finite dimensional Galois extension
In this section, we are going to prove the fundamental theorem for finite dimensional Galois extension.
Let F/K be an field extension, we define the Galois group of F over K, denoted GalF/K or GF/K or AutK(F ), as
GalF/K := {σ|σ ∈ AutF, σ|K = 1K}.
It’s clear that for σ ∈ GalF/K and u ∈ F algebraic over K with mini- mal polynomial p(x), then σ(u) satisfies the same minimal polynomial.
On the other hand, if F/K is normal, let u, v be two elements having the same minimal polynomial p(x), then we claim that there is an σ ∈ GalF/K such that σ(u) = v. To see this, we fix an algebraic closure K containing F . There is an K-isomorphism σ0 : K(u) → K(v) which extends to an embedding σ : F → K . Since F is normal over K, one has σ(F ) ⊂ F . And hence σ ∈ AutF .
Example 0.1. Consider the field F := Q(√3
2, ω) which is a splitting field of x3−2 over Q. Thus it’s normal over Q. One can check that the Galois group GalF/Q is generated by σ, τ that σ(√3
2) = √3
2ω, σ(ω) = ω, and τ (√3
2) = √3
2, τ (ω) = ω2. It’s easy to check that GalF/Q∼= S3. Example 0.2. Consider the field F := Q(√3
2) over Q. Then it’s easy to check that GalF/Q = {1F}.
There is a natural correspondence between subgroups of Galois groups and intermediate fields. To be precise, fix an extension F/K. Let H < G := GalF/K be a subgroup. One can define
H0 := {u ∈ F |σ(u) = u, ∀σ ∈ H}.
It’s clear that this is a field. On the other hand, given and intermediate field L such that K ⊂ L ⊂ F , then one can define
L0 := {σ ∈ GalF/K|σ(u) = u, ∀u ∈ L} = {σ ∈ GalF/K|σ|L= 1L}.
It’s easy to check the following properties:
Proposition 0.3. Let F/K be an extension with Galois group G. Let L be an intermediate field, i.e. K ⊂ L ⊂ F , and H < G is a subgroup.
(1) F0 = {1F}, K0 = G, and {1F}0 = F . (2) For any L, one has L ⊂ L00, L0 = L000. (3) For any H, one has H < H00, H0 = H000.
(4) For any intermediate fields L ⊂ M, one has M0 < L0. (5) For any subgroups J < H, one has H0 ⊂ J0.
1
Proof. Most of the proof follows directly from the definition. We only sketch the proof for L0 = L000.
By L ⊂ L00 and (4), one has
(L00)0 < L0. On the other hand, by (5), one has
L0 < (L0)00.
We are done. ¤
Proposition 0.4. There is a one-to-one correspondence between {L|K ⊂ L ⊂ F, L00 = L} ↔ {H|H < G, H00 = H}.
Proof. The correspondence is given by L 7→ L0 (or H 7→ H0).
To show the injective, one sees that if L01 = L02, then L1 = L001 = L002 = L2.
For any H with H00 = H, we take L = H0, then H = L0. It suffices to check that L00 = L. This follows from the fact that H000 = H0. ¤ In the proposition, one might expect that G0 = K. However, this is not always the case (see e.g. Example 2). For extension with this property, we call it Galois. It turns out that this naive definition is a very delicate one which leads to some nice properties.
Definition 0.5. An extension F/K is said to be Galois if (GalF/K)0 = K.
Theorem 0.6 (Fundamental theorem of finite dimensional Galois ex- tension). Let F/K be a finite dimensional Galois extension with Galois group G, then
(1) There is an one-to-one correspondence between {L|K ⊂ L ⊂ F } ↔ {H|H < G}.
(2) The corresponding degree are equal. That is, if K ⊂ L ⊂ M ⊂ F , then [M : L] = [L0 : M0]. And if J < H < G, then [H : J] = [J0 : H0].
(3) An intermediate field E is Galois over K if and only if E0C G.
And in this case, GalE/K ∼= G/E0. Proof. Step 1. [M : L] ≥ [L0 : M0].
We prove the case that M = L(u) for some u ∈ M and by induction on [M : L], we are done. Suppose now that M = L(u) and let p(x) be the minimal polynomial of u over L. Let S be the set of roots of p(x) in F . Then one has a map
Φ : L0 → S, σ 7→ σ(u).
One can check that Φ induces an injective map L0/M0 → S. Hence one has
[L0 : M0] = |L0/M0| ≤ |S| ≤ deg(p(x)) = [M : L].
Step 2. [H : J] ≥ [J0 : H0].
Let n = [H : J]. Suppose on the contrary that there are n + 1 elements u1, ..., un+1 ∈ J0 linearly independent over H0.
We consider the equation Pn+1
i=1 uixi = 0 in F Consider now a set of representative of H/J, denoted {e = σ1, ..., σn}. By applying σi to the above equation. Then one has a system of linear equations in F .
(∗)
σ1(u1)x1+ σ1(u2)x2+ ... + σ1(un+1)xn+1 = 0 σ2(u1)x1+ σ2(u2)x2+ ... + σ2(un+1)xn+1 = 0 ...
σn(u1)x1+ σn(u2)x2 + ... + σn(un+1)xn+1 = 0
Pick a solution in F with smallest number of non-zero ai’s, may assume it’s (a1, ..., as, 0..., 0) and a1 = 1.
If there is an τ ∈ H such that τ (a2) 6= a2, then by applying τ to the system (∗), one get the same system of equations with a solution (τ (a1), τ (a2), ..., τ (as), 0, ..., 0 . Hence
(a1, ..., as, 0..., 0) − (τ (a1), τ (a2), ..., τ (as), 0, ..., 0) = (0, a2− τ (a2), ..., 0) is a non-zero solution of smaller length. This is the required contradic- tion.
To find τ . We look at u1a1 + ... + usas = 0. Since {u1, ..., us} is independent over H0, not all a1 is in H0. We may assume that a2 6∈ H0. Hence there is a τ ∈ H such that τ (a2) 6= a2. We are done.
Step 3. We show that every intermediate field L, L00 = L. And every subgroup H < G, H00 = H.
By Step 1, one has
[L00 : K] = [L00 : K00] ≤ [K0 : L0] ≤ [L : K],
however, one has L ⊂ L00. Thus one has L = L00. Similarly, one can prove that H00= H by considering [H00: {1F}].
Step 4. [M : L] = [L0 : M0] and [H : J] = [J0 : H0].
This follows from [M : L] = [M : K]/[L : K] = [K0 : M0]/[K0 : L0] = [L0 : M0]. And the other one is similar.
Step 5. F/K is normal and separable.
Given u ∈ F , with minimal polynomial p(x) over K. As in the proof of Step 1. One has [K(u)0 : K0] ≤ |S| ≤ deg(p(x)) = [K(u) : K]. By Step 4, they are equalities. In particular, every root of p(x) is in F and there is no multiple roots. Thus F is normal and separable over K.
Step 6. If N C G, then N0 is stable. That is, for all σ ∈ G, σ(N0) ⊂ N0 (indeed = N0).
Since N C G, for all σ ∈ G and for all τ ∈ N, one has σ−1τ σ ∈ N.
Thus, σ−1τ σ(N0) = N0. It follows that τ σ(N0) = σ(N0), for all τ ∈ N.
Hence σ(N0) is fixed by all N and thus σ(N0) ⊂ N0.
Step 7. If E is a stable intermediate subfield. Then the restriction map GalF/K → GalE/K is well-defined and surjective.
Since E is stable, then σ|E ∈ GalE/K for any σ ∈ GalF/K. Moreover, let τ ∈ GalE/K, by the extension theorem, there is an extension τ : F → K. Since F is normal over K, τ is in fact an automorphism of F .
Step 8. If an intermediate field E is stable, then E/K is Galois.
To see this, it suffices to show that for any u ∈ E −K, there is an σ ∈ GalE/K such that σ(u) 6= u. Fix any F 3 v 6= u with the same minimal polynomial as u. There is an K-isomorphism σ0 : K(u) → K(v) such that σ(u) = v. σ can be extended to an embedding σ : F → K, which gives an automorphism of F . The restriction σ = σ|E gives an automorphism of E that σ(u) 6= u.
Step 9. If E/K is Galois, then E is stable.
One first notices that E/K is normal. For every σ ∈ GalF/K, σ gives an embedding σ|E : E → K. Since E/K is normal, σ|E is an automorphism of E. And hence E is stable under the Galois group GalF/K action.
Step 10. If E is stable, then E0 is normal.
This can be checked directly. For all σ ∈ G and τ ∈ E0 and for all u ∈ E,
σ−1τ σ(u) = σ−1τ (σ(u)) = σ−1σ(u) = u,
since σ(u) ∈ E. Therefore, σ−1τ σ ∈ E0. ¤ Remark 0.7. Some of the result we proved still true in a more general setting. We list some here:
(1) If F/K is an extension, and an intermediate field E is stable, then E0C GalF/K.
(2) Let F/K be an extension. If N C GalF/K, then H0 is stable.
(3) If F/K is Galois, and E is a stable intermediate field, then E is Galois over K. (finite-dimensional assumption is unnecessary here)
(4) An intermediate field E is algebraic and Galois over K, then E is stable.
We conclude this section with the following theorem concerning the relation between Galois extension, normal extension and splitting fields.
Theorem 0.8. Let F/K be an extension, then the following are equiv- alent
(1) F is algebraic and Galois over K.
(2) F is separable over K and F is a splitting field over K of a set S of polynomials.
(3) F is a splitting field of separable polynomials in K[X].
(4) F/K is normal and separable.
Proof. Fix u ∈ F with minimal polynomail p(x) over K. Let {u = u1, ..., ur} be distinct roots of p(x) in F . For any σ, then σ permutes {u = u1, ..., ur}. Thus f (x) := Qr
i=1(x − ui) is invariant under σ.
Hence f (x) ∈ K[x]. It follows that f (x) = p(x). This proved that (1) ⇒ (2), (3), (4).
One notices that (2) ⇔ (4). Thus it remains to show that (2) ⇒ (3), and (3) ⇒ (1).
For (2) ⇒ (3), let f (x) ∈ S and let g(x) be an monic irreducible component of f (x). Since f (x) splits in F , it’s clear that g(x) is an minimal polynomial of some element in F . Moreover, since F/K is separable, g(x) is separable. One sees that F is in fact a splitting field of such g(x)’s.
For (3) ⇒ (1), we first note that F/K is algebraic since F is a split- ting field. We shall prove that (4) ⇒ (1). The implication (3) → (4) follows from a general fact about separable extension that an algebraic extension F/K is separable if F is generated by separable elements.
To this end, pick any u ∈ F − K, with minimal polynomial p(x) of degree ≥ 2 and separable. Hence there is a different root, say v, of p(x) in F . It’s natural to consider the K-isomorphism σ : K(u) → K(v).
Which can be extended to ¯σ : F → K. Since F is normal, ¯σ is an automorphism of F , hence in GalF/K sending u to v 6= u. So F/K is Galois.
¤