Fields and the Galois theory

April 21, 2024

Table of Contents
1. Algebraic extensions of fields 2
1.1. Fields, prime subfield, characteristic 2
1.2. Extensions and subextensions 2
1.3. Finite extensions 3
1.4. Simple extensions 4
1.5. Towers of simple extensions 4
1.6. The composite of two finite extensions 5
1.7. Quadratic and biquadratic extensions 5
1.8. Algebraic extensions 6
2. Adjoining roots and splitting fields 7
2.1. Adjoining roots of polynomials and conjugate elements 7
2.2. The splitting field of a polynomial 8
2.3. The algebraic closure of a field 10
2.4. Separable and inseparable polynomials and extensions 11
3. Cyclotomic extensions and finite fields 12
3.1. Roots of unity and cyclotomic fields 12
3.2. Finite fields 14
4. Galois extensions and the Galois theorem 15
4.1. Embeddings of an extension and conjugate subextensions 15
4.2. Normal extensions 16
4.3. Galois extensions and Galois groups 17
4.4. Composites and towers of separable extensions 18
4.5. Examples of Galois groups 18
4.6. The fundamental Galois theorem 20
4.7. Examples of diagrams of subextensions and the corresponding Galois groups 23
5. Composites and towers of Galois extensions 24
5.1. The change of the basic field of a Galois extension 24
5.2. The composite of two extensions of which one is Galois 25
5.3. The composite of two Galois extensions 25
5.4. Free composites of Galois extensions 26
5.5. Composites of towers of Galois extensions 26
6. Some applications of the Galois theory 27
6.1. More methods of finding the minimal polynomial 27
6.2. The norm of algebraic elements 28
6.3. Abelian extensions √ 29
6.4. Subextensions of the real radical extension F ( n a)/F , a > 0, and the Galois group of xn − a 29
6.5. The theorem on a primitive element 29
6.6. p-extensions 30
6.7. The fundamental theorem of algebra 31
6.8. Constructions with ruler and compass 31
6.9. Linear independence of square roots of square free integers 33
6.10. The theory of symmetric rational functions 33
7. Solving polynomial equations in radicals 34

1
 7.1. Radical and polyradical extensions 34
7.2. Cyclic and polycyclic extensions 34
7.3. Radical and cyclic extensions 35
7.4. Solvability of polynomials in radicals 36
7.5. The alternating group and the discriminant 37
7.6. The Galois group and solution in radicals of cubics 37
7.7. The Galois group and solution in radicals of quartics 38
7.8. Computation of Galois groups 40
8. Introduction to transcendental extensions 42

1. Algebraic extensions of fields

1.1. Fields, prime subfield, characteristic

1.1.1. A field is a commutative division ring, that is, a commutative unital ring in which all nonzero elements
are units.
1.1.2. Examples of fields.
(i) Q, R, and C.
(ii) Fp = Zp = Z/(p), where p is a prime integer.
(iii) For any integral domain we have its field of fractions.
(iv) Here are special cases of (iii): for any field F we have the field F (x) of rational functions in one variable,
and for every n, the field F (x1 , . . . , xn ) of rational functions in n variables.
(v) Let R be a commutative ring and M be a maximal ideal in R; then R/M is a field.
(vi) A special case of (v): Let F be a field and let f ∈ F [x] be an irreducible polynomial. Then F [x]/(f ) is
a field.
1.1.3. Fields have no nontrivial ideals. Hence, factorization is not defined on fields, “quotient fields” do not
exist. Any (nonzero) homomorphism of fields is a monomorphism.
1.1.4. Let F be a field, and let P be the cyclic additive subgroup of F generated by 1. There are two cases:
Case 1. P is finite. Then P ∼ = Z for some prime p ∈ N, and P is a field isomorphic to F ; it is called the
p p
prime subfield of F . In this case we say that F has characteristic p, write char F = p, and say that F has
finite characteristic.
Case 2. P is infinite, ∼
= Z. Then P is contained in (and generates) a subfield of F isomorphic to Q, which is,
again, called the prime subfield of F . We say that F has characteristic 0 in this case and write char F = 0.
In both cases, of a finite and of zero characteristic, the prime subfield is the minimal subfield of F ,
contained in all other subfields of F .
1.2. Extensions and subextensions
K
1.2.1. If K is a field and F is a subfield of K, we say that K is an extension of F , and write K/F or .
F
(More exactly, an extension is a pair (K, F ) of fields with F ⊆ K.)
1.2.2. If F is a subfield of L and L is a subfield of K, then we say that L/F is a subextension of the extension
K/F .
1.2.3. The intersection of any family of subfields of a field K is a subfield of K; if all these fields are extensions
of a subfield F of K, then their intersection is an extension of F .
1.2.4. If K/F is an extension and S is a subset of K, F [S] denotes the F -algebra generated by S,
n o
F [S] = f (α1 , . . . , αn ) : n ≥ 0, f ∈ F [x1 , . . . , xn ], α1 , . . . , αn ∈ S .

If S is finite, S = {α1 , . . . , αn }, we write F [α1 , . . . , αn ] for F [S].

2
1.2.5. Let K/F be an extension and let S be a subset of K. Then F (S) is the minimal extension of F
that contains S; it is called the extension of F generated by S. (F (S) is the intersection of all extensions
of F that contain S.) F (S) contains the ring F [S] and is (isomorphic to) the field of fractions of F [S]:
F (S) = α/β : α, β ∈ F [S], β 6= 0 .
If K = F (S) for a finite set S, we say that the extension K/F is finitely generated . If S is a finite set,
S = {α1 , . . . , αn }, then we write F (α1 , . . . , αn ) for F (S).
1.2.6. A sequence Kn /Kn−1 / · · · /K1 /F of successive extensions is called a tower of extensions. Abusing
language, we also say in this situation that Kn is a tower of extensions.
1.2.7. If L1 and L2 are subfields of a field K, then the field L1 (L2 ) = L2 (L1 ) (the minimal extension of both
L1 and L2 ) is called the composite of L1 and L2 and is denoted by L1 L2 .
1.2.8. We have the following diamond diagram of extensions:
L1 L2
L1 L2
L1 ∩ L2 .
Notice that this is the minimal such diagram, in the sense that if
K
L1 L2
L
is another diagram of extensions with the same L1 and L2 , then K is an extension of L1 L2 and L is a subfield
of L1 ∩ L2 :
K
L1 L2
L1 L2
L1 ∩ L2
L.
1.3. Finite extensions
1.3.1. If K/F is an extension, then K is an F -vector space (and an F -algebra). The dimension dimF K of
K is called the degree of this extension, or the degree of K over F , and is denoted by [K : F ].
If [K : F ] < ∞, K/F is said to be a finite extension, and is said to be an infinite extension otherwise.
K
In diagrams of extensions, the degree n = [K : F ] appears this way: n
F.
1.3.2. An extension of degree 2 is said to be quadratic, of degree 3 cubic, of degree 4 quartic, of degree 5
quintic, etc.
1.3.3. Theorem. Let  K/L/F be a tower of extensions. If B is a basis of L over F and C is a basis of K
over L, then CB = γβ : γ ∈ C, β ∈ B is a basis of K over F .
P
Proof. Every α ∈ K is representable as a finite sum α = γ∈C αγ γ with αγ ∈ L for all γ. (It is assumed
P that
all but finitely many αγ are equalPto 0.) For each γ ∈ C, αγ is representable as a finite sum αγ = β∈B aγ,β β
with aγ,β ∈ F for all β. So, α = γ∈C aβ,γ γβ. So, the set CB spans K as an F -vector space.
β∈B P
Let’s now assume that a (finite) linear combination γ∈C aβ,γ γβ = 0 where aγ,β ∈ F for all γ and β.
P P P β∈B P
then γ∈C αγ γ = γ∈C β∈B aβ,γ γβ = 0, where for each γ, αγ = β∈B aβ,γ β ∈ L. This implies that
αγ = 0 for every γ. But then, for every γ, aβ,γ = 0 for every β. Hence, the set CB is linearly independent
over F .
1.3.4. Corollary. If K/L and L/F are finite extensions, then K/F is also finite, with [K : F ] = [K : L]
· [L : F ].

3
1.3.5. Corollary. If L/F is a subextension of a finite extension K/F , then both K/L and L/F are finite,
with [K : L] [K : F ] and [L : F ] [K : F ].

1.4. Simple extensions

1.4.1. An extension K/F is said to be simple if it is generated by a single element: K = F (α) for some
α ∈ K.
1.4.2. Let K/F be an extension and let α ∈ K. We then have an F -algebras homomorphism ϕ: F [x] −→ K
sending x to α and every f ∈ F [x] to f (α). The subring ϕ(F [x]) = {f (α), f ∈ F [x]} of K is denoted by
F [α], and we have F [α] ∼
= F [x]/ ker ϕ.
1.4.3. Let K/F be a simple extension, K = F (α), and let ϕ: F [x] −→ K be the homomorphism that maps
x to α. There can be two cases:
Case 1: ker ϕ 6= 0.
Then ker ϕ is a maximal ideal in F [x], generated by an irreducible polynomial p, F [α] is a field, so K = F [α].
Thus, K = f (α), f ∈ K[x], deg f ≤ n − 1 where n = deg p, with the basis {1, α, . . . , αn−1 } over F , and
[K : F ] = n.
In this case:
(i) We say that the element α is algebraic over F .
(ii) We call the irreducible polynomial p the minimal polynomial of α and denote it by mα,F or just mα .
The minimal polynomial mα,F of α is defined uniquely up to multiplication by scalars; it is usually assumed
that it is monic. We have mα (α) = 0, and f (α) = 0 for f ∈ K[x] iff mα f . mα is the only irreducible
polynomial such that mα (α) = 0.
(iii) We call the degree of mα (which is also the degree [K : F ]) the degree of α over F and denote it by
degF α.
Case 2: ker ϕ = 0. In this case K contains the copy F [α] = ϕ(F [x]) of the ring F [x], and is its field of
fractions, so that F (α) ∼
= F (x), the field of rational functions over F . We then have [K : F ] = ∞.
In this case, we say that α is transcendental over F .
1.4.4. If K/F is a finite extension then for every α ∈ K, degF α [K : F ].
1.4.5. Let K/F be a finite extension and let α ∈ K. Here are some methods of finding the minimal
polynomial mα,F of an element α algebraic over F :
(i) Find a “small” nonzero polynomial f satisfying f (α) = 0 and prove that it is irreducible.
(ii) Write the powers of α in coordinates with respect to a basis of K over F , and find the minimal linear
dependence relation between them.
(iii) The action of α on K by multiplication, u 7→ αu, is a linear transformation of the finite dimensional
F -vector space K; let’s denote it by T . Let K = W1 ⊕ · · · ⊕ Wd be the decomposition of K into a direct sum
of cyclic T -invariant subspaces, and let p1 , . . . , pd be the invariant factors of T . The actions of T on Wi are
all isomorphic, so all invariant factors are equal, p1 = . . . = pd , and the minimal polynomial of T (and so, of
α) is p1 .
(iv) It follows from (iii) that the characteristic polynomial cT of T is mdα . so, mT is the irreducible polynomial
for which cT = mdα .
(iv) See also subsection 6.1.1 below.

1.5. Towers of simple extensions

1.5.1. Any finitely generated extension K/F is a tower of simple extensions: if K = F (α1 , . . . , αn ) then
we have the tower K = Kn /Kn−1 / · · · /K1 /K0 = F , where for each i, Ki = F (α1 , . . . , αi ), and so, Ki =
Ki−1 (αi ).
1.5.2. If L/F is a subextension of an extension K/F and α ∈ K is algebraic over F , then α is algebraic over
L as well, and mα,L mα,F , so degL α = deg mα,L ≤ deg mα,F = degF α.

4
1.5.3. It follows from Theorem 1.3.3 by induction on n that
Theorem. If K = F (α1 , . . . , αn ) and α1 , . . . , αn are algebraic over F , then K = F [α1 , . . . , αn ] and

n
Y n
Y
[K : F ] = degF (α1 ,...,αi−1 ) αi ≤ degF αi .
i=1 i=1

Proof. We have
K = F (α1 )(α2 ) . . . (αn ) = F [α1 ][α2 ] . . . [αn ] = F [α1 , . . . , αn ].

The second part immediately follows from Corollary 1.3.4 and 1.5.2.

1.6. The composite of two finite extensions

1.6.1. If K/F is a finite extension, then it is generated by finitely many algebraic elements, and is a tower
of finite simple extensions.
1.6.2. Theorem. If L1 /F and L2 /F are two finite subextensions of an extension K/F , then their composite
L1 L2 is also a finite extension of F , with [L1 L2 : F ] ≤ [L1 : F ] · [L2 : F ]. If, as F -vector spaces, L1
is spanned by a set {α1 , . . . , αn } and L2 by a set {β1 , . . . , βm }, then L1 L2 is spanned over F by the set

αi βj , i = 1, . . . , n, j = 1, . . . , m .

 
Proof. We have L1 L2 = F α1 , . . . , αn , β1 , . . . , βm = F α1 , . . . , αn , β1 , . . . , βm , which is the F -algebra
generated by L1 and L2 . The elements of this algebra are linear combinations, Pn with coefficients Pfrom F,
m
of products αβ with α ∈ L1 and P β ∈ L 2 , and for such a product, if α = i=1 a i α i and β = j=1 b j βj
with ai , bj ∈ F , we have αβ = i,j ai bj αi βj . Hence, every element of L1 L2 is a linear combination of the
elements αi βj with coefficients from F .
1.6.3. Let L1 /F and L2 /F be two finite subextensions of an extension K/F , with [L1 : F ] = n and
[L2 : F ] = m. Then in the diamond diagram
L1 L2
m′ n′
L1 L2
n m
L1 ∩ L2 .

we have nm′ = n′ m, n′ ≤ n, and m′ ≤ m. If n and m are coprime, then n′ = n and m′ = m.

1.6.4. It follows that for two finite subextensions L1 /F and L2 /F of an extension K/F the F -algebras
homomorphism L1 ⊗F L2 −→ L1 L2 is surjective. In the case [L1 L2 : F ] = [L1 : F ] · [L2 : F ], this
is
 an isomorphism, and if {α1 , . . . , αn } is a basis of L1 over F and {β1 , . . . , βm } is a basis of L2 , then
αi βj , i = 1, . . . , n, j = 1, . . . , m is a basis of L1 L2 .

1.7. Quadratic and biquadratic extensions

1.7.1. An extension K/F with [K : F ] = 2 is said to be quadratic.
√
1.7.2. Let F be √ a field with char F 6= 2, and let K/F be a quadratic extension. Then K = F ( d) for some
d ∈ F . (Under d we understand any element δ of any extension of F such that δ 2 = d.) Indeed, let α
be any element of K \ F ; then degF α = 2, so α is a root of a quadratic polynomial p = x2 + ax + b with
2 2 2
a,
√ b ∈ F , α + aα + b = 0. Then√(α + a/2) = d,√where d = a /4 − b. (d is the discriminant of p(!).) So,
d = α + a/2 ∈ K, and K = F ( d). The set {1, d} is a basis of K over F .
√
1.7.3. Let F be a field with√char F 6= 2, and let√K/F be quadratic, K = F ( d). An element α ∈ K satisfies
√
α2 ∈ F iff α ∈ F or α ∈ F d (that is, √ α = b d for some b ∈ F ). Indeed, every α ∈ K has form a√+ b d
with a, b ∈ F , then α2 = a2 + b2 d + 2ab d, and we have α2 ∈ F iff ab = 0, so either d = a, or d = b d.
1.7.4. An extension K/F with [K : F ] = 4 is said to be quartic.

5
1.7.5. A quartic extension K/F is called biquadratic if it is representable as a composite of two √ quadratic
extensions,
√ K = L 1 L 2 with [L 1 : F ] = [L 2 : F ]
√ √
= 2. Assume that char F =
6 2, then
√ √ L 1 = F ( d1 ) and
L2 = F ( d2 ) for some d1 , d2 ∈ F , and K = F d1 , d2 for some d1 , d2 ∈ F , with d1 , d2 6∈ F ; for K/F
√ √
to be quartic √it is also necessary and sufficient that L1 6= L2 , that is, d2 6= c d1 with c ∈ F , which is
equivalent to d1 d2 6∈ F .
 √ √ √
The set 1, d1 , d2 , d1 d2 is a basis of K over F .
√ √
√
1.7.6. Let char F 6= 2 and K/F be biquadratic, K = F d1 , d2 . By 1.7.3, for α ∈ K we have α2 ∈ F ( d1 )
√ √ √ √ √ √ √
iff α ∈ F ( d1 ) or α ∈ F ( d√1 ) d2 , and α2√∈ F ( d2 ) iff √ α ∈ F ( d2 ) or α ∈ F ( d2 ) d1 . It follows that
α2 ∈ F iff α ∈ F , or α ∈ F d1 , or α √ ∈ F d 2 , or
√ √α ∈ F d1 d2 . (These are one-dimensional intersections
√
of √the two-dimensional
√ subspace F ( d 1 ) or F ( d 1 ) d 2 with the two-dimensional subspace F ( d2 ) or
F ( d2 ) d1 .) Since every nontrivial proper subextension of a biquadratic extension must be quadratic, here
is the complete diagram of all subextensions of K/F :

√ √

K= F d1 , d2
2
√ √ 2
2 √

F ( d1 ) F ( d2 ) F d1 d2
2
2 2
F

p √
1.7.7. Let now char F 6= 2 and
√ K/F be a tower √ of two quadratic extensions, K = F (α) where α = a+ b
for some a, b ∈ F such that b 6∈ F and α 6∈ F ( b).

K= F (α)
2
√
F ( b)
2
F

Claim. K/F is biquadratic iff a2 − b = c2 for some c ∈ F .

√ √
√ √ √ √
Proof. Let K/F be biquadratic, K = F d1 , d2 . Since ( b)2 ∈ F and b 6∈ F , we have b ∈ F d1 ,
√ √ √ √ √ √ √
F d2 or√ F √d1 d2 . W.l.o.g.
√ assume
√ that b ∈ F d1 . Then α2 =√ a + b ∈ F ( d1 ) and α 6∈ F ( √d1 ), so
α ∈ F ( d1 ) d2 , α = x √d2 + y d1√d2 for sme x, y ∈ F . Then a + b = α2 = x2 d2 + y 2 d1 d2 + 2xyd2 d1 , so
2 2
a = x d2 + y d1 d2 and b = 2xyd2 d1 , thus

2
2
a 2 − b = x 2 d2 + y 2 d1 d2 − 4x2 y 2 d22 d1 = x2 d2 − y 2 d1 d2 .

Conversely, assume that a2 − b = c2 for c ∈ F . Put d1 = 21 (a + c) and d2 = 12 (a − c); then a = d1 + d2 ,

c = d1 − d2 , b = a2 − c2 = 4d1 d2 , and we have

√ p p p
2
α2 = a + b = d1 + d2 + 2 d1 d2 = d1 + d2 ,

√ √
√ √
√ √

and α =
 √ √
± d 1 +
 d 2 . Hence, α ∈ F d 1 , d 2 , and so K ⊆ F d1 , d2 ; since [K : F ] = 4 and
√ √

F d1 , d2 : F ≤ 4, we obtain that K = F d1 , d2 .

1.8. Algebraic extensions

1.8.1. An extension K/F is said to be algebraic if every α ∈ K is algebraic over F , and is called transcendental
otherwise.

6
1.8.2. Theorem. Any finite extension is algebraic. An algebraic extension is finite iff it is finitely generated.
Moreover, an extension is finite if it is generated by finitely many algebraic elements.
Proof. If K/F is a finite extension, then every element of K has a finite degree over F , and so, is algebraic
over F . Also, K has a finite basis, is generated by the elements of this basis, so is finitely generated.
Conversely, if K/F is a finitely generated algebraic extension, or is only generated by finitely many
algebraic elements: K = F (α1 , . . . , αk ) where α1 , . . . , αk are algebraic over F , then K is the composite
K = F (α1 ) · · · F (αk ) of finite extensions, and so, is finite.
1.8.3. We know that towers and composites of finite extensions are finite. Since algebraic extensions are
unions of finite extensions, it follows that towers and composites of algebraic extensions are algebraic:
Theorem. If K/L and L/F are algebraic extensions, then K/F is algebraic. If L1 /F and L2 /F are two
algebraic subextensions of an extension of F , then (L1 L2 )/F is algebraic. Moreover, if an extension K/F is
generated by algebraic elements, then it is algebraic.
Proof. Let K/L and L/F be algebraic extensions, and let α ∈ K; then α is algebraic over L and we need
to show that α is algebraic over F . Let mα,L = xn + βn−1 xn−1 + · · · + β1 x + β0 ∈ L[x] be the minimal
polynomial of α over L; then it is also the minimal polynomial of α over the field L′ = F (β0 , . . . , βn−1 ). L′ /F
is a finitely generated algebraic extension, so it is finite; α is algebraic over L′ , so the extension L′ (α)/L′ is
also finite; thus, L′ (α)/F is finite, and α is algebraic over F .
Assume that K/F is generated by a set S of elements algebraic over F , and let α ∈ K. Then α is a
rational function, with coefficients from F , of finitely many elements β1 , . . . , βk of S. Thus, α ∈ F (β1 , . . . , βk ),
which is a finite extension of F ; so, α is algebraic over F .
Now let L1 /F and L2 /F be two algebraic subextensions of an extension K/F . Then the composite
extension (L1 L2 )/F is generated by elements of L1 and L2 , which are algebraic over F , so (L1 L2 )F is
algebraic.

1.8.4. Let K/F be an extension. Then the set E = α ∈ K : α is algebraic over F is a subfield of K:
indeed, for any α1 , α2 ∈ E, α1 ± α2 , α1 α2 , α1 /α2 are contained in the algebraic extension F (α1 , α2 ), and so,
are algebraic over F and are contained in E. Since E contains all elements of K algebraic over F , E/F is the
maximal algebraic subextension of K/F . Any element α ∈ K \ E is transcendental over E, since otherwise
it is algebraic over F . Thus, any extension K/F decomposes into a tower K/E/F where E/F is algebraic
and K/E is transcendental with no algebraic elements.
1.8.5. Real numbers, algebraic over Q, are called algebraic numbers. Algebraic numbers form a subfield A
of R. The field A is countable (it consists of roots of polynomials with rational coefficients, the set of such
polynomials is countable, and each polynomial has only finitely many roots), so “almost all” real numbers
are transcendental .

2. Adjoining roots and splitting fields

2.1. Adjoining roots of polynomials and conjugate elements

2.1.1. If K/F is an extension, f ∈ F [x] is a polynomial, and α ∈ K is such that f (α) = 0, we say that α is
a root of f . An element α ∈ K is a root of some nonzero f ∈ F [x] iff α is algebraic over F and the minimal
polynomial mα,F of α over F divides f . In any extension of F , a nonzero polynomial f ∈ F [x] cannot have
more than deg f roots.
2.1.2. If ϕ: A1 −→ A2 is a mapping and B ⊆ A1 ∩ A2 , we say that ϕ fixes B if ϕ(a) = a for every a ∈ B.
If K1 /F and K2 /F are two extensions of a field F , a homomorphism K1 /F −→ K2 /F , or a homomor-
phism K1 −→ K2 over F , is a homomorphism ϕ: K1 −→ K2 that fixes F :
ϕ
K1 −→ K2 .
F

A homomorphism of extensions is either an isomorphism, or a proper embedding.

7
2.1.3. Theorem. Let K1 /F and K2 /F be two extensions of a field F , let f ∈ F [x] be irreducible, and
let α1 ∈ K1 and α2 ∈ K2 be roots of f . Then both F (α1 ) and F (α2 ) are isomorphic to F [x]/(p) under
isomorphisms that fix F and map α1 and α2 to x, so F (α1 )/F ∼
= F (α2 )/F under an isomorphism that maps
α1 to α2 :
∼ F (α ),
F (α1 ) −→ α1 ↔ α2 .
2

Conversely, if ϕ: K1 /F −→ K2 /F is a homomorphism of extensions of a field F and α1 ∈ K1 is algebraic over

F , then α2 = ϕ(α1 ) ∈ K2 is also algebraic over F and has the same minimal polynomial, mα2 ,F = mα1 ,F .

Proof. This is very easy: Both F (α1 ) and F (α2 ) are isomorphic to F [x]/(p), where isomorphisms
ϕi : F [x]/(p) −→ F (αi ) are defined by f mod p −→ f (αi ), i = 1, 2. In particular, ϕi fix F and map x mod p
to αi .
ϕ1 ϕ2
F (α1 ) ←− F [x]/(p) −→ F (α2 ), α1 ↔ x mod p ↔ α2 .

So, ϕ2 ◦ϕ−1
1 is an isomorphism F (α1 ) −→ F (α2 ) that fixes F and maps α1 to α2 .

2.1.4. If K = F (α) where α is a root of an irreducible polynomial f ∈ F [x], we say that K is obtained from
F by adjoining a root of f . Such a field K is unique up to isomorphism.
2.1.5. Now let F be a field and f ∈ F [x] be an irreducible polynomial. Is there always a field, an extension of
F , where f has a root? (We know that this is so for polynomials over Q or R, any such polynomial has a root
in C.) Well, if we don’t have such an extension, we can always construct it artificially. Put K = F [x]/(f );
since p is irreducible and F [x] is a PID, the ideal (f ) is prime and maximal, and K is a field. Let α ∈ K be
the class of x modulo f in K, then p(α) = f (x) mod f = 0, so α is a root of f in K. Since f is irreducible,
f is the minimal polynomial of α over F . We therefore have the following result:

Theorem. For any irreducible polynomial f over a field F there exists a simple extension K = F (α) of F
such that f (α) = 0 and f is the minimal polynomial of α.

2.1.6. Any (not necessarily irreducible) nonconstant polynomial f ∈ F [x] also has a root in some extension
of F : indeed, it suffices to adjoin a root α of one of the irreducible factors of f , then f (α) = 0.
2.1.7. It follows that two nonconstant polynomials f1 , f2 ∈ F [x] are coprime iff they have a common root in
no extension of F . Indeed, if f1 and f2 have a common root α, then they both are divisible by the minimal
polynomial mα,F of α over F . Conversely, if f1 and f2 are not coprime, they have a common irreducible
divisor g ∈ F [x], and a root of g (which exists in some extension of F ) is a common root of f1 and f2 .
2.1.8. Let K/F be an extension. Two algebraic over F elements α1 , α2 ∈ K are said to be conjugate over
F if they are roots of the same irreducible polynomial p ∈ F [x], that is, if mα1 ,F = mα2 ,F .
Since mα,F has at most deg mα,F = degF α roots in K, an algebraic over F element α ∈ K has at most
degF α conjugates in K, counting itself.
2.1.9. If L/F is a subextension of an extension K/F , then for any element α ∈ K algebraic over F we have
mα,L mα,F . Hence, the set of conjugates of α over L is a subset of the set of conjugates of α over F .

2.2. The splitting field of a polynomial

2.2.1. If ϕ: F1 −→ F2 is a homomorphism of fields, then ϕ naturally extends, by putting ϕ(x) = x, to a
homomorphism F1 [x] −→ F2 [x] of the rings of polynomials over F1 and F2 . We will use this constantly.
2.2.2. We will need the following theorem, which is an obvious generalization of the theorem saying that
conjugate elements generate isomorphic extensions.

8
Theorem. Let ϕ: F1 −→ F2 be an isomorphism of two fields, let f1 be an irreducible polynomial over
F1 , let f2 = ϕ(f1 ), let α1 be a root of f1 and α2 be a root of f2 . Then ϕ extends to an isomorphism
F1 (α1 ) −→ F2 (α2 ) that maps α1 to α2 :
∼ F (α ), α ↔ α .
ϕ: F1 (α1 ) −→ 2 2 1 2

ϕ: F1 −→ ∼ F2
Conversely, if ϕ: K1 −→ K2 is a homomorphism of fields, F1 is a subfield of K1 , F2 = ϕ(F1 ), and α1 ∈ K1
is algebraic over F1 , then α2 = ϕ(α1 ) ∈ F2 is algebraic over F2 and mα2 ,F2 = ϕ(mα1 ,F1 ).
Indeed, if F1 and F2 are isomorphic, we may simply identify them, thus identify f1 and f2 and get an
isomorphism F1 (α1 ) −→ F2 (α2 ) of extensions.
2.2.3. Let K be a field, and let f ∈ K[x] be a nonconstant. We say that f completely splits in K if
f (x) = a(x − α1 ) · · · (x − αn ) for some a, α1 , . . . , αn ∈ K. (Informally, “all roots of f are in K”; more exactly,
no additional roots of f appear in any extension of K.)
2.2.4. Let F be a field and let f ∈ F [x] be a nonconstant polynomial. An extension K/F is said to be a
splitting field of f if this is the minimal extension where f splits completely; that is, f splits completely
in K and K is generated by the roots of f : K = F (α1 , . . . , αn ) such that f (x) = a(x − α1 ) · · · (x − αn ).
(Informally, K is obtained from F by adjoining all roots of f .) √ √

2.2.5. (i) The splittig field of the polynomial f (x) = x4 − 5x2 + 6 = (x2 − 2)(x2 − 3) ∈ Q[x] is Q 2, 3 .
√ √ √
√

(ii) The splittig field of the polynomial f (x) = x3 − 2 ∈ Q[x] is Q 3 2, ω 3 2, ω 2 3 2 = Q 3 2, ω , where
√ √
ω = −1+2 −3 = 3 1.
2.2.6. Theorem. For any field F and any nonconstant polynomial f ∈ F [x], a splitting field of f exists
and is unique up to an isomorphism over F . The degree of this field over F does not exceed (deg f )!.
Proof of the existing part. The splitting field is obtained by just adjoining the roots of f one-by-one.
More formally, we use induction on n = deg f . Let L be the extension of F obtained by adjoining a root
α of f , L = F (α). Over L, f factorizes, f (x) = (x − α)g(x), with g ∈ L[x]. By induction, there is a
splitting field of g: a field K = L(α1 , . . . , αn−1 ) such that g(x) = a(x − α1 ) · · · (x − αn−1 ). But then
f (x) = a(x − α)(x − α1 ) · · · (x − αn−1 ), so f completely splits in K, and K = F (α, α1 , . . . , αn−1 ), so K is the
splitting field of f . Moreover, we have [L : F ] = degF α ≤ n (since mα,F f ), and by induction hypothesis,
[K : L] ≤ (n − 1)!, so [K : F ] ≤ n!.
2.2.7. To prove the uniqueness part of Theorem 2.2.6 by induction, it should be generalized:
Theorem. Let ϕ: F1 −→ F2 be an isomorphism of two fields, let f1 ∈ F1 [x] and f2 = ϕ(f1 ), and let K1 and
K2 be a splitting fields of f1 and f2 respectively. Then ϕ extends to an isomorphism K1 −→ K2 :
∼ K
ϕ: K1 −→ 2

ϕ: F1 −→∼ F
2
which maps the set of roots of f1 in K1 onto the set of roots of f2 in K2 .
Proof. Let p1 be an irreducible factor of f1 and let p2 = ϕ(p1 ), then p2 is an irreducible factor of f2 . Let α1
be a root of p1 in K1 and α2 be a root of p2 in K2 , and let L1 = F1 (α1 ), L2 = F2 (α2 ) by Theorem 2.2.2, ϕ
extends to an isomorphism L1 −→ L2 with ϕ(α1 ) = α2 . We now have f1 (x) = (x − α1 )g1 (x) with g1 ∈ L1 [x]
ϕ
and f2 (x) = (x − α2 )g2 (x) with g2 ∈ L2 [x]; since gi (x) = fi (x)/(x − αi ), i = 1, 2, and x − α1 −→ x − α2 ,
we have ϕ(g1 ) = g2 . Now, K1 is the splitting field of g1 and K2 is the splitting field of g2 ; by induction on
deg f1 = deg f2 , ϕ extends to an isomoprhism K1 −→ K2 which maps the set of roots of g1 obto the set of
roots of g2 .
2.2.8. Let α be an algebraic element over a field F . Then the splitting field K of the minimal polynomial
of α “contains all conjugates of α”, in the sense that if E is any extension of K, all conjugates of α in E are
contained in K.
2.2.9. Given a family F of polynomials over a field F , a splitting field of F is a minimal extension of F
where all polynomials from F completely split. If F is finite, F = {f1 , . . . , fk }, then the splitting field of F
is the splitting field of the single polynomial f1 · · · fk . We will prove that the splitting field exists in the case
F is infinite below, after we construct the algebraic closure of F .

9
2.3. The algebraic closure of a field
2.3.1. A field K is said to be algebraically closed if every nonconstant polynomial from K[x] has a root
in K. In this case for any f ∈ K[x] we have f (x) = (x − α1 )f1 (x) = (x − α1 )(x − α2 )f2 (x) = · · · =
a(x − α1 ) · · · (a − αn ), that is, every polynomial from K[x] completely splits in K.
2.3.2. It is the fundamental theorem of algebra that C is algebraically closed.
2.3.3. If K is algebraically closed, then there are no algebraic elements over K, and K has no nontrivial
algebraic extensions: indeed, if α is algebraic over K, then mα,K ∈ K[x] splits in K, so it is linear, and
α ∈ K.
2.3.4. Let F be a field; an algebraic extension K/F is called an algebraic closure of F if every polynomial
from F [x] completely splits in K. An algebraic closure of F is often denoted by F .
2.3.5. Clearly, if K/F is an extension and K is algebraically closed, then the maximal subextension of K
that is algebraic over F is an algebraic closure of F . The converse is also true:
Theorem. For every field F , the algebraic closure of F is algebraically closed.
Proof. Let F be an algebraic closure of F , and let f ∈ F [x]. Let α be a root of f (in some extension of F );
then α is algebraic over F , which is algebraic over F , so α is algebraic over F . But mα,F completely splits
in F , so α ∈ F .
2.3.6. An algebraic closure F of F is just the splitting field of the set F [x] of all polynomials over F . Indeed,
every polynomial from F [x] splits in F . On the other hand, every element of F is algebraic over F , so is a
root of some polynomial from F [x]; hence, F is generated by the roots of polynomials from F [x].
Theorem. For every field F , the algebraic closure of F exists, and is unique up to isomorphism over F .
Proof of existence. To adjoin all roots of all polynomials we, of course, need Zorn’s lemma: Consider the
set of all algebraic extensions of F . The union of any chain of algebraic extensions of F is algebraic as well,
thus Zorn’s lemma applies and provides us with a maximal algebraic extension K of F . Every polynomial
from F [x] (as well as from K[x]) splits in K since otherwise we would have a nontrivial algebraic extension
of K, which would be an algebraic extension of F strictly larger than K.
Actually, this proof contains a mistake: “the set of all algebraic extensions of F ” does not actually exist.
To correct it, take a “big” set X containing F , – of cardinality strictly larger than the cardinality of the set
of all roots of all polynomials from F [x], – and only consider the extensions of F that are subsets of X. The
large cardinality of X guarantees that any algebraic extension L/F does not exhaust X, and so, if L′ is a
larger extension, a copy of L′ can be constructed from elements of X.
You can find a different, nice proof of this theorem in the textbook.
2.3.7. The uniqueness of the algebraic closure follows from the following proposition:
Proposition. If K/F is an extension and K is algebraically closed, then for any algebraic extensions L/F
there exists an embedding L/F −→ K/F . More generally: if K is algebraically closed, ϕ: F −→ K is an
embedding, and an extension L/F is algebraic, then ϕ extends to an embedding L −→ K.
Proof. Let’s start with the case L/F is finite. Take any α ∈ L \ F . The polynomial ϕ(mα,F ) splits in K;
let β ∈ K be a root of mα,F . ϕ the extends to an isomorphism F (α) −→ ϕ(F )(β) by ϕ(α) = β, and gives
an embedding F (α) −→ K. By induction on [L : F ], ϕ further extends to an embedding L −→ K.
Now consider the general case. Take the family of all embeddings ψ: N −→ K where N is a field
with F ⊆ N ⊆ L such that ψ |F = ϕ. Zorn’s lemma applies to this family and gives a maximal element
η: M −→ K. If M 6= L, take any α ∈ L \ M , and then η can be extended to an embedding M (α) −→ K,
which contradicts its maximality. Hence, M = L.
2.3.8. As a corollary, we obtain that every algebraic extension of F can be found in F .
Theorem. Every algebraic extension of a field F is isomorphic to a subextension of the algebraic closure F
of F .

10
2.3.9. Proof of uniqueness of the algebraic closure. Let K1 and K2 be two algeraic closures of F . By
Theorem 2.3.8, there is an embedding ϕ: K1 /F −→ K2 /F . Now, K2 is an alebraic extension of ϕ(K1 ), which
is isomorphic to K1 and is therefore algebraically closed; hence, K2 = ϕ(K1 ), thus ϕ is an isomorphism.
2.3.10. We now have:
Theorem. For any field F and any family F ⊆ F [x] of polynomials a splitting field of F exists and is
unique up to isomorphism.
Proof. Let K be the subfield of F generated by all the roots of all the polynomials from F; then K is the
splitting field of F. It is clearly the only splitting field of F contained in F . Now, any other splitting field
of F has a copy in F , which must be K. So, all splitting fields of F are isomorphic (as extensions of F ).
2.3.11. (i) The algebraic closure of R is C.
(ii) The algebraic closure of Q is not C (since C is not algebraic over Q), it is the field of all complex algebraic
numbers.
2.4. Separable and inseparable polynomials and extensions
2.4.1. Let F be a field. A nonconstant polynomial f ∈ F [x] is said to be separable if it has no multiple roots
in its splitting field (and so, in any extension of F ), and inseparable otherwise.
A polynomial f of degree n ≥ 1 is separable iff f has n distinct roots in its splitting field.
2.4.2. As we know, a root α of a polynomial f is a multiple root of f iff it is a root of the derivative f ′ of
f as well. Thus, a polynomial f is separable iff it has no common roots with its derivative f ′ , that is, iff f
and f ′ are coprime.
2.4.3. Let f be an irreducible polynomial over a field F . Then f and f ′ must be coprime, unless f ′ = 0.
This is impossible if char F = 0; thus, if char F = 0, every irreducible polynomial over K is separable.
But if char F = p 6= 0, this is possible: an irreducible polynomial f is inseparable iff it has form f (x) =
an xnp + an−1 x(n−1)p + · · · + a1 xp + a0 , that is, if f (x) = g(xp ) for some g ∈ F [x].
2.4.4. An element α algebraic over a field F is said to be separable over F if the minimal polynomial of α
is separable. α is separable iff it has exactly degF α conjugates over F (counting itself) in certain extension
of K (in the splitting field of its minimal polynomial).
2.4.5. An extension K/F is said to be separable if every α ∈ K is separable over F .
2.4.6. Non-separable extensions are said to be inseparable. An example of a inseparable extension is
Fp (t)/Fp (tp ): the polynomial xp − tp ∈ Fp (tp )[x] is irreducible and is the minimal polynomial of t ∈ Fp (t),
but is inseparable – it has a single root t of multiplicity p.
2.4.7. Theorem. If K/F is a separable extension, then for any subextension L/F of K/F , both L/F and
K/L are separable.
(The converse of this theorem is also true, but we cannot prove it yet.)
Proof. Any α ∈ L is also in K, and so is separable over F .
For any α ∈ K, we have mα,L mα,F , and since mα,F is separable, mα,L is separable too, so α is seprable
over L.
2.4.8. A field F is said to be perfect if every algebraic extension of F is separable.
2.4.9. Theorem.
√ Any field of characteristic zero is perfect. A field F of characteristic p is perfect iff for
every a ∈ F , p a ∈ F as well (that is, there exists b ∈ F such that bp = a).
Proof. In characteristic zero, every irreducible polynomial is separable.
√
Let char F = p. Assume that for every a ∈ F , p a ∈ F . Let f ∈ F [x] be an irreducible inseparable
polynomial, f (x) = an xnp + an−1 x(n−1)p + · · · + a1 xp + a0 . For each i, find bi ∈ F such that bpi = ai , and
put g(x) = bn xn + bn−1 xn−1 + · · · + b1 x + b0 . Then f (x) = bpn xnp + bpn−1 x(n−1)p + · · · + bp1 xp + bp0 = g(x)p ,
so f cannot be irreducible.
√
Converely, assume that there is a ∈ F such that p a is not in F . Then the polynomial f (x) = xp − a is
inseparable; let’s show it is irreducible. Adjoin a root α of f , that is, an element α such that αp = a; then
f (x) = xp − αp = (x − α)p . If f is reducible, let h ∈ F [x] be an irreducible factor of f . Then h(x) = (x − α)k

11
for some 1 ≤ k ≤ p − 1. Since α 6∈ F , we actually have 2 ≤ k ≤ p − 1; but then h is irreducible, has a
multiple root, and is not of the form h(x) = g(xp ) for some g ∈ F [x], contradiction.

2.4.10. Let F be a field of characteristic p. The mapping φ: F −→ F defined by φ(a) = ap is an endomor-

phism of F : for any a, b ∈ F , φ(a+b) = (a+b)p = ap +bp = φ(a)+φ(b), and φ(ab) = (ab)p = ap bp = φ(a)φ(b).
φ is called the Frobenius endomorphism of F .
2.4.11. By Theorem 2.4.9, F is a perfect field iff its Frobenius endomorphism is surjective, that is, is an
automorphism.
2.4.12. If F is a finite field, then its Frobenius endomorphism, being injective (as every (nonzero) field
homomorphism), is also surjective. So, every finite field is perfect.
2.4.13. The field F = Fp (t) of rational functions over the field Fp is non-perfect:
√ indeed, the polynomial
f (x) = xp − t ∈ F [x] is irreducible, but f ′ = 0. (f has a single root α = p t of multiplicity p: indeed,
f (x) = (x − α)p .)

3. Cyclotomic extensions and finite fields

3.1. Roots of unity and cyclotomic fields

Let F be a field.
3.1.1. For every n ∈ N, the elements a of F satisfying an = 1 are called the nth roots of unity or roots of
unity of degree n in F ; these are the roots of the polynomial xn − 1. The nth roots of unity which are not
dth roots of unity for d < n are called primitive nth roots of unity. Every root of unity of degree d n is an
nth root of unity, and every nth root of unity is a primitive dth roots of unity for some d n.
3.1.2. Lemma. Any finite subgroup of the multiplicative group of a field is cyclic.

Proof. Let G be a finite group of elements of a field F under multiplication, let |G| = m. Let k be the
senior invariant factor of G, so that ak = 1 for all a ∈ G. This means that the polynomial xk − 1 has ≥ m
roots in F , so k ≥ m. Hence, k = m and G has a unique cyclic component, that is, is cyclic.

Thus, the nth roots of unity form, under multiplication, a cyclic group. Since an = 1 for all elements of
this group, the order of the group divides n.
3.1.3. The splitting field of the polynomial xn − 1 ∈ F [x] is called the nth cyclotomic extension of F ; the
n-th cyclotomic extension of Q is called the nth cyclotomic field .
3.1.4. Let K be the nth cyclotomic extension of F , and let Gn be the group of roots of unity of degree n in
K. If char F = 0 or char F = p with p / n, then the polynomial xn − 1 is separable (it has no common roots
with its derivative nxn−1 ), so |Gn | = n.
If char F = p and n = pr m, (m, p) = 1, then the roots of unity of degree n are the roots of unity of
r r
degree m: ap m = 1 implies that φr (am ) = (am )p = 1 where φ is the Frobenius endomorphism, and so
am = 1. Hence, |Gn | = m.
3.1.5. From now on, let us assume that either char F = 0 or char F / n. Then xn − 1 is separable, |Gn | = n,
and the set of primitive nth roots of unity in K is the set of elements generating Gn . If ω is a primitive nth
root of unity, then the nth roots of unity are ω k , k = 0, 1, . . . , n − 1; thus, we have K = F (ω). The primitive
nth roots of unity are the elements ω k with k coprime with n; there are exactly ϕ(n) of them, where ϕ is
Euler’s totient function.
3.1.6. The nth roots of unity over Q are the complex numbers e2kπi/n , k = 0, 1, . . . , n − 1. ω = e2πi/n is a
primitive nth root of unity.
3.1.7. Let Pn be the set of primitive nth roots of unity over F (contained in the cyclotomic extension of
F ): Pn =Q ω k : 1 ≤ k ≤ n − 1, (k, n) = 1 , where ω is any primitive nth root of unity. The polynomial
Φn (x) = α∈Pn (x − α) is called the nth cyclotomic polynomial . Φn is monic, separable, and has degree
ϕ(n).

12
 Q
3.1.8. Theorem. For every n ∈ N, d|n Φd (x) = xn − 1.
Proof. We have Y Y Y Y
xn − 1 = (x − ω) = (x − α) = Φd (x).
ω:ω n =1 d|n α∈Pd d|n

3.1.9. Corollary. For every n ∈ N, the coefficients of Φn are in the prime subfield. In characteristic zero,
Φn ∈ Z[x] (has integer coefficients).
Proof. IfQby induction, for all d < n, Φd have their coefficients in the prime subfield, then so is Φn (x) =
(xn − 1)/ d|n Φd (x). If, in characteristic zero, the coefficients of Φd are in Z, then since they are all monic,
d<n
the coefficients of Φn are in Z by Gauss’s lemma.
Q
3.1.10. The formula Φn (x) = (xn − 1)/ d|n Φd (x) allows us to compute the cyclotomic polynomials
d<n
inductively. We have Φ1 (x) = x − 1, Φ2 (x) = (x2 − 1)/(x − 1) = x + 1, Φ3 (x) = (x3 − 1)/(x − 1) = x2 + x + 1,
Φ4 = (x4 − 1)/(x − 1)(x + 1) = x2 + 1.
Here are the initial cyclotomic polynomials:

Φ1 (x) = x − 1
Φ2 (x) = x + 1
Φ3 (x) = x2 + x + 1

Φ4 (x) = x2 + 1 = Φ2 (x2 )
Φ5 (x) = x4 + x3 + x2 + x + 1

Φ6 (x) = x2 − x + 1 = Φ3 (−x)
Φ7 (x) = x6 + · · · + x + 1

Φ8 (x) = x4 + 1 = Φ2 (x4 )


Φ9 (x) = x6 + x3 + 1 = Φ3 (x3 )


Φ10 (x) = x4 − x3 + x2 − x + 1 = Φ5 (−x)
Φ11 (x) = x10 + · · · + x + 1

Φ12 (x) = x4 − x2 + 1 = Φ6 (x2 )

and

Φ15 (x) = x8 − x7 + x5 − x4 + x3 − x + 1

Φ36 (x) = x12 − x6 + 1 = Φ6 (x6 ) = Φ3 (−x6 )
Φ105 (x) = 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

(Φ105 is the first cyclotomic polynomial that has a coefficient distinct from ±1 or 0. (Notice that 105 = 3·5·7.))
3.1.11. The proof of following facts are left as exercises:
(i) For any odd n ≥ 3, Φ2n (x) = Φn (−x).
r−1
(ii) For any prime p, Φp (x) = xp−1 + xp−2 + · · · + x + 1, and for any r ∈ N, Φpr (x) = Φp (xp ).
r pr−1
(iii) More generally, if p is prime and n = mp with p / m, then Φn (x) = Φpm (x ).
3.1.12. Theorem. For every n ∈ N, Φn is irreducible in Q[x]. Thus, in characteristic zero, all primitive
nth roots of unity are conjugate over Q.
Proof. By Gauss’s lemma, we only need to show that Φn is irreducible in Z[x]. Assume that it is reducible,
let Φn = f g where f, g ∈ Z[x] are nonconstant and monic. Let ω be a primitive root of unity of degree n,
then all roots of Φn have form ω k for some k with (k, n) = 1; some of them are roots of f and the other
are roots of g. There must be k and a prime p, both coprime with n, such that α = ω k is a root of f and
αp = ω kp is a root of g. Then α is a common root of f and g(xp ), so f (x) and g(xp ) are not coprime, and
have a common factor. Let f˜ = f mod p ∈ Fp [x] and g̃ = g mod p ∈ Fp [x], then f˜(x) and g̃(xp ) = g̃(x)p have
a common factor, so f˜ and g̃ have a common factor, and so Φ̃n = f˜g̃ is inseparable, contradiction.

13
3.1.13. So, in characteristic zero, for any n, Φn is the minimal polynomial of every primitive nth root of
unity. We therefore have:
Corollary. For every n ∈ N, the nth cyclotomic field has degree ϕ(n) over Q.

3.2. Finite fields

3.2.1. Any finite field K has pn elements, where p = char K and n = [K : Fp ].
3.2.2. Let K be a field of order pn . Then the multiplicative group K ∗ of K has pn − 1 elements, so for any
n n
nonzero α ∈ K we have αp −1 = 1, so for all α ∈ K we have αp = α. Hence, all pn elements of K are roots
n
of the polynomial xp − x, and K is the splitting field of this polynomial.
Conversely, given a prime p and a positive integer n, let K be the splitting field of the polynomial
n
f (x) = xp − x ∈ Fp [x]. Let S be the set of the roots of f in K. Since f ′ 6= 0, f is separable, so |S| = pn .
n n n n n
Next, S is a field: if α, β ∈ S, that is, αp = α and β p = β, then (α ± β)p = αp ± β p = α ± β,
n n n
(αβ)p = αp β p = αβ, and (α−1 )n = α−1 , so f (α + β) = f (αβ) = f (α−1 ) = 0. Hence, K = S, and
|K| = pn .
Since the splitting field of any polynomial exists and is unique up to isomorphism, we get:
Theorem. For every prime p and every n ∈ N there exists a unique, up to isomorphism, field of order pn ;
n
it is the splitting field of the polynomial f (x) = xp − x ∈ Fp [x], and consists of the roots of this polynomial.
The (unique up to isomorphism) field of cardinality q = pn is denoted by Fq .
3.2.3. Theorem. For any prime p and every n ∈ N,
(i) the field Fpn is a simple extension of its prime subfield Fp ;
(ii) there exists an irreducible polynomial of degree n in Fp [x].
Proof. The group F∗pn of nonzero elements of Fpn under multiplication is cyclic; let α be any generator of
this group. Then the powers of α run over the set of all nonzero elements of Fpn , so α generates this field,
Fpn = Fp (α). It also follows that degFp α = [Fpn : Fp ] = n, so mα,Fp ∈ Fp [x] is an irreducible polynomial of
degree n.
d n
3.2.4. Lemma. If d, n ∈ N and d n, then (xd − 1) (xn − 1) and for any r ∈ N, (xr −1 − 1) (xr −1 − 1).


Proof. xn − 1 = (xd )n/d − 1 = (xd − 1) (xd )n/d−1 + · · · + xd + 1 , so (xd − 1) (xn − 1). It follows that for
d n
any r ∈ N, (rd − 1) (rn − 1), so now (xr −1 − 1) (xr −1 − 1).
3.2.5. Theorem. For any prime p and n ∈ N, the field Fpn contains a single copy of the field Fpd for each
d dividing n, and has no other subfields.
It follows that the diagram of subextensions of Fpn looks exactly like the diagram of subgroups of Zn .
Proof. Let L be a subfield of Fpn of degree d over Fp . Then d divides n = Fpn : Fp . Thus, |L| = pd and
L∼
d
= Fpd . Hence, all elements α ∈ L are roots of the polynomial xp − x; since there are at most pd such
roots in Fpn , there may be at most one such subfield L of Fpn .
d n d
On the other hand, for every d n the polynomial xp −1 − 1 divides xp −1
− 1, thus all roots of xp −1
− 1,
which are just all nonzero elements of Fpd , are contained in Fpn .
3.2.6. The following theorem allows to find inductively the number of irreducible polynomials of degree n
in Fp [x].
P
Theorem. If ψ(n) is the number of monic irreducible polynomials of degree n in Fp [x], then d|n dψ(d) =
pn .
Proof. For every d dividing n let Pd be the set of monic irreducible polynomials from Fp [x] of degree d.
Every element of Fpn is a root of the polynomial f = mα,Fp from Pd for some d n; on the other hand, for
Q Q
every d n, every f ∈ Pd is separable and splits completely in Fpd and so in Fpn . So, d|n f ∈Pd f (x) =
Q pn
P
α∈Fpn (x − α) = x − x, and d|n dψ(d) = pn .

14
3.2.7. For any prime p, the fileds Fpn! , n ∈ N, form a nested sequence, Fp ⊆ Fp2! ⊆ Fp3! ⊆ Fp4! ⊆ · · ·.
S∞
The union of this sequence, Fp = n=1 Fpn! , is an algebraic extension of Fp that contains all roots of all
irreducible polynomials from Fp ; hence, it is the algebraic closure of Fp .

4. Galois extensions and the Galois theorem

4.1. Embeddings of an extension and conjugate subextensions

This section may be cumbersome, but it is a key to the Galois theory.
4.1.1. Let K/F and E/F be two extensions. A (nonzero, of course) homomorphism K/F −→ E/F (that is,
a homomorphism K −→ E which is identical on F ) is called an embedding of K/F to E/F , or an embedding
of K into E over F .
4.1.2. An isomorphism K/F −→ K/F is called an automorphism of K/F , or an automorphism of K over
F . The automorphisms of an extension K/F form a group, denoted by Aut(K/F ).
4.1.3. Let ϕ be an embedding of an extension K/F into an extension E/F . Then for any polynomial
f ∈ F [x], ϕ(f ) = f (since ϕ preserves the coefficients of f ). So, for any root α of f in K, ϕ(α) is a root of
f in E: f (ϕ(α)) = ϕ(f )(ϕ(α)) = ϕ(f (α)) = ϕ(0) = 0.
In particular, if K/F is a subextension of E/F , then any embedding of K/F into E/F maps every
element α ∈ K algebraic over F to a root of the minimal polynomial of α, that is, to a conjugate to α over
F.
4.1.4. If K/F is an algebraic extension, then any embedding ϕ: K/F −→ K/F is an automorphism of K/F .
Indeed, if K/F is finite, then ϕ must be surjective, so is an automorphism. In the general case, to show that
ϕ is surjective, let α ∈ K, and let L be the subfiled of K generated by the conjugates of α in K. Since ϕ
permutes the roots of mα,F , we have ϕ(L) ⊆ L, and since L/F is finite, ϕ(L) = L. So, α ∈ ϕ(L) ⊆ ϕ(K).
4.1.5. Let E/F be an extension, and let α be an element algebraic over F . The set of embeddings F (α)/F −→
E/F is in one-to-one correspondence with the set of roots of the minimal polynomial mα,F of α in E: each
embedding ϕ: F (α)/F −→ E/F is defined by ϕ(α), which must be a root of mα,F . There are at most degF α
embeddings of F (α)/F into E/F ; there are exactly degF α embeddings of F (α)/F into E/F iff mα,F is
separable and completely splits in E.
4.1.6. We will need a generalization of 4.1.5: Let ϕ: F1 −→ E be a homomorphism of fields, let F2 = ϕ(F1 ),
let α be an element algebraic over F1 , let f1 = mα,F1 and let f2 = ϕ(f1 ) ∈ F2 [x]. For any homomorphism
ψ: F1 (α) −→ E extending ϕ (that is, with ψ |F1 = ϕ), we have 0 = ψ(f1 (α)) = ϕ(f1 )(ψ(α)) = f2 (ψ(α)), so ψ
maps α to a root α′ of f2 , and is defined by α′ . Thus, the set of homomorphisms ψ: F1 (α) −→ E extending ϕ
is in one-to-one correspondence with the set of roots of f2 in E. There are at most deg f2 = deg f1 = degF1 α
such homomorphisms of F1 to E; there are exacly degF1 α such homomorphisms iff f2 is separable and
completely splits in E.
Note also that if ϕ is a homomorphism over a subfield F of F1 and F2 (that is, with ϕ|F = IdF ) and
f = mα,F , then f1 f , and so f2 = ϕ(f1 ) ϕ(f ) = f . So, if f completely splits in E, then f2 has roots in E,
and so, there is an embedding ψ: F1 (α) −→ E extending ϕ.
4.1.7. Theorem. Let E/F be an extension, let K/F be a finite extension, let n = [K : F ]. Then there are at
most n embeddings K/F −→ E/F . There are exactly n embeddings K/F −→ E/F iff K/F is separable and
for every α ∈ K, mα,F completely splits in E. For this, it suffices if there is a set of generators {α1 , . . . , αk }
of K/F such that for each i, mαi ,F is separable and completely splits in E.
Proof. Represent K as a tower of simple extensions, K = Lk /Lk−1 / · · · /L1 /L0 = F , where for each
i, Li = Li−1 (αi ) for some αi ∈ K. Then n = degF α1 · degL1 α2 · · · degLk−1 αk . By 4.1.5, there are
≤ degF α1 embeddings L1 /F −→ E/F ; by 4.1.6, each such embedding has ≤ degF α2 extensions to a
homomorphism L2 −→ E; etc., with the total number of embeddings K/F −→ E/F being ≤ degF α1 ·
degL1 α2 · · · degLk−1 αk = n.
Assume that K = F (α1 , . . . , αk ) where for each i, αi has exactly degF αi conjugates in E, and let
L0 = F and Li = Li−1 (αi ), i = 1, . . . , k. Then there are exactly degF α1 embeddings L1 /F −→ E/F . Let
ϕ be such an embedding. The minimal polynomial mα2 ,L1 is an irreducible divisor of mα2 ,F , and since ϕ

15
fixes F , ϕ(mα2 ,L1 ) is also an irreducible divisor of mα2 ,F , so it is separable and completely splits in E. Thus
by 4.1.6, ϕ has degL1 α2 extensions to a homomorphism L2 −→ E. And so on, with the total number of
embeddings K/F −→ E/F being equal to degF α1 · degL1 α2 · · · degLk−1 αk = n.
If there is an element α ∈ K for which mα,F has less than degF α roots in E, then in the argument
above, put α = α1 . We will then get that the total number of embeddings K/F −→ E/F is less than n.
4.1.8. If K/F is a subextension of E/F and ϕ: K/F −→ E/F is an embedding, then the extension ϕ(K)/F
is said to be conjugate to K/F . By Theorem 4.1.7, a subextension of degree n may have at most n conjugates
in an extension E/F .

4.2. Normal extensions

In the textbook the term “normal extension” is not introduced, replaced by “a splitting field”. (Indeed,
we will see that these two classes of extensions coincide.) But the notion of a normal extension is commonly
used, is advantageous, and is closely related to the notion of a normal subgroup, so I prefer to use it.
4.2.1. An algebraic extension K/F is said to be normal if for any α ∈ K “all conjugates of α are in K”,
that is, the minimal polynomial of α over F completely splits in K. Equivalently, K/F is normal if any
irreducible polynomial from F [x] that has a root in K completely splits in K, that is, K is the splitting field
of the minimal polynomials of all its elements.
K
In diagrams, the normality of an extension is indicated by a double line:
√
√ √
F √

4.2.2. The extensions Q 2 /Q and Q 2, 3 /Q are normal, the extension Q 3 2 /Q is not.
4.2.3. Every quadratic extension is clearly normal.
4.2.4. The following theorem is easy:
Theorem. (i) If L/F is a subextension of a normal extension K/F , then K/L is also normal.
(ii) If L1 and L2 are normal subextensions of an extension K/F , then the intersection (L1 ∩ L2 )/F is also
normal.
(Notice that in (i), the subextension L/F does not have to be normal!)
Proof. (i) For any α ∈ K, mα,L mα,F and mα,F splits completely in K, so mα,L also splits completely.
(ii) For every α ∈ L1 ∩ L2 , mα,F splits completely and all its roots are contained in both L1 and L2 , so in
L1 ∩ L2 .
4.2.5. Theorem. If an algebraic extension K/F is normal then for every extension E/K and every embed-
ding ϕ: K/F −→ E/F one has ϕ(K) ⊆ K (and so, ϕ is an automorphism of K/F ). Conversely, if there is
an extension E/K such that E/F is normal and for every embedding ϕ: K/F −→ E/F one has ϕ(K) ⊆ K,
then K/F is normal.
Proof. Let K/F be normal, E/K be an extension, and ϕ: K/F −→ E/F be an embedding. For any α ∈ K,
ϕ(α) is conjugate to α in E, and thus is contained in K. So, ϕ(K) ⊆ K.
In the other direction, assume that K/F is not normal, and let α ∈ K be such that mα,F does not
split completely in K. Let K be the algebraic closure of K, and let β ∈ K be a root of mα,F which is not
in K. There is an isomorphism ϕ: F (α) −→ F (β) that fixes F and maps α to β, and we can extend ϕ to
an embedding K −→ K. Since ϕ(α) = β 6∈ K, we have ϕ(K) 6⊆ K. Now, if E is an extension of K such
that E/F is normal, we may assume that E ⊆ K. Then ϕ(E) = E, so ϕ can be seen as an embedding
K/F −→ E/F with ϕ(K) 6⊆ K.
4.2.6. Theorem. Assume that an algebraic extension K/F is generated by a set S such that for every
α ∈ S, all conjugates of α over F are in K. (That is, the minimal polynomial of α over F splits in K.)
Then K/F is normal. In particular, the splitting field of any family F ⊆ F [x] is a normal extension of F .
Proof. For any extension E/K, any embedding ϕ: K/F −→ E/F maps all elements of S to their conjugates,
which are in K by assumption. Since S generates K, this implies that ϕ(K) ⊆ K.
4.2.7. As a corollary, we get:

16
Theorem. If L1 and L2 are normal subextensions of an extension K/F , then their composite (L1 L2 )/F is
also normal.
4.2.8. Theorem. For any algebraic extension K/F there exists a normal extension E/F containing K such
that no proper subextension of E/F containing K is normal. If K/F is finite, then E/F is also finite.
This extension E/F is called the normal closure of K/F .
Proof. E is just the splitting field of the set of the minimal polynomials of any set of generators of K over
F.
4.2.9. The following is an important property of normal extensions:
Theorem. Let K/F be a normal extension and let L/F be its subextension. Then every embedding L/F −→
K/F extends to an automorphism of K/F .
Proof. Let E be the algebraic closure of K. Any embedding ϕ: L/F −→ K/F , and so L/F −→ E/F ,
extends to an embedding ϕ: K/F −→ E/F . Since K/F is normal, ϕ(K) = K.

4.3. Galois extensions and Galois groups

4.3.1. Here is the central definition of the course: A finite normal separable extension is called a Galois
extension.
4.3.2. We have:
Theorem. A finite extension K/F is Galois iff | Aut(K/F )| = [K : F ].
Indeed, elements of Aut(K/F ) are just embeddings K/F −→ K/F , and we have exactly [K : F ] such
embeddings iff K/F is separable and normal.
4.3.3. If K/F is a Galois extension, then the group Aut(K/F ) is called the Galois group of K/F , and is
denoted by Gal(K/F ). By Theorem 4.3.2, Gal(K/F ) is a finite group of order [K : F ].
A Galois extension is called cyclic, abelian, nilpotent, or solvable, if its Galois group is cyclic, abelian,
nilpotent, or solvable respectfully.
4.3.4. The action of every element of the Galois group G = Gal(K/F ) of a Galois extension K/F is defined
by its action on the generators of K/F , which are mapped to some their conjugates. Thus G can be seen as
a subgroup of the group of permutations of a finite set of generators and their conjugates.
4.3.5. The action of the Galois group G = Gal(K/F ) of a Galois extension K/F on any set of elements of
K conjugate over F is transitive: indeed, if α, α′ ∈ K are conjugate over F , then, by 4.2.9, the isomorphism
F (α)/F −→ F (α′ )/F that maps α to α′ extends to an automorphism of K/F , that is, to an element of G.
4.3.6. For an extension to be Galois, it suffices if it is “Galois on generators”:
Theorem. (i) A finite extension K/F is Galois iff it is generated by elements separable over F whose all
conjugates over F are contained in K.
(ii) An extension K/F is Galois iff K is a splitting field of a separable polynomial from F [x].
4.3.7. If K is the splitting field of a separable polynomial f ∈ F [x], then the Galois group Gal(K/F ) is also
called the Galois group of f , and is denoted by Gal(f /F ) or just Gal(f ). Via its action on the roots of f ,
the group Gal(f ) is (isomorphic to) a subgroup of Sn for n = deg f .
If f is irreducible over F , then, by 4.3.5, Gal(f /F ) acts transitively on the set of the roots of f .
4.3.8. From the definition, the criteria above, and properties of normal and separable extensions we have:
Theorem. (i) If L/F is a subextension of a Galois extension K/F , then K/L is also Galois.
(ii) If L1 and L2 are Galois subextensions of an extension K/F , then their intersection (L1 ∩ L2 )/F is also
Galois.
(iii) If L1 and L2 are Galois subextensions of an extension K/F , then their composite (L1 L2 )/F is also
Galois.

17
4.3.9. If K/F is a finite separable extension, then its normal closure is a Galois extension: indeed, the
normal closure of K/F is generated by conjugates of separable elements, which all are also separable. It is
called the Galois closure of K/F .
The Galois closure of K/F is generated by the conjugates of K over F .

4.4. Composites and towers of separable extensions

Taking the normal closure of a finite separable extension converts it into a Galois extension; we may
now use this to obtain the properties of separable extensions that we were not able to prove before.
4.4.1. Theorem. If an algebraic extension K/F is generated by a set of elements separable over F , then
K/F is separable.
Proof. Let α ∈ K; we need to show that α is separable over F , and for this end we may replace K by
a subfield generated by finitely many of the (separable) generators of K/F , and thus assume that K/F is
finite. Let E/F be the Galois closure of K/F ; then E/F is separable, so K/F is separable.
4.4.2. Corollary. If L1 /F and L2 /F are separable subextensions of an extension K/F , then their composite
(L1 L2 )/F is also separable.
4.4.3. Theorem. If K/L and L/F are separable extensions, then K/F is separable.
Proof. Let α ∈ K; we have to prove that α is separable over F , so, we may assume that K = L(α). After
replacing L by the field generated by the coefficients of the polynomial p = mα,L , we may assume that L/F
is finite. Let n = [L : F ] and m = degL α = deg p. Let E/F be the normal closure of K/F . Then there are n
embeddings ϕ: L/F −→ E/F , and every such embedding can be extended to an embedding K/F −→ E/F
by mapping α to a root of ϕ(p). Since p is separable, the polynomial ϕ(p) is also separable; thus there
are m extensions of ϕ to an embedding K/F −→ E/F . So, totally we have nm = [K : F ] embeddings
K/F −→ E/F , which, by Theorem 4.1.7, implies that K/F is separable.
4.4.4. Let K/F be an algebraic extension, let L/F, L1 /F, L2 /F be subextensions of K/F . We have the
following:

finite separable normal Galois

If K is generated by “good” elements then K/F is “good” − + + −
If K is generated by finitely many “good” elements then K/F is “good” + + + +
If L1 /F and L2 /F are “good” then (L1 ∩ L2 )/F is “good” + + + +
If L1 /F and L2 /F are “good” then (L1 L2 )/F is “good” + + + +
If K/F is “good” then L/F is “good” + + − −
If K/F is “good” then K/L is “good” + + + +
If L/F and K/L are “good” then K/F is “good” + + − −

(Dependently on the column, an element α is assumed to be “good” if, respectively, degF α < ∞ (α is
algebraic, which is automatic since K/F is algebraic); α is separable; mα,F splits completely in K; and mα,F
is separable and splits completely in K. An extension is “good” if it is finite, separable, normal, and Galois
respectively.)

4.5. Examples of Galois groups

2
4.5.1. The Galois group√ of the
√ polynomial f (x) = x − 2 is isomorphic to Z2 : the only nonidentical element
of this group maps 2 7→ − 2. (Clearly, the Galois group of any separable quadratic extension is Z2 .)
4.5.2. The Galois group G of the polynomial f (x) = (x2 − 2)(x2 − 3) √ over Q√is isomorphic to V4 = Z2 .
2

Namely, G = {1, ϕ1 , ϕ2 , ϕ3 }, where the action of ϕi on the elements 2 and 3, generating the splitting
field of f , is given by
√ √ √ √ √ √
2 7→ − 2 2 7→ 2 2 7→ − 2
ϕ1 : √ √ , ϕ2 : √ √ , ϕ3 : √ √ .
3 7→ 3 3 7→ − 3 3 7→ − 3

(Clearly, the Galois group of any biquadratic extension in char 6= 2 is V4 .)

18
 √
4.5.3. Let ω = e2πi/3 and α = 3 2. The Galois group G of the polynomial f (x) = x3 − 2 over Q has order 6.
It acts as a group of all permutations of the roots α1 = α, α2 = ωα, α3 = ω 2 α of f , and so, is isomorphic to
S3 : G = {1, σ, σ 2 , τ1 , τ2 , τ3 } where σ =
√ (α1 , α2 , α3 ), τ1 = (α2 , α3 ), τ2 = (α1 , α3 ), τ3 = (α1 , α2 ). The action
of G on the elements ω = e2πi/3 and 3 2, generating the splitting field of f , is given by

ω 7→ ω ω 7→ ω 2 ω 7→ ω 2 ω 7→ ω 2
σ: , τ1 : , τ2 : , τ3 : .
α 7→ ωα α 7→ α α 7→ ω 2 α α 7→ ωα

4.5.4. Let ω be a primitive nth root of unity over Q; say, ω = e2πi/n . The Galois group G of the cyclotomic
extension Q(ω)/Q (and of the nth cyclotomic polynomials Φn ) has order ϕ(n), where ϕ is Euler’s totient
function.
 Every element of G is uniquely defined by its action on ω, and maps ω to ω k for some k ∈ Z∗n , thus
G = ηk : k ∈ Z∗n , where ηk (ω) = ω k . For any k, l ∈ Z∗n we have ηk (ηl (ω)) = ω kl , so ηk ηl = ηkl . Hence, G
is isomorphic to Z∗n .
4.5.5. Let L be the nth cyclotomic extension of Q, L = Q(ω) √ where ω = e2πi/n . Then the splitting field of
n n
the polynomial f = x − 2 over L is K = L(α) where α = 2. Let us assume that degL (α) = n (that is,
that xn − 2 is irreducible in L[x]), then the Galois group G = Gal(f ) = Gal(K/L) has order n. Since G acts
transitively on the roots of f , there exists σ ∈ G such that σ(α) = ωα. Then for any k, σ k (α) = ω k α, thus
σ has order n. So, G is cyclic, isomorphic to Zn , generated by σ.
4.5.6.√ Let K ⊂ C be the splitting field of the polynomial f = x8 − 2 ∈ Q[x],
√ then K = (α, ω) where α is the
real 2 and ω = e2πi/8 = 1+i
8
√ . Let G = Gal(f ) = Gal(K/Q). Notice that
2
2 is contained in both Q(α) and
√ 4
√ 7
Q(ω): 2 = α and 2 = ω + ω . Since ω 6∈ Q(α) (ω is not real), [K : Q(α)] = 2. We have the following
diagram:
2
K =4 Q(α, ω)
Q(α) Q(ω)
4 2
Q(α) ∩ Q(ω)
2
Q.
So, [K : Q] = 16. √
1+i
It is more convenient to use i = ω 2 as a generator instead of ω: since ω = √
2
and 2 ∈ Q(α), we have
ω ∈ Q(α, i), and so, Q(α, i) = K.
2
K =8 Q(α, i)
Q(α) Q(i)
8 2
Q.
Let us find the multiplication in G explicitly. α has 8 conjugates αω k , k = 0, . . . , 7, and i has two
conjugates ±i, so the total number of choices we have where to send α and i is 16; since |G| = [K : Q] = 16
as well, any choice of the image of α and, independently, of i gives rise to
√ an element of G. Define √ ϕ∈G
by ϕ(α) = αω, ϕ(i) = i, and ψ ∈ G by ψ(α) = α, ψ(i) = −i. Then ϕ( 2) = ϕ(α4 ) = α4 ω 4 = − 2, so
ϕ(ω) = −ω = ω 5 . So, under the action of ϕ, we have

α 7→ αω 7→ αω 6 7→ αω 7 7→ αω 4 7→ αω 5 7→ αω 2 7→ αω 3 7→ α and i 7→ i,
√
which√means that the order of ϕ in G is 8. The order of ψ is clearly equal to 2. Next, ψ( 2) = ψ(α4 ) =
α4 = 2, so ψ(ω) = 1−i
√ = ω 7 . Hence,
2

(ψϕψ −1 )(α) = (ψϕ)(α) = ψ(αω) = αω 7

and (ψϕψ −1 )(i) = i, so ψϕψ −1 = ϕ3 . Hence, G = ϕ, ψ ϕ8 = ψ 2 = 1, ψϕψ −1 = ϕ3 . This is the

semidihedral (or the quasidihedral) group SD16 , a semidirect product Z8 × Z2 . (There are no other relations
in G since |G| = 16.)

19
4.5.7. Let F be a field with char F 6= 2 and let f ∈ F q[x] be an√irreducible biquadratic
q polynomial, f =
4 2 1

1
√

2
x + ax + b. The roots of f are ±α, ±β where α = 2 −a + a − 4b and β = 2 −a − a − 4b , 2
√
√ = b. Since f is irreducible,√
with αβ degF α = degF β = 4. Let K be the splitting field of f , K = F (α, β) =
F (α, b), Since f is irreducible, δ = a2 − 4b 6∈ F , and so [F (δ) : F ] = 2. We therefore have the extensions
diagram
K = F (α, β)
x x

F (α) F (β)
2 2
F (δ)
2
F,
where x = 1 or 2, so either [K : F ] = 4 or [K : F ] = 8.
Let G = Gal(f ) = Gal(K/F ); then G is a subgroup of S4 of order 4 or 8 that acts transitively on the
set R = {α, −α, β, −β}. Any element ϕ ∈ G acts on the square
α— β

−β — −α
symmetrically with respect to the center of the square: if ϕ(λ) = γ, where λ, γ ∈ {±α, ±β}, then ϕ(−λ) = −γ
and ϕ(γ) = ±λ. So, ϕ preserves the square, and G is a subgroup of the dihedral group D8 . Since G acts on
R transitively, it is isomorphic to one of the groups V4 , Z4 , of D8 .
Assume that |G| = 4; this is the case iff F (α)√= F (β). Since F (α) is a quadratic extension of F (δ) and
both α2 , β 2 ∈ F (δ), we have that β ∈ F (δ)α, so b = αβ ∈ F (δ)α2 = F (δ); since also F (δ) is a quadratic
√ 2 √ √
extension of F and δ 2 , b ∈ F , either b ∈ F or b ∈ F δ. Let ϕ ∈ G be such that ϕ(α) = β. We
have two options: ϕ(β) = α or ϕ(β) = −α; in the first case ϕ is a reflection and G = {1, ϕ, ψ, ϕψ} ∼ = V4
2 3 ∼
where
√ ψ : α ↔ −β; in the second
√ case
√ ϕ is a
√ rotation
√ by π/2 and G = {1, ϕ, ϕ ϕ } = Z 4 . In the√first case,
ϕ( b) = ϕ(αβ) = −βα √ = √b and ψ(
√ b) = b, so b has no conjugates over F except itself, so b ∈ F ; in
the second case, ϕ(√ b) = − b, so b 6∈ F .√ √ √
We obtain: if b ∈ F then G ∼ = V4 ; if b/δ ∈ F , then G ∼ = Z4 ; if both
b, b/δ 6∈ F , then G ∼ = D8 .
4.5.8. Let K be a finite field, K = Fpn . The Galois group G = Gal Fpn /Fp has order [K : Fp ] = n. The
Frobenius automorphism φ of K fixes Fp , thus φ ∈ G. I claim that the order |φ| of φ is n; this implies that
G is cyclic, isomorphic to Zn , generated by φ. Indeed, the multiplictive group of K has a generator α, so
m
that the minimal k for which αk = 1 is k = pn − 1. Thus the minimal m for which φm (α) = αp = α is
m = n, so |φ| = n.

4.6. The fundamental Galois theorem

4.6.1. If K/F be a Galois extension, then for any subextension L/F of K/F , the extension K/L is also Galois,
and Gal(K/L) ≤ Gal(K/F ). We therefore have a mapping L 7→ Gal(K/L) from the set of subextensions
L/F of K/F to the set of subgroups H of G.
4.6.2. Let K be a field a let H be a group of automorphisms of K. An element α ∈ K is said to be fixed by
H if ϕ(α) = α for all ϕ ∈ H; a set S ⊆ K is said to be fixed by H is all elements of S are fixed by H. By
Fix(H) we denote the set of all elements of K fixed by H; this is a subfield of K, called the subfield of K
fixed by H.
If K/F is an extension and H ≤ Aut(K/F ), then Fix(H) is an extension of F . We therefore have a
mapping H 7→ Fix(H) from the set of subgroups H of G to the set of subextensions L/F of K/F .
4.6.3. The fundamental Galois theorem – short version. Let K/F be a Galois extension and let
G = Gal(K/F ). Then the mappings L 7→ Gal(K/L) and H 7→ Fix(H) are inverses of each other, and define
a one-to-one correspondence between the set of subextensions L/F of K/F and the set of subgroups H of G.
4.6.4. The proof of the Galois theorem is based on the following proposition:
Proposition. Let K be a field, let G be a finite group of automorphisms of K, and let F = Fix(G). Then
[K : F ] = |G|.

20
(It follows that the extension K/F is Galois.)
Proof. Let |G| = n and [K : F ] = m; let G = {ϕ1 , . . . , ϕn } and let {α1 , . . . , αm } be a basis of K over F .
(1) Assume that n > m. Consider the following system of m linear equations over K in n variables:

 ϕ1 (α1 )x1 + · · · + ϕn (α1 )xn = 0

..
 .

ϕ1 (αm )x1 + · · · + ϕn (αm )xn = 0.

Since n > m, this system has a nontrivial solution: there are β1 , . . . , βn ∈ K, not all zero, such that

 ϕ1 (α1 )β1 + · · · + ϕn (α1 )βn = 0

..
 .

ϕ1 (αm )β1 + · · · + ϕn (αm )βn = 0.

For any (a1 , . . . , am ) ∈ F m , adding these equalities with coefficients ai we get

Pm
Pm

ϕ1 i=1 a i α i β 1 + · · · + ϕn i=1 ai αi βn = 0.
Pm
But any element α of K is representable in the form α = i=1 ai αi for some a1 , . . . , am ∈ F ; so, we have
ϕ1 (α)β1 + · · · + ϕn (α)βn = 0 for all α ∈ K. (That is, we get that ϕi are linearly dependent over K,
β1 ϕ1 + · · · + βn ϕn = 0.) Choose a minimal such zero linear combination, in the sense of the number of
nonzero summands; w.l.o.g. we may assume that this is

ϕ1 (α)β1 + · · · + ϕr (α)βr = 0 (A)

for all α ∈ K, with β1 , . . . , βr 6= 0. (Notice that, clearly, r ≥ 2.)

Now find α0 ∈ K for which ϕ1 (α0 ) 6= ϕ2 (α0 ). For every α ∈ K we have

ϕ1 (α0 α)β1 + · · · + ϕr (α0 α)βr = ϕ1 (α0 )ϕ1 (α)β1 + · · · + ϕr (α0 )ϕr (α)βr = 0. (B)

Subtracting ϕ1 (α0 )(B) from (A) we get

ϕ2 (α0 ) − ϕ1 (α0 ) ϕ2 (α)β1 + · · · + ϕr (α0 ) − ϕ1 (α0 ) ϕr (α)βr = 0

for all α ∈ K, which is a nontrivial zero linear combination of ϕi having less than r nonzero summands,
contradiction.
(2) Now assume that m > n. Consider the following system of n linear equations over K in m variables:

 ϕ1 (α1 )x1 + · · · + ϕ1 (αm )xm = 0

..
 .

ϕn (α1 )x1 + · · · + ϕn (αm )xm = 0.

Since m > n, this system has a nontrivial solution: there are β1 , . . . , βm ∈ K, not all zero, such that

 ϕ1 (α1 )β1 + · · · + ϕ1 (αm )βm = 0

..
 .

ϕn (α1 )β1 + · · · + ϕn (αm )βm = 0.

This means that for every ϕ ∈ G, ϕ(α1 )β1 + · · · + ϕ(αm )βm = 0. Choose a minimal such zero linear
combination, in the sense of the number of nonzero summands; w.l.o.g. we may assume that this is

ϕ(α1 )β1 + · · · + ϕ(αr )βr = 0 (C)

21
for all ϕ ∈ G, with β1 , . . . , βr 6= 0. (Notice that, clearly, r ≥ 2.)
After dividing (C) by β1 we may assume that β1 = 1 ∈ F , so ψ(β1 ) = β1 for all ψ ∈ G. It cannot be
that all βi are in F , since we have α1 β1 + · · · + αr βr = 0 (using ϕ = 1), and αi are linearly independent over
F ; w.l.o.g. assume that β2 6∈ F . Find ψ ∈ G such that ψ(β2 ) 6= β2 . Applying ψ to (C), we get


ψ ϕ(α1 )β1 + · · · + ϕ(αr )βr = ψ(ϕ(α1 ))ψ(β1 ) + · · · + ψ(ϕ(αr ))ψ(βr ) = 0.
Since the products ψϕ for ϕ ∈ G run over all elements of G, we have that
ϕ(α1 )ψ(β1 ) + · · · + ϕ(αr )ψ(βr ) = 0 (D)
for all ϕ ∈ G. Subtracting (D) from (C) (and recalling that ψ(β1 ) = β1 ) we get



ϕ(α2 ) β2 − ψ(β2 ) + · · · + ϕ(αr ) βr − ψ(βr ) = 0
for all ϕ ∈ G, which is a nontrivial zero linear combination having less than r nonzero summands, contra-
diction.
4.6.5. Proof of the Galois theorem. Let L/F be a subextension of K/F , let H = Gal(K/L), and let
e = Fix(H). Since H fixes L we have L ⊆ L.
L e Let [K : L] = n, then |H| = n, and by Proposition 4.6.4,
e e
[K : L] = n; so, L = L.
Now let H be a subgroup of G, let L = Fix(H), and let H e = Gal(K/L). Since H fixes L, we have
H ≤ H.e Let |H| = n, then by Proposition 4.6.4, [K : L] = n, and |H|
e = n since K/L is Galois; so, H
e = H.

4.6.6. The fundamental Galois theorem – full version. Let K/F be a Galois extension and let
G = Gal(K/F ). Let L, L1 and L2 be subextensions of K/F and let H, H1 and H2 be the corresponding
subgroups of G (under the bijection L 7→ Gal(K/L)). Then
(i) |H| = [K : L] and |G : H| = [L : F ].
(ii) L1 ⊆ L2 iff H1 ≥ H2 , and in this case, [L2 : L1 ] = |H1 : H2 |. So, the diagram of subextensions of K/L
is isomorphic to the diagram of subgroups of G flipped upside down.
(iii) The subgroup H1 ∩ H2 corresponds to the composite L1 L2 and the subgroup hH1 , H2 i corresponds to the
intersection L1 ∩ L2 .
(iv) Every embedding of L/F into K/F is defined by an element of G; the set of embeddings of L/F into
K/F is in a one-to-one correspondence with the set G/H of left cosets of H in G.
(v) For any ϕ ∈ G, the subgroup of G corresponding to the conjugate ϕ(L) of L is the conjugate ϕHϕ−1 of
H. The number of conjugates of L/F in K/F equals |G : NG (H)|, where NG (H) is the normalizer of H in
G.
(vi) H is a normal subgroup of G iff L/F is a normal extension. In this case, L/F is Galois, the mapping
ϕ 7→ ϕ|L defines a homomorphism Gal(K/F ) −→ Gal(L/F ) and induces an isomorphism G/H ∼ = Gal(L/F ).
Proof. (i) Since H = Gal(K/L), we have |H| = [K : L]. Now, |G : H| = |G|/|H| = [K : F ]/[K : L] = [L : F ].
(ii) If L1 ⊆ L2 , then every element of H2 = Gal(K/L2 ) fixes L1 , so is contained in H1 = Gal(K/L1 ).
Conversely, if H2 ≤ H1 , then L1 = Fix(H1 ) ⊆ Fix(H2 ) = L2 . And in this case, [L2 : L1 ] = [K : L1 ]/[K :
L2 ] = |H1 |/|H2 | = |H1 : H2 |. Hence, the diagram of subextensions of K/L is the same as the diagram of
subgroups of G, only flipped upside down (and even the numbers near the edges of the diagram, that is, the
degrees of subextensions and the indices of subgroups, are the same).
(iii) It follows that for “the minimal diamond” diagram of L1 and L2 corresponds to that of H1 and H2 :
L1 L2 H1 ∩ H2
n1 n2 n1 n2

L1 L2 H1 H2
m1 m2 m1 m2
L1 ∩ L2 hH1 , H2 i
so that L1 L2 (the minimal field containing L1 and L2 ) correspond to H1 ∩ H2 (the maximal subgroup of
both H1 and H2 , and L1 ∩ L2 (the maximal subfield of both L1 and L2 ) correspond to hH1 , H2 i (the minimal
group containing both H1 and H2 ).

22
(iv) Since K/F is normal, every embedding L/F −→ K/F extends to an embedding K/F −→ K/F , that
is, an element of G. Two elements ϕ, ψ of G define the same embedding L −→ K iff ϕ−1 ψ is identical on L,
that is, iff ϕ−1 ψ ∈ Gal(K/L) = H, that is, iff ϕH = ψH.
(v) We have ψ ∈ Gal(K/ϕ(L)) iff ψ(ϕ(α)) = ϕ(α) for every α ∈ L iff ϕ−1 ψϕ ∈ Gal(K/L) = H. So,
Gal(K/ϕ(L)) = ϕHϕ−1 , a conjugate of H. Thus, the conjugates of L/F are in one-to-one correspondence
with the conjugates of H, which, in their turn, are in one-to-one correspondence with left cosets of N (H) in
G.
(vi) H is a normal subgroup of G iff it has no conjugates in G but itself, iff L/F has no conjugates in K/F
but itself, iff every embedding ϕ of L/F into K/F preserves L, ϕ(L) = L, iff L/F is normal, and so Galois.
In this case the mapping ϕ 7→ ϕ|L defines a homomorphism η: G = Gal(K/F ) −→ Gal(L/F ). Since every
automorphism of L/F extends to an automorphism K/F , η is surjective. The kernel of η consists of elements
of G that fix L, that is, ker(η) = Gal(K/L) = H. Hence, Gal(L/F ) ∼ = G/H.

4.7. Examples of diagrams of subextensions and the corresponding Galois groups

√ √
4.7.1. The diagram of subextensions of the biquadratic extension Q( 2, 3)/Q, along with the diagram of
subgroups of its Galois group:

√ √
Q( 2, 3) 2
1 2
2 2
2 2
√ √ √
Q( 2) Q( 3) Q( 6) hϕ2 i hϕ1 i hϕ3 i
2 2 2
2 2 2
Q V4 = hϕ1 , ϕ2 i.

√ √
It follows that Q( 2, 3) contains no other subfields!
4.7.2. The diagram of
√ subextensions of the splitting field K/Q of the polynomial x3 − 2, K = Q(ω, α), where
2πi/3 3
ω=e and α = 2, with the diagram of subgroups of its Galois group:

Q(ω, α) 1
3 2 3 2 2
2 2 2
Q(ω) Q(α) Q(ωα) Q(ω 2 α) hσi hτ1 i hτ2 i hτ3 i
3 3 3
2 3 3
2 3
Q S3 = hσ, τ1 i.

 
α7→α
α7→ω 2 α α7→ωα

α7→ωα
where σ : (ω7 →ω ) fixes ω, τ1 : ω7→ω 2 fixes α, τ2 : ω7→ω 2 fixes ωα, and τ3 : ω7→ω2 fixes ω 2 α. Notice that
the subextension Q(ω)/Q is normal (as the corresponding subgroup hσi), and the subextensions Q(α)/Q,
Q(ωα)/Q, Q(ω 2 α)/Q are all conjugate (as the corresponding subgroups hτ1 i, hτ2 i, hτ3 i).
4.7.3. For any prime p and n ∈ N, Gal(F n /F ) = hφi ∼ = Z , where φ is the Frobenius automorphism. So,
p p n
the diagram of subfields of Fpn is the same as the diagram of subgroups of the cyclic group Zn : for every
d
d n, the subfield Fpd corresponds to the subgroup hdi of Zn : α ∈ Fpd iff ϕd (α) = αp = α. The subgroup
∼ ∼
hdi is isomorphic to Zn/d , and Gal(Fpd /Fp ) = Zn /Zn/d = Zd .
4.7.4. Let K ⊂ C be the splitting√field of x8 − 2 ∈ Q[x] and G = Gal(K/Q). As we know from 4.5.6,
K = Q(α, ω) = Q(α, i), where α = 8 2 and ω = 1+i √ , G is generated by ϕ : α 7→ ωα, i 7→ i, and ψ : α 7→ α,
2
i 7→ −i, and has the presentation G = ϕ, ψ : ϕ = ψ 2 = 1, ψϕψ −1 = ϕ3 . (So, G ∼
8
= Z8 × Z2 ∼= SD16 .) Let’s
7 7
find all normal subgroups of G = {1, ϕ, . . . , ϕ , ψ, ϕψ, . . . , ϕ ψ}. These are, of course, 1 and G itself. Next,
these are hϕi and its cyclic subgroups hϕ2 i and hϕ4 i. The identity ϕ−1 ψϕ = ϕ2 ψ implies that the elements
ϕk ψ split into two conguacy classes, {ψ, ϕ2 ψ, ϕ4 ψ, ϕ6 ψ} and {ϕψ, ϕ3 ψ, ϕ5 ψ, ϕ7 ψ}. If a normal subgroup N
contains an element from one of these classes, then it contains all other elements of this class, so contains ϕ2 .
If N also contains an element ϕk with an odd k, or contains an element of the other class, then it contains
ϕ and coincides with G. Hence, we may only have, and do have, two more normal subgroups in G: hϕ2 , ψi

23
and hϕ2 , ϕψi. Thus, the complete diagram of normal subgroups of G is
1
2
hϕ4 i
2
hϕ2 i
2 2
2

hϕ2 , ψi hϕi hϕ2 , ϕψi

2
2 2
G.
By duality, we must have an isomorphic diagram of normal subextensions of K/Q:
K
2
L5
2
L4
2 2
2
L2 L1 L3
2
2 2
Q.
Let’s find the fields Li :
(i) Since i is fixed by ϕ, we have L1 = Q(i). (We need just one element of degree 2 to generate L1 .)
√ √
(ii) Both ϕ2 and ψ fix 2, so L2 = Q( 2).
√ √
(iii) Both ϕ2 and ϕψ fix i 2, so L3 = Q(i 2).
√ √
(iv) ϕ2 fixes i and 2, so L4 = Q(i, 2). (Indeed, L4 is fixed by ϕ2 and has degree 4 over Q.)
√ √
(v) ϕ4 fixes i and 4 2 (since ϕ4 (α) = αω 4 = −α), so L5 = Q(i, 4 2).
So, the diagram of all normal subextensions of K/Q is
K
2
√
Q(i, 4 2)
2 √
Q(i, 2)
2 2
2
√ √
Q( 2) Q(i) Q(i 2)
2
2 2
Q
√
√

Let’s find the Galois groups of K over the fields Li . The groups Gal K/Q(i, 4 2) and Gal K/Q( 2, i)
are generated by ϕ4 and ϕ2 ,
and are isomorphic to Z2 and to Z4 respectively.
The group Gal K/Q(i) is generated by ϕ and is isomorphic to Z8 .
√

The group Gal K/Q( 2) is ϕ2 , ψ (ϕ2 )4 = ψ 2 = 1, ψϕ2 ψ −1 = (ϕ2 )−1 and is isomorphic to D8 .
√

The group Gal K/Q(i 2) is hϕ2 , ϕψi. Let’s put a = ϕ2 and b = ϕψ, then a and b satisfy a4 = 1;
b2 = ϕψϕψ = ϕ4 = a2 , and so b4 = 1; bab−1 = ϕψϕ2 ψ −1 ϕ−1 = ϕ6 = a3 = a−1 . These relations define the
group Q8 = a, b a4 = b4 = 1, a2 = b2 , bab−1 = a−1 .

5. Composites and towers of Galois extensions

This is a rather technical section, which results will be helpful below.

5.1. The change of the basic field of a Galois extension

5.1.1. Theorem. Let K/F be a Galois extension and L be any subfield of an extension of K. Then KL/F L
is also Galois, and Gal(KL/F L) is (isomorphic to) a subgroup of Gal(K/F ).

24
Proof. The generators of K over F also generate KL over F L, so KL/F L is finite. These generators are
all separable over F , so over F L. All conjugates of these elements over F , and so, over F L, are in K, so
KL/F L is normal.
Any element ϕ of Gal(KL/F L) fixes F and defines an embedding ϕ|K : K/F −→ KL/F ; since K/F is
normal, ϕ(K) = K, so ϕ|K ∈ Gal(K/F ). We therefore have a homomorphism Gal(KL/F L) −→ Gal(K/F ),
ϕ 7→ ϕ|K . If ϕ|K is trivial, then ϕ fixes K, and since ϕ fixes L too, ϕ fixes KL, that is, is identical. Hence,
the homomorphism ϕ 7→ ϕ|K is injective, and Gal(KL/F L) is isomorphic to a subgroup of Gal(K/F ).

5.2. The composite of two extensions of which one is Galois

5.2.1. Theorem. Let a Galois extension K/F be a composite K = L1 L2 of two subextensions L1 /F
and L2 /F such that L1 ∩ L2 = F and L1 /F is normal. Then Gal(K/F ) ∼
= Gal(K/L1 ) × Gal(L1 /F ),
Gal(K/L2 ) ∼
= Gal(L1 /F ), and [K : F ] = [L1 : F ] · [L2 : F ].

Proof. Let G = Gal(K/F ), and let H1 , H2 ≤ G be the subgroups corresponding to L1 and L2 , H1 =

Gal(K/L1 ) and H2 = Gal(K/L2 ). Then H1 is normal in G, H1 H2 = G, and H1 ∩ H2 = 1, so G = H1 × H2 ,
and Gal(L1 /F ) ∼
= G/H1 ∼
= H2 :

H1 ∩ H2 = 1 L1 L2 = K
m n m n

H1 H2 L1 L2
n m n m
H1 H2 = G L1 ∩ L2 = F .

Let n = [L1 : F ] and m = [L2 : F ]; then n = |G : H1 | = |H2 | = [K : L2 ], m = |G : H2 | = |H1 | = [K : L1 ],

and [K : F ] = nm.

5.2.2. Example. Let K be the splitting field of an irreducible polynomial

√ xn − a ∈ Q[x] for some n ∈ N.
Then K = Q(ω, α) where ω is a primitive nth root of unity and α = a. So, K is the composite, K = L1 L2 ,
n

of the fields L1 = Q(ω) and L2 = Q(α). The extension L2 /Q is not, generally speaking, normal, and has
degree n. The cyclotomic extension L1 /Q is normal, of degree ϕ(n), and we have Gal(L1 /Q) ∼ = Z∗n and
∼
Gal(K/L1 ) = Zn . It need not be that the intersection L1 ∩ L2 = Q, but if it is (say, if (n, ϕ(n)) = 1), then
Gal(K/Q) ∼ = Zn × Z∗n .

5.3. The composite of two Galois extensions

5.3.1. Theorem. Let an extension K/F be a composite K = L1 L2 of two Galois subextensions L1 /F
and L2 /F with L1 ∩ L2 = F . Then K/F is Galois, and we have Gal(K/F ) ∼ = Gal(K/L1 ) × Gal(L1 /F ),
Gal(K/L2 ) ∼
= Gal(L1 /F ), Gal(K/L1 ) ∼
= Gal(L2 /F ), and [K : F ] = [L1 : F ] · [L2 : F ]:

H1 ∩ H2 = 1 L1 L2 = K
m n m n

H1 H2 L1 L2
n m n m
H1 H2 = G L1 ∩ L2 = F .

2 2
5.3.2. Examples. (i) Let K√ be the splitting field
√ of the polynomial f (x) = (x − 2)(x − 3) ∈ Q[x]. Then
K = L1 L2 where L1 = Q( 2) and L2 = Q( 3) are normal extensions of Q, and L1 ∩ L2 = Q. Hence,
Gal(K/F ) ∼
= Gal(L1 /Q) × Gal(L2 /Q) ∼= Z2 × Z2 .
(ii) Let K
√ be the splitting field of the polynomial f (x) = (x2 − 2)(x 3
√
− 3) ∈√Q[x].√Then K = L1 L2 , where
3 2πi/3 3
L1 = Q( 2) and L2 is the splitting field of x − 3, L2 = Q e , 3 = Q( −3, 3 3). Both L1 and L2 are
normal extensions of Q, and L1 ∩ L2 = Q, so Gal(K/F ) ∼ = Gal(L1 /Q) × Gal(L2 /Q) ∼= Z 2 × S3 .

25
5.3.3. Now let an extension K/F be a composite of two Galois subextensions L1 /F and L2 /F with L1 ∩L2 6=
F . Then K/F is Galois; let G = Gal(K/F ), H1 = Gal(K/L1 ) and H2 = Gal(K/L2 ). By Theorem 5.3.1 we
have the diagrams
H1 ∩ H2 = 1 L1 L2 = K
m n m n

H1 H2 L1 L2
n m n m
H1 H2 L1 ∩ L2
d d
G F
where H1 H2 = H1 × H2 , and [K : F ] = [L1 : F ] · [L2 : F ]/[L1 ∩ L2 : F ].
Let N1 = Gal(L1 /F ) and N2 = Gal(L2 /F ). Then N1 ∼ = G/H1 , and H1 ∼
= (H1 H2 )/H2 is isomorphic to
a (normal) subgroup of N2 ∼= G/H2 , that is, G “is made of” N1 and a subgroup of N2 .
5.3.4. Here is a more detailed description of the group G from 5.3.3. We have a natural homomorphism
η: G −→ N1 × N2 , ϕ 7→ (ϕ|L1 , ϕ|L2 ), which is injective since L1 L2 = K. η is not, however, surjective: if


ϕ1 = ϕ|L1 and ϕ2 = ϕ|L2 , then ϕ1 |L1 ∩L2 = ϕ2 |L1 ∩L2 . Let D = Gal (L1 ∩ L2 )/F , then D = G/(H1 H2 ) ∼ =

∼

N1 / (H1 H2 )/H2 = N2 / (H1 H2 )/H1 is a common factor of N1 and N2 ; let τ1 : N1 −→ D and τ1 : N2 −→ D
be the factorization mappings. Then the image of η lies in the subgroup

N1 ×D N2 = (ϕ1 , ϕ2 ) : τ1 (ϕ1 ) = τ2 (ϕ2 )

of N1 × N2 , called the relative direct product of the groups N1 and N2 with respect to their common factor
D. Comparing their cardinalities, we find that G ∼ = N1 × D N 2 .
5.3.5. Example. Let K be the splitting field of the polynomial f (x) = (x3 − 2)(x3 − 3) ∈ Q[x]. √ Then
√
3 3
K = L1 L2 , where
√ L√1 and L 2 are the splitting fields of x − 2 and of x − 3 respectively,
√ L 1 = Q( −3, 3 2)
and L1 = Q( −3, 3 3). Both L1 and L2 are normal extensions of Q, L1 ∩ L2 = Q( −3), so Gal(K/F ) ∼ =
Gal(L1 /Q) ×Gal((L1 ∩L2 )/Q) Gal(L2 /Q) ∼ = S3 × Z 2 S3 .

5.4. Free composites of Galois extensions

5.4.1. Let us say that a composite K = L1 · · · Ln of algebraic extensions Li /F is free if the natural
epimorphism L1 ⊗F Q · · · ⊗F Ln −→ K is an isomorphism. If Li /F are all finite, this is equivalent to
n
having [K : F ] = i=1 [L
i : F ]. In the case K = L1 · · · Ln is a free composite, for each i we have
Li ∩ L1 · · · Li−1 Li+1 · · · Ln = F .
5.4.2. Let K/F be a Galois extension whose Galois group G = Gal(K/F ) is a direct product, G = H1 ×
· · · × Hk , of subgroups H1 , . . . , Hn . Then each of Hi is normal in G. For each i put Ni = H1 × · · · × Hi−1 ×
Hi+1 × · · · × Hn ; then Ni are normal subgroups of G with G/Ni ∼ = Hi , and N1 ∩ · · · ∩ Nn = 1. For each
i = 1, . . . , n, let Li Q
= Fix(Ni ); then L
Qn i /F are Galois extensions with Gal(Li /F ) ∼ = G/Ni ∼
= Hi . We have
n
L1 · · · Ln = K and i=1 [Li : F ] = i=1 |Hi | = |G|, so K is a free composite of L1 , . . . , Ln .
5.4.3. Conversely, if an extension K/F is a composite, K = L1 · · · Ln , of Galois
subextensions Li /F with
Gal(Li /F ) = Hi , i = 1, . . . , n, such that for each i, Li ∩ L1 · · · Li−1 Li+1 · · · Ln = F , then by 5.3.1, K/F is
Galois with Gal(K/F ) ∼ = H1 × · · · × Hn , and is a free composite of L1 /F, . . . , Ln /F .

5.5. Composites of towers of Galois extensions

5.5.1. Let K/F be a Galois extension, and assume that K is a tower,

K = Ln /Ln−1 / · · · /L1 /L0 = F, (5.1)

of Galois extensions, that is, with Li /Li−1 being Galois for all i. For each i, let Hi = Gal(K/Li ); then G
has the subnormal series
1 = Hn ≤ Hn−1 ≤ · · · ≤ H1 ≤ H0 = G, (5.2)
where for each i, H /H ∼
i−1 = Gal(L /L ).
i i i−1
Conversely, if K/F is a Galois extension whose Galois group G possesses a subnormal series (5.2),
then K/F is representable as a tower of Galois extensions (5.1), where for each i, Li = Fix(Hi ), and
Gal(Li /Li−1 ) ∼
= Hi−1 /Hi .

26
5.5.2. Let K = Ln /Ln−1 / · · · /L1 /L0 = F and K ′ = L′m /L′m−1 / · · · /L′1 /L′0 = F be two towers of Galois
extensions, contained in a common field. Then the composite KK ′ is representable as the tower

KK ′ = (Ln L′m )/(Ln L′m−1 )/ · · · /(Ln L′1 )/Ln /Ln−1 / · · · /L1 /L0 = F.


of Galois extensions, and by Theorem 5.1.1 or by 5.1.1, for each j the group Gal (Ln L′j )/(Ln L′j−1 ) is
isomorphic to a subgroup of Gal(L′j /L′j−1 ). We obtain:
Theorem. The composite of two towers of Galois extensions, with Galois groups N1 , . . . , Nr , is a tower of
Galois extensions, whose Galois groups are subgroups of N1 , . . . , Nr .
5.5.3. If K/L and L/F are Galois extensions, the extension K/F may not be Galois. By Theorem 4.4.3,
K/F is separable; let E/F be the Galois closure of K/F , let Gal(E/F ) = {ϕ1 , . . . , ϕn }. E is generated by
the conjugates of K, so E is the composite E = K1 · · · Kn where for each i, Ki = ϕi (K). Since the extension
L/F is normal, for each i, ϕi (L) = L, so Ki is an extension of L, and we have the commutative diagram

∼ K
ϕi :K −→ i

∼ L.
L −→
F

(We cannot say, however, that Ki /L is isomorphic to K/L since ϕi does not, generally speaking, fix L.)
Since ϕ is an isomorphism, Gal(Ki /L) ∼= Gal(K/L).
√ √
5.5.4. Example. Let α = 4 2, so that α2 = 2. Let K = K1 = Q(α) and L = Q(α2 ). The extensions K/L
and L/Q are quadratic
√ and so Galois, but the extension K/Q is not. The conjugates of α over Q are ±α,
±iα (where i = −1), and the Galois closure of K/Q is Q(α, iα) = K1 K2 where K2 = Q(iα). The field
K2 is also a quadratic extension of L, the minimal polynomial of its generator iα over L is x2 + α2 . The
homomorphism ϕ that produces the commutative diagram
∼ K
ϕ:K −→ 2

∼ L
L −→
Q

is defined by ϕ(α) = iα, and maps α2 to −α2 .

5.5.5. Let K = Ln /Ln−1 / · · · /L1 /L0 = F be a tower of Galois extensions. By Theorem 4.4.3, K/F is sepa-
rable; let E/F be the Galois closure of K/F , and let G = Gal(E/F ). Then E/F is the composite of the exten-
sions ϕ(K)/F , ϕ ∈ G, and for each ϕ, this extension

is the tower ϕ(K) = ϕ(Ln )/ϕ(Ln−1 )/ · · · /ϕ(L1 )/ϕ(L0 ) =
F of Galois extensions with Gal ϕ(Li )/ϕ(Li−1 ) ∼ = Gal(Li /Li−1 ) for all i. By Theorem 5.5.2, we obtain:
Theorem. If K/F is a tower of Galois extensions, with Galois groups N1 , . . . , Nr , then the Galois closure
E/F of K/F is also a tower of Galois extensions, whose Galois groups are subgroups of N1 , . . . , Nr . It
follows that Gal(E/F ) has a subnormal series with factors being subgroups of N1 , . . . , Nr .

6. Some applications of the Galois theory

6.1. More methods of finding the minimal polynomial

6.1.1. Let α be a separable algebraic element over a field F . Construct a Galois extension K/F that contains
α and find G = Gal(K/F ). Find the orbit Gα = {α1 , . . . , αn } of α under the actionQnof G; then α1 , . . . , αn
are all the conjugates ofQα over F , and the minimal polynomial of α is mα,F (x) = i=1 (x − αi ).
We also have that ϕ∈Gal(K/F ) (x − ϕ(α)) = mkα,F for some k. Comparing the degrees, we see that
Q
deg mα,F = [F (α) : F ] and deg ϕ∈Gal(K/F ) (x − ϕ(α)) = [K : F ], so k = [K : F ]/[F (α) : F ] = [K : F (α)].

27
 √ √ √ √

Example. Let’s find the minimal polynomial of α = 2 + 3 over Q. We have α ∈ K = Q 2, 3 , whose
√ √ √ √
the Galois group G = Gal(K/Q) ∼ = V4 acts by 2 7→ ± 2 and 3 7→ ± 3. The set of conjugates of α is
the orbit of α under the action of G, which is
n √ √ √ √ √ √ √ √ o
α1 = 2+ 3, α2 = − 2 + 3, α3 = 2 − 3, α4 = − 2 − 3 .

The minimal polynomial of α is therefore

√
√

f (x) = (x−α1 )(x−α2 )(x−α3 )(x−α4 ) = x2 −(5+2 6) x2 −(5−2 6) = x4 −10x2 +25−24 = x4 −10x2 +1.

And without any computations, since the orbit of α has 4 elements, degQ α = 4.
6.1.2. Let α be a separable algebraic element over a field F , let K/F be a Galois extension that contains
α, let L/F be a normal subextension of K/F , and let p = mα,L , the minimal polynomial of α over L. The
minimal polynomial mα,F of α over F splits over L to a product or irreducible polynomials, the minimal
polynomials of their roots; for every ϕ ∈ Gal(K/F ), ϕ(p) is contained and is irreducible in L[x], and its
roots are conjugate of α; thus, these are the other irreducible factors of mα,F over L. (In particular, all these
factors have the same degree.) Since L/F is normal, ϕ ∈ Gal(K/F ) induce elements of Gal(L/F ); so, mα,F
is the product of distinct polynomials ϕ(p), ϕ ∈ Gal(L/F ).
Q
We also have that ϕ∈Gal(L/F ) ϕ(p) = mkα,F for some k. Comparing the degrees, we see that deg mα,F =
Q
[F (α) : F ] and deg ϕ∈Gal(L/F ) ϕ(p) = [L : F ] · [L(α) : L], so k = [L : F ] · [L(α) : L]/[F (α) : F ] = [L(α) :
F (α)].
√ √ √ √
Example.
√ Again, let F = Q and α = 2 + 3. Take L = Q( 2). √ Then p(x)√ = mα,L (x) = (x − √2)2 − 3 =
x2 − 2 √2x − 1. We have L(α) = Q(α), so mα,Q = pϕ(p) where ϕ : 2 7→ − 2, so mα,Q = (x2 − 2 2x − 1) ·
(x2 + 2 2x − 1) = x4 − 2x2 + 1 − 8x2 = x4 − 10x2 + 1.

6.2. The norm of algebraic elements

6.2.1. Let K/F be a Galois extension,Q let G = Gal(K/F ). For every α ∈ K define the norm of α (in K and
over F ) by the formula NK/F (α) = ϕ∈G ϕ(α). In the case degF α = [K : F ], that is, when K = F (α), all
elements ϕ(α), ϕ ∈ G, are distinct, and NK/F (α) is the product of all the conjugates of α; otherwise, it is
the product of all conjugates of α to the power of [K : F (α)].
For any α, since NK/F (α) is fixed by G, we have NK/F (α) ∈ F . From the very definition, NK/F is
multiplicative: NK/F (αβ) = NK/F (α)NK/F (β). Hence, NK/F is a homomorphism of the multiplicative
groups K ∗ −→ F ∗ (and NK/F (0) = 0).
√ √
6.2.2. Example. If K/F √ is a quadratic
√ extension, K = F ( D), then for α = a + b D ∈ K, a, b ∈ F , we
have NK/F (α) = (a + b D)(a − b D) = a2 − b2 D. This is just the norm that was so helpful to us in the
first semester when we dealt with the rings of quadratic integers.
6.2.3. Now let K/F be a finite separable extension, let [K : F ] = n. Let E/F be any Galois extension
containing K,Q let G = Gal(E/F ) and H = Gal(E/K). For α ∈ K, define the norm of α in K over F by
NK/F (α) = ϕ∈G/H ϕ(α). (Where G/H is the set of left cosets of H in G; it is not a group if H is not
normal in G.) Notice that for ϕ, ψ from the same left coset of H in G, ϕ(α) = ψ(α), so the formula above
defines NK/F (α) well.
In the case E = K, this coincides with the definition in 6.2.1.
6.2.4. We have the following:
Proposition. (i) NK/F does not depend on the choice of the extension E.
(ii) NK/F (α) ∈ F for all α ∈ K.
(iii) NK/F is a multiplicative function from K to F , NK/F (α1 α2 ) = NK/F (α1 )NK/F (α2 ) for any α1 , α2 ∈ K.
n/d
(iv) For α ∈ K, let mα,F (x) = xd + · · · + a1 x + a0 ; then NK/F (α) = (−1)n a0 .

28
(v) For α ∈ K, let T be the linear transformation of K defined by multiplication by α, T (β) = αβ. Then
NK/F (α) = det T .
Proof. (ii) follows from the fact that for every α ∈ K, NK/F (α) is fixed by G.
(iii) is clear from the definition. Q
The product of all conjugates of α is (−1)d a0 , where d = degQ F α. In the product ϕ∈G ϕ(α) each
conjugate of α appears |G|/d times, and in the product NK/F (α) = ϕ∈G/H ϕ(α) it appears (|G|/|H|)/d =
n/d
n/d times. So, NK/F (α) = (−1)n a0 , which proves (iv).
(i) follows from (iv).
n/d
Finally, in (v), the characteristic polynomial of T is cT = mα,F . Hence, det T , which is the constant
n
term of cT times (−1) , equals NK/F (α) by (iv).

6.3. Abelian extensions

Let K/F be an abelian extension, that is, a Galois extension whose Galois group G is abelian.
6.3.1. Every subgroup of G is normal, so every subextension of K/F is normal. This implies that for every
α ∈ K, F (α) contains all conjugates of α.
6.3.2. By the fundamental theory of abelian groups, G is a direct product of cyclic subgroups, G = H1 ×
· · · × Hk , H i ∼
= Zni for some ni ∈ N, i = 1, . . . , k. By 5.4, K is a free composite
of cyclic subextensions:
K = L1 · · · Ln where for every i, Li = Fix H1 ×
· · · × Hi−1 × Hi+1 × · · · × Hk , Li /F are Galois with
Gal(Li /F ) ∼ = Hi , and Li ∩ L1 · · · Li−1 Li+1 · · · Lk = F , i = 1, . . . , k.
6.3.3. Any cyclotomic extension is abelian, so is a free composite of cyclic subextensions.
√
6.4. Subextensions of the real radical extension F ( n a)/F , a > 0, and the Galois group of xn − a
Let F be a real field (that is, F ⊆ R), let a ∈ F , a > 0, let n ∈ N, and assume that the polynomial
xn − a is irreducible in F [x].
√ √
6.4.1. Claim. The only subextensions of F ( n a)/F are √ subextensions of the form F ( d a)/F√with d n. In
particular, the only nontrivial normal subextension of F ( n a)/F , and only if n is even, is F ( a)/F .
√
Proof. Let α = n a ∈ R and K = F (α). Let L/F be a subextension of K/F , with [L : F ] = d; then
degL α = n/d. Let β be the product of the conjugates of α over L, then β ∈ L, All conjugates of α
over F have form √ ω k α, where ω√= e2πi/n ; so, β = αn/d ω r for some r. √ But since β ∈ R, ω r = ±1, so
n/d
β = ±α = ± a. Since degF a = d = [L : F ], we obtain that L = F ( d a). (The polynomial xd − a is
d d

irreducible in F [x] since otherwise xn − a = (xn/d )d − a would also be reducible.)

√ √
If d ≥ 3, then F ( d a)/F is not normal, since d a has nonreal conjugates.
√
6.4.2. Let α = n a ∈ R, ω = e2πi/n , K = F (α), and N = F (ω); then KN = F (ω, α) is the splitting field
of xn − a. Since N/F is abelian, √ the extension (K ∩ N )/F is a normal subextension of K/F , so either
K ∩ N = F , or, if n is even and a ∈ N , is a quadratic extension.
If F = Q, in the first case we have [KN : F ] = nϕ(n) and Gal(xn − a) = Gal(KN/Q) ∼ = Zn × Z∗n , and
in the second case [KN : Q] = nϕ(n)/2.
√


6.4.3. Examples. (i) The splitting field of f = x8 − 3 ∈ Q[x] is KN where K = Q 8 3 and N = Q 1+i √
2
.
It is easy to see that K ∩ N = Q, so Gal(f /Q) ∼ = Z8 × Z∗8 , |G| = 32.
√


(ii) The splitting field of f = x8 − 2 ∈ Q[x] is KN where K = Q 8 2 and N = Q 1+i √
2
. This time
√
K ∩ N = Q( 2), and |G| = 16. (G is the semidihedral group SD16 .)

6.5. The theorem on a primitive element

6.5.1. We will need the following nice fact:
Proposition. Any finite separable extension has only finitely many subextensions.
Proof. Any such extension is contained in a Galois extension (the Galois closure thereof), which has only
finitely many subextensions (corresponding to (finitely many) subgroups of its Galois group).
6.5.2. An element α of an algebraic extension K/F is said to be primitive if K = F (α).

29
6.5.3. Theorem. Every finite separable extension is simple (that is, possesses a primitive element).
Proof. Let K/F be a finite separable extension. Consider two cases.
(i) F is finite. In this case K is also a finite field, and is a simple extension of Fp .
(ii) F is infinite. Then the union of any finite collection of proper subspaces of K viewed as an F -vector
space is a proper subset of K. Since K has only finitely many proper subfields that are subextensions of
K/F , there is an element α ∈ K that is not contained in any of these subfields (see Lemma below). Hence,
F (α) = K.
Lemma. If F is an infinite field and V is an F -vector space, then V is not a union of its proper subspaces.
Sn Sn
Proof. Assume S that V = i=1 Vi for n ≥ 2. We may assume that V1 6⊆ i=2 Vi , otherwise we can exclude
n
V1 . Let α ∈ V1 \ i=2 Vi and β 6∈ V1 , and consider the line L = αt+β(1−t), t ∈ F . We have L∩V1 = {α};
L is infinite, so there is i ≥ 2 and distinct t1 , t2 ∈ F such that αt1 + β(1 − t1 ), αt2 + β(1 − t2 ) ∈ Vi . But then
α ∈ Vi , contradiction.
6.5.4. Corollary. Let L/F be a finite separable extension of degree n, and let K/F be the Galois closure
of L/F . Then Gal(K/F ) is (isomorphic to) a subgroup of Sn .
Proof. Let α be a primitive element of L with respect to F . Then every conjugate of L in K is generated by
a conjugate of α, and K is generated by the set A of conjugates of α. Any automorphism of K/F is defined
by its actions on A. Since |A| = degF α = [L : F ] = n, Gal(K/F ) is isomorphic to a subgroup of Sn .
6.5.5. Theorem 6.5.3 does not hold for inseparable extensions: take K = Fp (x, y) (the field of rational
functions in two variables over Fp ) and F = Fp (xp , y p ); then [K : F ] = p2 , but for every element h ∈ K,
hp ∈ F , so [F (h) : F ] ≤ p. (Indeed, for any f ∈ Fp [x, y], f (x, y)p = f (xp , y p ) ∈ F and for any f /g ∈ K,
(f /g)p = f p /g p ∈ F too.)

6.6. p-extensions
Let p be a prime integer.
6.6.1. A Galois extension K/F is said to be a p-extension if [K : F ] = pn for some n ∈ N (and so, Gal(K/F )
is a p-group).
A finite extension is a p-extension if it is contained in a Galois p-extension. For Galois extensions
these two definitions coincide: if K/F is Galois and is a subextension of a Galois p-extensions E/F , then
Gal(K/F ) is a quotient group of Gal(E/F ) and so, is a p-group.
√

6.6.2. The degree of any p-extension is a power of p. The converse is not true: the extension Q 3 2 /Q has
degree 3 but is not a 3-extension.
6.6.3. Theorem. An extension is a p-extension iff it is a tower of cyclic Galois extensions of degree p.
Proof. Let K/F be a p-extension, let K ⊆ E such that E/F is Galois with G = Gal(E/F ) being a p-group.
Let H = Gal(E/K), then H ≤ G, |H| = pk for some k. By Sylow’s theory, or by the theory of p-groups, there
is a series H = Hk ≤ Hk+1 · · · ≤ Hn = G of subgroups of G such that |Hi | = pi for each i. Since, for each i,
|Hi : Hi−1 | = p, Hi−1 is a normal subgroup of Hi , with Hi /Hi−1 ∼
= Zp . For each i, let Li = Fix(Hi ), then we
have a tower K = Lk /Lk+1 / · · · /Ln = F such that for every i, Li−1 /Li is Galois with Gal(Li−1 /Li ) ∼= Zp .
Conversely, let K = L0 /L1 / · · · /Ln = F , where for every i, Li−1 /Li is Galois with the Galois group
∼
= Zp . Then K/F is seprable; let E be the Galois closure of K; then E is a composite of towers isomorphic
to the tower of K, so is itself a tower of Galois extensions with Galois groups isomorphic to a subgroup of
Zp (which is either Zp or is trivial), so [E : F ] = pn .
We also see from the proof that a finite extension is a p-extension iff it is separable and its Galois closure
is a p-extension.
6.6.4. Theorem. (i) If K/F is a p-extension and L/F is a subextension of K/F , then both K/L and L/F
are p-extension.
(ii) If L1 /F and L2 /F are p-subextensions of an extension K/F , then their composite L1 L2 /F is a p-
extension.

30
(iii) If K/L and L/F are p-extensions, then K/F is also a p-extension.
6.6.5. Since any separable quadratic extension is Galois, an extension is a 2-extension iff it is a tower of
separable quadratic extensions; we will say that it is polyquadratic in this case.

6.7. The fundamental theorem of algebra

6.7.1. The fundamental theorem of algebra can be proved with the help of the Galois theory:
Theorem. C is the algebraic closure of R (and so, is algebraically closed).
Proof. The proof is based on two facts:
(i) Any polynomial over R of odd degree has a root in R and so, is either linear or reducible in R[x].
√ √ √
(ii)p
For any a ∈ R, a ∈ R p if a > 0 and R( a) ∼ = C if a < 0. For any z ∈ C, z ∈ C (for z = a + bi, w =
√ √
√1
2
a2 + b2 + a + i sign(b)
√
2
a2 + b2 − a, where all radicals are ≥ 0). Since every (separable) quadratic
extension is obtained by adjoining a square root, this implies that C is the only (up to isomorphism) quadratic
extension of R and has no quadratic extensions itself.
I’ll show that every nontrivial finite extension L of R is isomorphic to C. Let K/R be the Galois closure
of L/R, let G = Gal(K/R). Let H be the Sylow 2-subgroup of G and let N = Fix(H). Then for every
α ∈ N , the polynomial mα,R has odd degree, which is, by (i), impossible unless α ∈ R. Hence, N = R and
H = G. Then G is a 2-group, K/R is a 2-extension and is a tower K = L0 /L1 / · · · /Lk = R of quadratic
extensions. By (ii) we see that either K = R or K/R ∼ = C/R (and so, L = R or L/R ∼ = C/R).

6.8. Constructions with ruler and compass

6.8.1. Given a set S of points on the Euclidean plane, of cardinality ≥ 2, the following constructions with
ruler and compass are allowed to produce new points and add them to S: (i) connecting two of the points
by a straight line; (ii) drawing a circle centered at one of the points and passing through another; (iii)
finding (and adding to S) the points of intersection of two lines, of a line and a circle, or of two circles
already constructed. The points constructible this way are said to be constructible (from S, with ruler and
compass).

b
b b b

b
b b
b
b

6.8.2. Let S be a set of (more than one) points on the plane. Let us introduce a Cartesian coordinate system
on the plane (using points of S as the origin and a unit coordinate vector). A real number is said to be
constructible (from S) if it represents a coordinate of a constructible (from S) point. Clearly, a point on the
plane is constructible (from S) iff both its coordinates are constructible (from S) numbers.
6.8.3. It is easy to see that the coordinates of the numbers constructible from a set S form a field: if we
have points whose (first or second) coordinates are a and b, then we can construct points whose√(say, first)
coordinates are a + b, a − b, ab, or a/b. Moreover, we can also construct a point with coordinate a, so this
field is closed under taking quadratic extensions (“is quadratically closed”).
Let F be the field generated by the coordinates of the points of S. If a real number a is constructible
from S, we will also say that a is constructible over F . A real number is said to be just constructible if it is
constructible over Q.
6.8.4. Let S be a set of points in the plane and F be the field generated by the coordinates of the points from
S. The coordinates of any new point obtained from the points of S by the operations (i)-(iii) are solutions
of either a linear or a quadratic equation with coefficients from the field, generated by the coordinates of S,
and so, either belong to F or to a quadratic extension of F . Hence, we have:
Proposition. A real number is constructible over a real (that is, contained in R) field F iff it is contained
in a real polyquadratic extension of F .

31
6.8.5. The restriction that the constructible numbers must be real is inconvenient, and we can get rid of
it. A complex number γ = α + βi, α, β ∈ R, is said to be constructible over a real field F iff both α and
β are constructible over F . (Thus, if we interprete the plane as the complex plane, a complex number is
constructible iff the correponding point is constructible.) Then we also have:
Proposition. A complex number is constructible over a real field F iff it is contained in a polyquadratic
extension of F .
Proof. If a complex number γ = α + βi is constructible, that is, both α and β are contained in towers of real
quadratic extensions, then the composite of these towers is also a real polyquadratic tower, which contains
both α and β, and γ is contained in the (quadratic) extension of this tower obtained by adjoining i.
Conversely, assume that γ = α + βi is contained in a tower K = Ln /Ln−1 / . . . /L0 = F (i) of quadratic
extensions of F (i). For every j let L′j be the complex conjugate of Lj , let Mj = Lj L′j , and let Nj = Mj ∩ R.
Then we have the tower N = Nn /Nn−1 / . . . /N0 = F of real extensions, and α, β ∈ N . I claim that for every
j, the extension
√ Nj /Nj−1 is a tower of at most two quadratic extensions. Indeed, let Lj = Lj−1 (z) where
z = x + yi = c for some c = a + bi ∈ Lj−1 , then L′j = L′j−1 (z̄) and Mj = Mj−1 (z, z̄) = Mj−1 (x, y). So
Nj = Nj−1 (x, y). (There is a basis in Mj over Mj−1 consisting of elements of the form xr y s for some r, s,
and a linear combination of such elementsq √ is in Nj iff
all the coefficients are real, that is, are from Nj−1 .)
2 2
But x − y = a and 2xy = b, so x = a2 + b2 + a /2 and y = b/2x, with a, b ∈ Nj−1 .

6.8.6. From 6.6.5 we obtain:

Theorem. A complex number α is constructible over a real field F iff the splitting field of mα,F is a
2-extension.
6.8.7. As a corollary we see that the following problems are non-solvable with ruler and compass:
(i) Squaring a circle: Construct a square that has the same area as the unit circle; in other words, construct
a ∈ R such that a2 = π.
√
Indeed, the number π is transcendental.
(ii) Doubling the cube: Construct a cube having the volume of two unit cubes; or, in coordinates: find an
a ∈ R such that a3 = 2.
√
Indeed, 3 2 is not contained in any 2-extension.
(iii) Trisecting an angle: Given an angle θ, construct the angle θ/3. This problem is equivalent to the
problem of constructing the cubic root of a complex number of absolute value 1. It is solvable for some θ
(for θ = π for instance), but non-solvable in general. (For example, the angle of π/3 is not trisectable, as we
will see right below.)
6.8.8. Consider the problem of constructing (over Q) of a regular n-gon. This problem is equivalent to
constructing the complex number ω = e2πi/n , that is, a primitive root of unity of degree n. We know that
the extension Q(ω)/Q is Galois of degree ϕ(n) (where ϕ is Euler’s totient function); we therefore have:
Proposition. A regular n-gon is constructible iff ϕ(n) is a power of 2.
6.8.9. Let n = 2r pr11 · · · prl l where pi are distinct odd primes, r ≥ 0 and ri ≥ 1. Then ϕ(n) = 2s (p1 − 1)pr11 −1
· · · (pl − 1)prl l −1 where s = 0 if r = 0 and s = r − 1 if r ≥ 1, and for ϕ(n) to be a power of 2 it is necessary
that for each i, ri = 1 and pi − 1 is a power of 2.
Prime integers of the form 2r + 1, r ∈ N, are called Fermat’s primes. We therefore have:
Proposition. ϕ(n) is a power of 2, and so a regular n-gon is constructible, iff n = 2r p1 · · · pl where pi are
distinct Fermat’s primes.
Examples of Fermat’s primes are 3 = 21 + 1, 5 = 22 + 1, 17 = 24 + 1, 257 = 28 + 1, 216 + 1 = 65537.
(It is not directly related but is worth mentioning that if an integer of the form 2r + 1, r ∈ N, is prime,
then r = 2s for some integer s ≥ 0. Indeed, if r =
km for an odd k, then 2r + 1 is divisible by 2m + 1:
2r + 1 = (2m + 1) 2(k−1)m − 2(k−2)m + · · · − 2m + 1 .)

32
6.9. Linear independence of square roots of square free integers
√ √

6.9.1. Theorem. Let p1 , . . . , pn be distinct prime integers, and let K = Q p1 , . . . , pn . Then K/Q is a
√ √
free composite Q( p1 ) · · · Q( p1 ), [K : Q] = 2n , and Gal(K/Q) ∼= Zn2 .
Proof. We will
prove this by induction on n: assume that the assertion holds for some n, let K =
√ √ √ √
Q p1 , . . . , pn , and let p be a prime integer distinct from p1 , . . . , pn . If p 6∈ K, then K∩Q( p) = Q, both
√ √ √
∼ n
K/Q and Q( p)/Q are normal, so K( p)/Q is their free composite, and Gal K( p)/Q = Z2 × Z2 = Zn+1 2 .
√
So, it suffices to show that p ∈ K is impossible.
√ √
If p ∈ K, then Q( p)/Q is a subextension of K/Q of degree 2, and so, corresponds to a subgroup
of Gal(K/Q) of index 2. But Gal(K/Q) ∼ = Zn2 has 2n − 1 such subgroups, so √ K contains 2n − 1 quadratic
subextensions, and they are all known: these are the extensions of √ the form Q( m)/Q where m = pi1 · · · pik
√
for some 1 ≤√k ≤ n and 1 ≤ i1 < · · · < ik ≤ n. So, Q( p) = Q( m) for some m of this form. This implies
√
that p = c m for some c ∈ Q, so p = c2 m, which is clearly impossible.
√
6.9.2. Theorem. The set m : m is a square free positive integer is linearly independent over Q.
Proof. Let Q be a finite set of square-free positive integers, and let Q p1 , . . . , pn be the set of all prime divisors
of the elements
√ of Q. Then every element of Q has form p S = pi for some S ⊆ {1, . . . , n}. But the
√ √ i∈S

set B = pS : S ⊆ {1, . . . , n} is a basis of K = Q p1 , . . . , pn as a Q-vector space, so it is linearly
independent.
√ √
√ √
6.9.3. Let p1 , . . . , pn be distinct prime integers, let K = Q p1 , . . . , pn , and let α = c1 p1 + · · · + cn pn
for some nonzero c1 , . . . , cn ∈ Q.
Claim. α is a primitive element of K/Q, K = Q(α).
√ √
Proof. α has 2n distinct conjugates in K, ±c1 p1 ± · · · ± cn pn , so [Q(α) : Q] = 2n = [K : Q], so
K = Q(α).

6.10. The theory of symmetric rational functions

6.10.1. A polynomial, or a rational function, h(x1 , . . . , xn ) in variables x1 , . . . , xn is said to be symmetric if
it is invariant under any permutation of x1 , . . . , xn : h(xσ(1) , . . . , xσ(n) ) = h(x1 , . . . , xn ).
6.10.2. The polynomials s1 (x1 , . . . , xn ) = x1 + · · · + xn , s2 (x1 , . . . , xn ) = x1 x2 + x1 x3 · · · + xn−1 xn , . . .,
sn (x1 , . . . , xn ) = x1 · · · xn are called the elementary symmetric polynomials.
6.10.3. If a monic polynomial f (x) = xn + an−1 xn−1 + . . . + a0 , of degree n, has roots α1 , . . . , αn , then, up to
the sign, the coefficients of f are just the elementary symmetric polynomials of αi : an−1 = −s1 (α1 , . . . , αn ),
an−2 = s2 (α1 , . . . , αn ), . . ., a0 = (−1)n sn (α1 , . . . , αn ).
6.10.4. The fundamental theorem on symmetric polynomials says that, over any ring R, the symmetric
polynomials are uniquely representable as polynomials in the elementary symmetric polynomials:
Theorem. For every symmetric
polynomial h ∈ R[x1 , . . . , xn ] there exists a unique g ∈ R[y1 , . . . , yn ] such
that h(x1 , . . . , xn ) = g s1 , . . . , sn .
It follows that the ring of elementary symmetric polynomials in n variables is isomorphic to the ring of
polynomials in n variables.
6.10.5. Using the Galois theory, we can obtain a slightly weaker result. Let F be a field and n ∈ N. Define
K = F (x1 , . . . , xn ), the field of rational functions over F in variables x1 , . . . , xn . The symmetric group Sn
acts on K by permuting the variables xi ; the field L = Fix(Sn ) is the field of symmetric rational functions.
By Proposition 4.6.4, [K : L] = |Sn | = n!.
On the other hand, K is the splitting field of the so-called generic polynomial

G(x) = (x − x1 ) · · · (x − xn ) = xn − s1 xn−1 + · · · + (−1)n sn ,

whose coefficients lie in the field L′ = F (s1 , . . . , sn ); by Theorem 2.2.6, [K : L′ ] ≤ n!. Since L′ ⊆ L, we get
that L = L′ .

33
6.10.6. We have proved:
The fundamental theorem on symmetric functions. For any field F and n ∈ N, every symmetric
rational function h ∈ F (x1 , . . . , xn ) is representable in the form


h(x1 , . . . , xn ) = g s1 (x1 , . . . , xn ), . . . , sn (x1 , . . . , xn )

for some g ∈ F (y1 , . . . , yn ).

6.10.7. We have also obtained that for any field F and n ∈ N, the generic polynomial G(x) = xn − s1 xn−1 +
· · · + (−1)n sn ∈ L[x], where L = F (s1 , . . . , sn ) is the field of symmetric rational functions in n variables, has
Gal(G/L) ∼ = Sn .
6.10.8. As another corollary of 6.10.6 we have:
Theorem. Let F be a field, let f (x) = xn + a1 xn−1 + . . . + an ∈ F [x], and let α1 , . . . , αn be the roots of f
(in the splitting field of f ). Then for any symmetric polynomial h in n variables, h(α1 , . . . , αn ) ∈ F .

7. Solving polynomial equations in radicals

7.1. Radical and polyradical extensions

√
7.1.1. Let F be a field, let a ∈ F , and n ∈ N. By n a we will denote any element α of√an extension of F
such that αn = a. An extension K/F is said to be radical (or simple radical ) if K = F ( n a) for some a ∈ F
and n ∈ N.
7.1.2. An extension K/F is said to be polyradical (or an extension by radicals, or a root extension) if it is
a tower of radical extensions.
7.1.3. Clearly, any tower and any composite of polyradical extensions is polyradical.

7.2. Cyclic and polycyclic extensions

7.2.1. An extension K/F is said to be cyclic if it is a Galois extension with a cyclic Galois group.
7.2.2. Theorem. If L/F is a subextension of a cyclic extension K/F , then both K/L and L/F are cyclic.
Proof. Subgroups and quotient groups of a cyclic group are cyclic.
7.2.3. A (finite, separable) extension is said to be polycyclic if it is a tower of cyclic extensions.
7.2.4. Theorem. (i) A composite and a tower of polycyclic extensions is polycyclic.
(ii) If an extension is polycyclic then its Galois closure is polycyclic.
Proof. (i) If K1 and K2 are towers of Galois extensions with Galois groups H1 , . . . , Hn , then their composite
(assuming it is defined) is also a tower of Galois extensions with Galois groups being subgroups of Hi . So,
if Hi are all cyclic, then K1 K2 is a polycyclic extension.
If K/L and L/F are towers of cyclic extensions, then so is K/F .
(ii) If K/F is polycyclic, then the Galois closure of K/F is a composite of conjugates of K/F , which all are
isomorphic to K/F and so, are polycyclic.
7.2.5. A group is said to be polycyclic if it possesses a finite subnormal series with cyclic factors. It is easy to
see that a finite group is solvable iff it is polycyclic. (If a group has a subnormal series with abelian factors,
then this series can be refined to a series with cyclic factors.)
7.2.6. We have:
Theorem. A Galois extension K/F is polycyclic (=solvable) iff Gal(K/F ) is a polycyclic group.
Proof. K/F is a tower of cyclic extensions Galois extension iff G = Gal(K/F ) has a subnormal series with
cyclic factors.
In contrast with p-extensions, it is not however true that a non-Galois extension is polycyclic if its Galois
closure is polycyclic. (Since it is not true that every subgroup of a finite polycyclic group is a member of a
subnormal series with cyclic factors.)

34
7.3. Radical and cyclic extensions
We are now going to convince ourselves that (under certain conditions) radical and cyclic extensions
are the same!
7.3.1. Theorem. Let n ∈ N, let F be a field that contains √ all n-th roots of unity (that is, the polynomial
xn − 1 splits completely in F ), and let a ∈ F be such that n a is separable √ over F . (This is so, for instance,
if F is a perfect field, or if n is not divisible by char F .) Then K = F ( n a) is a cyclic extension of F of
degree dividing n.
√
Proof. Let α = n a, a ∈ F \ {0}, and ω be a generator of the group of nth roots of 1. α is a root of
the separable polynomial xn − a, whose all roots are ω k α, k ∈ Zn ; so, all conjugates of α have this form.
Since α is separable and ω ∈ F , K = F (α)/F is a Galois extension; let G = Gal(K/F ). We have a
mapping η: G −→ Zn , η(ϕ) = k such that ϕ(α) = ω k ϕ; since ϕ is defined by its action on α, η is injective.
For ϕk , ϕl ∈ G such that η(ϕk ) = k and η(ϕl ) = l, that is, ϕk (α) = ω k α and ϕl (α) = ω l α, we have
ϕk ϕl (α) = ϕk (ω l α) = ω k+l α (notice that G fixes ω), so η is a group homomorphism. So, G is isomorphic to
a subgroup of Zn .
7.3.2. In order to show that, conversely, cyclic extensions are radical, we will need “Lagrange’s resolvent”,
Let n ∈ N, let F be a field, let ω ∈ F be an nth root of unity. Let K/F be a cyclic extension of degree n,
and let ϕ be a generator of Gal(K/F ). For α ∈ K, the Lagrange resolvent (α, ω) is the element of K defined
by
(α, ω) = α + ωϕ(α) + ω 2 ϕ2 (α) + · · · + ω n−1 ϕn−1 (α).


7.3.3. Lemma. In the notation of 7.3.2, for any α ∈ K, ϕ (α, ω) = ω −1 (α, ω), and (α, ω)n ∈ F .
Proof. We have



ϕ (α, ω) = ϕ(α)+ωϕ2 (α)+ω 2 ϕ3 (α)+· · ·+ω n−1 ϕn (α) = ω −1 ωϕ(α)+ω 2 ϕ2 (α)+ω 3 ϕ3 (α)+· · ·+ω n ϕn (α)
= ω −1 (α, ω)


since ω n = 1 and ϕn (α) = α. Thus, ϕ (α, ω)n = ω −n (α, ω)n = (α, ω)n . Since ϕ generates Gal(K/F ), the
whole group fixes (α, ω)n , so (α, ω)n ∈ F .
7.3.4. In the process of proving the Fundamental theorem of the Galois theory, we had the following fact:
Lemma. The set of automorphisms of any field is linearly independent: for any field K, any distinct
ϕ1 , . . . , ϕn ∈ Aut(K), and any β1 , . . . , βn ∈ K not all zero, β1 ϕ1 + · · · + βn ϕn 6= 0.
Proof. Assume that the assertion is wrong and let β1 ϕ1 + β2 ϕ2 + · · · + βn ϕn = 0 be a minimal linear
dependence relation with all βi 6= 0. For every α ∈ K we then have

β1 ϕ1 (α) + β2 ϕ2 (α) + · · · + βn ϕn (α) = 0. (∗)

Let γ ∈ K be such that ϕ1 (γ) 6= ϕ2 (γ). (Clearly, n ≥ 2.) For every α ∈ K we now have

0 = β1 ϕ1 (γα) + β2 ϕ2 (γα) + · · · + βn ϕn (γα) = β1 ϕ1 (γ)ϕ1 (α) + β2 ϕ2 (γ)ϕ2 (α) + · · · + βn ϕn (γ)ϕn (α),

that is,
β1 ϕ1 (γ)ϕ1 + β2 ϕ2 (γ)ϕ2 + · · · + βn ϕn (γ)ϕn = 0. (∗∗)
Subtracting (∗∗) − ϕ1 (γ)(∗), we obtain that

β2 (ϕ2 (γ) − ϕ1 (γ))ϕ2 + · · · + βn (ϕn (γ) − ϕ1 (γ))ϕn = 0,

which contradicts the minimality of (∗).

7.3.5. We can now prove:
Theorem. Let n ∈ N, let F be a field that contains a primitive nth
√ root of unity ω, and let K/F be a cyclic
extension of degree n. Then K/F is a radical extension, K = F ( n a) for some a ∈ F .

35
(The existence of a primitive nth root of unity implies that char F / n.)
Proof. Let ϕ be a generator of Gal(K/F ). By Lemma 7.3.4, the automorphisms 1, ϕ, . . . , ϕn−1 of K/F are
linearly independent, so 1 + ωϕ + ω 2 ϕ2 + · · · + ω n−1 ϕn−1 6= 0, and there is α ∈ K such that γ = (α, ω) 6= 0.
By Lemma 7.3.3, a = γ n ∈ F . Since ω is a primitive root, √ ϕk (γ) = ω −k γ are distinct for k = 0, . . . , n − 1,
so γ has n conjugates, degF γ = n, and K = F (γ) = F ( a). n

7.3.6. Applying Theorem 7.3.5 and Theorem 7.3.1 to towers, we obtain:

Theorem. (i) If K/F is a polycyclic extension of degree n with char F / n and F contains a primitive nth
root of unity, then K/F is polyradical.
(ii) If K/F is a tower of separable radical extensions, with radicals of degrees n1 , . . . , nk , and for each i, F
contains all ni -th roots of unity, then K/F is polycyclic.

7.4. Solvability of polynomials in radicals

7.4.1. An element α, algebraic over F , is expressible by radicals (or can be solved for in terms of radicals)
over F if α is contained in a polyradical extension of F . A polynomial f ∈ F [x] is said to be solvable in
radicals (or by radicals) if all roots of f are expressible by radicals. Notice that if f is irreducible and one of
its roots is expressible in radicals, then f is solvable in radicals.
7.4.2. The following great theorem is the main goal of the Galois theory:
Theorem. Let F be a field and f ∈ F [x].
(i) If f is separable and the group Gal(f /F ) is solvable of order not divisible by char F , then f is solvable in
radicals.
(ii) If f is solvable in radicals, whose degrees, if F is not perfect, are not divisible by char F , then Gal(f /F )
(makes sense and) is solvable.
Proof. (i) Let K be the splitting field of f , let ω be a primitive root of degree n = | Gal(f /F )|. The
extension K(ω)/F (ω) is Galois, whose Galois group is a subgroup of Gal(K/F ) = Gal(f /F ) and is therefore
solvable. By Theorem 7.3.6, K(ω)/F (ω) is polyradical, and since F (ω)/F is a radical extension, K(ω)/F is
also polyradical. Hence, f is solvable in radicals.
(ii) Now assume that all roots of f are contained in a field K such that K/F is a tower of radical extensions
where F is perfect or the degrees n1 , . . . , nk of radicals are not divisible by char F , so that K/F is separable.
Let ω be a generator of the group generated by the roots of unity of degrees n1 , . . . , nk ; then the extension
K(ω)/F (ω) is also polyradical, and so, polycyclic (in particular, Galois). Since F (ω)/F is also polycyclic
(its Galois group is a subgroup of Z∗n for some n), K(ω)/F is polycyclic. Let E/F be the Galois closure of
K(ω)/F , then E/F is polycylic, and so, the group Gal(E/F ) is solvable. Let L ⊆ E be the splitting field of
f ; then Gal(L/F ) is a quotient group of Gal(E/F ), so it is solvable as well.
7.4.3. Corollary. If char F 6= 2, 3, every polynomial f ∈ F [x] of degree ≤ 4 is solvable in radicals. The
general polynomial of degree ≥ 5 is not solvable in radicals.
(This means that there is no general “symbolic” formula that allows to express the roots of a polynomial of
degree ≥ 5 in radicals.)
Proof. For any polynomial f of degree n, Gal(f ) is a subgroup of Sn . The groups S2 , S3 , S4 are solvable,
so all their subgroups are solvable, so all polynomials of degree ≤ 4 are solvable in radicals.
For n ≥ 5, Sn is not solvable. The general polynomial of degree n ≥ 5 has Galois group Sn and is not
therefore solvable in radicals.
7.4.4. We see that there is no general “symbolic” formula in radicals for roots of quintic; it would however be
nice to have a concrete example of a “numerical” plynomial unsolvable in radicals. The following proposition
allows to construct such examples:
Proposition. Any irreducible polynomial of degree 5 over Q that has three real and two non-real roots has
Galois group isomorphic to S5 and so, is unsolvable in radicals.
(Actually, same idea allows to construct polynomials of degree p with Gal ∼
= S for every prime p.)
p

36
Proof. Let f ∈ Q[x] be such a polynomial and let G = Gal(f /Q). Let α ∈ C be a root of f , let K ⊆ C be
the splitting field of f . Consider G as a subgroup of S5 via its action on the roots of f . f has two non-real
complex roots, so the complex conjugation is a transposition in S5 . Since Q(α) ⊆ K and [Q(α) : Q] = 5,
the degree [K : Q] is divisible by 5, so 5 |G|. Hence, G contains an element of order 5, which may only be
a 5-cycle. But (recall that) any transposition and a 5-cycle in S5 generate S5 , so G = S5 .
Example. f (x) = x5 − 4x + 2 ∈ Q[x] is irreducible, has three real and two non-real roots, so Gal(f, Q) ∼=S , 5
and f is unsolvable in radicals.
7.4.5. To express an element in radicals, we need first to adjoin to our basic field certain roots of unity.
Applying the Galois theory to cyclotomic extensions,
√ we√can√express by radicals the roots of unity themselves,
√3
√
−1+ −3 4
√ √ √
5 −1+ 5
√
10+2 5 6
√
1+ −3 8
√ √ √
2+ −2
like 1 = , 1 = −1, 1 = + , 1 = , 1 = . (However, at least
2 4 4 2 √n
2
theoretically, I don’t see why these expressions are better than “the radical” 1.)

7.5. The alternating group and the discriminant

7.5.1. Let f ∈ F [x] be a separable polynomial over a field F and let α1 , . . . , αn be the roots of f . The
Galois group Q Gal(f ), through its action on the set {α1 , . . . , αn }, is identified with a subgroup of Sn . The
product δ = i<j (αi − αj ) is fixed by even permutations from Sn , and switches sign under the action of
odd permutations.
7.5.2. D = δ 2 is a symmetric polynomial of α1 , . . . , αn , so, is a polynomial in the coefficients of f , and is
contained in F . It is called the discriminant of f , and is denoted by Disc(f ) or D(f ). Notice that D(f ) = 0
iff f has a multiple root.
(i) For a quadratic polynomial f (x) = x2 +ax+b = (x−α1 )(x−α2 ), D(f ) = (α2 −α1 )2 = (α2 +α1 )2 −4α1 α2 =
a2 − 4b.
(ii) For a cubic polynomial f (x) = x3 + ax2 + bx + c, D(f ) = a2 b2 + 18abc − 4b3 − 4a3 c − 27c2 . Replacing f
by the cubic g(x) = f (x − a/3)2 = x3 + px + q doesn’t change the discriminant, D(f ) = D(g) = −4p3 − 27q 2 .
(iii) For a quartic polynomial f (x) = x4 +px2 +qx+r, D(f ) = 16p4 r−4p3 q 2 −128p2 r2 +144pq 2 r−27q 4 +256r3 .
7.5.3. Theorem. The Galois group p Gal(f ) of a separable polynomial f ∈ F [x] of degree n is contained in
the alternating group An iff δ = D(f ) ∈ F . If δ 6∈ F , then F (δ) is the quadratic extension of F fixed by
Gal(f ) ∩ An .
Proof. Gal(f ) ≤ An iff all elements of Gal(f ) fix δ, that is, iff δ ∈ F . If this is not so, then Gal(f ) ∩ An is
a subgroup of Gal(f ) of index 2, so fixes a quadratic subextension L of the splitting field of f ; but F (δ)/F
is quadratic and fixed by Gal(f ) ∩ An , so L = F (δ).
2
7.5.4. Let f ∈ F [x] be p a quadratic polynomial, f (x) = x + ax + b, over a field F of characteristic 6= 2.
By Theorem 7.5.3, if p , otherwise Gal(f ) ∼
D(f ) ∈ F , then Gal(f ) is trivial, and f splits in Fp = Z2 and the



splitting field of f is F D(f ) . (And indeed, the roots of f are 21 −a ± D(f ) .)

7.6. The Galois group and solution in radicals of cubics

Let F be a field of characteristic 6= 2, 3. Let f = x3 +a2 x2 +a1 x+a0 ∈ F [x] be a monic irreducible cubic
polynomial, let α1 , α2 , α3 be the roots of f , let K = F (α1 , α2 , α3 ) be the splitting field of f , let G = Gal(f ),
and let D = D(f ).
7.6.1. After replacing x + a2 /3 by x, f takes the form f (x) = x3 + px + q; this operation changes neither
K, nor G, nor D.
7.6.2. G is isomorphic to a subgroup of S3 that acts transitively √ on the set {α1 , α2 , α3 } of the roots of f ;
hence, either G ∼= S3 or G ∼ = A3 ∼= Z3 . We have G ≤ A3 iff D ∈ F , so G ∼ = Z3 in this case, and otherwise
G∼= S3 .
7.6.3. Here are two examples of irreducible cubics over Q and their Galois groups:
x3 − 2 has D = −108, so G ∼ = S3 ;
x3 − 3x + 1 has D = 92 , so G ∼ = Z3 .
7.6.4. In the case F ⊆ R, if all roots of f are real, then D = (α1 − α2 )2 (α1 − α3 )2 (α2 − α3 )2 is positive; if
f has one real and two non-real roots, then D < 0. So, if D < 0 then G ∼ = S3 (and if D > 0 then G ∼ = S3 or
Z3 ).

37
 √
−1+ −3
7.6.5. To find a formula for the roots of f in radicals, adjoin to F a primitive 3rd root of unity ω = :
√ √ 2
replace F by F ( −3) and K by K( −3); assume that f is still irreducible. We have the tower

K
3
√
F ( D)
1 or 2
F
√ √
Since K is a cyclic cubic extension of F ( D), it is radical, K = F (γ) for some γ such that γ 3 ∈ F ( D). γ can
be found as the Lagrange resolvent of one of generators of K, say, of the root α1 of f : γ = α1 + ωα2 + ω 2 α3 .
Let also γ ′ = α1 √ + ω 2 α2 + ωα3 , then γ ′ =√τ (γ) where τ is the transposition (2, 3) (or rather (α √ 2 , α3 )).
We have γ ∈ F ( D), that is, γ 3 = a + b D for some a, b ∈ F . Then (γ ′ )3 = τ (γ 3 ) = a − b D, and
3

so, a = 21 (γ 3 + (γ ′ )3 ) and b = 2√1D (γ 3 − (γ ′ )3 ). γ 3 and (γ ′ )3 are fixed by any even permutation from S3
√
and are switched by any odd permutation, and D changes sign under an odd permutation; so, a and b
3
are symmetric polynomials√ in α1 , α2 , α3 and√can be found: for f (x) = √ x + px + q computations give that
27 3 3 27 3 ′ 3 27 3
a = − 2 q and b = 2 −3, so γ = − 2 q + 2 −3D, (γ ) = − 2 q − 2 −3D,
q √ q √
γ= 3
− 27
2 q+
3
2 −3D and γ′ = 3
− 27
2 q−
3
2 −3D.

Since α1 + α2 + α3 = 0, we get

α1 = 13 (γ + γ ′ ), α2 = 31 (ωγ + ω 2 γ ′ ), α3 = 13 (ω 2 γ + ωγ ′ ).
√ √
3 . . . and γ ′ = 3 . . . are not independent: they must satisfy
(Here, the cubic roots γ =

γγ ′ = α12 + α22 + α32 + (ω + ω 2 )(α1 α2 + α1 α3 + α2 α3 ) = (α1 + α2 + α3 )2 − 3(α1 α2 + α1 α3 + α2 α3 ) = −3p,

so that γ ′ = −3p/γ.) These formulas for the roots of a cubic are called Cardano’s formulas.
7.6.6. Casus irreducibilis. For f ∈ Q[x], even if D > 0 and thus all three roots of f are real, none of them is
expressible by radicals in R only: the radical formulas for each root will necessarily involve
√ non-real complex
numbers. Indeed, assume that a root α of f lies in a tower Kn /Kn−1 / . . . /K1 /Q( D)/Q of real radical
extensions and no root of f is contained in Kn−1 . As we know, any real subextension of a real radical
extension is also radical, so Kn−1 (α)/Kn−1 is radical, and we may assume that Kn = Kn−1 (α). √ Since√f has
no roots in Kn−1 , f is irreducible over Kn−1 , so [Kn : Kn−1 ] = degKn−1 (α) = 3. Since Q(α, D)/Q( D) is
abelian, Kn contains all roots of f and so, Kn /Kn−1 is normal. But then Kn ∋ ω = e2πi/3 6∈ R, contradiction.

7.7. The Galois group and solution in radicals of quartics

Let F be a field of characteristic 6= 2, 3. Let f = x4 + a3 x3 + a2 x2 + a1 x + a0 ∈ F [x] be a monic
irreducible quartic polynomial with roots α1 , α2 , α3 , α4 , let K = F (α1 , α2 , α3 , α4 ) be the splitting field of f ,
let G = Gal(f ), and let D = D(f ).
7.7.1. After replacing x + a3 /4 by x, f takes the form f (x) = x4 + px2 + qx + r; this operation changes
neither K, nor G, nor D.
7.7.2. G is a subgroup of the group S4 acting on the set {α1 , α2 , α3 , α4 } of the roots of f , and the action of
G on this set is transitive. Here is the list of subgroups of S4 that act transitively:
(i) S4 itself, of order 24;
(ii) the alternating group A4 , of order 12;
(iii) three conjugate subgroups H1 = (1, 3, 2, 4), (1, 2) , H2 = (1, 2, 3, 4), (1, 3) , H3 = (1, 2, 4, 3), (1, 4) of
order 8, isomorphic to D8 ;

(iv) the normal subgroup V = 1, (1, 2)(3, 4), (1, 3)(2, 4), (1, 4)(2, 3) , of order 4, isomorphic to the Klein
4-group V4 ∼ = Z22 ;
(v) and three conjugate cyclic subgroups C1 = (1, 3, 2, 4) , C2 = (1, 2, 3, 4) , C3 = (1, 2, 4, 3) , of order 4,
isomorphic to Z4 .

38
7.7.3. The group S4 is solvable and has the normal series 1 ≤ V ≤ A4 ≤ S4 , with A4 /V ∼ = Z3 and
S4 /A4 ∼
= Z2 . Hence, G, as a subgroup√of S4 , has the normal series 1 ≤ (V ∩ G) ≤ (A4 ∩ G) ≤ G, and the
corresponding tower for K is K/L/F ( D)/F where L = Fix(V ∩ G):

1 K
V ∩G L
√ (7.1)
A4 ∩ G F ( D)
G F

7.7.4. To determine L, let θ1 = (α1 + α2 )(α3 + α4 ), θ2 = (α1 + α3 )(α2 + α4 ), θ3 = (α1 + α4 )(α2 + α3 ).

(Another variant is θ1 = α1 α2 + α3 α4 , θ2 = α1 α3 + α2 α4 , θ3 = α1 α4 + α2 α3 .) G permutes the elements θi ,
thus the polynomial
R(x) = (x − θ1 )(x − θ2 )(x − θ3 )

is contained in F [x]. The polynomial R is called the cubic resolvent of f .

Computations show that for f (x) = x4 + px2 + qx + r, R(x) = x3 − 2px2 + (p2 − 4r)x + q 2 .
Lemma. The discriminant D(R) of the cubic resolvent R of f equals the discriminant D(f ) of f .
Proof. θ1 − θ2 = α1 α3 + α1 α4 + α2 α3 + α2 α4 − α1 α2 − α1 α4 − α3 α2 − α3 α4 = α1 α3 + α2 α4 − α1 α2 − α3 α4
= (α1 − α4 )(α3 − α2 ), similarly θ1 − θ3 = −(α1 − α3 )(α2 − α4 ) and θ2 − θ3 = −(α1 − α2 )(α3 − α4 ). Hence,
Y Y
D(R) = (θi − θj )2 = (αi − αj )2 = D(f ).
1≤i<j≤3 1≤i<j≤4

Hence, if f is separable, then R is separable and θi are all distinct. The stabilizer of θ1 in G is the group
H1 ∩ G, of θ2 is H2 ∩ G, and of θ3 is H3 ∩ G. Since H1 ∩ H2 ∩ H3 = V , in the diagram (7.1) we have that
L = F (θ1 , θ2 , θ3 ) = Fix(V ∩ G).
7.7.5. Theorem. Let us interpret G as a subgroup of S4 and use notation from 7.7.2. Let R be the cubic
resolvent of f .
√
(i) If R is irreducible and D 6∈ F , then G = S4 .
√
(ii) If R is irreducible and D ∈ F , then G = A4 .
(iii) If R splits completely in F , then G = V (and is isomorphic to V4 ).
√
(iv) If R splits over F into a linear and quadratic polynomials and f is irreducible over F ( D), then G is
one of the groups Hi (and is isomorphic to D8 ).
√
(v) If R splits over F into a linear and quadratic polynomials and f is reducible over F ( D), then G is one
of the groups Ci (and is isomorphic to Z4 ).
√
Proof. (i) In this case Gal(L/F ) = G/(V ∩ G) ∼ = S . Also, since D 6∈ F , G 6≤ A . From the list in 7.7.2,
3 4
only S4 has these properties.
√
(ii) In this case Gal(L/F ) = G/(V ∩ G) ∼ = Z3 , and since D ∈ F , G ≤ A4 . From the list in 7.7.2, only A4
has these properties.
(iii) In this case Gal(L/F ) = G/(V ∩ G) is trivial, so G ≤ V , so G = V .
(iv,v) In these cases G fixes one of θi , so G ≤ Hi , so G = Hi or Ci . The group Hi ∩ A4 = V acts√transitively
on the set {α1 , α2 , α3 , α4 }, whereas Ci ∩ A4 ∼
= Z2 does not. Hence if f is irreducible over F ( D) (and so
all αi are conjugate over this field), then G = Hi ; and if not, then G = √ Ci . (Notice also that Ci ∩ A4 is a
product of two transpositions, thus in the case G = Ci , f splits over F ( D) into a product of two quadratic
polynomials.)
7.7.6. When f is “biquadratic”, f = x4 + px2 + r, the cubic resolvent R(x) = (x2 − 2px + p2 − 4r)x is always
reducible, so G ∼
6 S4 or A4 . (We knew this!)
=

39
7.7.7. Here are examples of irreducible quartics, over Q, representing all isomorphism types of Galois groups:
x4 − x − 1 has D = −283, R = x3 + 4x + 1 is irreducible, G ∼ = S4 ;
x4 + x + 1 has D = 229, R = x3 − 4x + 1 is irreducible, G ∼ = S4 ;
x4 + 8x + 12 has D = 5762 , R = x3 − 16x + ∼
√ 16 is√irreducible, G = A4 ;
x − 10x + 1 (the minimal polynomial of 2 + 3) has D = −260744, R = x(x + 8)(x + 12), G ∼
4 2
= V4 ;
x4 + 36x + 63 has D = 43202 , R = (x + 18)(x − 6)(x − 12), G ∼ = V√ 4;
x4 − 2 has D = −211 , R = (x2 + 8)x, x4 − 2 is irreducible over Q( −2), G ∼ = D8 ;
x4 + 3x + 3 has D = 21 · 152 , R = (x − 3)(x2 + 3x − 3), G ∼ = D8 ;
x4 − 4x2 + 2 has D = −19 · 28 , R = (x2 + 8x + 8)x, G ∼ = Z4 ;
x4 + 5x + 5 has D = 5 · 552 , R = (x + 5)(x2 − 5x + 5), G ∼ = Z4 .
7.7.8. The roots θi of the cubic resolvent R of f are expressible in radicals with the help of Cardano’s formulas.
Assume that f√(x) = x4 +px2 +qx+r, then α1 +α2 +α3 +α4 = 0, and θ1 = (α1 +α2 )(α3 +α4 ) = −(α1 +α2 )2 ,
so α1 + α2 = −θ1 . Using this and similar equalities, we now get

p p
p p p p

1 1
α1 = 2 −θ1 + −θ3 ,−θ2 + α2 = 2 −θ1 − −θ2 − −θ3 ,
p p p
p p p

1 1
α3 = 2 − −θ1 + −θ2 − −θ3 , α4 = 2 − −θ1 − −θ2 + −θ3 .

7.8. Computation of Galois groups

7.8.1. There exist ingeneous and effective algorithms for computing Galois groups; here is one of them
(which belongs to van der Waerden). Let K/F be a Galois extension, let G = Gal(K/F ). Let {α1 , . . . , αn }
be a G-invariant set of (pairwise distinct) generators of K. (If K is a splitting field of f ∈ F [x] it is
natural to take as these generators the roots of f .) Let’s identify G with a subgroup of Sn acting on
this set. Let Fe = F (x1 , . . . , xn ) and K
e = K(x1 , . . . , xn ) = Fe (α1 , . . . , αn ), where xi are free variables;
e e e e
consider the extension K/F . K/F is Galois with Gal(tK/tF ) ≤ Gal(K/F ); since [K e : Fe ] = [K : F ],
Gal(tK/tF ) = Gal(K/F ).
For each σ ∈ Sn let rσ = ασ(1) x1 + · · · + ασ(n) xn ∈ K[x];e in particular, r1 = α1 x1 + · · · + αn xn . The
group Sn acts on the set {rσ , σ ∈ Sn } by permuting αi :

ρ(rσ ) = αρσ(1) x1 + · · · + αρσ(n) xn = rρσ , ρ ∈ Sn .

(Or: Sn acts on K e by permuting xi , under which action ρ(rσ ) = Pn ασ(i) xρ(i) = Pn ασρ−1 (i) xi = rσρ−1 .)
i=1 i=1
The conjugates of r1 in K[x] e are the elements σ(r1 ) = rσ with σ ∈ G, so r1 has |G| distinct conjugates (and
so, Ke = Fe (r1 )). Let g(x) = Q
σ∈Sn (x−rσ ); since g is invariant under all permutations of α1 , . . . , αn , we have
g ∈ Fe [x]. Let g = g1 · · · gk be the factorization of g into a product of irreducible factors in Fe [x]. (Finding
this factorization is the computational part of the algorithm.) W.l.o.g., let xQ− r1 be a factor of g1 , so that g1
is the minimal polynomial of r1 . Let H be the subset of Sn such that g1 = σ∈H (x − rσ ); then {rσ , σ ∈ H}
is just the set of all the conjugates of r1 , hence, H = G. We can also interprete H as follows: Sn acts on the
set {g1 , . . . , gk }, and G = H is the stabilizer of g1 under this action.
7.8.2. The Galois group of an integer polynomial can sometimes be found by reducing the polynomial modulo
distinct primes. Let f ∈ Z[x] be a monic separable polynomial; let D = D(f ), then D is an integer. Let p
be a prime integer not dividing D; consider the polynomial f¯ = f mod p ∈ Fp [x].

Dedekind’s theorem. As groups of permutations of the roots of f and the corresponding roots of f¯,
Gal(f¯/Fp ) is a subgroup of Gal(f /Q).

(If it is hard to calculate D to check if p / D, this calculation can be avoided: we have D = 0 mod p iff f¯ is
inseparable.)

40
Proof. Let α1 , . . . , αn be the roots of f . Put R = Z[α1 , . . . , αn ]. p is not a unit in R. (Indeed, R is a finitely
generated Z-module, as Z is a ED any its submodule must also be finitely generated, but the ring Z[p−1 ] is
not finitely generated as a Z-module.) So, p is contained in a maximal ideal P of R. Let L = R/P , then
L is a field in which p = 0, so L is a finite extension of Fp . L is generated by ᾱ1 , . . . , ᾱn (the images of
α1 , . .. , αn ) and is a splitting field of the polynomial f¯ (the image of f in Fp [x]). Let G = Gal(f /Q) and let
H = ϕ ∈ G : ϕ(P ) = P . Then H acts on L, so we have a homomorphism η: H −→ Gal(L/Fp ) = Gal(f¯).
Since D(f¯) = D(f ) mod p 6= 0, f¯ is separable, and the elements ᾱi are distinct. Since every element of H is
defined by its action on αi , and so on ᾱi , η is injective and we identify H with a subgroup of Gal(f¯). To
prove that H = Gal(f¯) it suffices to show that |H| ≥ | Gal(f¯)|.
Let α ∈ R be such that its image ᾱ ∈ L generates L, L = Fp (ᾱ). Any two of the ideals ϕ(P ),
ϕ ∈ G, either coincide or are comaximal, thus, by the Chinese remainder theorem, there is β ∈ R such that
β = α mod P and β = 0 mod ϕ−1 (P ) for all ϕ ∈ G \ H. We then haveQϕ(β) = ϕ(α) mod ϕ(P ) = ϕ(α) mod P
for all ϕ ∈ H and ϕ(β) = 0 mod P for all ϕ ∈ G \ H. Let g(x) = ϕ∈G (x − ϕ(β)), then g ∈ Q[x]; after
replacing βQby dβ (and α by dα) for a suitable d ∈ N we may assume that g ∈ Z[x]. The image of g in Fp [x]
is x|G|−|H| ϕ∈H (x − ϕ(ᾱ)) and is divisible by mᾱ , so |H| ≥ degFp (ᾱ) = [L : Fp ] = | Gal(f¯)|.

7.8.3. As the group Gal(f¯/Fp ) is cyclic and transitive on each set of conjugate roots of f¯, we get:
Theorem. For each prime integer p not dividing D, if h1 · · · hk is the factorization of f mod p into irreducible
factors with ni = deg hi , i = 1, . . . , k, then Gal(f /Q) contains an element of the cycle type (n1 , . . . , nk ).
7.8.4. Examples. (i) Let f = x4 + 3x2 − 3x − 2 ∈ Z[x]. Let G = Gal(f /Q). Since f = x4 − 2 =
(x2 +x+2)(x2 +2x+2) mod 3, G contains a permutation of the cycle type (2, 2). Since f = x(x3 +x+1) mod 2,
f contains a permutation of cycle type (1, 3), that is, a 3-cycle. It also follows that f is irreducible: f hasno
root since it doesn’t have a root modulo 3, and f is not a product of two quadrics since it is not such a
product modulo 2. Hence, G ∼ = S4 or A4 ; since (as we can compute) D(f ) < 0, we get that G ∼ = S4 .
(ii) Let f = x6 + x4 + x + 3 ∈ Z[x], let G = Gal(f /Q). Then f = (x + 1)(x2 + x + 1)(x3 + x + 1) mod 2,
f = (x + 6)(x5 + 5x4 + 4x3 + 9x2 + x + 6) mod 11, f = (x2 + 8x + 1)(x2 + 9x + 10)(x2 + 9x + 12) mod 13. It
follows that f is irreducible: if f = f1 f2 , then taking f modulo 11 we see that f must have a linear factor,
but this disagrees with the decomposition of f modulo 13. By Theorem 7.8.3, G contains permutations ρ of
cycle type (1, 2, 3) and σ of cycle type (1, 5); σ is a 5-cycle and ρ3 is a transposition. So, G is a transitive
subgroup of S6 that contains a 5-cycle and a transposition; it is easy to see that such a subgroup must
coincide with S6 .
7.8.5. We can use Theorem 7.8.3 to construct, for every n ∈ N, a polynomial f ∈ Z[x] with Gal(f /Q) ∼ = Sn .
Let f2 ∈ Z2 [x] be an irreducible (monic) polynomial of degree n; let f3 ∈ Z3 [x] be a separable monic
polynomial of degree n which is the product of a linear polynomial and an irreducible polynomial of degree
n − 1; and let f5 ∈ Z5 [x] be a separable monic polynomial of degree n that is the product of an irreducible
quadratic polynomial, an irreducible polynomial of an odd degree n − 2 or n − 3, and a linear polynomial in
the case n is even and ≥ 2 (in the case n = 2, f5 is just a quadratic polynomial). Let f ∈ Z[x] be a monic
polynomial of degree n such that f = f2 mod 2, f = f3 mod 3, and f = f5 mod 5; it exists by the Chinese
remainder theorem. (Say, f = −15f˜2 + 10f˜3 + 6f˜5 , where f˜p ∈ Z[x] are such that fp = f˜p mod p, p = 2, 3, 5.)
Let G = Gal(f /Q). Since f2 is irreducible, f is irreducible. By Theorem 7.8.3 with p = 3, G contains an
(n − 1)-cycle, and with p = 5, G contains a product τ ρ of a transposition τ and a m-cycle ρ with an odd
m; then (τ ρ)m = τ is also in G. A simple lemma says that if a subgroup of Sn is transitive and contains an
(n − 1)-cycle and a transposition then it coincides with Sn ; so, G = Sn .
7.8.6. The discriminant of a polynomial (which is so important for determining its Galois group) can some-
times be found with the help of a resultant. For two polynomials f, g ∈ F [x], Q f (x) = a(x − α1 ) · · · (x − αn )
and g(x) = b(x − β1 ) · · · (x − βm ), the resultant of f and g is Res(f, g) = am bn i,j (αi − βj ). Since Res(f, g)
is invariant under any permutation of αi -s and of βj -s, we have Res(f, g) ∈ F . Clearly, Res(f, g) = 0 iff f
and g have a common root.
Qn Qm
Directly from the definition, Res(f, g) = am i=1 g(αi ) = (−1)nm bn j=1 f (βj ). In particular, if f is
linear, f (x) = a(x − α), then Res(f, g) = am g(α).
7.8.7. We have:

41
Lemma. If f is a monic polynomial of degree n, then D(f ) = (−1)n(n−1)/2 Res(f, f ′ ).
Qn Pn Q Q
Proof. Let f (x) = i=1 (x − αi ), then f ′ (x) = i=1 j6=i (x − αj ), and for every i, f ′ (αi ) = j6=i (αi − αj ).
So,
n
Y n Y
Y Y Y
Res(f, f ′ ) = f ′ (αi ) = (αi − αj ) = (αi − αj ) = (−1)n(n−1)/2 (αi − αj )2 = (−1)n(n−1)/2 D(f ).
i=1 i=1 j6=i i6=j i<j

Qm
7.8.8. From the identity Res(f, g) = (−1)nm bn j=1 f (βj ) it follows that if f1 is a polynomial of degree n1
such that f1 = f mod g, then Res(f, g) = (−1)(n−n1 )m bn−n1 Res(f1 , g).
7.8.9. Let’s use 7.8.8 to compute the discriminant D(f ) for polynomials of the form f = xn + ax + b. We
have f ′ (x) = nxn−1 + a, so f (x) = n1 xf ′ (x) + n−1
n ax + b, so

2

D(f ) = (−1)n(n−1)/2 Res(f, f ′ ) = (−1)n(n−1)/2 nn−1 (−1)(n−1) Res n−1n ax + b, f
′

n−1 ′

n−1 
n−1 
= (−1)(n+2)(n−1)/2 nn−1 n−1n a
bn
f − a(n−1) = (−1)(n+2)(n−1)/2 nn−1 n−1
n a
bn
n − a(n−1) +a
= (−1)n(n−1)/2 nn bn−1 + (−1)(n+2)(n−1)/2 (n − 1)n−1 an .

For n = 5 this is D(f ) = 55 b4 + 44 a5 .

8. Introduction to transcendental extensions

The theory of transcendental extensions resembles the theory of modules over integral domains, with
linear dependence replaced by algebraic (polynomial) dependence. Let F be a field.
8.0.1. A set A of elements of an extension K/F is said to be algebraically dependent over F if for some
α1 , . . . , αk ∈ A there is a nonzero polynomial f ∈ F [x1 , . . . , xk ] such that f (α1 , . . . , αk ) = 0; A is said to be
algebraically independent otherwise.
8.0.2. If a set A is algebraically dependent, f (α1 , . . . , αk ) = 0 with f ∈ F [x1 , . . . , xk ] \ F [x1 , . . . , xk−1 ], then
αk is algebraic over F (α1 , . . . , αk−1 ). The converse is also
true, and we see that A is algebraically dependent
iff there is α ∈ A such that α is algebraic over F A \ {α} .
8.0.3. Let K/F be an extension; a maximal algebraically independent over F subset B of K is called a
transcendence base of K/F . A set B ⊆ K is a transcendence base of K/F iff K/F (B) is an algebraic
extension.
8.0.4. Using Zorn’s lemma, we can easily prove:
Theorem. For any extension K/F , a transcendence base of K/F exists.
8.0.5. We can also prove that all transcendence bases of K/F have the same cardinality. (We need “a
polynomial” analogue of the replacement theorem to do this.) The cardinality of a transcendence base of
K/F is called the transcendence degree of K/F .
8.0.6. An extension K/F is said to be purely transcendental if it has a transcendence base B such that
K = F (B). In this case, K is isomorphic to the field of rational functions in variables xα , α ∈ B.
8.0.7. We obtain that every extension K/F is a tower, K/L/F , where L/F is purely transcendental and
K/L is algebraic.

42