3.3 The fundamental theorem of Galois theory

Let EE be an extension of FF. A subextension of E/FE/F is an extension M/FM/F with MEM\subset E, i.e., a field MM with FMEF\subset M\subset E. When EE is Galois over FF, the subextensions of E/FE/F are in one-to-one correspondence with the subgroups of Gal(E/F)\operatorname{Gal}(E/F). More precisely, there is the following statement.

Theorem 3.16 (Fundamental theorem of Galois theory)

Let EE be a Galois extension of FF with Galois group GG. The map HEHH\mapsto E^{H} is a bijection from the set of subgroups of GG to the set of subextensions of E/FE/F,

{subgroups H of G}1:1{subextensions FME},\{\text{subgroups }H\text{ of }G\}\overset{1\colon 1}{\leftrightarrow}\{\text{% subextensions }F\subset M\subset E\},

with inverse MGal(E/M)M\mapsto\operatorname{Gal}(E/M). Moreover,

  1. (a)

    the correspondence is inclusion-reversing: H1H2EH1EH2;H_{1}\supset H_{2}\iff E^{H_{1}}\subset E^{H_{2}};

  2. (b)

    indexes equal degrees: (H1:H2)=[EH2:EH1](H_{1}\colon H_{2})=[E^{H_{2}}\colon E^{H_{1}}];

  3. (c)

    σHσ1σM\sigma H\sigma^{-1}\leftrightarrow\sigma M, i.e., EσHσ1=σ(EH)E^{\sigma H\sigma^{-1}}=\sigma(E^{H}); Gal(E/σM)=σGal(E/M)σ1.\operatorname{Gal}(E/\sigma M)=\sigma\operatorname{Gal}(E/M)\sigma^{-1}.

  4. (d)

    HH is normal in GEHG\iff E^{H} is normal (hence Galois) over FF, in which case

    Gal(EH/F)G/H.\operatorname{Gal}(E^{H}/F)\simeq G/H.

Proof.

For the first statement, we have to show that HEHH\mapsto E^{H} and MGal(E/M)M\mapsto\operatorname{Gal}(E/M) are inverse maps. Let HH be a subgroup of GG. Then, Corollary 3.5 shows that Gal(E/EH)=H\operatorname{Gal}(E/E^{H})=H. Let M/FM/F be a subextension. Then EE is Galois over MM by (3.13), which means that EGal(E/M)=ME^{\operatorname{Gal}(E/M)}=M\,.

(a) We have the obvious implications,

H1H2EH1EH2Gal(E/EH1)Gal(E/EH2).H_{1}\supset H_{2}\implies E^{H_{1}}\subset E^{H_{2}}\implies\operatorname{Gal% }(E/E^{H_{1}})\supset\operatorname{Gal}(E/E^{H_{2}}).

As Gal(E/EHi)=Hi\operatorname{Gal}(E/E^{H_{i}})=H_{i}, this proves (a).

(b) Let HH be a subgroup of GG. According to 3.11b,

(Gal(E/EH):1)=[E:EH].(\operatorname{Gal}(E/E^{H})\colon 1)=[E\colon E^{H}].

This proves (b) in the case H2=1H_{2}=1, and the general case follows, using that

(H1:1)\displaystyle(H_{1}\colon 1)
=(H1:H2)(H2:1)\displaystyle\overset{\phantom{1.20}}{=}(H_{1}\colon H_{2})(H_{2}\colon 1)\quad
[E:EH1]\displaystyle[E\colon E^{H_{1}}]
=1.20[E:EH2][EH2:EH1].\displaystyle\overset{\text{\ref{ef10}}}{=}[E\colon E^{H_{2}}][E^{H_{2}}\colon E% ^{H_{1}}].

(c) For τG\tau\in G and αE\alpha\in E,

τα=αστσ1(σα)=σα.\tau\alpha=\alpha\iff\sigma\tau\sigma^{-1}(\sigma\alpha)=\sigma\alpha.

Therefore, τ\tau fixes MM if and only if στσ1\sigma\tau\sigma^{-1} fixes σM\sigma M\,, and so σGal(E/M)σ1=Gal(E/σM)\sigma\operatorname{Gal}(E/M)\sigma^{-1}=\operatorname{Gal}(E/\sigma M). This shows that σGal(E/M)σ1\sigma\operatorname{Gal}(E/M)\sigma^{-1} corresponds to σM.\sigma M.

(d) Let HH be a normal subgroup of GG. Because σHσ1=H\sigma H\sigma^{-1}=H for all σG\sigma\in G, we must have σEH=EH\sigma E^{H}=E^{H} for all σG\sigma\in G, i.e., the action of GG on EE stabilizes EHE^{H}. We therefore have a homomorphism

σσ|EH:GAut(EH/F)\sigma\mapsto\sigma|E^{H}\colon G\rightarrow\operatorname{Aut}(E^{H}/F)

whose kernel is HH. As (EH)G/H=F(E^{H})^{G/H}=F, we see that EHE^{H} is Galois over FF (by Theorem 3.9) and that G/HGal(EH/F)G/H\simeq\operatorname{Gal}(E^{H}/F) (by 3.11b).

Conversely, suppose that MM is normal over FF, and let α1,,αm\alpha_{1},\ldots,\alpha_{m} generate MM over FF. For σG\sigma\in G, σαi\sigma\alpha_{i} is a root of the minimal polynomial of αi\alpha_{i} over FF, and so lies in MM. Hence σM=M\sigma M=M, and this implies that σHσ1=H\sigma H\sigma^{-1}=H (by (c)).

Remark 3.17

Let E/FE/F be a Galois extension, so that there is an order reversing bijection between the subextensions of E/FE/F and the subgroups of GG. From this, we can read off the following results.

(a) Let M1,M2,,MrM_{1},M_{2},\ldots,M_{r} be subextensions of E/FE/F, and let HiH_{i} be the subgroup corresponding to MiM_{i} (i.e., Hi=Gal(E/Mi)H_{i}=\operatorname{Gal}(E/M_{i})). Then (by definition) M1M2MrM_{1}M_{2}\cdots M_{r} is the smallest field containing all MiM_{i}; hence it must correspond to the largest subgroup contained in all HiH_{i}, which is Hi\bigcap H_{i}. Therefore

Gal(E/M1Mr)=H1Hr.\operatorname{Gal}(E/M_{1}\cdots M_{r})=H_{1}\cap...\cap H_{r}.

(b) Let HH be a subgroup of GG and let M=EHM=E^{H}. The largest normal subgroup contained in HH is N=σGσHσ1N=\bigcap\nolimits_{\sigma\in G}\sigma H\sigma^{-1} (see GT, 4.1), and so ENE^{N} is the smallest normal extension of FF containing MM. Note that, by (a), ENE^{N} is the composite of the fields σM\sigma M. It is called the normal, or Galois, closure of MM in EE.

Proposition 3.18

Let EE and LL be extensions of FF contained in some common field. If E/FE/F is Galois, then EL/LEL/L and E/ELE/E\cap L are Galois, and the map

σσ|E:Gal(EL/L)Gal(E/EL)\sigma\mapsto\sigma|E\colon\operatorname{Gal}(EL/L)\rightarrow\operatorname{% Gal}(E/E\cap L)

is an isomorphism.

Proof.

Because EE is Galois over FF, it is the splitting field of a separable polynomial

fF[X]f\in F[X]. Then ELEL is the splitting field of ff over LL, and EE is the splitting field of ff over ELE\cap L. Hence EL/LEL/L and E/ELE/E\cap L are Galois. Every automorphism σ\sigma of ELEL fixing the elements of LL maps roots of ff to roots of ff, and so σE=E\sigma E=E. There is therefore a homomorphism

σσ|E:Gal(EL/L)Gal(E/EL).\sigma\mapsto\sigma|E\colon\operatorname{Gal}(EL/L)\rightarrow\operatorname{% Gal}(E/E\cap L)\text{.}

If σGal(EL/L)\sigma\in\operatorname{Gal}(EL/L) fixes the elements of EE, then it fixes the elements of ELEL, and hence is the identity map. Thus, σσ|E\sigma\mapsto\sigma|E is injective. If α\alpha\in EE is fixed by all σGal(EL/L)\sigma\in\operatorname{Gal}(EL/L), then αEL\alpha\in E\cap L. By Corollary 3.5,

EL{EL}E{E}L{L}EL{E\cap L}F{F}====

this implies that the image of σσ|E\sigma\mapsto\sigma|E is Gal(E/EL)\operatorname{Gal}(E/E\cap L).

Corollary 3.19

Suppose, in the proposition, that LL is finite over FF. Then

[EL:F]=[E:F][L:F][EL:F].[EL\colon F]=\frac{[E\colon F][L\colon F]}{[E\cap L\colon F]}\text{.}

Proof.

According to Proposition 1.20,

[EL:F]=[EL:L][L:F],[EL\colon F]=[EL\colon L][L\colon F],

but

[EL:L]=3.18[E:EL]=1.20[E:F][EL:F].[EL\colon L]\overset{\ref{ft18f}}{=}[E\colon E\cap L]\overset{\ref{ef10}}{=}% \frac{[E\colon F]}{[E\cap L\colon F]}\text{.}

Proposition 3.20

Let E1E_{1} and E2E_{2} be extensions of FF contained in some common field. If E1E_{1} and E2E_{2} are Galois over FF, then E1E2E_{1}E_{2} and E1E2E_{1}\cap E_{2} are Galois over FF, and the map

σ(σ|E1,σ|E2):Gal(E1E2/F)Gal(E1/F)×Gal(E2/F)\sigma\mapsto(\sigma|E_{1},\sigma|E_{2})\colon\operatorname{Gal}(E_{1}E_{2}/F)% \rightarrow\operatorname{Gal}(E_{1}/F)\times\!\operatorname{Gal}(E_{2}/F)

is an isomorphism of Gal(E1E2/F)\operatorname{Gal}(E_{1}E_{2}/F) onto the subgroup

H={(σ1,σ2)σ1|E1E2=σ2|E1E2}H=\{(\sigma_{1},\sigma_{2})\mid\sigma_{1}|E_{1}\cap E_{2}=\sigma_{2}|E_{1}\cap E% _{2}\}

of Gal(E1/F)×Gal(E2/F)\operatorname{Gal}(E_{1}/F)\times\!\operatorname{Gal}(E_{2}/F).

Proof: Let aE1E2a\in E_{1}\cap E_{2}, and let ff be its minimal polynomial over FF. Then ff has

degf\deg f distinct roots in E1E_{1} and degf\deg f distinct roots in E2E_{2}. Since ff can have at most degf\deg f roots in E1E2E_{1}E_{2}, it follows that it has degf\deg f distinct roots in E1E2E_{1}\cap E_{2}. This shows that E1E2E_{1}\cap E_{2} is normal and separable over FF, and hence Galois (3.9). As E1E_{1} and E2E_{2} are Galois over FF, they are splitting fields for separable polynomials f1,f2F[X]f_{1},f_{2}\in F[X]. Now E1E2E_{1}E_{2} is a splitting field for lcm(f1,f2)\textrm{lcm}(f_{1},f_{2}), and hence it also is Galois over FF. The map σ(σ|E1,σ|E2)\sigma\mapsto(\sigma|E_{1},\sigma|E_{2}) is clearly an injective homomorphism, and its image is contained in HH. We’ll prove that the image is the whole of HH by counting.

E1E2{E_{1}E_{2}}E1{E_{1}}E2{E_{2}}E1E2{E_{1}\cap E_{2}}F{F}

From the fundamental theorem,

Gal(E2/F)Gal(E2/E1E2)Gal(E1E2/F),\frac{\operatorname{Gal}(E_{2}/F)}{\operatorname{Gal}(E_{2}/E_{1}\cap E_{2})}% \simeq\operatorname{Gal}(E_{1}\cap E_{2}/F)\text{,}

and so, for each σ1Gal(E1/F)\sigma_{1}\in\operatorname{Gal}(E_{1}/F), σ1|E1E2\sigma_{1}|E_{1}\cap E_{2} has exactly [E2:E1E2][E_{2}\colon E_{1}\cap E_{2}] extensions to an element of Gal(E2/F)\operatorname{Gal}(E_{2}/F). Therefore,

(H:1)=[E1:F][E2:E1E2]=[E1:F][E2:F][E1E2:F],(H\colon 1)=[E_{1}\colon F][E_{2}\colon E_{1}\cap E_{2}]=\frac{[E_{1}\colon F]% \cdot[E_{2}\colon F]}{[E_{1}\cap E_{2}\colon F]},

which equals [E1E2:F][E_{1}E_{2}\colon F] by (3.19)..\hfill\square