The fundamental theorem of Galois theory
The Galois group of an extension is a finite group, while the fields between the two fields seem to have no obvious list. Can the subgroups list those fields exactly, and tell which of them are Galois in turn?
Where you are 4 in Galois theory
Why this exists
In Example 2 of Field automorphisms and the Galois group the one nonidentity automorphism of \displaystyle \mathbb{Q}(\sqrt[4]{7}) fixed exactly \displaystyle \mathbb{Q}(\sqrt{7}), a field strictly between. Nothing hands us a list of the fields between F and E, but the Galois group is finite. When an extension has as many automorphisms as its degree allows, every intermediate field is the set fixed by a subgroup, every subgroup arises, and the arithmetic of the group (indices, normal subgroups, quotients) becomes the arithmetic of the fields (degrees, normal extensions, their Galois groups). Questions about towers of fields, and so about solving equations, become questions about finite groups.
Definition
In plain words: a set of automorphisms picks out the elements it leaves alone, and those form a field; an extension is Galois when it is the splitting field of a polynomial with no repeated roots.
Throughout, E/F is an extension of fields.
- An intermediate field is a field K with F \subseteq K \subseteq E.
- For a set S of automorphisms of E, the fixed field is E^S = \{a \in E : \sigma(a) = a \text{ for every } \sigma \in S\}.
- A finite extension E/F is Galois if E is a splitting field over F of a separable polynomial in F[x], one with no repeated root (Field automorphisms and the Galois group). If \operatorname{char} F = 0 or F is finite, every splitting field E of a nonconstant f \in F[x] is Galois: the product of the distinct monic irreducible factors of f is separable with the same roots (proof of Corollary 7 there).
- For intermediate fields K_1, K_2, the compositum K_1K_2 is the smallest subfield of E containing both, the intersection of all such subfields.
- As in Splitting fields, a finite E/F is normal if every irreducible polynomial in F[x] with a root in E splits over E.
Intuition
Two maps run between the intermediate fields and the subgroups of G = \operatorname{Gal}(E/F): a field K goes to the automorphisms that fix it, \operatorname{Gal}(E/K), and a subgroup H goes to what it fixes, E^H. Both reverse inclusion: more automorphisms fix fewer elements, and a larger field is fixed by fewer automorphisms. So the lattice of fields, E on top, should look like the subgroup lattice turned upside down, \{e\} on top and G at the bottom. The theorem says that for a Galois extension this is exact: the two maps are inverse bijections, and degrees are indices.
Results
Throughout, F \subseteq E are fields.
Lemma 1 (fixed fields). (a) For any set S of automorphisms of E, E^S is a subfield of E. (b) If H_1 \subseteq H_2 \subseteq \operatorname{Gal}(E/F), then F \subseteq E^{H_2} \subseteq E^{H_1}. (c) If F \subseteq K_1 \subseteq K_2 \subseteq E, then \operatorname{Gal}(E/K_2) \le \operatorname{Gal}(E/K_1).
Theorem 2 (Artin’s bound). If G is a finite group of automorphisms of E, then [E : E^G] \le |G|.
Corollary 3. If G is a finite group of automorphisms of E, then \operatorname{Gal}(E/E^G) = G and [E : E^G] = |G|.
Theorem 4 (Galois extensions). Let E/F be finite and G = \operatorname{Gal}(E/F). These are equivalent: (a) E/F is Galois; (b) |G| = [E : F]; (c) E^G = F. When they hold, for each \gamma \in E, with \gamma_1, \dots, \gamma_r the distinct elements \sigma(\gamma) for \sigma \in G,
Theorem 5 (the fundamental theorem of Galois theory). Let E/F be Galois with G = \operatorname{Gal}(E/F). (a) The maps K \mapsto \operatorname{Gal}(E/K) and H \mapsto E^H are inverse bijections between the intermediate fields and the subgroups of G, and both reverse inclusion. (b) If K corresponds to H, then [E : K] = |H| and [K : F] = [G : H]. (c) If K corresponds to H and \sigma \in G, then \sigma(K) corresponds to \sigma H\sigma^{-1}. (d) If K corresponds to H, these are equivalent: H \trianglelefteq G; \sigma(K) = K for every \sigma \in G; K/F is Galois; K/F is normal. When they hold, \sigma \mapsto \sigma|_K is a surjective homomorphism G \to \operatorname{Gal}(K/F) with kernel H, so \operatorname{Gal}(K/F) \cong G/H.
Corollary 6. Let E/F be Galois with group G. (a) There are finitely many intermediate fields. (b) If K_1, K_2 correspond to H_1, H_2, then K_1K_2 corresponds to H_1 \cap H_2, and K_1 \cap K_2 to \langle H_1, H_2 \rangle.
Why it is true
“AU” is Field automorphisms and the Galois group, “SF” Splitting fields, “AE” Algebraic elements and minimal polynomials.
Proof of Lemma 1. (a) Each automorphism fixes 0 and 1 and preserves differences, products and inverses, so if it fixes a and b, it fixes a - b, ab and, for b \ne 0, b^{-1}. (b) and (c) are immediate from the definitions. \square
Proof of Theorem 2. Let G = \{\sigma_1, \dots, \sigma_n\} with \sigma_1 = \mathrm{id}_E, put K = E^G, and take \alpha_1, \dots, \alpha_{n+1} \in E. It is enough to find c_j \in K, not all 0, with \sum_j c_j\alpha_j = 0: then no n + 1 elements of E are independent over K, and [E : K] \le n.
The n equations \sum_{j=1}^{n+1} \sigma_i(\alpha_j)\,y_j = 0 in n + 1 unknowns have a nonzero solution in E^{n+1} (row reduction works over any field; Bases and dimension, proved). Choose a nonzero solution y with the fewest nonzero entries, scaled so that some y_k = 1. For \tau \in G, applying \tau to equation i gives \sum_j \tau\sigma_i(\alpha_j)\,\tau(y_j) = 0, and as i runs over 1, \dots, n, \tau\sigma_i runs over all of G; so (\tau(y_j))_j is again a solution. The difference (y_j - \tau(y_j))_j is a solution that is 0 wherever y is, and also at k, so it has fewer nonzero entries than y and is 0. Thus every y_j is fixed by every \tau, so y_j \in K, and the equation for \sigma_1 = \mathrm{id}_E reads \sum_j y_j\alpha_j = 0 with y_k = 1. \square
Proof of Corollary 3. Each element of G fixes E^G, so G \le \operatorname{Gal}(E/E^G). By AU Theorem 4 and Theorem 2, |G| \le |\operatorname{Gal}(E/E^G)| \le [E : E^G] \le |G|, so equality holds throughout. \square
Proof of Theorem 4. (a) \Rightarrow (b): if E is a splitting field of f with distinct roots, SF Corollary 5 gives exactly [E : F] isomorphisms E \to E fixing F, and they are the elements of G. (b) \Rightarrow (c): let K = E^G \supseteq F. By Corollary 3, [E : K] = |G| = [E : F], and the tower law gives [K : F] = 1.
(c) \Rightarrow (a), with the formula. G is finite (AU Theorem 4). For \gamma \in E let g = \prod_i (x - \gamma_i). Each \tau \in G permutes \gamma_1, \dots, \gamma_r, since \tau\sigma runs over G as \sigma does and \tau is injective; so applying \tau to the coefficients of g gives \prod_i (x - \tau(\gamma_i)) = g, and the coefficients lie in E^G = F. As g(\gamma) = 0, m_\gamma \mid g. Each \gamma_i is a root of m_\gamma (AU Theorem 2(a)), and they are distinct, so g \mid m_\gamma (The remainder and factor theorems); both are monic, so m_\gamma = g. Now let \beta_1, \dots, \beta_n be a basis of E over F and f the product of the distinct polynomials among m_{\beta_1}, \dots, m_{\beta_n}. Each splits over E with distinct roots, and distinct monic irreducible polynomials share no root (AE Theorem 1(d)), so f is separable. Its roots lie in E and include the \beta_j, which generate E, so E is a splitting field of f over F (SF Lemma 2(b)). \square
Proof of Theorem 5. Let E be a splitting field over F of the separable f. (a) Both maps reverse inclusion (Lemma 1). For a subgroup H, Corollary 3 gives \operatorname{Gal}(E/E^H) = H. For an intermediate K, E is a splitting field of f over K (SF Lemma 2(c)), so E/K is Galois, and Theorem 4(c) for E/K gives E^{\operatorname{Gal}(E/K)} = K. (b) As E/K is Galois, [E : K] = |\operatorname{Gal}(E/K)| = |H| (Theorem 4(b)). The tower law and Lagrange's theorem give [K : F] = [E : F]/[E : K] = |G|/|H| = [G : H]. (c) \sigma(K) is an intermediate field. For \tau \in G: \tau fixes \sigma(K) elementwise iff \tau\sigma(a) = \sigma(a) for all a \in K, iff \sigma^{-1}\tau\sigma \in H, iff \tau \in \sigma H\sigma^{-1}. (d) If H \trianglelefteq G (Normal subgroups), then by (c) \sigma(K) and K have the same subgroup, so \sigma(K) = K by (a). If \sigma(K) = K for all \sigma, then \rho(\sigma) = \sigma|_K lies in \operatorname{Gal}(K/F), and \rho is a homomorphism with kernel \operatorname{Gal}(E/K) = H. By the first isomorphism theorem (The isomorphism theorems) its image has |G|/|H| = [K : F] elements, using (b), while |\operatorname{Gal}(K/F)| \le [K : F] (AU Theorem 4). So \rho is onto, \operatorname{Gal}(K/F) \cong G/H (Quotient groups), and |\operatorname{Gal}(K/F)| = [K : F], so K/F is Galois (Theorem 4). A Galois extension is a splitting field, hence normal (SF Theorem 6). Finally let K/F be normal, a \in K and \sigma \in G. Then \sigma(a) is a root of m_a (AU Theorem 2(a)), which splits over K because it is irreducible with the root a in K; its roots in E are its roots in K (SF Lemma 2(a)), so \sigma(a) \in K. Thus \sigma(K) \subseteq K, and as \sigma is injective and F-linear both have dimension [K : F] over F, so \sigma(K) = K. By (c) and (a), \sigma H\sigma^{-1} = H for every \sigma. \square
Proof of Corollary 6. (a) A finite group has finitely many subgroups. (b) The fixed field of an automorphism is a subfield (Lemma 1(a)), so it contains K_1K_2 iff it contains K_1 and K_2: \operatorname{Gal}(E/K_1K_2) = H_1 \cap H_2. The automorphisms fixing a given a \in E form a subgroup, so a is fixed by \langle H_1, H_2 \rangle iff it is fixed by H_1 and by H_2, iff a \in K_1 \cap K_2. \square
Worked examples
A field with three quadratic subfields
Find every intermediate field of \displaystyle E = \mathbb{Q}(\sqrt{3}, \sqrt{7}) over \mathbb{Q}.
SolutionE is a splitting field of (x^2 - 3)(x^2 - 7), whose roots \displaystyle \pm\sqrt{3}, \pm\sqrt{7} are distinct, so E/\mathbb{Q} is Galois. If \displaystyle \sqrt{7} = a + b\sqrt{3} with a, b \in \mathbb{Q}, then \displaystyle 7 = a^2 + 3b^2 + 2ab\sqrt{3} forces ab = 0; b = 0 makes \displaystyle \sqrt{7} rational and a = 0 makes (3b)^2 = 21, both impossible (Irrational numbers). So [E : \mathbb{Q}] = 2 \cdot 2 = 4 and |G| = 4 (Theorem 4). Each element of G sends \displaystyle \sqrt{3} \mapsto \pm\sqrt{3} and \displaystyle \sqrt{7} \mapsto \pm\sqrt{7} (AU Theorem 2(a)) and is determined by the two signs, so all four sign choices occur. Let \sigma change the sign of \displaystyle \sqrt{3} only and \tau that of \displaystyle \sqrt{7} only. Then G = \{e, \sigma, \tau, \sigma\tau\}, each nonidentity element has order 2, and by Lagrange's theorem the proper nontrivial subgroups are \langle\sigma\rangle, \langle\tau\rangle, \langle\sigma\tau\rangle.
By Theorem 5(b) each of their fixed fields has degree [G : H] = 2 over \mathbb{Q}. As \sigma fixes \displaystyle \sqrt{7}, \displaystyle \mathbb{Q}(\sqrt{7}) \subseteq E^{\langle\sigma\rangle}, and equal degrees force equality. Likewise \displaystyle E^{\langle\tau\rangle} = \mathbb{Q}(\sqrt{3}). Since \displaystyle \sigma\tau(\sqrt{3}\sqrt{7}) = (-\sqrt{3})(-\sqrt{7}), \displaystyle E^{\langle\sigma\tau\rangle} = \mathbb{Q}(\sqrt{21}). By Theorem 5(a) the intermediate fields are exactly \mathbb{Q}, \displaystyle \mathbb{Q}(\sqrt{3}), \displaystyle \mathbb{Q}(\sqrt{7}), \displaystyle \mathbb{Q}(\sqrt{21}) and E. G is abelian, so every subgroup is normal, and each of these fields is Galois over \mathbb{Q} (Theorem 5(d)).
A cubic with three real roots
Let f = x^3 - 4x - 1 with splitting field E over \mathbb{Q}. Find G = \operatorname{Gal}(E/\mathbb{Q}), every intermediate field, and which of them are Galois over \mathbb{Q}.
Solutionf(1) = -4 and f(-1) = 2, so f has no rational root (The rational root theorem) and the cubic is irreducible (Irreducible polynomials and Eisenstein's criterion, Theorem 1). The values f(-2) = -1, f(-1) = 2, f(0) = -1, f(2) = -1, f(3) = 14 give real roots r_1 \in (2, 3), r_2 \in (-1, 0), r_3 \in (-2, -1), so E \subseteq \mathbb{R}. Numbering the roots 1, 2, 3 makes G a subgroup of S_3 (AU Theorem 2(b)), and 3 = [\mathbb{Q}(r_1) : \mathbb{Q}] divides [E : \mathbb{Q}] = |G|.
Let \delta = (r_1 - r_2)(r_1 - r_3)(r_2 - r_3). For a cubic x^3 + px + q, writing \delta^2 in terms of r_1 + r_2 + r_3 = 0, r_1r_2 + r_1r_3 + r_2r_3 = p and r_1r_2r_3 = -q gives \delta^2 = -4p^3 - 27q^2, here 256 - 27 = 229. As 229 is prime, \displaystyle \delta = \pm\sqrt{229} \notin \mathbb{Q}. A permutation of the roots multiplies \delta by its sign (Permutation groups and cycle notation). If G \le A_3, then G would fix \delta, putting \delta in E^G = \mathbb{Q} (Theorem 4(c)). So G has an odd element and order divisible by 3: G = S_3.
Its subgroups are \{e\}, \langle(2\ 3)\rangle, \langle(1\ 3)\rangle, \langle(1\ 2)\rangle, A_3 and S_3. (2\ 3) fixes r_1, so \mathbb{Q}(r_1) \subseteq E^{\langle(2\ 3)\rangle}, both of degree 3 = [S_3 : \langle(2\ 3)\rangle]: they are equal. Likewise (1\ 3) gives \mathbb{Q}(r_2) and (1\ 2) gives \mathbb{Q}(r_3). A_3 fixes \delta, and [\mathbb{Q}(\delta) : \mathbb{Q}] = 2 = [S_3 : A_3], so \displaystyle E^{A_3} = \mathbb{Q}(\sqrt{229}). These six fields are drawn above; the three cubic ones are distinct (Theorem 5(a)), though all are real.
A_3 is normal, so \displaystyle \mathbb{Q}(\sqrt{229})/\mathbb{Q} is Galois with group S_3/A_3 \cong \mathbb{Z}_2: the odd permutations restrict to \displaystyle \sqrt{229} \mapsto -\sqrt{229}. But (1\ 2)(2\ 3)(1\ 2) = (1\ 3), so \langle(2\ 3)\rangle is not normal and \mathbb{Q}(r_1)/\mathbb{Q} is not normal: f has the root r_1 there but does not split, although all its roots are real. By Theorem 5(c), (1\ 2) carries \mathbb{Q}(r_1) to \mathbb{Q}(r_2), matching the conjugate subgroup \langle(1\ 3)\rangle. Finally \langle(2\ 3)\rangle \cap A_3 = \{e\}, so Corollary 6(b) gives \displaystyle E = \mathbb{Q}(r_1, \sqrt{229}).
Seventh roots of unity
Let \displaystyle \zeta = \omega_7 = \operatorname{cis}\frac{2\pi}{7} and E = \mathbb{Q}(\zeta). Find G = \operatorname{Gal}(E/\mathbb{Q}) and every intermediate field.
SolutionThe roots of x^7 - 1 are the distinct \zeta^k, 0 \le k \le 6 (Roots of a complex number and roots of unity), so E is its splitting field and E/\mathbb{Q} is Galois. \zeta is a root of \Phi_7(x) = x^6 + x^5 + \dots + 1, irreducible by Irreducible polynomials and Eisenstein's criterion (Theorem 6), so [E : \mathbb{Q}] = 6 with basis 1, \zeta, \dots, \zeta^5, and also \zeta, \zeta^2, \dots, \zeta^6 (its product with \zeta). Each \sigma \in G sends \zeta to a root \zeta^k of \Phi_7, 1 \le k \le 6, and is determined by k; as |G| = 6, every k occurs: \sigma_k(\zeta) = \zeta^k. Since \sigma_j\sigma_k(\zeta) = \zeta^{jk}, k \mapsto \sigma_k is an isomorphism \mathbb{Z}_7^\times \to G, and the powers of 3 modulo 7 are 3, 2, 6, 4, 5, 1, so G = \langle\sigma_3\rangle is cyclic of order 6. It has one subgroup of each order dividing 6 (Cyclic groups): \{e\}, H_2 = \{e, \sigma_6\}, H_3 = \{e, \sigma_2, \sigma_4\} and G. So exactly two fields lie strictly between, of degrees 3 and 2 (Theorem 5(b)).
\sigma_6(\zeta) = \zeta^{-1} = \bar\zeta, so \sigma_6 is complex conjugation on E, and \displaystyle c = \zeta + \zeta^6 = 2\cos\frac{2\pi}{7} lies in E^{H_2}. In the basis \zeta, \dots, \zeta^6, \sigma_2(c) = \zeta^2 + \zeta^5 \ne c, so c \notin E^G = \mathbb{Q}; as [E^{H_2} : \mathbb{Q}] = 3 is prime, E^{H_2} = \mathbb{Q}(c).
\sigma_2 permutes \zeta, \zeta^2, \zeta^4 cyclically, so \eta = \zeta + \zeta^2 + \zeta^4 lies in E^{H_3}, while \sigma_3(\eta) = \zeta^3 + \zeta^6 + \zeta^5 = \eta' \ne \eta. From \Phi_7(\zeta) = 0, \eta + \eta' = -1, and multiplying out with \zeta^7 = 1,
So \eta is a root of x^2 + x + 2, \displaystyle \eta = \frac{-1 \pm \sqrt{7}\,i}{2}, and \displaystyle E^{H_3} = \mathbb{Q}(\eta) = \mathbb{Q}(\sqrt{7}\,i). (The sign is +: the imaginary part of \eta is \displaystyle \sin\frac{2\pi}{7} + \sin\frac{4\pi}{7} - \sin\frac{\pi}{7} > 0.) The intermediate fields are \mathbb{Q}, \displaystyle \mathbb{Q}(\sqrt{7}\,i), \mathbb{Q}(c) and E. As G is abelian, both middle fields are Galois over \mathbb{Q}, with groups G/H_3 \cong \mathbb{Z}_2 and G/H_2 \cong \mathbb{Z}_3.
Stabilizers and minimal polynomials
Let E/F be Galois with group G and \gamma \in E. Prove that \operatorname{Gal}(E/F(\gamma)) = \{\sigma \in G : \sigma(\gamma) = \gamma\}, that [F(\gamma) : F] is the number of distinct \sigma(\gamma), and that F(\gamma) = E exactly when only e fixes \gamma. Then find the minimal polynomial over \mathbb{Q} of c = \zeta + \zeta^6 from Example 3.
SolutionProof. Let S = \{\sigma \in G : \sigma(\gamma) = \gamma\}. Every element of F(\gamma) = F[\gamma] (AE Theorem 1(e)) is a polynomial expression in \gamma with coefficients in F, so \sigma \in G fixes F(\gamma) elementwise iff it fixes \gamma: \operatorname{Gal}(E/F(\gamma)) = S. By Theorem 5(b), [F(\gamma) : F] = [G : S], and \sigma(\gamma) = \tau(\gamma) iff \tau^{-1}\sigma \in S iff \sigma S = \tau S, so [G : S] is the number of distinct images. Finally F(\gamma) = E iff S = \operatorname{Gal}(E/E) = \{e\}, by Theorem 5(a). \square
For c, \sigma_k(c) = \zeta^k + \zeta^{-k} takes three values, c_1 = c, c_2 = \zeta^2 + \zeta^5 and c_3 = \zeta^3 + \zeta^4, so [\mathbb{Q}(c) : \mathbb{Q}] = 3, as Example 3 found, and Theorem 4 gives m_c = (x - c_1)(x - c_2)(x - c_3). Using \zeta^7 = 1: c_1 + c_2 + c_3 = -1; c_1c_2 = c_1 + c_3, c_1c_3 = c_2 + c_3 and c_2c_3 = c_1 + c_2, which sum to -2; and c_1c_2c_3 = (c_1 + c_3)c_3 = (c_2 + c_3) + (\zeta^6 + 2 + \zeta) = 1. So
Boundaries
- Not Galois. For \displaystyle E = \mathbb{Q}(\sqrt[4]{7}) (AU Example 2), the group \{e, \sigma\} fixes \displaystyle \mathbb{Q}(\sqrt{7}), so \mathbb{Q} and \displaystyle \mathbb{Q}(\sqrt{7}) both go to the whole group: K \mapsto \operatorname{Gal}(E/K) is not injective.
- Inseparable polynomials. For F = \mathbb{Z}_3(t) and E = F(s) with s^3 = t (AU Exercise 13), [E : F] = 3 but the group is trivial, so its fixed field is E, not F: Theorem 4(c) fails.
- Infinite extensions. For infinite algebraic extensions such as \overline{\mathbb{Q}}/\mathbb{Q} the correspondence holds only for subgroups closed in the Krull topology, beyond this course.
- Galois at the top, not always at the bottom. E/K is Galois for every intermediate K, but K/F only when H is normal: \mathbb{Q}(r_1) in Example 2 is not.
Where this goes next
A formula in radicals for the roots of f puts them in a tower of fields, each obtained from the last by adjoining an n-th root. Once suitable roots of unity are adjoined, each step is Galois with an abelian group, and Theorem 5(d) turns the tower into a chain of subgroups, each normal in the next with abelian quotient: the Galois group of f is a solvable group. Why the quintic has no formula finds a rational quintic whose group is S_5, which is not solvable.
Exercises
Intermediate fields
- List every intermediate field of \displaystyle E = \mathbb{Q}(\sqrt{5}, i) over \mathbb{Q} and the subgroup of \operatorname{Gal}(E/\mathbb{Q}) fixing each.
- Let \displaystyle E = \mathbb{Q}(\sqrt{6}, \sqrt{10}). Show that [E : \mathbb{Q}] = 4, that \displaystyle \sqrt{15} \in E, and list the fields strictly between \mathbb{Q} and E.
- With E as in Exercise 2, use Theorem 4 to find the minimal polynomial over \mathbb{Q} of \displaystyle \sqrt{6} + \sqrt{10}, and conclude that it generates E.
- Let \alpha be a root of x^3 - 7x - 7. Show that \beta = 3\alpha^2 - 5\alpha - 14 is another root, that \mathbb{Q}(\alpha) is a splitting field, and that no field lies strictly between \mathbb{Q} and \mathbb{Q}(\alpha).
- Let \displaystyle \zeta = \omega_5 = \operatorname{cis}\frac{2\pi}{5}. Show that \operatorname{Gal}(\mathbb{Q}(\zeta)/\mathbb{Q}) is cyclic of order 4 and find the one field strictly between \mathbb{Q} and \mathbb{Q}(\zeta).
- Let \displaystyle E = \mathbb{Q}(\sqrt{2}, \sqrt{3}, \sqrt{5}), of degree 8 over \mathbb{Q}. How many intermediate fields have degree 2 over \mathbb{Q}, and how many degree 4? List those of degree 2.
- List the fields between \mathbb{F}_3 and \mathbb{F}_{81} and the subgroup of \operatorname{Gal}(\mathbb{F}_{81}/\mathbb{F}_3) fixing each.
Using the correspondence
- In Example 3, which intermediate fields are subfields of \mathbb{R}?
- In Example 3, which \sigma_k restrict to the nonidentity element of \displaystyle \operatorname{Gal}(\mathbb{Q}(\sqrt{7}\,i)/\mathbb{Q})? Which restrict to the element of \operatorname{Gal}(\mathbb{Q}(c)/\mathbb{Q}) sending c to \zeta^2 + \zeta^5?
- In Example 2, with \delta = (r_1 - r_2)(r_1 - r_3)(r_2 - r_3), use Example 4 to show that E = \mathbb{Q}(r_1 + \delta).
Proofs
- Let E/F be Galois with abelian group. Prove that every intermediate field K is Galois over F with abelian group.
- Let E/F be Galois and K_1, K_2 intermediate fields with K_1/F Galois. Prove that [K_1K_2 : K_2] = [K_1 : K_1 \cap K_2].
- Let E/F be Galois with group G, and K an intermediate field with group H. Prove that the fields \sigma(K), \sigma \in G, number exactly [G : N_G(H)].
- Let E/F be Galois with F infinite. Using Corollary 6(a), prove that E = F(\gamma) for some \gamma \in E.
- In a chalet after a ski day, Yuna argues: “\displaystyle \operatorname{Gal}(\mathbb{Q}(\sqrt[3]{12})/\mathbb{Q}) is trivial, and the fundamental theorem matches intermediate fields with subgroups, so there is only one intermediate field, and \displaystyle \mathbb{Q}(\sqrt[3]{12}) = \mathbb{Q}.” Kofi is sure the conclusion is false. Find the flaw, and list the true intermediate fields.
Answers
- \displaystyle \mathbb{Q}(\sqrt{5}) \subseteq \mathbb{R} misses i, so [E : \mathbb{Q}] = 4, and E splits (x^2 - 5)(x^2 + 1). As in Example 1, G = \{e, \sigma, \tau, \sigma\tau\} with \displaystyle \sigma(\sqrt{5}) = -\sqrt{5}, \sigma(i) = i, and \tau complex conjugation. G gives \mathbb{Q}, \{e\} gives E, \langle\sigma\rangle gives \mathbb{Q}(i), \langle\tau\rangle gives \displaystyle \mathbb{Q}(\sqrt{5}), and \langle\sigma\tau\rangle gives \displaystyle \mathbb{Q}(\sqrt{5}\,i), as \displaystyle \sigma\tau(\sqrt{5}\,i) = (-\sqrt{5})(-i).
- If \displaystyle \sqrt{10} = a + b\sqrt{6} with a, b \in \mathbb{Q}, then ab = 0; b = 0 makes \displaystyle \sqrt{10} rational and a = 0 gives (3b)^2 = 15. So [E : \mathbb{Q}] = 4, and \displaystyle \sqrt{6}\sqrt{10} = 2\sqrt{15} puts \displaystyle \sqrt{15} in E. As in Example 1 the group is the four sign changes, and its three subgroups of order 2 fix \displaystyle \mathbb{Q}(\sqrt{6}), \displaystyle \mathbb{Q}(\sqrt{10}), \displaystyle \mathbb{Q}(\sqrt{15}), the only fields strictly between.
- The images \displaystyle \pm\sqrt{6} \pm \sqrt{10} are four distinct numbers, so m is the product of the four factors \displaystyle x - (\pm\sqrt{6} \pm \sqrt{10}). Pairing by the sign of \displaystyle \sqrt{6} gives (x^2 - 4)^2 - 24x^2 = x^4 - 32x^2 + 16. Only e fixes \displaystyle \sqrt{6} + \sqrt{10}, so by Example 4 it generates E.
- The cubic is Eisenstein at 7, so [\mathbb{Q}(\alpha) : \mathbb{Q}] = 3. With \alpha^3 = 7\alpha + 7 and \alpha^4 = 7\alpha^2 + 7\alpha: \beta^2 = 4\alpha^2 - 7\alpha - 14, \beta^3 = 21\alpha^2 - 35\alpha - 91, and \beta^3 - 7\beta - 7 = 0. \beta \ne \alpha, else \alpha has degree at most 2. The roots sum to 0, so the third is -3\alpha^2 + 4\alpha + 14, which is -\alpha - \beta. So \mathbb{Q}(\alpha) is a splitting field, its group has prime order 3 (Theorem 4), with no subgroups but \{e\} and itself, and by Theorem 5(a) the only intermediate fields are \mathbb{Q} and \mathbb{Q}(\alpha).
- \Phi_5 is irreducible (Irreducible polynomials and Eisenstein's criterion, Theorem 6), and as in Example 3 the group is \{\sigma_1, \dots, \sigma_4\} \cong \mathbb{Z}_5^\times, cyclic since the powers of 2 are 2, 4, 3, 1. Its one subgroup of order 2 is \{e, \sigma_4\}, fixing \theta = \zeta + \zeta^4. With \theta' = \sigma_2(\theta) = \zeta^2 + \zeta^3: \theta + \theta' = -1 and \theta\theta' = \zeta^3 + \zeta^4 + \zeta + \zeta^2 = -1, so \theta is a root of x^2 + x - 1. As \displaystyle \theta = 2\cos\frac{2\pi}{5} > 0, \displaystyle \theta = \frac{-1 + \sqrt{5}}{2}, and the field is \displaystyle \mathbb{Q}(\sqrt{5}).
- Seven of each. An element of G is determined by its signs on \displaystyle \sqrt{2}, \sqrt{3}, \sqrt{5}, and |G| = 8, so G \cong \mathbb{Z}_2 \times \mathbb{Z}_2 \times \mathbb{Z}_2, a vector space over \mathbb{Z}_2. Subgroups of order 2 (one per nonzero element) give degree 4; subgroups of index 2, the kernels of the 7 nonzero linear maps to \mathbb{Z}_2, give degree 2. The fields \displaystyle \mathbb{Q}(\sqrt{d}), d = 2, 3, 5, 6, 10, 15, 30, lie in E and are distinct: as in Exercise 2, \displaystyle \mathbb{Q}(\sqrt{d}) = \mathbb{Q}(\sqrt{d'}) would make \displaystyle \sqrt{dd'} rational. So they are all seven.
- \mathbb{F}_{81} splits x^{81} - x, so it is Galois, with group \langle\operatorname{Fr}\rangle cyclic of order 4 (Finite fields, Theorem 7). The subgroups \langle\operatorname{Fr}\rangle, \langle\operatorname{Fr}^2\rangle, \{\mathrm{id}\} fix \mathbb{F}_3, \{a : a^9 = a\} = \mathbb{F}_9 (of degree [G : \langle\operatorname{Fr}^2\rangle] = 2) and \mathbb{F}_{81}: the fields are \mathbb{F}_3 \subset \mathbb{F}_9 \subset \mathbb{F}_{81}.
- Only \mathbb{Q} and \mathbb{Q}(c). K \subseteq \mathbb{R} iff conjugation \sigma_6 fixes K, iff \sigma_6 \in H; the subgroups containing \sigma_6 are H_2 and G. \displaystyle \mathbb{Q}(\sqrt{7}\,i) is not real.
- By Theorem 5(d) the restriction has kernel H_3, so the nonidentity element comes from \sigma_3, \sigma_5, \sigma_6. For \mathbb{Q}(c), \sigma_k(c) = \zeta^k + \zeta^{-k} equals \zeta^2 + \zeta^5 for k = 2, 5: \sigma_2 and \sigma_5.
- Suppose \sigma(r_1 + \delta) = r_1 + \delta. If \sigma is even, \sigma(\delta) = \delta, so \sigma(r_1) = r_1, and the only even permutation fixing 1 is e. If \sigma is odd, \sigma(r_1) - r_1 = 2\delta; but all roots lie in (-2, 3), so \displaystyle |\sigma(r_1) - r_1| < 5 < 2\sqrt{229}. So only e fixes r_1 + \delta, and Example 4 gives E = \mathbb{Q}(r_1 + \delta).
- Proof. Let K correspond to H. Every subgroup of an abelian group is normal, so by Theorem 5(d) K/F is Galois with \operatorname{Gal}(K/F) \cong G/H, and aH \cdot bH = abH = baH = bH \cdot aH. \square
- Proof. Let K_i correspond to H_i. By Theorem 5(d), H_1 \trianglelefteq G, so H_1H_2 is a subgroup, hence equal to \langle H_1, H_2 \rangle, and H_1H_2/H_1 \cong H_2/(H_1 \cap H_2) (The isomorphism theorems). By Corollary 6(b), Theorem 5(b) and the tower law, [K_1K_2 : K_2] = [E : K_2]/[E : K_1K_2] = |H_2|/|H_1 \cap H_2| and [K_1 : K_1 \cap K_2] = |H_1H_2|/|H_1|, which are equal. \square
- Proof. G acts on the intermediate fields by \sigma \cdot K = \sigma(K) (Group actions, orbits and stabilizers). By Theorem 5(c) and (a), \sigma(K) = K iff \sigma H\sigma^{-1} = H, so \operatorname{Stab}(K) = N_G(H), and the orbit of K has [G : N_G(H)] elements. \square
- Proof. By induction on the number of generators of E = F(\beta_1, \dots, \beta_m), it suffices to show F(\alpha, \beta) = F(\gamma) for \alpha, \beta \in E. The fields F(\alpha + c\beta), c \in F, are intermediate fields of E/F, finitely many, while F is infinite; so F(\alpha + c\beta) = F(\alpha + c'\beta) = L for some c \ne c'. Then (c - c')\beta \in L, so \beta \in L and \alpha \in L, and F(\alpha, \beta) = L. \square
- The theorem needs E/\mathbb{Q} Galois, and it is not: x^3 - 12 is Eisenstein at 3, irreducible with a root in \displaystyle \mathbb{Q}(\sqrt[3]{12}) \subseteq \mathbb{R}, but its other two roots are not real, so the extension is not normal. Both \mathbb{Q} and \displaystyle \mathbb{Q}(\sqrt[3]{12}) go to the trivial group, so the map is not injective. The degree 3 is prime, so by the tower law the intermediate fields are \mathbb{Q} and \displaystyle \mathbb{Q}(\sqrt[3]{12}).