Chapter 8 · Fields and Galois Theory

8.3 Galois Theory

The symmetries of a field extension form a group, and that group sees everything: intermediate fields correspond to subgroups, normal extensions to normal subgroups, and solvability by radicals to solvability of the group. We build the theory from automorphisms and fixed fields, prove the fundamental theorem, and compute the Galois groups of $\mathbb{Q}(\sqrt2,\sqrt3)$, $\mathbb{Q}(\sqrt[3]{2},\omega)$, $\mathbb{Q}(\sqrt[4]{2},i)$, the cyclotomic fields and the finite fields.

The polynomial $x^4-10x^2+1$ has the four roots $\pm\sqrt2\pm\sqrt3$. There are $24$ ways to permute four objects, but only four of them "respect the algebra" of these roots: whatever we do to $\sqrt2+\sqrt3$ must also be done consistently to $-(\sqrt2+\sqrt3)$, to $(\sqrt2+\sqrt3)^2=5+2\sqrt6$, and to every other polynomial expression in the roots. The admissible permutations are exactly "change the sign of $\sqrt2$", "change the sign of $\sqrt3$", both, or neither. Galois's insight (1830) was that this small group of symmetries — the Galois group — controls the whole extension $\mathbb{Q}(\sqrt2,\sqrt3)/\mathbb{Q}$: its three subgroups of order $2$ correspond to the three quadratic subfields $\mathbb{Q}(\sqrt2)$, $\mathbb{Q}(\sqrt3)$, $\mathbb{Q}(\sqrt6)$, and nothing else lies in between. In general, intermediate fields correspond to subgroups (the fundamental theorem of Galois theory), normal subextensions to normal subgroups, and an equation is solvable by radicals precisely when its group can be built from abelian pieces. That last statement explains why there is a formula for the roots of every quartic but none for the general quintic.

CSIR NET asks about this material in a few recurring shapes: "the order of the Galois group of the splitting field of $x^4-3$ over $\mathbb{Q}$ is…", "how many subfields does $\mathbb{Q}(\zeta_{11})$ or $\mathbb{F}_{2^{12}}$ have?", "which of these extensions of $\mathbb{Q}$ is Galois?", "which intermediate fields are normal?", "is the Galois group abelian, cyclic, $S_3$, $D_4$?", "is this quintic solvable by radicals?". Every one of them reduces to three skills: computing $[K:F]$ with the tower law, describing automorphisms by where they send generators, and reading the subgroup lattice of a small group upside down. We use freely §8.1 (degrees, the tower law, minimal polynomials, $F(\alpha)\cong F[x]/\langle m\rangle$, splitting fields, separability), §8.2 (finite fields), §7.3 (cyclotomic polynomials) and Chapter 6 (Lagrange, normal subgroups and quotients, group actions, Sylow and Cauchy). Throughout, $K/F$ means that $F$ is a subfield of $K$, $\omega=e^{2\pi i/3}$, $\zeta_n=e^{2\pi i/n}$, and number fields are subfields of $\mathbb{C}$.

Automorphisms and Fixed Fields

📘 Definition 8.10 — Automorphisms of an extension

Let $K/F$ be a field extension. An automorphism of $K$ is a ring isomorphism $\sigma:K\to K$. It is an $F$-automorphism if $\sigma(a)=a$ for every $a\in F$. The $F$-automorphisms form a group under composition, the automorphism group $$\operatorname{Aut}(K/F)=\{\sigma\in\operatorname{Aut}(K):\ \sigma|_F=\mathrm{id}_F\}.$$

More generally, if $L$ is another field containing $F$, an $F$-embedding $K\to L$ is a ring homomorphism fixing $F$ pointwise. Every ring homomorphism between fields is injective (its kernel is an ideal not containing $1$, hence $0$), so an $F$-embedding identifies $K$ with a subfield of $L$.

Every automorphism fixes the prime subfield: $\sigma(1)=1$ forces $\sigma(n\cdot1)=n\cdot1$ and $\sigma(a/b)=\sigma(a)/\sigma(b)$. So $\operatorname{Aut}(K)=\operatorname{Aut}(K/\mathbb{Q})$ for every field of characteristic $0$ and $\operatorname{Aut}(K)=\operatorname{Aut}(K/\mathbb{F}_p)$ in characteristic $p$. Also, an $F$-automorphism is an $F$-linear map of the $F$-vector space $K$; this is what makes counting arguments possible.

📐 Theorem 8.29 — Automorphisms permute roots

Let $K/F$ be an extension and $\sigma:K\to L$ an $F$-embedding (for instance $\sigma\in\operatorname{Aut}(K/F)$).

  1. For $f\in F[x]$ and $\alpha\in K$, $\sigma(f(\alpha))=f(\sigma(\alpha))$. Hence $\sigma$ maps roots of $f$ to roots of $f$; in particular $\sigma(\alpha)$ is a root of the minimal polynomial of $\alpha$ over $F$ (a conjugate of $\alpha$).
  2. If $K=F(\alpha_1,\dots,\alpha_n)$, then $\sigma$ is determined by $\sigma(\alpha_1),\dots,\sigma(\alpha_n)$.
  3. If $K=F(\alpha_1,\dots,\alpha_n)$ where $\alpha_1,\dots,\alpha_n$ are all the distinct roots in $K$ of some $f\in F[x]$, then each $\sigma\in\operatorname{Aut}(K/F)$ permutes $\{\alpha_1,\dots,\alpha_n\}$, and $\sigma\mapsto(\text{this permutation})$ is an injective homomorphism $\operatorname{Aut}(K/F)\to S_n$.

Proof. (1) If $f=\sum a_ix^i$ with $a_i\in F$, then $\sigma\big(\sum a_i\alpha^i\big)=\sum\sigma(a_i)\sigma(\alpha)^i=\sum a_i\sigma(\alpha)^i$, because $\sigma$ is a ring homomorphism fixing each $a_i$. If $f(\alpha)=0$ then $f(\sigma(\alpha))=\sigma(0)=0$.

(2) The set of quotients $p(\alpha_1,\dots,\alpha_n)/q(\alpha_1,\dots,\alpha_n)$ with $p,q\in F[x_1,\dots,x_n]$ and $q(\alpha_1,\dots,\alpha_n)\ne0$ is a subfield of $K$ containing $F$ and every $\alpha_i$, so it is all of $K$. By (1) applied coefficientwise, $\sigma$ sends such a quotient to $p(\sigma\alpha_1,\dots,\sigma\alpha_n)/q(\sigma\alpha_1,\dots,\sigma\alpha_n)$, which depends only on the values $\sigma(\alpha_i)$.

(3) By (1), $\sigma$ maps the finite set $R=\{\alpha_1,\dots,\alpha_n\}$ into itself; it is injective, hence a bijection of $R$. Writing $\pi_\sigma$ for this permutation, $\pi_{\sigma\tau}=\pi_\sigma\pi_\tau$ since both sides are restrictions of $\sigma\tau$ to $R$. If $\pi_\sigma=\mathrm{id}$, then $\sigma$ fixes every generator and $\sigma=\mathrm{id}$ by (2). ∎

🔁 Recalled from §8.1 — extending isomorphisms and primitive elements

Let $\varphi:F\to F'$ be an isomorphism of fields and $f\mapsto f^\varphi$ the induced isomorphism $F[x]\to F'[x]$. We shall use three results proved in §8.1.

  1. (R1) Simple extensions. If $\alpha$ is algebraic over $F$ with minimal polynomial $m$ and $L\supseteq F'$ is a field, the embeddings $\psi:F(\alpha)\to L$ extending $\varphi$ correspond bijectively, via $\psi\mapsto\psi(\alpha)$, to the roots of $m^\varphi$ in $L$; so there are at most $[F(\alpha):F]$ of them. With $\varphi=\mathrm{id}$: to define an $F$-embedding of $F(\alpha)$ it is enough to name a conjugate of $\alpha$, and if that conjugate lies in $F(\alpha)$ the embedding is an automorphism (an injective $F$-linear map of a finite-dimensional space to itself).
  2. (R2) Splitting fields. If $K$ is a splitting field of $f$ over $F$ and $K'$ one of $f^\varphi$ over $F'$, then $\varphi$ extends to an isomorphism $K\to K'$, in at most $[K:F]$ ways, and in exactly $[K:F]$ ways when $f$ is separable. In particular, if $K$ is the splitting field of $f\in F[x]$, then $|\operatorname{Aut}(K/F)|\le[K:F]$, with equality when $f$ is separable.
  3. (R3) Primitive element theorem. Every finite separable extension is simple, $K=F(\gamma)$; for infinite $F$ and $K=F(\alpha,\beta)$ with $\beta$ separable, $\gamma=\alpha+c\beta$ works for all but finitely many $c\in F$. For example $\mathbb{Q}(\sqrt2,\sqrt3)=\mathbb{Q}(\sqrt2+\sqrt3)$ (Example 2) and $\mathbb{Q}(\sqrt[3]2,\omega)=\mathbb{Q}(\sqrt[3]2+\omega)$ (Exercise III.9).
📘 Definition 8.11 — Fixed field
For a field $K$ and a set $H$ of automorphisms of $K$, the fixed field of $H$ is $$K^H=\{a\in K:\ \sigma(a)=a\ \text{ for all }\sigma\in H\}.$$
📐 Theorem 8.30 — The Galois connection

Let $K$ be a field.

  1. $K^H$ is a subfield of $K$ for every $H\subseteq\operatorname{Aut}(K)$; if $H\subseteq\operatorname{Aut}(K/F)$ then $F\subseteq K^H$.
  2. Both assignments reverse inclusions: $H_1\subseteq H_2\Rightarrow K^{H_1}\supseteq K^{H_2}$, and $E_1\subseteq E_2\Rightarrow\operatorname{Aut}(K/E_1)\supseteq\operatorname{Aut}(K/E_2)$.
  3. $H\subseteq\operatorname{Aut}(K/K^H)$ and $E\subseteq K^{\operatorname{Aut}(K/E)}$.
  4. For $\sigma\in\operatorname{Aut}(K)$: $K^{\sigma H\sigma^{-1}}=\sigma(K^H)$ and $\operatorname{Aut}(K/\sigma(E))=\sigma\operatorname{Aut}(K/E)\sigma^{-1}$.

Proof. (1) If $\sigma(a)=a$ and $\sigma(b)=b$, then $\sigma(a\pm b)=a\pm b$, $\sigma(ab)=ab$ and $\sigma(a^{-1})=a^{-1}$; and $0,1$ are fixed. (2) and (3) are immediate from the definitions: fixing everything in $H_2$ includes fixing everything in $H_1$, and every $\sigma\in H$ fixes $K^H$ pointwise. (4) For $a\in K$: $\sigma(a)\in K^{\sigma H\sigma^{-1}}$ iff $\sigma\tau\sigma^{-1}(\sigma(a))=\sigma(a)$ for all $\tau\in H$ iff $\sigma(\tau(a))=\sigma(a)$ for all $\tau$ iff $a\in K^H$. Similarly $\rho$ fixes $\sigma(E)$ pointwise iff $\rho\sigma(e)=\sigma(e)$ for all $e\in E$ iff $\sigma^{-1}\rho\sigma\in\operatorname{Aut}(K/E)$. ∎

The question Galois theory answers is: when are the inclusions in (3) equalities? For a finite group $H$ the first one always is (Artin's theorem below); the second one is an equality exactly for Galois extensions.

✏️ Example 1 — Four automorphism groups
Determine (a) $\operatorname{Aut}(\mathbb{Q}(\sqrt[3]{2})/\mathbb{Q})$; (b) $\operatorname{Aut}(\mathbb{Q}(\sqrt2,\sqrt3)/\mathbb{Q})$ and the fixed fields of its subgroups; (c) $\operatorname{Aut}(\mathbb{R})$; (d) $\operatorname{Aut}(\mathbb{C}/\mathbb{R})$.
(a) $\sigma(\sqrt[3]2)$ must be a root of $x^3-2$ lying in $\mathbb{Q}(\sqrt[3]2)\subset\mathbb{R}$ (Theorem 8.29). The only real cube root of $2$ is $\sqrt[3]2$, so $\sigma(\sqrt[3]2)=\sqrt[3]2$ and $\sigma=\mathrm{id}$. Thus $\operatorname{Aut}=\{1\}$ although $[\mathbb{Q}(\sqrt[3]2):\mathbb{Q}]=3$, and the fixed field of the whole group is $\mathbb{Q}(\sqrt[3]2)\ne\mathbb{Q}$. The other roots $\omega\sqrt[3]2,\omega^2\sqrt[3]2$ exist, but not inside this field.
(b) $K=\mathbb{Q}(\sqrt2,\sqrt3)$ has degree $4$: $\sqrt3\notin\mathbb{Q}(\sqrt2)$, since $(a+b\sqrt2)^2=3$ with $a,b\in\mathbb{Q}$ forces $2ab=0$, and then $a^2=3$ or $2b^2=3$, impossible in $\mathbb{Q}$. $K$ is the splitting field of the separable $(x^2-2)(x^2-3)$, so $|\operatorname{Aut}(K/\mathbb{Q})|=4$ by (R2). An automorphism is determined by $\sqrt2\mapsto\pm\sqrt2$, $\sqrt3\mapsto\pm\sqrt3$: four choices, so all occur. Let $\sigma$ change the sign of $\sqrt2$ only and $\tau$ that of $\sqrt3$ only. Every non-identity element has order $2$, so the group is $\{1,\sigma,\tau,\sigma\tau\}\cong V_4$. On the basis $1,\sqrt2,\sqrt3,\sqrt6$, $$\sigma(a+b\sqrt2+c\sqrt3+d\sqrt6)=a-b\sqrt2+c\sqrt3-d\sqrt6,$$ so $K^{\langle\sigma\rangle}=\mathbb{Q}(\sqrt3)$; likewise $K^{\langle\tau\rangle}=\mathbb{Q}(\sqrt2)$, $K^{\langle\sigma\tau\rangle}=\mathbb{Q}(\sqrt6)$ and $K^{V_4}=\mathbb{Q}$.
(c) Let $\sigma\in\operatorname{Aut}(\mathbb{R})$; it fixes $\mathbb{Q}$. If $x\gt0$ then $x=y^2$ with $y\ne0$ and $\sigma(x)=\sigma(y)^2\gt0$. Hence $x\lt y\Rightarrow\sigma(x)\lt\sigma(y)$: $\sigma$ preserves order. For rationals $r\lt x\lt s$ we get $r\lt\sigma(x)\lt s$; letting $r,s\to x$ gives $\sigma(x)=x$. So $\operatorname{Aut}(\mathbb{R})=\{\mathrm{id}\}$.
(d) $\sigma(i)^2=\sigma(-1)=-1$, so $\sigma(i)=\pm i$, and $\sigma(a+bi)=a\pm bi$: $\operatorname{Aut}(\mathbb{C}/\mathbb{R})=\{\mathrm{id},\ \text{conjugation}\}$, of order $2=[\mathbb{C}:\mathbb{R}]$.
Lesson $|\operatorname{Aut}(K/F)|\le[K:F]$ always (Theorem 8.31 below), with equality in (b) and (d) but not in (a). The difference is whether $K$ contains all the conjugates of its elements.
📐 Theorem 8.31 — Artin's theorem

Let $K$ be a field, $G$ a finite group of automorphisms of $K$, and $F=K^G$.

  1. Every $\alpha\in K$ is algebraic and separable over $F$, and its minimal polynomial is the orbit polynomial $$m_\alpha(x)=\prod_{\beta\in G\alpha}(x-\beta),$$ where $G\alpha=\{\sigma(\alpha):\sigma\in G\}$. So $[F(\alpha):F]=|G\alpha|$, which divides $|G|$.
  2. $[K:F]=|G|$ and $\operatorname{Aut}(K/F)=G$.

Consequently $|\operatorname{Aut}(K/F)|$ divides $[K:F]$ for every finite extension $K/F$.

Proof. (1) Let $G\alpha=\{\beta_1=\alpha,\beta_2,\dots,\beta_k\}$ (distinct). Each $\tau\in G$ permutes $G\alpha$, since $\tau(\sigma(\alpha))=(\tau\sigma)(\alpha)$ and $\tau$ is injective. Applying $\tau$ to the coefficients of $m_\alpha=\prod(x-\beta_i)$ therefore only permutes the factors, so every coefficient is fixed by every $\tau$: $m_\alpha\in F[x]$. It is monic, has $\alpha$ as a root and has distinct roots. If $g\in F[x]$ and $g(\alpha)=0$, then $g(\sigma(\alpha))=\sigma(g(\alpha))=0$ for all $\sigma$ (Theorem 8.29), so each $\beta_i$ is a root of $g$ and $m_\alpha\mid g$. Hence $m_\alpha$ is the minimal polynomial of $\alpha$; it is separable, and $\deg m_\alpha=|G\alpha|=[G:G_\alpha]$ divides $|G|$ by the orbit–stabiliser theorem (§6.3).

(2) By (1), $K/F$ is separable and algebraic, and every element has degree at most $n=|G|$ over $F$. Choose $\alpha\in K$ with $[F(\alpha):F]$ as large as possible. If some $\beta\in K$ were not in $F(\alpha)$, then $F(\alpha,\beta)/F$ would be finite and separable, so $F(\alpha,\beta)=F(\gamma)$ by the primitive element theorem (R3), with $[F(\gamma):F]\gt[F(\alpha):F]$ — contradicting the choice of $\alpha$. So $K=F(\alpha)$ and $[K:F]=|G\alpha|\le n$. On the other hand $G\subseteq\operatorname{Aut}(K/F)$, and by (R1) (with $\varphi=\mathrm{id}$, $L=K$) there are at most $[F(\alpha):F]$ $F$-automorphisms of $K=F(\alpha)$. Thus $$n=|G|\le|\operatorname{Aut}(K/F)|\le[K:F]\le n,$$ so all are equal and $G=\operatorname{Aut}(K/F)$.

For the last sentence: if $K/F$ is finite, $G=\operatorname{Aut}(K/F)$ is finite (each $\sigma$ is determined by the images of finitely many generators, each of which has finitely many conjugates). By (2), $|G|=[K:K^G]$, and $[K:F]=[K:K^G]\,[K^G:F]$ by the tower law. ∎

Part (1) is also a computing tool: the minimal polynomial of $\alpha$ over a fixed field is the product of $x-\beta$ over the orbit of $\alpha$.

✏️ Example 2 — Orbit polynomials and a primitive element of $\mathbb{Q}(\sqrt2,\sqrt3)$
Let $K=\mathbb{Q}(\sqrt2,\sqrt3)$, $G=\{1,\sigma,\tau,\sigma\tau\}$ as in Example 1, and $\gamma=\sqrt2+\sqrt3$. (a) Find the minimal polynomial of $\gamma$ over $\mathbb{Q}$ and deduce $K=\mathbb{Q}(\gamma)$. (b) Find the minimal polynomial of $\gamma$ over $\mathbb{Q}(\sqrt2)$ and over $\mathbb{Q}(\sqrt6)$. (c) Write $\sqrt2$ and $\sqrt3$ as polynomials in $\gamma$.
(a) The orbit $G\gamma=\{\pm\sqrt2\pm\sqrt3\}$ has four distinct elements, so by Artin's theorem (with $K^G=\mathbb{Q}$) $$m_\gamma=\big((x-\sqrt2)^2-3\big)\big((x+\sqrt2)^2-3\big),$$ which is $(x^2-1-2\sqrt2x)(x^2-1+2\sqrt2x)=(x^2-1)^2-8x^2=x^4-10x^2+1$. Since $[\mathbb{Q}(\gamma):\mathbb{Q}]=4=[K:\mathbb{Q}]$ and $\mathbb{Q}(\gamma)\subseteq K$, $K=\mathbb{Q}(\gamma)$.
(b) $\mathbb{Q}(\sqrt2)=K^{\langle\tau\rangle}$, and the $\langle\tau\rangle$-orbit of $\gamma$ is $\{\sqrt2+\sqrt3,\sqrt2-\sqrt3\}$: the minimal polynomial is $(x-\sqrt2)^2-3=x^2-2\sqrt2\,x-1$. Over $\mathbb{Q}(\sqrt6)=K^{\langle\sigma\tau\rangle}$ the orbit is $\{\gamma,-\gamma\}$ (as $\sigma\tau(\gamma)=-\sqrt2-\sqrt3$), giving $x^2-\gamma^2=x^2-(5+2\sqrt6)$.
(c) $\gamma^3=2\sqrt2+6\sqrt3+9\sqrt2+3\sqrt3=11\sqrt2+9\sqrt3$. Subtracting $9\gamma$ gives $\gamma^3-9\gamma=2\sqrt2$, so $$\sqrt2=\tfrac12(\gamma^3-9\gamma),$$ and $$\sqrt3=\gamma-\sqrt2=\tfrac12(11\gamma-\gamma^3).$$ Check numerically: $\gamma\approx3.14626$, $\tfrac12(\gamma^3-9\gamma)\approx1.41421$.

Normal, Separable and Galois Extensions

Recall from §8.1 that a polynomial is separable if it has no repeated root in a splitting field, an algebraic element is separable if its minimal polynomial is, and an algebraic extension is separable if all its elements are. Over fields of characteristic $0$ and over finite fields every algebraic extension is separable; the standard inseparable example is $\mathbb{F}_p(t)/\mathbb{F}_p(t^p)$, where $x^p-t^p=(x-t)^p$.

📘 Definition 8.12 — Normal extension
An algebraic extension $K/F$ is normal if every irreducible polynomial in $F[x]$ that has one root in $K$ splits into linear factors in $K[x]$. Equivalently: $K$ contains all the conjugates over $F$ of each of its elements.
📘 Definition 8.13 — Galois extension, Galois group
A finite extension $K/F$ is Galois if $|\operatorname{Aut}(K/F)|=[K:F]$. Its automorphism group is then called the Galois group and written $\operatorname{Gal}(K/F)$. If $f\in F[x]$ is separable with splitting field $K$, the Galois group of $f$ over $F$ is $\operatorname{Gal}(K/F)$; by (R2) it does not depend, up to isomorphism, on the choice of splitting field.
📐 Theorem 8.32 — Four characterisations of Galois extensions

For a finite extension $K/F$ with $G=\operatorname{Aut}(K/F)$ the following are equivalent:

  1. $K/F$ is Galois, i.e. $|G|=[K:F]$;
  2. $F$ is exactly the fixed field: $K^G=F$;
  3. $K$ is the splitting field over $F$ of a separable polynomial;
  4. $K/F$ is normal and separable.

When they hold, the conjugates over $F$ of any $\alpha\in K$ are exactly its images $\sigma(\alpha)$, $\sigma\in G$, and its minimal polynomial over $F$ is $\prod_{\beta\in G\alpha}(x-\beta)$.

Proof. $G$ is finite (Theorem 8.31). Put $F_0=K^G\supseteq F$; Artin's theorem gives $[K:F_0]=|G|$.

(i)$\Rightarrow$(ii): $[K:F]=|G|=[K:F_0]$, so $[F_0:F]=1$ by the tower law. (ii)$\Rightarrow$(iv): by Artin's theorem (1) with $K^G=F$, the minimal polynomial of each $\alpha\in K$ is $\prod_{\beta\in G\alpha}(x-\beta)$, which has distinct roots, all in $K$. So $\alpha$ is separable, and an irreducible $p\in F[x]$ with the root $\alpha$ is a constant times this product and splits in $K$. (iv)$\Rightarrow$(iii): write $K=F(\alpha_1,\dots,\alpha_r)$ and let $f$ be the product of the distinct minimal polynomials $m_{\alpha_i}$. Each is separable, and distinct monic irreducibles have no common root (their gcd is $1$), so $f$ is separable. By normality $f$ splits in $K$, and $K$ is generated over $F$ by roots of $f$, so $K$ is a splitting field of $f$. (iii)$\Rightarrow$(i) is (R2). The final sentence was shown in (ii)$\Rightarrow$(iv). ∎

So in characteristic $0$ (and over finite fields), "Galois" simply means "finite and normal". The same extension argument shows that a finite extension is normal if and only if it is the splitting field of some polynomial. Typical cases:

📐 Theorem 8.33 — The Galois group as a group of permutations of the roots

Let $f\in F[x]$ be separable of degree $n$ with roots $\alpha_1,\dots,\alpha_n$ in its splitting field $K$, and $G=\operatorname{Gal}(K/F)$.

  1. $G$ is isomorphic to a subgroup of $S_n$ (acting on the roots); so $[K:F]=|G|$ divides $n!$.
  2. The $G$-orbits on $\{\alpha_1,\dots,\alpha_n\}$ are exactly the root sets of the irreducible factors of $f$.
  3. $f$ is irreducible if and only if $G$ acts transitively on the roots; in that case $n$ divides $|G|$, so $n\le[K:F]\le n!$.

Proof. (1) is Theorem 8.29(3) with Lagrange's theorem in $S_n$. (2) The orbit of $\alpha_i$ is the set of roots of its minimal polynomial (Theorem 8.32), which is the irreducible factor of $f$ having $\alpha_i$ as a root. (3) By (2), one orbit means one irreducible factor; as $f$ is separable this factor occurs once, so $f$ is irreducible up to a constant. Then the orbit–stabiliser theorem gives $n=|G\alpha_1|=[G:G_{\alpha_1}]$, which divides $|G|$. (Directly: $[K:F]=[K:F(\alpha_1)]\cdot n$.) ∎

The permutations that occur are those compatible with all the algebraic relations among the roots. For $x^4-2$, with roots $\sqrt[4]2,\,i\sqrt[4]2,\,-\sqrt[4]2,\,-i\sqrt[4]2$ at the corners of a square, the relation "the sum of opposite roots is $0$" must be preserved, and only the $8$ symmetries of the square survive. Interactive 1 shows the Galois group of several polynomials acting on their roots.

Interactive 1 — The Galois Group Acting on the Roots

Choose a polynomial. Its roots are plotted in the complex plane and numbered $1,\dots,n$. Each element $\sigma$ of the Galois group is shown by arrows from every root to its image (a green ring marks a fixed root). The readouts give $\sigma$ on the generators of $K$, the permutation in cycle notation, its order and the fixed field of $\langle\sigma\rangle$. Press Play to run through the whole group and notice how few of the $n!$ permutations occur.

σ—
σ on generators—
as a permutation—
order of σ—
fixed field of ⟨σ⟩—
G and |G| = [K : ℚ]—
📐 Theorem 8.34 — Galois groups of finite fields
Let $q$ be a prime power and $n\ge1$. Then $\mathbb{F}_{q^n}/\mathbb{F}_q$ is Galois, and $\operatorname{Gal}(\mathbb{F}_{q^n}/\mathbb{F}_q)$ is cyclic of order $n$, generated by the Frobenius automorphism $\varphi_q(x)=x^q$. In particular $\operatorname{Gal}(\mathbb{F}_{p^n}/\mathbb{F}_p)=\langle\varphi\rangle\cong\mathbb{Z}_n$ with $\varphi(x)=x^p$.

Proof. Let $K=\mathbb{F}_{q^n}$. By §8.2, $K$ is the splitting field over $\mathbb{F}_p$ — hence also over $\mathbb{F}_q\subseteq K$ — of $x^{q^n}-x$, whose derivative is $-1$, so it is separable. Thus $K/\mathbb{F}_q$ is Galois of degree $n$ (since $q^n=|K|=q^{[K:\mathbb{F}_q]}$). The map $\varphi_q$ is a power of the Frobenius $x\mapsto x^p$, so it is a ring homomorphism ($(a+b)^p=a^p+b^p$ in characteristic $p$); it is injective, hence bijective on the finite set $K$; and it fixes $\mathbb{F}_q$, because $a^{q-1}=1$ for $a\in\mathbb{F}_q^\times$ by Lagrange. So $\varphi_q\in G$. If $\varphi_q^k=\mathrm{id}$ with $1\le k$, every element of $K$ is a root of $x^{q^k}-x$, which has at most $q^k$ roots, so $q^n\le q^k$ and $k\ge n$. Hence $\varphi_q$ has order at least $n=|G|$, so exactly $n$, and $G=\langle\varphi_q\rangle$. ∎ (For $q=p$ this is the description of all automorphisms of $\mathbb{F}_{p^n}$ in §8.2; the point here is the relative version over any subfield $\mathbb{F}_q$.)

✏️ Example 3 — Prove or disprove
Decide each statement. (a) If $F\subseteq E\subseteq K$ and both $E/F$ and $K/E$ are normal, then $K/F$ is normal. (b) If $K/F$ is Galois and $F\subseteq E\subseteq K$, then $K/E$ is Galois. (c) In the same situation, $E/F$ is Galois. (d) For every finite extension, $|\operatorname{Aut}(K/F)|$ divides $[K:F]$. (e) Every extension of degree $2$ is Galois.
(a) False Take $\mathbb{Q}\subset\mathbb{Q}(\sqrt2)\subset\mathbb{Q}(\sqrt[4]2)$. Each step has degree $2$, and a degree-$2$ extension is normal: if $\alpha$ has minimal polynomial $x^2+bx+c$, the other root is $-b-\alpha\in F(\alpha)$. But $x^4-2$ is irreducible over $\mathbb{Q}$ (Eisenstein at $2$) with the root $\sqrt[4]2\in\mathbb{Q}(\sqrt[4]2)\subset\mathbb{R}$, while its root $i\sqrt[4]2$ is not real. Normality is not transitive.
(b) True If $K$ is the splitting field over $F$ of a separable $f$, then $K$ is also the splitting field of $f$ over $E$: $f$ splits in $K$, and $K=F(\text{roots})\subseteq E(\text{roots})\subseteq K$. Apply Theorem 8.32(iii).
(c) False $K=\mathbb{Q}(\sqrt[3]2,\omega)$ is Galois over $\mathbb{Q}$ (splitting field of $x^3-2$), but $E=\mathbb{Q}(\sqrt[3]2)$ is not normal over $\mathbb{Q}$ (Example 1(a)). Which intermediate fields are Galois is decided by the fundamental theorem below: exactly those whose group is a normal subgroup.
(d) True This is the last part of Artin's theorem: $|\operatorname{Aut}(K/F)|=[K:K^{\operatorname{Aut}(K/F)}]$, a factor of $[K:F]$ by the tower law. For $\mathbb{Q}(\sqrt[4]2)/\mathbb{Q}$ it equals $2$, a proper divisor of $4$.
(e) False It is true in characteristic $\ne2$, but fails in characteristic $2$. Let $F=\mathbb{F}_2(t)$ and $K=F(\sqrt t)$, obtained by adjoining a root $u$ of $x^2-t$. This polynomial is irreducible ($t$ is not a square in $\mathbb{F}_2(t)$: in $(a/b)^2=t$ the degrees of the two sides have different parity), so $[K:F]=2$; but $x^2-t=(x-u)^2$, so $u$ is inseparable and $\operatorname{Aut}(K/F)=\{1\}$. The extension is normal but not Galois.

The Fundamental Theorem of Galois Theory

📐 Theorem 8.35 — Fundamental theorem of Galois theory

Let $K/F$ be a finite Galois extension with group $G=\operatorname{Gal}(K/F)$. For an intermediate field $F\subseteq E\subseteq K$ write $E^*=\operatorname{Gal}(K/E)=\operatorname{Aut}(K/E)$, and for a subgroup $H\le G$ write $H^*=K^H$.

  1. Correspondence. $E\mapsto E^*$ and $H\mapsto H^*$ are mutually inverse, inclusion-reversing bijections between the intermediate fields of $K/F$ and the subgroups of $G$.
  2. Degrees. $K/E$ is Galois, $[K:E]=|E^*|$ and $[E:F]=[G:E^*]$.
  3. Conjugation. For $\sigma\in G$, $\sigma(E)^*=\sigma E^*\sigma^{-1}$: conjugate subgroups correspond to conjugate (isomorphic) intermediate fields.
  4. Normality. The following are equivalent: $E/F$ is Galois; $E/F$ is normal; $\sigma(E)=E$ for every $\sigma\in G$; $E^*$ is a normal subgroup of $G$. In that case restriction $\sigma\mapsto\sigma|_E$ is a surjective homomorphism $G\to\operatorname{Gal}(E/F)$ with kernel $E^*$, so $$\operatorname{Gal}(E/F)\cong G/\operatorname{Gal}(K/E).$$
  5. Lattice. $(E_1E_2)^*=E_1^*\cap E_2^*$ and $(E_1\cap E_2)^*=\langle E_1^*,E_2^*\rangle$, where $E_1E_2$ is the compositum (the smallest subfield containing both).

Proof. (2) $K/E$ is Galois by Example 3(b), so $|\operatorname{Gal}(K/E)|=[K:E]$, and $[E:F]=[K:F]/[K:E]=|G|/|E^*|=[G:E^*]$.

(1) Inclusion reversal is Theorem 8.30(2). For an intermediate field $E$, $E^{**}=K^{\operatorname{Gal}(K/E)}=E$ by Theorem 8.32(ii) applied to the Galois extension $K/E$. For a subgroup $H$, $H^{**}=\operatorname{Aut}(K/K^H)=H$ by Artin's theorem, since $H$ is finite. So the two maps are inverse to each other.

(3) is Theorem 8.30(4) with $H=E^*$.

(4) Let $H=E^*$. If $E/F$ is normal and $\sigma\in G$, then for $\alpha\in E$, $\sigma(\alpha)$ is a root of the minimal polynomial of $\alpha$, which splits in $E$; so $\sigma(E)\subseteq E$, with equality since $\sigma$ is an injective $F$-linear map of the finite-dimensional space $E$. If $\sigma(E)=E$ for all $\sigma$, then $\sigma H\sigma^{-1}=\sigma(E)^*=E^*=H$ by (3): $H\trianglelefteq G$. If $H\trianglelefteq G$, then $\sigma(E)^*=\sigma H\sigma^{-1}=H=E^*$, so $\sigma(E)=E$ by the injectivity in (1). In this situation each $\sigma\in G$ restricts to an element $\sigma|_E\in\operatorname{Aut}(E/F)$; restriction is a homomorphism whose kernel consists of the $\sigma$ fixing $E$ pointwise, namely $H$. Its image has order $|G|/|H|=[E:F]$, while $|\operatorname{Aut}(E/F)|\le[E:F]$ (Theorem 8.31). Hence restriction is onto, $|\operatorname{Aut}(E/F)|=[E:F]$, $E/F$ is Galois, and $\operatorname{Gal}(E/F)\cong G/H$ by the first isomorphism theorem (§6.2). Finally Galois implies normal (Theorem 8.32), which closes the cycle of implications.

(5) An automorphism fixes $E_1E_2$ pointwise iff it fixes $E_1\cup E_2$ pointwise (the fixed set is a field), i.e. iff it lies in $E_1^*\cap E_2^*$. And $K^{\langle H_1,H_2\rangle}=K^{H_1}\cap K^{H_2}$, because an element is fixed by the subgroup generated by $H_1\cup H_2$ iff it is fixed by every element of $H_1$ and of $H_2$; apply this with $H_i=E_i^*$ and use (1). ∎

Draw the subgroup lattice with $\{1\}$ at the top and $G$ at the bottom, and the field lattice with $K$ at the top and $F$ at the bottom: the correspondence is then a horizontal reflection, and the index of each edge on the group side equals the degree of the matching edge on the field side. Interactive 2 draws both lattices.

🧭 Problem-solving strategy — computing a Galois correspondence
  1. Identify $K$ as a splitting field of a separable polynomial (so $K/F$ is Galois) and compute $[K:F]=|G|$ with the tower law.
  2. Describe each $\sigma$ by its values on generators — roots, radicals, roots of unity — using (R1) to know which choices are allowed. Identify $G$ by its multiplication or as permutations of the roots (Theorem 8.33).
  3. List the subgroups (cyclic subgroups first; Lagrange, Sylow and Cauchy tell you which orders occur).
  4. Find each fixed field. $[K:K^H]=|H|$, so you need an element $\theta$ fixed by $H$ whose $G$-orbit has exactly $[G:H]$ elements; then $K^H=F(\theta)$. Good candidates: $\sum_{h\in H}h(\alpha)$, $\prod_{h\in H}h(\alpha)$, or a square root you can see. A guaranteed recipe: if $K=F(\alpha)$, then $K^H$ is generated over $F$ by the coefficients of $\prod_{h\in H}(x-h(\alpha))$, the minimal polynomial of $\alpha$ over $K^H$.
  5. Read off normality: normal subgroups $\leftrightarrow$ Galois subextensions; conjugate subgroups $\leftrightarrow$ conjugate fields, which are isomorphic but different.

Why does the recipe in step 4 work? Let $E_0\subseteq K^H$ be the field generated over $F$ by those coefficients. Then $\alpha$ is a root of a polynomial of degree $|H|$ over $E_0$, so $[K:E_0]=[E_0(\alpha):E_0]\le|H|=[K:K^H]$, which forces $E_0=K^H$.

Interactive 2 — The Galois Correspondence, Lattice by Lattice

Left: the subgroups $H$ of $G$, with $\{1\}$ at the top. Right: the intermediate fields $E=K^H$, with $K$ at the top. Heights are proportional to $\log|H|$, so a pair sits at the same level, and each edge carries the same number on both sides (an index on the left, a degree on the right). A green border marks a normal subgroup and the matching Galois subextension; amber marks a non-normal one. Click a box or press Play to step through the pairs.

H—
|H| = [K : E]—
[G : H] = [E : F]—
E = K^H—
H normal? Gal(E/F)—
✏️ Example 4 — The splitting field of $x^3-2$
Let $K$ be the splitting field of $x^3-2$ over $\mathbb{Q}$. (a) Show $K=\mathbb{Q}(\sqrt[3]2,\omega)$ and $[K:\mathbb{Q}]=6$. (b) Show $\operatorname{Gal}(K/\mathbb{Q})\cong S_3$ and describe its elements. (c) List all intermediate fields. (d) Which of them are Galois over $\mathbb{Q}$?
(a) The roots are $\alpha_0=\sqrt[3]2$, $\alpha_1=\omega\sqrt[3]2$, $\alpha_2=\omega^2\sqrt[3]2$, so $K=\mathbb{Q}(\alpha_0,\alpha_1,\alpha_2)=\mathbb{Q}(\sqrt[3]2,\omega)$ (as $\omega=\alpha_1/\alpha_0$). $[\mathbb{Q}(\sqrt[3]2):\mathbb{Q}]=3$ by Eisenstein; $\omega\notin\mathbb{Q}(\sqrt[3]2)\subset\mathbb{R}$ and $\omega^2+\omega+1=0$, so $[K:\mathbb{Q}(\sqrt[3]2)]=2$ and $[K:\mathbb{Q}]=6$.
(b) $G$ embeds in $S_3$ (Theorem 8.33) and $|G|=6$, so $G\cong S_3$: every permutation of the roots occurs. Each element is determined by $\sqrt[3]2\mapsto\omega^k\sqrt[3]2$ ($k=0,1,2$) and $\omega\mapsto\omega^{\pm1}$. Let $\sigma:\sqrt[3]2\mapsto\omega\sqrt[3]2,\ \omega\mapsto\omega$ (the $3$-cycle $\alpha_0\to\alpha_1\to\alpha_2\to\alpha_0$) and $\tau$ = complex conjugation ($\alpha_0$ fixed, $\alpha_1\leftrightarrow\alpha_2$). Then $\sigma^3=\tau^2=1$ and $\tau\sigma\tau=\sigma^{-1}$; the elements are $1,\sigma,\sigma^2,\tau,\sigma\tau,\sigma^2\tau$.
(c) $S_3$ has six subgroups: $\{1\}$, three of order $2$, $\langle\sigma\rangle$ of order $3$, and $G$. Fixed fields (degree over $\mathbb{Q}$ = index):
$\langle\tau\rangle\mapsto\mathbb{Q}(\sqrt[3]2)$ (degree $3$; $\tau$ fixes the real root);
$\langle\sigma\tau\rangle\mapsto\mathbb{Q}(\omega^2\sqrt[3]2)$, since $\sigma\tau(\alpha_2)=\sigma(\alpha_1)=\alpha_2$;
$\langle\sigma^2\tau\rangle\mapsto\mathbb{Q}(\omega\sqrt[3]2)$, since $\sigma^2\tau(\alpha_1)=\sigma^2(\alpha_2)=\alpha_1$;
$\langle\sigma\rangle\mapsto\mathbb{Q}(\omega)=\mathbb{Q}(\sqrt{-3})$ (degree $2$; $\sigma$ fixes $\omega$).
Together with $\mathbb{Q}$ and $K$ these are all six intermediate fields — there are no others.
(d) The normal subgroups of $S_3$ are $\{1\}$, $A_3=\langle\sigma\rangle$ and $S_3$, so the Galois subextensions are $K$, $\mathbb{Q}(\omega)$ and $\mathbb{Q}$, with $\operatorname{Gal}(\mathbb{Q}(\omega)/\mathbb{Q})\cong S_3/A_3\cong\mathbb{Z}_2$. The three cubic fields correspond to the three conjugate subgroups of order $2$; they are isomorphic to one another (each is $\mathbb{Q}[x]/\langle x^3-2\rangle$) but different, and $\sigma$ carries $\mathbb{Q}(\sqrt[3]2)$ to $\mathbb{Q}(\omega\sqrt[3]2)$, matching $\sigma\langle\tau\rangle\sigma^{-1}=\langle\sigma^2\tau\rangle$.
Check Lattice (5): $\mathbb{Q}(\sqrt[3]2)\cap\mathbb{Q}(\omega\sqrt[3]2)$ corresponds to $\langle\tau,\sigma^2\tau\rangle=G$, so it is $\mathbb{Q}$; their compositum corresponds to $\langle\tau\rangle\cap\langle\sigma^2\tau\rangle=\{1\}$, so it is $K$. Indeed both $\sqrt[3]2$ and $\omega=\omega\sqrt[3]2/\sqrt[3]2$ lie in the compositum.
✏️ Example 5 — $x^4-2$: the dihedral group and a change of base field
Let $K$ be the splitting field of $x^4-2$ over $\mathbb{Q}$ and $r=\sqrt[4]2$. (a) Show $K=\mathbb{Q}(r,i)$, $[K:\mathbb{Q}]=8$ and $\operatorname{Gal}(K/\mathbb{Q})\cong D_4$. (b) List the ten intermediate fields. (c) Find the Galois group of $x^4-2$ over $\mathbb{Q}(i)$, over $\mathbb{Q}(\sqrt2)$ and over $\mathbb{R}$.
(a) The roots are $i^jr$ ($j=0,1,2,3$), so $K=\mathbb{Q}(r,i)$. $[\mathbb{Q}(r):\mathbb{Q}]=4$ (Eisenstein at $2$), $\mathbb{Q}(r)\subset\mathbb{R}$ and $i\notin\mathbb{R}$, so $[K:\mathbb{Q}]=8$. An automorphism is determined by $r\mapsto i^kr$ and $i\mapsto\pm i$: at most $8$ choices, and $|G|=8$, so all occur. Let $\sigma:r\mapsto ir,\ i\mapsto i$ (order $4$: $r\mapsto ir\mapsto-r\mapsto-ir\mapsto r$) and $\tau$ = complex conjugation. Then $\tau\sigma\tau(r)=\tau\sigma(r)=\tau(ir)=-ir=\sigma^{-1}(r)$ and $\tau\sigma\tau(i)=i$, so $\tau\sigma\tau=\sigma^{-1}$: $G=\langle\sigma,\tau\rangle\cong D_4$, the symmetry group of the square of roots. In general $\sigma^k\tau$ sends $r\mapsto i^kr$ and $i\mapsto-i$.
(b) $D_4$ has $10$ subgroups. Fixed fields (check each element is fixed; the degree is the index):
order $4$: $\langle\sigma\rangle\mapsto\mathbb{Q}(i)$; $\langle\sigma^2,\tau\rangle\mapsto\mathbb{Q}(\sqrt2)$ ($\sqrt2=r^2$); $\langle\sigma^2,\sigma\tau\rangle\mapsto\mathbb{Q}(\sqrt{-2})$ ($\sigma\tau(ir^2)=(-i)(ir)^2=ir^2$);
order $2$: $\langle\sigma^2\rangle\mapsto\mathbb{Q}(\sqrt2,i)$; $\langle\tau\rangle\mapsto\mathbb{Q}(r)$; $\langle\sigma^2\tau\rangle\mapsto\mathbb{Q}(ir)$ ($\sigma^2\tau(ir)=(-i)(-r)=ir$); $\langle\sigma\tau\rangle\mapsto\mathbb{Q}((1+i)r)$; $\langle\sigma^3\tau\rangle\mapsto\mathbb{Q}((1-i)r)$;
and $\{1\}\mapsto K$, $G\mapsto\mathbb{Q}$. For instance $\sigma\tau((1+i)r)=(1-i)\,ir=(1+i)r$, and the $G$-orbit of $(1+i)r$ is $\{\pm(1+i)r,\pm(1-i)r\}$, of size $4=[G:\langle\sigma\tau\rangle]$; its minimal polynomial is $x^4+8$.
Normality The normal subgroups are $\{1\}$, $\langle\sigma^2\rangle$ (the centre), the three of order $4$ (index $2$) and $G$. So $\mathbb{Q}(i)$, $\mathbb{Q}(\sqrt2)$, $\mathbb{Q}(\sqrt{-2})$ and $\mathbb{Q}(\sqrt2,i)$ are Galois over $\mathbb{Q}$, while the four quartic fields $\mathbb{Q}(r)$, $\mathbb{Q}(ir)$, $\mathbb{Q}((1\pm i)r)$ are not; they come in two conjugate pairs, since $\sigma\langle\tau\rangle\sigma^{-1}=\langle\sigma^2\tau\rangle$ and $\sigma\langle\sigma\tau\rangle\sigma^{-1}=\langle\sigma^3\tau\rangle$.
(c) Over $E$, the splitting field of $x^4-2$ is $EK$, and for $E\subseteq K$ its group is $\operatorname{Gal}(K/E)=E^*$. Over $\mathbb{Q}(i)$: $\langle\sigma\rangle\cong\mathbb{Z}_4$, and $x^4-2$ stays irreducible since $[K:\mathbb{Q}(i)]=4$. Over $\mathbb{Q}(\sqrt2)$: $\langle\sigma^2,\tau\rangle\cong V_4$, and $x^4-2=(x^2-\sqrt2)(x^2+\sqrt2)$. Over $\mathbb{R}$: the splitting field is $\mathbb{C}$ and the group is $\mathbb{Z}_2$. The Galois group of a polynomial depends on the base field.
📐 Theorem 8.36 — Cyclotomic extensions
Let $n\ge1$ and $\zeta=\zeta_n=e^{2\pi i/n}$. Then $\mathbb{Q}(\zeta)$ is the splitting field of $x^n-1$ over $\mathbb{Q}$, $[\mathbb{Q}(\zeta):\mathbb{Q}]=\varphi(n)$, and $$(\mathbb{Z}/n\mathbb{Z})^\times\to\operatorname{Gal}(\mathbb{Q}(\zeta)/\mathbb{Q}),\qquad a\mapsto\sigma_a,\quad\sigma_a(\zeta)=\zeta^a,$$ is an isomorphism. In particular the Galois group is abelian, and every subfield of $\mathbb{Q}(\zeta_n)$ is Galois over $\mathbb{Q}$.

Proof. The roots of $x^n-1$ are $1,\zeta,\dots,\zeta^{n-1}$, all in $\mathbb{Q}(\zeta)$, so $\mathbb{Q}(\zeta)$ is its splitting field; $x^n-1$ is separable since $\gcd(x^n-1,nx^{n-1})=1$. The minimal polynomial of $\zeta$ is the cyclotomic polynomial $\Phi_n$, irreducible over $\mathbb{Q}$ of degree $\varphi(n)$ (§7.3), so $[\mathbb{Q}(\zeta):\mathbb{Q}]=\varphi(n)$. Each $\sigma\in G$ sends $\zeta$ to a root of $\Phi_n$, i.e. to a primitive $n$-th root of unity $\zeta^a$ with $\gcd(a,n)=1$, and $a$ is determined modulo $n$. The map $\sigma\mapsto a\bmod n$ is injective because $\sigma$ is determined by $\sigma(\zeta)$, and multiplicative because $\sigma_a\sigma_b(\zeta)=\sigma_a(\zeta^b)=\zeta^{ab}$. Both groups have order $\varphi(n)$, so it is an isomorphism. Subgroups of an abelian group are normal, so Theorem 8.35(4) applies to every subfield. ∎

The subfields of $\mathbb{Q}(\zeta_n)$ are therefore read off from the subgroups of $(\mathbb{Z}/n)^\times$. For a prime $p$ this group is cyclic of order $p-1$ (§5.3), so $\mathbb{Q}(\zeta_p)$ has exactly one subfield of each degree $d\mid p-1$.

✏️ Example 6 — Subfields of $\mathbb{Q}(\zeta_7)$ and of $\mathbb{F}_{64}$
(a) Find all subfields of $\mathbb{Q}(\zeta_7)$, with explicit generators and minimal polynomials. (b) Use Theorem 8.34 to find all subfields of $\mathbb{F}_{64}$, and count the elements $\alpha\in\mathbb{F}_{64}$ with $\mathbb{F}_2(\alpha)=\mathbb{F}_{64}$.
(a) group $G\cong(\mathbb{Z}/7)^\times$, cyclic of order $6$ generated by $3$ ($3,2,6,4,5,1$ are its powers mod $7$). Its subgroups are $\{1\}$, $\langle\sigma_6\rangle$ (order $2$; $\sigma_6=\sigma_{-1}$ is complex conjugation), $\langle\sigma_2\rangle=\{\sigma_1,\sigma_2,\sigma_4\}$ (order $3$) and $G$. So $\mathbb{Q}(\zeta_7)$ has exactly four subfields: $\mathbb{Q}$, one cubic, one quadratic, and $\mathbb{Q}(\zeta_7)$.
(a) quadratic Let $\eta=\zeta+\zeta^2+\zeta^4$ (sum over the subgroup $\{1,2,4\}$ of exponents). It is fixed by $\sigma_2,\sigma_4$, and $\sigma_3(\eta)=\zeta^3+\zeta^6+\zeta^5=\eta'$. Since $\eta+\eta'=\zeta+\dots+\zeta^6=-1$ and $\eta\eta'$ is a sum of nine powers $\zeta^{a+b}$ that equals $3+(\zeta+\dots+\zeta^6)=2$, $\eta$ is a root of $x^2+x+2$, so $\eta=\tfrac{-1+\sqrt{-7}}2$ (its imaginary part $\sin\tfrac{2\pi}7+\sin\tfrac{4\pi}7+\sin\tfrac{8\pi}7\approx1.3229=\tfrac{\sqrt7}2$ is positive). The quadratic subfield is $\mathbb{Q}(\sqrt{-7})$.
(a) cubic The fixed field of complex conjugation is $\mathbb{Q}(\zeta)\cap\mathbb{R}$; it contains $\theta=\zeta+\zeta^{-1}=2\cos\tfrac{2\pi}7$, whose $G$-orbit $\{2\cos\tfrac{2\pi}7,2\cos\tfrac{4\pi}7,2\cos\tfrac{6\pi}7\}$ has three elements. So the cubic subfield is $\mathbb{Q}(\cos\tfrac{2\pi}7)$, and the orbit polynomial is $x^3+x^2-2x-1$ (from $\theta^3=\zeta^3+\zeta^{-3}+3\theta$ and $\theta^2=\zeta^2+\zeta^{-2}+2$ we get $\theta^3+\theta^2=(\zeta^2+\zeta^3+\zeta^4+\zeta^5)+3\theta+2=(-1-\theta)+3\theta+2=2\theta+1$).
(b) $\operatorname{Gal}(\mathbb{F}_{64}/\mathbb{F}_2)=\langle\varphi\rangle\cong\mathbb{Z}_6$. Its subgroups are $\langle\varphi^d\rangle$ for $d\mid6$, and the fixed field of $\langle\varphi^d\rangle$ is $\{x:x^{2^d}=x\}=\mathbb{F}_{2^d}$. So the subfields are $\mathbb{F}_2,\mathbb{F}_4,\mathbb{F}_8,\mathbb{F}_{64}$ — one for each divisor of $6$, recovering the subfield theorem of §8.2; $\mathbb{F}_{16}$ and $\mathbb{F}_{32}$ are not subfields. By Theorem 8.35(5), $\mathbb{F}_4\cap\mathbb{F}_8\leftrightarrow\langle\varphi^2,\varphi^3\rangle=G$, so $\mathbb{F}_4\cap\mathbb{F}_8=\mathbb{F}_2$, and $\mathbb{F}_4\mathbb{F}_8=\mathbb{F}_{64}$. An element generates $\mathbb{F}_{64}$ iff it lies in no maximal proper subfield, $\mathbb{F}_4$ or $\mathbb{F}_8$; $|\mathbb{F}_4\cup\mathbb{F}_8|=4+8-2=10$, so $54$ elements are generators. Each is a root of a monic irreducible sextic, which has $6$ roots: there are $54/6=9$ irreducible sextics over $\mathbb{F}_2$, as Gauss's formula (§7.3) says.

Computing Galois Groups; Solvability by Radicals

To identify a Galois group as a subgroup of $S_n$ one needs invariants that can be computed from the coefficients. The simplest is the discriminant, which detects whether all the permutations are even.

📘 Definition 8.14 — Discriminant
Let $f\in F[x]$ be monic of degree $n$ with roots $\alpha_1,\dots,\alpha_n$ in a splitting field. Put $\delta=\prod_{i\lt j}(\alpha_i-\alpha_j)$ and define the discriminant $\Delta(f)=\delta^2$. Then $\Delta\ne0$ iff $f$ is separable. For quadratics and cubics: $$\Delta(x^2+bx+c)=b^2-4c,\qquad\Delta(x^3+px+q)=-4p^3-27q^2,$$ and $\Delta(x^3+ax^2+bx+c)=a^2b^2-4b^3-4a^3c-27c^2+18abc$.

If $\sigma\in\operatorname{Gal}(K/F)$ induces the permutation $\pi$ of the roots, then $\sigma(\delta)=\prod_{i\lt j}(\alpha_{\pi(i)}-\alpha_{\pi(j)})=\operatorname{sgn}(\pi)\,\delta$, because each pair appears once, reversed exactly for the inversions of $\pi$. Hence $\sigma(\Delta)=\Delta$ for all $\sigma$, and $\Delta\in K^G=F$: the discriminant is a polynomial in the coefficients.

📐 Theorem 8.37 — The discriminant and $A_n$
Let $\operatorname{char}F\ne2$ and let $f\in F[x]$ be separable of degree $n$ with Galois group $G\le S_n$. Then $G\subseteq A_n$ if and only if $\Delta(f)$ is a square in $F$. More precisely, the fixed field of $G\cap A_n$ is $F(\delta)=F(\sqrt{\Delta})$.

Proof. By the computation above, $\sigma(\delta)=\operatorname{sgn}(\sigma)\delta$, and $\delta\ne-\delta$ because $\delta\ne0$ and $\operatorname{char}F\ne2$. So $\delta$ is fixed by $\sigma$ iff $\sigma$ is even, and $\delta\in K^G=F$ iff $G\subseteq A_n$. The square roots of $\Delta$ in $K$ are $\pm\delta$, so $\Delta$ is a square in $F$ iff $\delta\in F$. Finally $\delta$ is fixed by $G\cap A_n$, so $F(\delta)\subseteq K^{G\cap A_n}$. If $G\subseteq A_n$ both fields are $F$. Otherwise $[G:G\cap A_n]=2$, so $[K^{G\cap A_n}:F]=2$, and $\delta\notin F$ gives $[F(\delta):F]=2$; the fields coincide. ∎

📐 Theorem 8.38 — Galois groups of cubics
Let $\operatorname{char}F\ne2$ and let $f\in F[x]$ be an irreducible separable cubic with discriminant $\Delta$. If $\Delta$ is a square in $F$, then $\operatorname{Gal}(f)\cong A_3\cong\mathbb{Z}_3$, the splitting field has degree $3$, and it equals $F(\alpha)$ for any one root $\alpha$. Otherwise $\operatorname{Gal}(f)\cong S_3$ and the splitting field has degree $6$, containing the unique quadratic subfield $F(\sqrt\Delta)$.

Proof. $G\le S_3$ is transitive (Theorem 8.33), so $3$ divides $|G|$ and $G$ is $A_3$ or $S_3$; Theorem 8.37 decides which. In the first case $[K:F]=3=[F(\alpha):F]$, so $K=F(\alpha)$. In the second, $F(\sqrt\Delta)$ is the fixed field of $A_3$, the only subgroup of index $2$. ∎ (This recovers, with the group in view, the description of splitting fields of cubics in §8.1.)

Irreducibility is essential: $(x-1)(x^2+1)$ has $\Delta=-16$, not a square, yet its Galois group is $\mathbb{Z}_2$, not $S_3$. For an irreducible quartic $G$ is one of the five transitive subgroups of $S_4$: $S_4$ ($x^4-x-1$), $A_4$ ($x^4+8x+12$, §7.3), $D_4$ ($x^4-2$), $\mathbb{Z}_4$ ($x^4-4x^2+2$, Exercise II.9) or $V_4$ ($x^4-10x^2+1$).

📘 Definition 8.15 — Solvable group
A group $G$ is solvable if it has a chain of subgroups $$\{1\}=G_0\trianglelefteq G_1\trianglelefteq\cdots\trianglelefteq G_k=G$$ in which each $G_{i-1}$ is normal in $G_i$ and each quotient $G_i/G_{i-1}$ is abelian.
📐 Theorem 8.39 — Solvable groups
  1. Subgroups and homomorphic images of solvable groups are solvable. If $N\trianglelefteq G$ and both $N$ and $G/N$ are solvable, then $G$ is solvable.
  2. Abelian groups, $p$-groups, $S_3$ and $S_4$ are solvable.
  3. For $n\ge5$, neither $S_n$ nor $A_n$ is solvable.

Proof. (1) Subgroups: for $H\le G$ put $H_i=H\cap G_i$. Then $H_{i-1}=H_i\cap G_{i-1}\trianglelefteq H_i$, and by the second isomorphism theorem $H_i/H_{i-1}\cong H_iG_{i-1}/G_{i-1}\le G_i/G_{i-1}$, which is abelian. Images: if $\psi:G\to Q$ is onto, put $Q_i=\psi(G_i)$; then $Q_{i-1}\trianglelefteq Q_i$, and $G_i\to Q_i/Q_{i-1}$ is onto and kills $G_{i-1}$, so $Q_i/Q_{i-1}$ is a quotient of the abelian group $G_i/G_{i-1}$. Extensions: by the correspondence theorem a chain for $G/N$ is $M_0/N\trianglelefteq\dots\trianglelefteq M_l/N$ with $N=M_0\le\dots\le M_l=G$, and $M_j/M_{j-1}\cong(M_j/N)/(M_{j-1}/N)$ is abelian by the third isomorphism theorem (§6.2). Put a chain for $N$ below it.

(2) An abelian group has the chain $\{1\}\trianglelefteq G$. A $p$-group $G\ne1$ has nontrivial centre $Z$ (§6.3); $Z$ is abelian and $G/Z$ is a smaller $p$-group, so induction and (1) apply. $S_3$: $\{1\}\trianglelefteq A_3\trianglelefteq S_3$ with quotients $\mathbb{Z}_3,\mathbb{Z}_2$. $S_4$: $\{1\}\trianglelefteq V_4\trianglelefteq A_4\trianglelefteq S_4$, where $V_4=\{1,(12)(34),(13)(24),(14)(23)\}$ is normal in $S_4$ (it is a union of conjugacy classes), with quotients $V_4$, $\mathbb{Z}_3$, $\mathbb{Z}_2$.

(3) Claim: if $n\ge5$, $H\le S_n$ contains every $3$-cycle and $N\trianglelefteq H$ with $H/N$ abelian, then $N$ contains every $3$-cycle. Indeed, for $x,y\in H$ the image of $xyx^{-1}y^{-1}$ in the abelian group $H/N$ is trivial, so $xyx^{-1}y^{-1}\in N$. Given a $3$-cycle $(a\,b\,c)$, choose $d,e$ distinct from each other and from $a,b,c$ (possible as $n\ge5$), and let $x=(a\,b\,d)$, $y=(a\,c\,e)$. Following each of $a,b,c,d,e$ through $y^{-1}$, then $x^{-1}$, $y$, $x$ shows $$xyx^{-1}y^{-1}=(a\,b\,c),$$ which therefore lies in $N$. Now if $G=S_n$ or $A_n$ had a chain as in Definition 8.15, then $G_k=G$ contains all $3$-cycles, and by the claim so do $G_{k-1},G_{k-2},\dots,G_0=\{1\}$ — absurd. ∎

(One can show that $A_5$, of order $60$, is the smallest non-solvable group; compare its simplicity, §6.4.)

📘 Definition 8.16 — Radical extension; solvable by radicals
Let $\operatorname{char}F=0$. An extension $L/F$ is radical if there is a tower $F=L_0\subseteq L_1\subseteq\dots\subseteq L_m=L$ with $L_i=L_{i-1}(a_i)$ and $a_i^{n_i}\in L_{i-1}$ for some $n_i\ge1$. A polynomial $f\in F[x]$ is solvable by radicals if its splitting field is contained in some radical extension of $F$ — that is, if its roots can be obtained from the coefficients by field operations and extraction of $n$-th roots.
📐 Theorem 8.40 — Galois's theorem on solvability
Let $\operatorname{char}F=0$ and $f\in F[x]$. Then $f$ is solvable by radicals if and only if its Galois group is solvable. In particular every polynomial of degree at most $4$ is solvable by radicals.

Proof sketch. ($\Rightarrow$) Suppose the splitting field $K$ of $f$ lies in a radical extension $L$, with exponents $n_i$, and let $N=\prod n_i$. Enlarging $L$, we may assume it contains a primitive $N$-th root of unity $\zeta$ and is Galois over $F$: adjoin $\zeta$ first, then all conjugates of the $a_i$ (each conjugate of $a_i$ again has its $n_i$-th power in the previous layer). Now (i) $F(\zeta)/F$ has abelian Galois group: $\sigma\mapsto a$ with $\sigma(\zeta)=\zeta^a$ is an injective homomorphism into $(\mathbb{Z}/N)^\times$, exactly as in Theorem 8.36; and (ii) once $\zeta$ is present, each step $E(a)/E$ with $a^n\in E$, $n\mid N$, is Galois with abelian group: every conjugate of $a$ is $\zeta^ja$, so $\sigma(a)=\zeta^{j_\sigma}a$, and $\sigma\mapsto j_\sigma\bmod n$ is an injective homomorphism into $\mathbb{Z}_n$. By the fundamental theorem these layers give a chain of subgroups of $\operatorname{Gal}(L/F)$ with abelian quotients, so $\operatorname{Gal}(L/F)$ is solvable, and $\operatorname{Gal}(K/F)$, a quotient of it by Theorem 8.35(4), is solvable by Theorem 8.39(1).

($\Leftarrow$) Adjoin a primitive $|G|$-th root of unity; the Galois group of $f$ over the new base is a subgroup of $G$, still solvable, so it has a chain with cyclic quotients of prime order $p$. Each corresponding step is a cyclic extension of degree $p$ of a field containing the $p$-th roots of unity, and such an extension is $E(a)$ with $a^p\in E$ (Kummer theory, using the Lagrange resolvent $a=\sum_{j}\zeta_p^{-j}\sigma^j(\beta)$). For $\deg f\le4$, $G$ is a subgroup of $S_4$, solvable by Theorem 8.39. ∎

📐 Theorem 8.41 — An irreducible polynomial of prime degree with two non-real roots
Let $p$ be prime and $f\in\mathbb{Q}[x]$ irreducible of degree $p$ with exactly two non-real roots in $\mathbb{C}$. Then $\operatorname{Gal}(f)\cong S_p$. For $p\ge5$, $f$ is therefore not solvable by radicals.

Proof. $G\le S_p$ is transitive, so $p$ divides $|G|$ (Theorem 8.33), and by Cauchy's theorem (§6.4) $G$ has an element of order $p$. In $S_p$ an element of order $p$ is a $p$-cycle, since its cycle lengths are at most $p$, sum to $p$ and have least common multiple $p$. The splitting field $K\subseteq\mathbb{C}$ is mapped to itself by complex conjugation (which permutes the roots of the real polynomial $f$), so conjugation restricts to an element of $G$; it fixes the $p-2$ real roots and swaps the two non-real ones: a transposition. It remains to show that a $p$-cycle $c$ and a transposition $(a\,b)$ generate $S_p$. Some power $c^k$ sends $a$ to $b$, and $c^k$ is again a $p$-cycle because $p$ is prime. Number the roots so that $c^k=(1\,2\,\cdots\,p)$ and $(a\,b)=(1\,2)$. Conjugating, $c^{kj}(1\,2)c^{-kj}=(j{+}1\ \,j{+}2)$, which gives all adjacent transpositions, and these generate $S_p$ (every permutation can be sorted by adjacent swaps). For $p\ge5$, $S_p$ is not solvable (Theorem 8.39), so Theorem 8.40 applies. ∎

✏️ Example 7 — Cubics, an unsolvable quintic, and a solvable one
Find the Galois groups over $\mathbb{Q}$ of (a) $x^3-3x+1$; (b) $x^3-x-1$; (c) $x^5-4x+2$; (d) $x^5-2$. Which are solvable by radicals?
(a) Candidates $\pm1$ give $-1$ and $3$, so the cubic is irreducible. $\Delta=-4(-3)^3-27=108-27=81=9^2$, a square: $G\cong\mathbb{Z}_3$ and the splitting field is $\mathbb{Q}(\rho)$ for any root. With $\rho=2\cos\tfrac{2\pi}9$, the identity $2\cos2\theta=(2\cos\theta)^2-2$ shows the roots are $\rho,\ \rho^2-2=2\cos\tfrac{4\pi}9$ and $2-\rho-\rho^2=2\cos\tfrac{8\pi}9$, all polynomials in $\rho$. The generator of $G$ is $\rho\mapsto\rho^2-2$ (see Interactive 1).
(b) Irreducible ($\pm1$ give $-1,-1$). $\Delta=-4(-1)^3-27(-1)^2=4-27=-23$, not a square in $\mathbb{Q}$ (it is negative): $G\cong S_3$, the splitting field has degree $6$ and contains $\mathbb{Q}(\sqrt{-23})$. (A negative discriminant also means exactly one real root.)
(c) Eisenstein at $2$ ($2\mid0,0,0,-4,2$ and $4\nmid2$): irreducible. Values $f(-2)=-22$, $f(-1)=5$, $f(1)=-1$, $f(2)=26$ show three real roots, in $(-2,-1)$, $(0,1)$ — note $f(0)=2$ — and $(1,2)$. And $f'(x)=5x^4-4$ has only two real zeros, so by Rolle's theorem $f$ has at most three real roots. Exactly two roots are non-real, so $G\cong S_5$ by Theorem 8.41: not solvable by radicals. No formula built from $+,-,\times,\div$ and $\sqrt[n]{\ }$ expresses its roots.
(d) The splitting field is $K=\mathbb{Q}(\alpha,\zeta_5)$ with $\alpha=\sqrt[5]2$. Both $[\mathbb{Q}(\alpha):\mathbb{Q}]=5$ and $[\mathbb{Q}(\zeta_5):\mathbb{Q}]=4$ divide $[K:\mathbb{Q}]\le5\cdot4$, and they are coprime, so $[K:\mathbb{Q}]=20$. The elements are $\alpha\mapsto\zeta^b\alpha$, $\zeta\mapsto\zeta^a$ ($b\in\mathbb{Z}_5$, $a\in\mathbb{Z}_5^\times$), acting on the root $\zeta^j\alpha$ as $j\mapsto aj+b$. The subgroup $N=\operatorname{Gal}(K/\mathbb{Q}(\zeta_5))\cong\mathbb{Z}_5$ is normal (as $\mathbb{Q}(\zeta_5)/\mathbb{Q}$ is Galois) with $G/N\cong\mathbb{Z}_4$, so $G$ is solvable — as it must be, since the roots $\zeta_5^j\sqrt[5]2$ are radicals. Degree $5$ alone does not make a polynomial unsolvable.

Ruler and compass. The same ideas settle Gauss's problem of the regular polygons. A real number is constructible iff it lies at the top of a tower of quadratic extensions of $\mathbb{Q}$ (§8.1). Since $\mathbb{Q}(\zeta_n)$ is Galois over $\mathbb{Q}$ with abelian group of order $\varphi(n)$, the regular $n$-gon is constructible iff $\varphi(n)$ is a power of $2$: necessity because $[\mathbb{Q}(\cos\tfrac{2\pi}n):\mathbb{Q}]=\varphi(n)/2$ must be a power of $2$; sufficiency because an abelian group of order $2^k$ has a chain of subgroups of successive index $2$, which the fundamental theorem turns into a tower of quadratic extensions (the Research–Teaching Nexus below carries this out for $n=17$). Equivalently $n=2^a p_1\cdots p_r$ with distinct Fermat primes $p_i=2^{2^m}+1$.

⚠️ Counterexamples and pitfalls
  1. $|\operatorname{Aut}(K/F)|$ can be smaller than $[K:F]$. $\mathbb{Q}(\sqrt[3]2)$: $1$ versus $3$; $\mathbb{Q}(\sqrt[4]2)$: $2$ versus $4$; $\mathbb{F}_p(t)/\mathbb{F}_p(t^p)$: $1$ versus $p$. Equality is the definition of Galois.
  2. Normality is not transitive. $\mathbb{Q}\subset\mathbb{Q}(\sqrt2)\subset\mathbb{Q}(\sqrt[4]2)$: both steps have degree $2$, hence are normal, but the whole is not.
  3. Intermediate fields of a Galois extension. $K/E$ is always Galois, $E/F$ only when $\operatorname{Gal}(K/E)\trianglelefteq\operatorname{Gal}(K/F)$: $\mathbb{Q}(\sqrt[3]2)\subset\mathbb{Q}(\sqrt[3]2,\omega)$.
  4. Degree $2$ means normal, not necessarily Galois. In characteristic $2$, $\mathbb{F}_2(\sqrt t)/\mathbb{F}_2(t)$ is inseparable with trivial automorphism group.
  5. The splitting field degree is neither $n$ nor $n!$ in general. $x^3-2$: $6$; $x^3-3x+1$: $3$; $x^4+1$: $4$; $x^4-2$: $8$; $x^4-x-1$: $24$; $(x^2-2)(x^2-8)$: only $2$, since $\sqrt8=2\sqrt2$.
  6. The Galois group depends on the base field. $x^4-2$: $D_4$ over $\mathbb{Q}$, $\mathbb{Z}_4$ over $\mathbb{Q}(i)$, $V_4$ over $\mathbb{Q}(\sqrt2)$, $\mathbb{Z}_2$ over $\mathbb{R}$.
  7. Isomorphic is not equal. $\mathbb{Q}(\sqrt[3]2)\cong\mathbb{Q}(\omega\sqrt[3]2)$ as fields, but they are different subfields of $\mathbb{C}$ (one is real, one is not), matching conjugate but different subgroups.
  8. Finite fields: cyclic, and divisors not inequalities. $\operatorname{Gal}(\mathbb{F}_{p^n}/\mathbb{F}_p)\cong\mathbb{Z}_n$ (not $(\mathbb{Z}/n)^\times$, which is the cyclotomic answer over $\mathbb{Q}$); $\mathbb{F}_{p^m}\subseteq\mathbb{F}_{p^n}$ iff $m\mid n$, so $\mathbb{F}_{16}\not\subseteq\mathbb{F}_{64}$.
  9. The fundamental theorem needs Galois. For $\mathbb{Q}(\sqrt[3]2)/\mathbb{Q}$ there are two intermediate fields but the trivial group has one subgroup.
  10. Degree $\ge5$ does not mean unsolvable. $x^5-2$ (group of order $20$) and $x^5-1$ are solvable by radicals; unsolvability needs a non-solvable group, such as $S_5$ for $x^5-4x+2$.
  11. All subfields Galois does not force $G$ abelian. Every subgroup of the quaternion group $Q_8$ is normal, and Galois extensions of $\mathbb{Q}$ with group $Q_8$ exist; all their intermediate fields are Galois over $\mathbb{Q}$.
  12. The cubic discriminant test needs irreducibility. $(x-1)(x^2+1)$ has non-square $\Delta=-16$ but group $\mathbb{Z}_2$.
  13. $\operatorname{Aut}(\mathbb{C})$ is not $\{\mathrm{id},\text{conjugation}\}$. Those are the only continuous automorphisms (or those fixing $\mathbb{R}$); using Zorn's lemma one gets uncountably many others. By contrast $\operatorname{Aut}(\mathbb{R})$ is trivial (Example 1).

Exercises

I
Automorphisms, fixed fields and normality.
  • 1 Find $\operatorname{Aut}(\mathbb{Q}(\sqrt[4]2)/\mathbb{Q})$ and its fixed field. Is the extension Galois?
  • 2 Let $\alpha=\sqrt{1+\sqrt2}$. Find the minimal polynomial of $\alpha$ over $\mathbb{Q}$, show $[\mathbb{Q}(\alpha):\mathbb{Q}]=4$, and find $\operatorname{Aut}(\mathbb{Q}(\alpha)/\mathbb{Q})$. Is $\mathbb{Q}(\alpha)/\mathbb{Q}$ normal?
  • 3 Prove that every extension of degree $2$ of a field of characteristic $\ne2$ is Galois.
  • 4 Let $K=\mathbb{Q}(\sqrt2,\sqrt3,\sqrt5)$. Show $[K:\mathbb{Q}]=8$, $\operatorname{Gal}(K/\mathbb{Q})\cong\mathbb{Z}_2^3$, and find the number of intermediate fields of each degree.
  • 5 Let $\zeta=\zeta_5$. Show that there is an automorphism $\sigma$ of $\mathbb{Q}(\zeta)$ with $\sigma(\zeta)=\zeta^2$, and that $\sigma$ has order $4$.
  • 6 Let $K=\mathbb{F}_p(t)$ and $F=\mathbb{F}_p(t^p)$. Show $[K:F]=p$, $K/F$ is normal but not separable, and $\operatorname{Aut}(K/F)=\{1\}$.
  • 7 Let $K/F$ be finite Galois and $F\subseteq E\subseteq K$. Prove that there are exactly $[E:F]$ distinct $F$-embeddings $E\to K$.
  • 8 Find $\operatorname{Aut}(\mathbb{Q}(\sqrt[6]2)/\mathbb{Q})$ and its fixed field.
  • 9 Show that $\mathbb{Q}(\sqrt[3]2)$ and $\mathbb{Q}(\omega\sqrt[3]2)$ are isomorphic fields but different subfields of $\mathbb{C}$.
1 $r=\sqrt[4]2$ is real, so $\sigma(r)$ is a real root of $x^4-2$: $\pm r$. Both occur by (R1), so $\operatorname{Aut}=\{1,\rho\}$ with $\rho(r)=-r$. Since $\rho(a+br+cr^2+dr^3)=a-br+cr^2-dr^3$, the fixed field is $\mathbb{Q}(r^2)=\mathbb{Q}(\sqrt2)$. Not Galois: $2\lt4$.
2 $\alpha^2=1+\sqrt2$, so $(\alpha^2-1)^2=2$: $\alpha$ is a root of $x^4-2x^2-1$. $[\mathbb{Q}(\sqrt2):\mathbb{Q}]=2$, and $1+\sqrt2$ is not a square in $\mathbb{Q}(\sqrt2)$: $(a+b\sqrt2)^2=1+\sqrt2$ gives $2ab=1$, $a^2+2b^2=1$, hence $2a^4-2a^2+1=0$, which has no real root. So $[\mathbb{Q}(\alpha):\mathbb{Q}]=4$ and $x^4-2x^2-1$ is the minimal polynomial. Its roots are $\pm\alpha$ (real) and $\pm i\sqrt{\sqrt2-1}$ (non-real). As $\mathbb{Q}(\alpha)\subset\mathbb{R}$, $\operatorname{Aut}=\{1,\alpha\mapsto-\alpha\}$, of order $2$, with fixed field $\mathbb{Q}(\sqrt2)$. Not normal: the minimal polynomial has a root in $\mathbb{Q}(\alpha)$ but does not split there.
3 $K=F(\alpha)$ for any $\alpha\in K\setminus F$, with minimal polynomial $x^2+bx+c$. Its other root $-b-\alpha$ lies in $K$, and differs from $\alpha$ (else $\alpha=-b/2\in F$, using $2\ne0$). So $K$ is the splitting field of a separable polynomial, and Theorem 8.32 applies; $\operatorname{Gal}(K/F)=\{1,\alpha\mapsto-b-\alpha\}$.
4 $E=\mathbb{Q}(\sqrt2,\sqrt3)$ is Galois with group $V_4$, so its only quadratic subfields are $\mathbb{Q}(\sqrt2),\mathbb{Q}(\sqrt3),\mathbb{Q}(\sqrt6)$ (one per subgroup of index $2$). If $\sqrt5\in E$, $\mathbb{Q}(\sqrt5)$ would be one of them; but $\mathbb{Q}(\sqrt d)=\mathbb{Q}(\sqrt5)$ forces $5d$ to be a rational square, false for $d=2,3,6$. So $[K:\mathbb{Q}]=8$. $K$ is the splitting field of $(x^2-2)(x^2-3)(x^2-5)$; the $8$ sign changes of $\sqrt2,\sqrt3,\sqrt5$ all occur, each of order $\le2$: $G\cong\mathbb{Z}_2^3$. Subgroups of $\mathbb{Z}_2^3=\mathbb{F}_2^3$ are subspaces: $1$ of order $1$, $7$ of order $2$, $7$ of order $4$, $1$ of order $8$. So there are $7$ quartic subfields, $7$ quadratic subfields $\mathbb{Q}(\sqrt d)$, $d\in\{2,3,5,6,10,15,30\}$, and $16$ intermediate fields in all.
5 $\zeta^2$ is a root of the minimal polynomial $\Phi_5$ of $\zeta$, so by (R1) there is a $\mathbb{Q}$-embedding $\sigma:\mathbb{Q}(\zeta)\to\mathbb{Q}(\zeta)$ with $\zeta\mapsto\zeta^2$; it is an injective linear map of a $4$-dimensional $\mathbb{Q}$-space to itself, hence bijective. $\sigma^k(\zeta)=\zeta^{2^k}$: $\zeta^2,\zeta^4,\zeta^3,\zeta$, so $\sigma$ has order $4$ and $\operatorname{Gal}(\mathbb{Q}(\zeta_5)/\mathbb{Q})=\langle\sigma\rangle\cong\mathbb{Z}_4$.
6 Put $s=t^p$. $t$ is a root of $x^p-s$, irreducible over $\mathbb{F}_p(s)$ by Eisenstein's criterion for the prime $s$ of the PID $\mathbb{F}_p[s]$ (with Gauss's lemma, valid in any UFD). So $[K:F]=p$. In $K[x]$, $x^p-s=(x-t)^p$: $K$ is the splitting field of $x^p-s$ (normal), but $t$ is a $p$-fold root (inseparable). An $F$-automorphism sends $t$ to a root of $(x-t)^p$, i.e. to $t$, so it is the identity.
7 Every $F$-embedding $\psi:E\to K$ extends to an automorphism of $K$: $K$ is the splitting field over $E$ and over $\psi(E)$ of the same separable $f\in F[x]$, so (R2) applies. Hence the embeddings are the restrictions $\sigma|_E$, $\sigma\in G$, and $\sigma|_E=\sigma'|_E$ iff $\sigma^{-1}\sigma'\in\operatorname{Gal}(K/E)$. Their number is $[G:\operatorname{Gal}(K/E)]=[E:F]$.
8 $\beta=\sqrt[6]2$ has minimal polynomial $x^6-2$ (Eisenstein), whose real roots are $\pm\beta$. So $\operatorname{Aut}=\{1,\beta\mapsto-\beta\}$, and the fixed field consists of the combinations of even powers of $\beta$: $\mathbb{Q}(\beta^2)=\mathbb{Q}(\sqrt[3]2)$, of degree $3$ — consistent with $[\mathbb{Q}(\beta):\mathbb{Q}(\sqrt[3]2)]=2=|\operatorname{Aut}|$.
9 Both $\sqrt[3]2$ and $\omega\sqrt[3]2$ are roots of the irreducible $x^3-2$, so $\mathbb{Q}(\sqrt[3]2)\cong\mathbb{Q}[x]/\langle x^3-2\rangle\cong\mathbb{Q}(\omega\sqrt[3]2)$ via $\sqrt[3]2\mapsto\omega\sqrt[3]2$. They are different subsets: the first lies in $\mathbb{R}$, the second contains the non-real number $\omega\sqrt[3]2$.
II
Computing Galois groups over $\mathbb{Q}$.
  • 1 Find the splitting field and the Galois group of $(x^2-2)(x^2-8)$.
  • 2 Find the Galois group of $x^4-5x^2+6$, as a subgroup of $S_4$. Is it transitive?
  • 3 Show that the splitting field of $x^4+1$ is $\mathbb{Q}(i,\sqrt2)$ and find its Galois group over $\mathbb{Q}$ and over $\mathbb{Q}(i)$.
  • 4 Show that $x^3-3x+1$ has Galois group $\mathbb{Z}_3$, and verify directly that $\rho^2-2$ is a root whenever $\rho$ is.
  • 5 Find the Galois group and the degree of the splitting field of $x^3-x-1$.
  • 6 Find the Galois group of $x^3+x^2-2x-1$ and identify its roots.
  • 7 Find the Galois group of $\Phi_9=x^6+x^3+1$ and its subfields of degrees $2$ and $3$.
  • 8 Show that $x^5-6x+3$ is not solvable by radicals.
  • 9 Let $\alpha=\sqrt{2+\sqrt2}$, a root of $x^4-4x^2+2$. Show that $\mathbb{Q}(\alpha)$ is the splitting field and that the Galois group is cyclic of order $4$.
1 The roots are $\pm\sqrt2,\pm2\sqrt2$, so the splitting field is $\mathbb{Q}(\sqrt2)$ and the group is $\mathbb{Z}_2$ — not $V_4$.
2 $x^4-5x^2+6=(x^2-2)(x^2-3)$, splitting field $\mathbb{Q}(\sqrt2,\sqrt3)$, group $V_4$. Numbering the roots $\sqrt2,-\sqrt2,\sqrt3,-\sqrt3$ as $1,2,3,4$ it is $\{1,(12),(34),(12)(34)\}$ — not transitive (two orbits, matching the two irreducible factors, Theorem 8.33), and different from the normal $V_4$ of double transpositions.
3 The roots are $\zeta_8^{1,3,5,7}$, so the splitting field is $\mathbb{Q}(\zeta_8)$; and $\zeta_8=\tfrac{1+i}{\sqrt2}$ with $i=\zeta_8^2$, $\sqrt2=\zeta_8+\zeta_8^7$, so $\mathbb{Q}(\zeta_8)=\mathbb{Q}(i,\sqrt2)$, of degree $4=\varphi(8)$. The group is $(\mathbb{Z}/8)^\times\cong V_4$. Over $\mathbb{Q}(i)$: $x^4+1=(x^2-i)(x^2+i)$, the splitting field is the same, of degree $2$ over $\mathbb{Q}(i)$, and the group is $\langle\sigma_5\rangle\cong\mathbb{Z}_2$ ($\sigma_5$ fixes $i=\zeta^2$ and negates $\sqrt2$).
4 Irreducible ($f(\pm1)=-1,3$); $\Delta=-4(-3)^3-27=81=9^2$, so $G=A_3\cong\mathbb{Z}_3$ (Theorem 8.38). Directly: with $\rho^3=3\rho-1$, $$(\rho^2-2)^3-3(\rho^2-2)+1=\rho^6-6\rho^4+9\rho^2-1,$$ and substituting $\rho^4=3\rho^2-\rho$, $\rho^6=(3\rho-1)^2=9\rho^2-6\rho+1$ gives $9\rho^2-6\rho+1-18\rho^2+6\rho+9\rho^2-1=0$.
5 Irreducible; $\Delta=4-27=-23$, not a square: $G\cong S_3$, degree $6$.
6 Irreducible ($f(1)=-1$, $f(-1)=1$). With $a=1$, $b=-2$, $c=-1$: $\Delta=4+32+4-27+36=49=7^2$, so $G\cong\mathbb{Z}_3$. The roots are $2\cos\tfrac{2\pi}7,2\cos\tfrac{4\pi}7,2\cos\tfrac{6\pi}7$ (Example 6), and the splitting field is the real cubic subfield of $\mathbb{Q}(\zeta_7)$.
7 $G\cong(\mathbb{Z}/9)^\times=\{1,2,4,5,7,8\}$, cyclic of order $6$ generated by $2$ (powers $2,4,8,7,5,1$). The unique quadratic subfield is $\mathbb{Q}(\zeta_9^3)=\mathbb{Q}(\omega)=\mathbb{Q}(\sqrt{-3})$; the unique cubic subfield is $\mathbb{Q}(\zeta_9+\zeta_9^{-1})=\mathbb{Q}(\cos\tfrac{2\pi}9)$, the splitting field of $x^3-3x+1$ from II.4.
8 Eisenstein at $3$: irreducible of degree $5$. $f(-2)=-17$, $f(-1)=8$, $f(1)=-2$, $f(2)=23$: three real roots; $f'=5x^4-6$ has two real zeros, so at most three real roots. Exactly two non-real roots, so $G\cong S_5$ (Theorem 8.41), which is not solvable (Theorem 8.39), hence not solvable by radicals (Theorem 8.40).
9 Eisenstein at $2$ gives degree $4$. The roots are $\pm\alpha,\pm\beta$ with $\beta=\sqrt{2-\sqrt2}$, and $\alpha\beta=\sqrt2=\alpha^2-2$, so $\beta=(\alpha^2-2)/\alpha\in\mathbb{Q}(\alpha)$: $\mathbb{Q}(\alpha)$ is the splitting field and $|G|=4$. Let $\sigma(\alpha)=\beta$. Then $\sigma(\sqrt2)=\sigma(\alpha^2-2)=\beta^2-2=-\sqrt2$, and $\sigma(\beta)=\sigma(\sqrt2)/\sigma(\alpha)=-\sqrt2/\beta=-\alpha$. So $\sigma^2(\alpha)=-\alpha\ne\alpha$, $\sigma$ has order $4$, and $G\cong\mathbb{Z}_4$. (Indeed $\alpha=2\cos\tfrac\pi8$ and $\mathbb{Q}(\alpha)$ is the real subfield of $\mathbb{Q}(\zeta_{16})$.)
III
The fundamental theorem at work.
  • 1 Find all subfields of $\mathbb{Q}(\zeta_5)$; show the quadratic one is $\mathbb{Q}(\sqrt5)$ by finding the minimal polynomial of $\zeta+\zeta^4$.
  • 2 Show $\operatorname{Gal}(\mathbb{Q}(\zeta_{12})/\mathbb{Q})\cong V_4$ and identify the three quadratic subfields together with their subgroups.
  • 3 How many subfields does $\mathbb{Q}(\zeta_{13})$ have, and of which degrees?
  • 4 Show $(\mathbb{Z}/16)^\times\cong\mathbb{Z}_2\times\mathbb{Z}_4$, count the subfields of $\mathbb{Q}(\zeta_{16})$ and name its quadratic subfields.
  • 5 Let $K/F$ be Galois with cyclic group of order $n$. Prove that for each $d\mid n$ there is exactly one intermediate field of degree $d$ over $F$, and that every intermediate field is Galois over $F$.
  • 6 Let $K/\mathbb{Q}$ be Galois of degree $12$. Show that $K$ has subfields of degree $3$ and $4$ over $\mathbb{Q}$. Must it have one of degree $2$?
  • 7 In the splitting field of $x^4-2$ (Example 5), prove that the fixed field of $\langle\sigma\tau\rangle$ is $\mathbb{Q}((1+i)\sqrt[4]2)$ and find the minimal polynomial of $(1+i)\sqrt[4]2$.
  • 8 Using the lattice in Example 5, find $\mathbb{Q}(\sqrt[4]2)\cap\mathbb{Q}(i\sqrt[4]2)$ and the compositum $\mathbb{Q}(\sqrt[4]2)\,\mathbb{Q}(i\sqrt[4]2)$.
  • 9 Let $K/F$ be Galois with group $G$. Prove that $\theta\in K$ satisfies $K=F(\theta)$ if and only if $\sigma(\theta)\ne\theta$ for every $\sigma\ne1$ in $G$. Use it to show $\mathbb{Q}(\sqrt[3]2,\omega)=\mathbb{Q}(\sqrt[3]2+\omega)$.
1 $G=\langle\sigma_2\rangle\cong\mathbb{Z}_4$, with the unique subgroup $\{1,\sigma_4\}$ of order $2$: subfields $\mathbb{Q}\subset E\subset\mathbb{Q}(\zeta_5)$, $[E:\mathbb{Q}]=2$. $E$ contains $\theta=\zeta+\zeta^4$, and $\sigma_2(\theta)=\zeta^2+\zeta^3\ne\theta$. Sum: $\theta+\sigma_2\theta=-1$; product: $\zeta^3+\zeta^4+\zeta^6+\zeta^7=\zeta^3+\zeta^4+\zeta+\zeta^2=-1$. So $\theta$ is a root of $x^2+x-1$, $\theta=\tfrac{-1+\sqrt5}2=2\cos72^\circ$, and $E=\mathbb{Q}(\sqrt5)$.
2 $(\mathbb{Z}/12)^\times=\{1,5,7,11\}$ and $5^2\equiv7^2\equiv11^2\equiv1$: $V_4$. With $\zeta=\zeta_{12}=\tfrac{\sqrt3+i}2$: $i=\zeta^3$ is fixed by $\sigma_a$ iff $3a\equiv3\pmod{12}$ iff $a\in\{1,5\}$, so $\{1,\sigma_5\}\leftrightarrow\mathbb{Q}(i)$; $\omega=\zeta^4$ is fixed iff $a\equiv1\pmod3$, so $\{1,\sigma_7\}\leftrightarrow\mathbb{Q}(\sqrt{-3})$; $\sqrt3=\zeta+\zeta^{11}$ is fixed by $\sigma_{11}$, so $\{1,\sigma_{11}\}\leftrightarrow\mathbb{Q}(\sqrt3)$. Also $\mathbb{Q}(\zeta_{12})=\mathbb{Q}(i,\sqrt3)$.
3 $G\cong(\mathbb{Z}/13)^\times$, cyclic of order $12$, has one subgroup for each divisor of $12$. So there are $6$ subfields, exactly one of each degree $1,2,3,4,6,12$ over $\mathbb{Q}$ (the quadratic one is $\mathbb{Q}(\sqrt{13})$, as Gauss sums show).
4 $(\mathbb{Z}/16)^\times=\{\pm1,\pm3,\pm5,\pm7\}$; $5$ has order $4$ ($5,9,13,1$), $-1\notin\langle5\rangle$, so the group is $\{\pm1\}\times\langle5\rangle\cong\mathbb{Z}_2\times\mathbb{Z}_4$. Its subgroups: $1$ trivial, $3$ of order $2$ ($\{1,7\},\{1,9\},\{1,15\}$), $3$ of order $4$ ($\langle3\rangle=\{1,3,9,11\}$, $\langle5\rangle$, $\{1,7,9,15\}$), and the whole group — $8$ subfields. The quadratic subfields are the fixed fields of the order-$4$ subgroups: $\langle5\rangle$ fixes $i=\zeta^4$ ($4a\equiv4\pmod{16}$ iff $a\equiv1\pmod4$), $\{1,7,9,15\}$ fixes $\sqrt2=\zeta^2+\zeta^{-2}$ ($a\equiv\pm1\pmod8$), and $\langle3\rangle$ fixes $\sqrt{-2}=\zeta^2+\zeta^6$ ($a\equiv1,3\pmod8$). So they are $\mathbb{Q}(i)$, $\mathbb{Q}(\sqrt2)$, $\mathbb{Q}(\sqrt{-2})$.
5 A cyclic group of order $n$ has exactly one subgroup of each order $m\mid n$ (§6.1). An intermediate field of degree $d$ corresponds to a subgroup of index $d$, i.e. order $n/d$: exactly one. Subgroups of an abelian group are normal, so every intermediate field is Galois over $F$ (Theorem 8.35(4)), with cyclic group $G/H$.
6 $|G|=12=2^2\cdot3$. By Sylow's theorems $G$ has subgroups of order $4$ and $3$; their fixed fields have degrees $[G:H]=3$ and $4$. A subfield of degree $2$ needs a subgroup of order $6$, which need not exist: $A_4$ has none. The splitting field of $x^4+8x+12$ has group $A_4$ (§7.3), so it has no quadratic subfield. (Subfields of degree $6$ always exist: Cauchy gives a subgroup of order $2$.)
7 $\theta=(1+i)r$: $\sigma\tau(\theta)=\sigma((1-i)r)=(1-i)ir=(1+i)r$, so $\theta\in K^{\langle\sigma\tau\rangle}$, a field of degree $4$. The $G$-orbit of $\theta$: $\sigma(\theta)=(1+i)ir=-(1-i)r$, $\sigma^2(\theta)=-\theta$, $\sigma^3(\theta)=(1-i)r$ — four elements, so $[\mathbb{Q}(\theta):\mathbb{Q}]=4$ and $\mathbb{Q}(\theta)=K^{\langle\sigma\tau\rangle}$. The orbit polynomial is $(x^2-\theta^2)(x^2-(1-i)^2r^2)=(x^2-2i\sqrt2)(x^2+2i\sqrt2)=x^4+8$.
8 $\mathbb{Q}(r)\leftrightarrow\langle\tau\rangle$ and $\mathbb{Q}(ir)\leftrightarrow\langle\sigma^2\tau\rangle$. The intersection corresponds to $\langle\tau,\sigma^2\tau\rangle=\{1,\sigma^2,\tau,\sigma^2\tau\}$, so it is $\mathbb{Q}(\sqrt2)$ (indeed $r^2=\sqrt2$ and $(ir)^2=-\sqrt2$). The compositum corresponds to $\langle\tau\rangle\cap\langle\sigma^2\tau\rangle=\{1\}$, so it is all of $K$ (it contains $ir/r=i$).
9 $F(\theta)$ is an intermediate field, and $F(\theta)=K$ iff $\operatorname{Gal}(K/F(\theta))=\{1\}$ (Theorem 8.35(1)) iff no $\sigma\ne1$ fixes $\theta$ (fixing $\theta$ is the same as fixing $F(\theta)$). For $\theta=\sqrt[3]2+\omega$: $\sigma^k\tau^e(\theta)=\omega^k\sqrt[3]2+\omega^{\pm1}$, and $\omega^k\sqrt[3]2+\omega^{\pm1}=\sqrt[3]2+\omega$ would give $(\omega^k-1)\sqrt[3]2=\omega-\omega^{\pm1}$; for $k=0$ this needs the sign $+$, i.e. $\sigma=1$, and for $k\ne0$ the left side has modulus $\sqrt3\sqrt[3]2\approx2.18$ while the right side has modulus $0$ or $\sqrt3$. So $\theta$ is a primitive element.
IV
Finite fields, solvability and constructions.
  • 1 Describe $\operatorname{Gal}(\mathbb{F}_{5^6}/\mathbb{F}_{25})$: its order and a generator.
  • 2 How many subfields does $\mathbb{F}_{2^{24}}$ have? How many of them contain $\mathbb{F}_{16}$?
  • 3 Count the $\alpha\in\mathbb{F}_{81}$ with $\mathbb{F}_3(\alpha)=\mathbb{F}_{81}$, and deduce the number of monic irreducible quartics over $\mathbb{F}_3$.
  • 4 Let $\varphi$ be the Frobenius of $\mathbb{F}_{p^n}$. Prove that the fixed field of $\varphi^k$ is $\mathbb{F}_{p^{\gcd(k,n)}}$.
  • 5 Prove that $S_4$ is solvable by exhibiting a chain and checking each quotient.
  • 6 Prove that every polynomial of degree $\le4$ over $\mathbb{Q}$ is solvable by radicals, and give an irreducible quintic over $\mathbb{Q}$ that is.
  • 7 Find $[K:\mathbb{Q}]$ for the splitting field $K$ of $x^5-2$, and a chain showing that $\operatorname{Gal}(K/\mathbb{Q})$ is solvable.
  • 8 Which regular $n$-gons with $n\in\{7,9,15,17,20,51\}$ are constructible with ruler and compass?
  • 9 Let $f\in\mathbb{Q}[x]$ be an irreducible quartic with exactly two real roots. Prove that its Galois group has order $8$ or $24$, and give an example of each.
1 $[\mathbb{F}_{5^6}:\mathbb{F}_{25}]=6/2=3$, so by Theorem 8.34 the group is cyclic of order $3$, generated by $x\mapsto x^{25}$. Inside $\operatorname{Gal}(\mathbb{F}_{5^6}/\mathbb{F}_5)=\langle\varphi\rangle\cong\mathbb{Z}_6$ it is the subgroup $\langle\varphi^2\rangle$.
2 Subfields $\leftrightarrow$ subgroups of $\mathbb{Z}_{24}$ $\leftrightarrow$ divisors of $24$: there are $\tau(24)=8$, namely $\mathbb{F}_{2^d}$, $d\in\{1,2,3,4,6,8,12,24\}$. $\mathbb{F}_{16}=\mathbb{F}_{2^4}\subseteq\mathbb{F}_{2^d}$ iff $4\mid d$: $d=4,8,12,24$, so $4$ of them.
3 The proper subfields of $\mathbb{F}_{3^4}$ are $\mathbb{F}_3\subset\mathbb{F}_9$, so the generators are the $81-9=72$ elements outside $\mathbb{F}_9$. Each has an irreducible quartic minimal polynomial with $4$ distinct roots in $\mathbb{F}_{81}$, so there are $72/4=18$ monic irreducible quartics — Gauss's formula gives $\tfrac{3^4-3^2}{4}=18$.
4 In the cyclic group $\langle\varphi\rangle$ of order $n$, $\langle\varphi^k\rangle=\langle\varphi^d\rangle$ with $d=\gcd(k,n)$ (both are the subgroup of order $n/d$). Its fixed field is $\{x:x^{p^d}=x\}$, the set of roots of $x^{p^d}-x$ in $\mathbb{F}_{p^n}$, which is the subfield $\mathbb{F}_{p^d}$ (as $d\mid n$).
5 $\{1\}\trianglelefteq V_4\trianglelefteq A_4\trianglelefteq S_4$. $V_4$ is abelian; it is normal in $S_4$ because it consists of $1$ and the full conjugacy class of double transpositions. $A_4/V_4$ has order $3$ and $S_4/A_4$ order $2$, so both are cyclic, hence abelian.
6 The Galois group of $f$ embeds in $S_n$ with $n\le4$ (Theorem 8.33); $S_4$ is solvable, so are its subgroups (Theorem 8.39), and Theorem 8.40 applies. Irreducible solvable quintics: $x^5-2$ (IV.7), or the minimal polynomial $x^5+x^4-4x^3-3x^2+3x+1$ of $2\cos\tfrac{2\pi}{11}$, whose roots $2\cos\tfrac{2\pi k}{11}$ ($k=1,\dots,5$) generate the real subfield of $\mathbb{Q}(\zeta_{11})$, with cyclic group $\mathbb{Z}_{10}/\{\pm1\}\cong\mathbb{Z}_5$.
7 $[K:\mathbb{Q}]=20$ (Example 7(d)). Chain: $\{1\}\trianglelefteq N\trianglelefteq G$ with $N=\operatorname{Gal}(K/\mathbb{Q}(\zeta_5))\cong\mathbb{Z}_5$ and $G/N\cong\operatorname{Gal}(\mathbb{Q}(\zeta_5)/\mathbb{Q})\cong\mathbb{Z}_4$. Explicitly $G$ is the group of maps $j\mapsto aj+b$ on $\mathbb{Z}_5$.
8 Constructible iff $\varphi(n)$ is a power of $2$: $\varphi(7)=6$ no; $\varphi(9)=6$ no; $\varphi(15)=8$ yes; $\varphi(17)=16$ yes; $\varphi(20)=8$ yes; $\varphi(51)=\varphi(3)\varphi(17)=32$ yes. So $15,17,20,51$.
9 $G\le S_4$ is transitive, so $4\mid|G|$; complex conjugation fixes the two real roots and swaps the non-real pair, so $G$ contains a transposition. The transitive subgroups of $S_4$ have order $4$ ($\mathbb{Z}_4$, $V_4$), $8$ ($D_4$), $12$ ($A_4$) or $24$; $\mathbb{Z}_4=\langle(1\,2\,3\,4)\rangle$, the normal $V_4$ and $A_4$ contain no transposition. So $G\cong D_4$ or $S_4$. Examples: $x^4-2$ (roots $\pm\sqrt[4]2$ real, $\pm i\sqrt[4]2$ not; group $D_4$) and $x^4-x-1$ ($f(-1)=1$, $f(0)=-1$, $f(2)=13$ give two real roots, and $f'=4x^3-1$ has one real zero, so exactly two; its group is $S_4$).
🎯 Exam corner CSIR Part BCSIR Part CGATE MA

E1. CSIR Part B GATE MA Let $K$ be the splitting field of $x^4-3$ over $\mathbb{Q}$. The order of $\operatorname{Gal}(K/\mathbb{Q})$ is (A) $4$; (B) $6$; (C) $8$; (D) $24$.

Idea The roots are $\pm\sqrt[4]3,\pm i\sqrt[4]3$, so $K=\mathbb{Q}(\sqrt[4]3,i)$.
Degree $x^4-3$ is Eisenstein at $3$, so $[\mathbb{Q}(\sqrt[4]3):\mathbb{Q}]=4$; this field is real, so adjoining $i$ doubles the degree: $|G|=[K:\mathbb{Q}]=8$ (the group is $D_4$, as for $x^4-2$). (A) is the degree of one root; (D) $=4!$ would need every permutation; (B) is not even divisible by $4$, although $G$ is transitive on four roots.
Answer (C) $8$.

E2. CSIR Part B The number of fields $E$ with $\mathbb{Q}\subsetneq E\subsetneq\mathbb{Q}(\zeta_{11})$ is (A) $1$; (B) $2$; (C) $3$; (D) $4$.

Idea $\operatorname{Gal}(\mathbb{Q}(\zeta_{11})/\mathbb{Q})\cong(\mathbb{Z}/11)^\times\cong\mathbb{Z}_{10}$ (Theorem 8.36), whose subgroups correspond to the divisors $1,2,5,10$ of $10$.
Count Four subfields in all; excluding $\mathbb{Q}$ and $\mathbb{Q}(\zeta_{11})$ leaves the quadratic field $\mathbb{Q}(\sqrt{-11})$ and the real quintic field $\mathbb{Q}(\cos\tfrac{2\pi}{11})$. (D) counts the two ends as well.
Answer (B) $2$.

E3. CSIR Part B Which of the following extensions of $\mathbb{Q}$ is Galois? (A) $\mathbb{Q}(\sqrt[3]5)$; (B) $\mathbb{Q}\big(\sqrt{1+\sqrt3}\big)$; (C) $\mathbb{Q}(\cos\tfrac{2\pi}9)$; (D) $\mathbb{Q}(\sqrt[4]3)$.

(A), (D) Real fields generated by one root of $x^3-5$, resp. $x^4-3$, both irreducible with non-real roots: not normal.
(B) $\alpha=\sqrt{1+\sqrt3}$ has minimal polynomial $x^4-2x^2-2$ (Eisenstein at $2$), with the non-real roots $\pm i\sqrt{\sqrt3-1}$; $\mathbb{Q}(\alpha)\subset\mathbb{R}$, so it is not normal.
(C) $\mathbb{Q}(\cos\tfrac{2\pi}9)$ is a subfield of $\mathbb{Q}(\zeta_9)$, whose group is abelian, so it is Galois (it is the splitting field of $x^3-3x+1$, Example 7(a)).
Answer (C).

E4. CSIR Part C Let $K$ be the splitting field of $x^3-5$ over $\mathbb{Q}$. Which of the following statements are true? (One or more options may be correct.) (A) $[K:\mathbb{Q}]=6$. (B) $\operatorname{Gal}(K/\mathbb{Q})$ is abelian. (C) $K$ has exactly one subfield $E$ with $[E:\mathbb{Q}]=2$. (D) $\mathbb{Q}(\sqrt[3]5)/\mathbb{Q}$ is a Galois extension.

(A) True $K=\mathbb{Q}(\sqrt[3]5,\omega)$; degree $3\cdot2$ exactly as in Example 4.
(B) False $G\cong S_3$ (an irreducible cubic with $\Delta=-675$, not a square), which is non-abelian.
(C) True Quadratic subfields correspond to subgroups of index $2$; $S_3$ has exactly one, $A_3$, whose fixed field is $\mathbb{Q}(\omega)=\mathbb{Q}(\sqrt{-3})$. (Note $\mathbb{Q}(\sqrt\Delta)=\mathbb{Q}(\sqrt{-675})=\mathbb{Q}(\sqrt{-3})$.)
(D) False $\langle\tau\rangle$ is not normal in $S_3$; concretely $x^3-5$ has one root in the real field $\mathbb{Q}(\sqrt[3]5)$ but does not split there.
Answer (A) and (C).

E5. CSIR Part C GATE MA Let $K=\mathbb{F}_{2^{12}}$. Which of the following are true? (A) $\operatorname{Gal}(K/\mathbb{F}_2)$ is cyclic of order $12$. (B) $K$ has exactly $6$ subfields (including $\mathbb{F}_2$ and $K$). (C) $K$ contains a subfield with $256$ elements. (D) Exactly $4020$ elements $\alpha\in K$ satisfy $\mathbb{F}_2(\alpha)=K$.

(A) True Theorem 8.34: generated by $x\mapsto x^2$.
(B) True One subfield $\mathbb{F}_{2^d}$ for each $d\mid12$: $d=1,2,3,4,6,12$.
(C) False $256=2^8$ and $8\nmid12$.
(D) True $\alpha$ fails to generate iff it lies in a maximal proper subfield, $\mathbb{F}_{2^6}$ or $\mathbb{F}_{2^4}$, whose intersection is $\mathbb{F}_{2^2}$: $|\mathbb{F}_{64}\cup\mathbb{F}_{16}|=64+16-4=76$, and $4096-76=4020$. (Check: $4020/12=335$, the number of irreducible polynomials of degree $12$ over $\mathbb{F}_2$ by Gauss's formula.)
Answer (A), (B) and (D).

E6. CSIR Part C Let $K/\mathbb{Q}$ be a Galois extension with $\operatorname{Gal}(K/\mathbb{Q})\cong S_4$. Which of the following are true? (A) $K$ has exactly one subfield $E$ with $[E:\mathbb{Q}]=2$. (B) $K$ has a subfield $E$ with $[E:\mathbb{Q}]=6$ and $E/\mathbb{Q}$ Galois. (C) $K$ has a subfield $E$ with $[E:\mathbb{Q}]=3$ and $E/\mathbb{Q}$ Galois. (D) Every subfield $E$ of $K$ with $[E:\mathbb{Q}]=4$ is Galois over $\mathbb{Q}$.

(A) True Degree-$2$ subfields $\leftrightarrow$ subgroups of index $2$; the only one in $S_4$ is $A_4$.
(B) True $V_4\trianglelefteq S_4$ has index $6$, so $E=K^{V_4}$ is Galois of degree $6$ with group $S_4/V_4\cong S_3$.
(C) False It would need a normal subgroup of order $8$. The subgroups of order $8$ are the three Sylow $2$-subgroups ($\cong D_4$), which are conjugate, hence not normal. (Subfields of degree $3$ exist, but are not Galois.)
(D) False Degree-$4$ subfields $\leftrightarrow$ subgroups of order $6$: the four point stabilisers $\cong S_3$, which are conjugate and not normal. For instance, for $x^4-x-1$ the field $\mathbb{Q}(\alpha)$ generated by one root is of this kind.
Answer (A) and (B).
🧑‍🏫 For the teacher

Suggested plan — 3 lectures × 60 min.

  1. Automorphisms and fixed fields. Definitions 8.10–8.11 and Theorems 8.29–8.30; do Example 1 on the board (ask the class to predict $\operatorname{Aut}(\mathbb{Q}(\sqrt[3]2)/\mathbb{Q})$ before computing it). Recall (R1)–(R3) from §8.1 in five minutes. Prove Artin's theorem in full (it is short given the primitive element theorem). Example 2 as the first use of orbit polynomials. Project Interactive 1 on $x^4-10x^2+1$ and $x^3-2$.
  2. Galois extensions and the fundamental theorem. Theorem 8.32 (prove the cycle of implications), Theorem 8.33, Theorem 8.34 for finite fields. Example 3 as a "true or false" discussion. State and prove Theorem 8.35. Use Interactive 2 with $V_4$ and $S_3$, then work Example 4 with the lattice on the board.
  3. Examples and solvability. Example 5 ($D_4$; let Interactive 2 confirm the ten fields and Interactive 1 show the square of roots). Theorem 8.36 and Example 6 (switch Interactive 2 to $\mathbb{Z}_6$ and $\mathbb{Z}_{12}$). Discriminants and cubics (Theorems 8.37–8.38), solvable groups (prove the $3$-cycle argument of Theorem 8.39(3)), Galois's theorem as a statement with the sketch, Theorem 8.41 and Example 7. End with E1, E4 and E6 done live.

Discussion questions. (1) Show that the Galois group of $f$ is exactly the set of permutations of the roots that preserve every polynomial relation $P(\alpha_1,\dots,\alpha_n)=0$ with coefficients in $F$. Why is this a useful way to think but an awkward way to compute? (2) For $\mathbb{Q}(\sqrt[3]2)/\mathbb{Q}$ the correspondence fails. What is the smallest Galois extension containing $\mathbb{Q}(\sqrt[3]2)$, and how does the correspondence look there? (3) Is every finite group the Galois group of some extension of $\mathbb{Q}$? (Known for abelian groups, $S_n$, $A_n$ and solvable groups; open in general — the inverse Galois problem.)

Misconceptions to watch for. "$|\operatorname{Aut}(K/F)|=[K:F]$ always"; "normal over normal is normal"; "every intermediate field of a Galois extension is Galois over the base"; "the splitting field of a degree-$n$ polynomial has degree $n$ (or $n!$)"; "$\operatorname{Gal}(\mathbb{F}_{p^n}/\mathbb{F}_p)\cong(\mathbb{Z}/n)^\times$"; "$\mathbb{F}_{p^m}\subseteq\mathbb{F}_{p^n}$ when $m\le n$"; reversing the direction of the correspondence (bigger subgroup, bigger field); "degree $5$ means unsolvable"; applying the discriminant test to a reducible cubic.

Suggested homework. I(1, 2, 4, 6, 8), II(1, 3, 5, 8, 9), III(1, 2, 4, 6, 7), IV(2, 3, 7, 8, 9); Exam corner E2, E3, E5.

🔬 Research–Teaching NexusResearch-tutored · optional, about 60–90 minutes

Research question. How did Gauss, at eighteen, know that the regular $17$-gon can be drawn with ruler and compass — and how does the Galois group of $\mathbb{Q}(\zeta_{17})$ turn that knowledge into an explicit formula for $\cos\tfrac{2\pi}{17}$ using only square roots?

Why it matters. The structure Gauss exploited — the cyclic group $(\mathbb{Z}/p)^\times$ permuting the $p$-th roots of unity — is the same structure behind Rader's algorithm for fast Fourier transforms of prime length, used in signal and image processing, and behind the number-theoretic transforms that make lattice-based post-quantum cryptography (now being standardised internationally) fast. Seeing a Galois group cut a degree-$16$ problem into four quadratic steps is the clearest illustration of the fundamental theorem in action.

What to do.

  1. Check that $3$ is a primitive root modulo $17$ and list $3^k\bmod17$ for $k=0,\dots,15$. Explain why $G=\operatorname{Gal}(\mathbb{Q}(\zeta_{17})/\mathbb{Q})=\langle\sigma_3\rangle\cong\mathbb{Z}_{16}$ has the chain $G\supset\langle\sigma_3^2\rangle\supset\langle\sigma_3^4\rangle\supset\langle\sigma_3^8\rangle\supset\{1\}$, and what tower of fields it gives.
  2. For each subgroup $H=\langle\sigma_3^m\rangle$ ($m=2,4,8$) form the Gauss periods $\eta=\sum_{h\in H}h(\zeta)$ and their images under $\sigma_3$. Compute them numerically (Python: cmath.exp(2j*cmath.pi/17)).
  3. Show that the two $8$-term periods are the roots of $x^2+x-4$, that each pair of $4$-term periods is the pair of roots of a quadratic over $\mathbb{Q}(\sqrt{17})$, and that the $2$-term periods $2\cos\tfrac{2\pi k}{17}$ satisfy quadratics over the previous field. Assemble a nested-radical formula for $\cos\tfrac{2\pi}{17}$ and check it to $12$ decimal places.
  4. List all $n\le100$ for which the regular $n$-gon is constructible, and explain why $n=7$ and $n=9$ fail in terms of the Galois group.

Tools: Python (cmath, sympy.totient, sympy.nsimplify), Interactive 2 for the lattice picture, a ruler and compass for the brave.

Hand in: A two-page note: the tower of fields with the quadratic equations at each step, your formula for $\cos\tfrac{2\pi}{17}$ with its numerical check, and the list from step 4.

Going further. Read how the $257$-gon and the $65537$-gon were constructed in the nineteenth century; look up the Kronecker–Weber theorem (every abelian extension of $\mathbb{Q}$ lies in some $\mathbb{Q}(\zeta_n)$); and explore the inverse Galois problem.

Instructor note: the powers $3^k\bmod17$ are $1,3,9,10,13,5,15,11,16,14,8,7,4,12,2,6$. The $8$-term periods are $\tfrac{-1\pm\sqrt{17}}{2}\approx1.561553,\,-2.561553$ (sum $-1$, product $-4$). The $4$-term periods are $\approx2.049481,\,0.344151,\,-0.487928,\,-2.905704$; the pair $2.049481,-0.487928$ has sum $\tfrac{-1+\sqrt{17}}2$ and product $-1$, and the pair $0.344151,-2.905704$ has sum $\tfrac{-1-\sqrt{17}}2$ and product $-1$. Finally $2\cos\tfrac{2\pi}{17}\approx1.864944$ and $2\cos\tfrac{8\pi}{17}\approx0.184537$ have sum $2.049481$ and product $0.344151$ (the other $4$-term period). The constructible $n\le100$ ($n\ge3$) are $3,4,5,6,8,10,12,15,16,17,20,24,30,32,34,40,48,51,60,64,68,80,85,96$ — $24$ values. For $n=7$ and $9$, $\varphi(n)=6$, so $\cos\tfrac{2\pi}n$ has degree $3$ and cannot lie in a tower of quadratic extensions.

🧪 Self-check quiz

0/5 correct

Q1. The order of $\operatorname{Aut}(\mathbb{Q}(\sqrt[3]2)/\mathbb{Q})$ is

Q2. $\operatorname{Gal}(\mathbb{F}_{p^n}/\mathbb{F}_p)$ is isomorphic to

Q3. The degree over $\mathbb{Q}$ of the splitting field of $x^3-3x+1$ is

Q4. How many fields $E$ satisfy $\mathbb{Q}\subseteq E\subseteq\mathbb{Q}(\zeta_7)$ (both ends included)?

Q5. Which of these groups is not solvable?