also be a p-cycle since p is prime. Thus we can assume WLOG that σ = (1 2 · · · p). (More precisely:
there is an inner automoprhism of Sp which sends the original σ and τ to these new cycles.)
Then σ k−1 τ σ −(k−1) = (k k + 1), and the set of such transpositions with k = 1, . . . , p − 1 generate
Sp . 
Example. Let f = x5 − 6x + 3 ∈ Q[x]. This is irreducible over Q by Eisenstein, using p = 3.
Furthermore, pf has exactly 3 real roots. To show this, note that D(f ) = 5x4 − 6 has exactly two
real roots (at ± 4 6/5), neither of which is repeated. Thus the derivative of f changes sign exactly
twice over R. We have
lim f (x) = −∞, f (0) = 3 > 0, f (1) = −2 < 0, lim f (x) = +∞,
x→−∞ x→+∞
p p
and since − 4 6/5 < 0 < 1 < 4 6/5, we can deduce using the intermediate value theorem that f has
exactly 3 real roots.
Thus, by the preceding proposition, the Galois group of the splitting field is isomorphic to S5 ,
which is not solvable. Therefore f is not solvable by radicals.

45. Linear independence of characters

For any field L, and any set S, the set of all functions F(S, L) is itself a vector space over L, by
“pointwise” addition and scalar multiplication.
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 42

Proposition (Linear independence of “characters”). Let G be a group, L a field, and let χ1 , . . . , χn

be distinct homomorphisms G → L× . Then, viewed as functions G → L, they form a linearly
independent subset of F(G, L).
P
Proof. We show that if a1 , . . . , an ∈ L are such that j aj χj = 0, then all aj = 0. This amounts to
P
showing: if j aj χj (g) = 0 for all g ∈ G, then all aj = 0.
P
Key observation: for any non-trivial linear dependence j aj χj = 0 and h ∈ G, we can produce
P
a new linear dependence j aj χj (h)χj = 0, since
0 = a1 χ1 (hg) + · · · + an χn (hg) = a1 χ1 (h)χ1 (g) + · · · + an χn (h)χn (g).
P
Then, by subtracting φn (h) j aj φj = 0 from this new dependence, we get



a1 φ1 (h) − φn (h) φ1 + · · · + an−1 φn−1 (h) − φn (h) φn−1 .
We argue by contradiction, i.e., suppose there exists a non-trivial linear dependence nj=1 aj χj = 0.
P

Choose among these one with shortest length P m, i.e., smallest (positive) number of non-zero
coefficients. By relabelling this has the form m j=1 aj χj with a1 , . . . , am non-zero.
• If m = 1, then a1 χ1 = 0, whence a1 χ(1) = a1 (where 1 ∈ G is the identity element),
contradicting a1 6= 0.
• If m ≥ 2, then since χ1 6= χm we can find h ∈ H such that χ1 (h) 6= χm (h). Then using the
argument above we get a new linear dependence




a1 χ1 (h) − χm (h) χ1 + · · · + an−1 χm−1 (h) − χm (h) χm−1 , a1 χ1 (h) − χm (h) 6= 0.
This contradicts the minimality.

Consider fields K and L. The set Emb(K, L) of embeddings is a subset of the set of all functions
F(K, L).
Corollary (Linear independence of embeddings). The subset Emb(K, L) of F(K, L) is linearly
independent over L.
Proof. Restriction along K × ⊂ K defines a surjective L-linear map F(K, L)  F(K × , L). To show
EmbF (K, L) ⊆ F(K, L) is linearly independent, it suffices to show that its image in F(K × , L) is
linearly independent. (Note that distinct embeddings still differ as functions K × → L, since they
all send 0 7→ 0.)
But a field homomorphism φ : K → L restricts to a “character”, i.e., a group homomorphism
φ : K × → L× . Thus the claim results from linear independence of characters applied to G = K × . 

46. Kummer theory

A cyclic extension is a finite Galois extension with cyclic Galois group. It turns out we can cyclic extension
completely classify cyclic extensions over fields which have “enough” roots of unity: in this case
they are exactly the root-extensions.
Theorem (Kummer theory). Let F be a field which contains a primitive nth root of unity ζ.
Consider a finite Galois extension L/F with [L : F ] = n. The following are equivalent.
(1) G = Gal(L/F ) is a cyclic group. √
(2) There exists c ∈ F such that f = xn − c ∈ Irred(F ) and L = F ( n c).
Proof. (2) =⇒ (1). We have basically already done this: Since F contains a primitive nth root of
√ √
unity, we have an injective homomorphism ρ : G  Z/n defined by g( n c) = ζ ρ(g) n c. Since f is
irreducible, we have [L : F ] = |G| = n, so this is an isomorphism.
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 43

(1) =⇒ (2). Fix a primitive nth root of unity ζ ∈ F , and a generator σ ∈ G of the cyclic group.
For α ∈ L, we define a function λ : L → L by
n−1
X
λ(α) := ζ −k σ k (α) = α + ζ −1 σ(α) + ζ −2 σ 2 (α) + · · · + ζ −(n−2) σ n−2 (α) + ζ −(n−1) σ n−1 (α) ∈ L.
k=0
We can compute that
σ(Lλ(α)) = σ(α) + ζ −1 σ 2 (α) + ζ −2 σ 3 (α) + · · · + ζ −(n−2) σ n−1 (α) + ζ −(n−1) α
= ζλ(α).
Therefore
n
σ(λ(α) = ζ n λ(α)n = λ(α)n ,
so λ(α)n ∈ LG = F .
Suppose we find α ∈ L such that β := L(α) 6= 0. Then c := β n ∈ F , and also σ k (β) = ζ k β, which
√
means that β is not fixed by any non-identity element of G, so L = F (β). This exhibits L = F ( n c)
as desired.
The existence of such an α is given by linear independence of embeddings. If no such α exists,
then λ(α) = 0 for all α ∈ L, whence we have a non-trivial linear dependence
id +ζ −1 σ + · · · + · · · + ζ −(n−1) σ n−1 = 0
of elements of G. But G ⊆ Emb(L, L) ⊆ F(L, L), which is an L-linearly indpendent subset, so this
is impossible. 
Remark. The hypothesis about having a primitive root of unity in F is necessary. In particular, if
L/F is a cyclic Galois extension of prime degree p such that F does not contain a primitive pth
root of unity, then there is no α ∈ L r F such that αp ∈ F . (Exercise: prove this.)
An example is the splitting field L of f = x3 + x2 − 2x − 1 ∈ Irred(Q). As we have seen,
G = Aut(L/Q) is cyclic of order 3. In fact, L ⊆ Q(ζ7 ), where ζ7k + ζ7−k for k = 1, 2, 3 are the roots
of f . However, there is no α ∈ L r Q such that α3 ∈ Q. (Exercise: why is there no such α?)

47. Classification of solvable polynomials

We can use Kummer theory to give a criterion for radical extensions, as long as the base field W 7 Dec
has “enough” roots of unity.
Proposition. Let L/F be a Galois extension with solvable Galois group G, with n = |G|. If F
contains a primitive nth root of unity, then L/F is a radical extension.
Proof. Since G is solvable, there exists a chain of subgroups
G = G0 ⊇ G1 ⊇ G2 ⊇ · · · ⊇ Gr = {e}
with each Gj E Gj−1 and Gj−1 /Gj is finite cyclic. Note that |Gj−1 : Gj | divides n.
We have a corresponding chain of fixed fields
F = L0 ⊆ L1 ⊆ L2 ⊆ · · · ⊆ Lr = L,
where Lj = LGj . We have that each Lj /Lj−1 is finite Galois with cyclic Galois group of order nj
dividing n. By hypothesis, F contains a primitive nj th root of unity, so by Kummer theory we have
√
Lj = Lj−1 ( nj cj ). 
We can now give a complete classification of which polynomials f ∈ F [x] are solvable over radicals.
Proposition. Let F be a field of characteristic 0, and let L/F be a finite Galois extension with
Galois group G = Gal(L/F ). Then TFAE.
(1) There exists a finite extension R/L such that R/F is a radical extension.
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 44

(2) G is a solvable group.

Corollary. A polynomal f ∈ F [x] over a field of characteristic 0 is solvable by radicals iff the Galois
group of its splitting field is solvable.
Proof of proposition. (1) =⇒ (2). We can replace R/F by its normal closure, which will be a radical
Galois extension. Then G0 := Gal(R/F ) is solvable, as we have seen. Since L/F is Galois, G ≈ G0 /N
where N = Gal(R/F ), so G is solvable.
(2) =⇒ (1). Let n = [L : F ].
We suppose G is solvable. Write n = [L : F ] = |G|, and form an extension L ⊆ L(ζ) where ζ is a
primitive (n!)th root of unity. Thus, F (ζ) contains a primitive kth root of unity for all 1 ≤ k ≤ n.
We are going to show L(ζ)/F is a radical extension.
We have the following diagram of fields.
L(ζ)
≤n

F (ζ) L
n

F
We know L/F and F (ζ)/F are normal, and so are splitting fields of polynomials f and xn − 1 in
F [x]. Therefore L(ζ)/F is also a splitting field of (xn − 1)f in F [x] and so is also normal. Thus all
extensions pictured are Galois.
The cyclotomic extenion F (ζ)/F is certainly radical, so it suffices to show L(ζ)/F (ζ) is radical.
We have a corresponding diagram of Galois groups.
{e}

H N
G=G0 /N

G0
Since L(ζ) = F (ζ)L is a composite extension, from the Galois correspondence we have H ∩ N = {e},
and thus the homomorphism H → G0 /N ≈ G by h 7→ hN is injective. Since H = Gal(L(ζ)/F (ζ))
is isomorphic to a subgroup of the solvable group G it is also solvable.
Also, since L(ζ) = F (ζ)L, we have that k = [L(ζ) : F (ζ)] ≤ n = [L : F ]. By construction F
contains a primitive kth root of unity, so the previous proposition applies to show L(ζ)/F (ζ) is
solvable. 
Exercise. The theorem requires that L/F be a Galois extension. This is necessary. Give an example
of a finite extension L/Q such that Aut(L/Q) is solvable, but L/Q is not contained in any radical
extension of Q.

48. Ordered fields and square roots

Recall that an ordered field F is a pair (F, F>0 ), where F>0 is a subset of F for which:
• 1 ∈ F>0 , and a, b ∈ F>0 imply a + b, ab ∈ F>0 , and
• for all a ∈ F , exactly one of the following hold: (i) a ∈ F>0 , (ii) −a ∈ F>0 , (iii) a = 0.
The subset F>0 is thus the subset of positive elements. This implies that that F must have positive elements
characteristic 0, since 1 + · · · + 1 ∈ F>0 . Given this you can define an order relation a < b (equivalent
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 45

to b − a ∈ F>0 ), and show it has the usual properties. In particular, you have that (F × )2 ⊆ F>0 ,
i.e., non-zero squares are always positive.
In some cases (e.g., R, but not Q), all positive elements are squares.
Proposition. Let F be a field, and form L = F (i) where i is a root of x2 + 1 ∈ F [x]. TFAE.
(1) F has the structure of an ordered field with F>0 = (F × )2 .
/ F and L× = (L× )2 .
(2) i ∈
If these are true, L has no quadratic extensions.
Proof. (1) =⇒ (2). Since −1 ∈ / F>0 , clearly i ∈
/ F . Every element of F has a squareroot in L: by
hypothesis if a ≥ 0, while if a < 0 then a = (bi)2 with b2 = −a, b ∈ F . √
For u = a + bi ∈ L, a, b ∈ F , b 6= 0, we can choose a square root r = a2 + b2 ∈ F>0 . Then
r r
a+r a−r
+ ∈L
2 2
is a squareroot of u, where the signs on the squareroots are chosen so that
r r r
a+r a−r −b2 bi
= =+ .
2 2 4 2
(This is basically the same thing as the “half-angle p formula” from trigonometry:
p cos2 (θ/2) = (1 +
2
cos θ)/2 and sin (θ/2) = (1 − cos θ)/2, so eiθ/2 = (1 + cos θ)/2 + i (1 − cos θ)/2 if −π ≤ θ ≤ π.)
(2) =⇒ (1). Note that since i ∈ / F , we have x2 + 1 ∈ Irred(F ). Therefore char F = 6 2, since
2 2
otherwise x + 1 = (x + 1) . Also, [L : F ] = 2.
We set F>0 := (F × )2 and show it has the properties of a set of positive elements.
• Clearly 1 ∈ F>0 , and F>0 is closed under multiplication since a2 b2 = (ab)2 .
• Suppose a, b ∈ F . Then a + bi = (c + di)2 for some c, d ∈ F , so
2
a2 + b2 = (a + bi)(a − bi) = (c + di)2 (c − di)2 = (c + di)(c − di) = (c2 + d2 )2 ,


so F>0 is closed under addition.

• Suppose a ∈ F × . Then
a = (c + di)2 = (c2 − d2 ) + (2cd)i
for some c, d ∈ F not both 0. Since i ∈ / F we have 2cd = 0, whence (since char F 6= 2) either
(i) a = c2 or (ii) a = −d2 . If a, −a ∈ F>0 = (F × )2 then −1 = (−a)/a ∈ (F × )2 , but this is
impossible since i ∈ / F . Thus if a 6= 0 then either a or −a is in F>0 but not both.
The final statement about quadratic extensions is clear: every degree 2 polynomial over L
has a root in L, because L has squareroots and char L 6= 2, so the quadratic formula applies.


49. Real closed fields

A real closed field is an ordered field R such that (i) every positive element of R has a square real closed field
root in R, and (ii) every polynomial of odd degree over R has a root in R.
Thus in this case we have we have R>0 = (R× )2 , i.e., positive elements are exactly the non-zero
squares in R.
Example. The real numbers are a real closed field. The proofs of (i) and (ii) use the intermediate
value theorem: f (x) = x2 − a with a > 0 has limx→±∞ f (x) = +∞, and f (0) < 0, whence f
has roots, while g(x) of odd degree is such that limx→+∞ g(x) and limx→−∞ g(x) are infinite with
opposite signs.
Example. The field F = Qalg ∩ R consisting of real numbers which are algebraic is real closed.
 LECTURE NOTES (PART 3), MATH 500 (FALL 2022) 46

Example. A Puiseux series over F in some variable x is an expression of the form Puiseux series

X∞
f= ck xk/n , n ≥ 1, k0 ∈ Z, ck ∈ F.
k≥k0

Let F {x} denote the set of Puiseux series. This set natually a commutative ring, via the “obvious”
operations. In fact, it is a field.
If R is real closed, then so is R{x}. Positive elements of R{x} are non-zero series f = k≥k0 ck xk/n
P
such that the smallest non-zero ck0 is positive.

50. Proof of the fundamental theorem of algebra

Proposition. A field R is a real closed field iff C := R(i) is algebraically closed and i ∈
/ R.
Corollary. The complex numbers are algebraically closed.
Proof. ⇐=: Since C is algebraically closed it has all squareroots, and thus since i ∈ / R, as we have
shown R has the structure of an ordered field with R>0 = (R× )2 .
So all positive elements of R have a squareroot by definition. Finally, all f ∈ Irred(R) must have
degree 1 or 2 since [C : R] = 2, so any odd degree f ∈ R[x] must have a linear factor.
(1) =⇒ (2). Now we suppose R is real closed. We already have showed that i ∈ / C and that C
has no degree 2 extenions. We need to show C is algebraically closed.
Recall that a p-group is a finite group with order pk for some k. We use the following facts. p-group

• For every prime p, every finite group G has a subgroup P ≤ G which is a p-group, and such
that p - |G : P |. (First Sylow theorem.)
• Every non-trivial p-group has a subgroup of index p.
In fact, I only need these facts for p = 2.
We suppose K/C is a finite extension. This will be contained in a finite Galois extenion L/R.
We will show [L : R] = 2, whence C = K and thus C is algebraically closed. Let G = Aut(L : R)
and H = Aut(L : C), so |G : H| = 2.
Every non-trivial simple extension R(α)/R has even degree, since there are no f ∈ Irred(R) with
odd degree > 1. Therefore every proper subgroup of G has even index. In particular, consider a
2-Sylow subgroup P ≤ G. Since |G : P | is odd, we must have G = P , i.e., |G| = 2k for some k ≥ 1,
whence |H| = 2k−1 .
If |H| > 1, then there exists a subgroup A ≤ H with |H : A| = 2, and hence a degree 2 extension
C = LH ⊆ LK . But this is impossible because every quadratic polynomial in C splits. 
Since the reals are clearly a real closed field, this gives another proof of the algebraic closure of C.
Department of Mathematics, University of Illinois at Urbana-Champaign, Urbana, IL
Email address: rezk@illinois.edu