Field Theory: Extensions and Towers

Basics

remark:

Galois is defined as normal and separable.

definition (Simple Extensions):

An extension L/k is simple iff L=K(α) for some αL.

theorem (Primitive Element Theorem):

Every finite separable extension is simple.

corollary:

GF(pn) is a simple extension over Fp.

definition (Algebraic Extension):

A field extension L/k is algebraic iff every αL is the root of some polynomial fk[x].

theorem (Finite Extensions are Algebraic):

Every finite extension is algebraic.

proof (that finite extensions are algebraic):

If K/F and [K:F]=n, then pick any αK and consider 1,α,α2,.... This yields n+1 elements in an n-dimensional vector space, and thus there is a linear dependence f(α):=nj=1cjαj=0. But then α is the root of the polynomial f.

Normal Extensions

definition (Normal Field Extension):

Let L/k be an extension. Then TFAE:

  • L/k is normal.

  • Every irreducible polynomial fk[x] that has one root in L has all of its roots in L, and thus splits in L[x].

    • So if αL then every Galois conjugate αkL as well.. Thus either f splits in L or f has no roots in L.
example (of normal extensions):

    
  • Useful trick: if [L:k]=2 then L/k is automatically normal.

  • Useful trick: if L/K/k, then K/k is normal iff Gal(L/K) .

  • K \coloneqq{\mathbf{Q}}(2^{1\over 3}) is not normal, since K\subset {\mathbf{R}} but (x^3-2) = \prod_k x-\zeta_3^k 2^{1\over 3} with \zeta_3, \zeta_3^2 \in {\mathbf{C}}.

    • Another reason: an embedding \sigma: K\to \overline{k} can send 2^{1\over 3} to any other root of x^3-2.
  • {\mathbf{Q}}(\sqrt 2, \sqrt 3) is normal over {\mathbf{Q}}, since it it is finite and splits f(x) \coloneqq(x^2-2)(x^2-3), which is a separable polynomial.

  • L \coloneqq{\mathbf{Q}}(2^{1\over 4}) is not normal, since it is finite but not the splitting field of any polynomial.

  • {\mathbf{Q}}(\zeta_k) is normal for \zeta_k any primitive kth root of unity.

  • A normal non-separable extension: { \mathbf{F} }_p(x, y) _{/ {{ \mathbf{F} }_p (x^p, y^p)}}. This has finite degree p^2 but infinitely many subfields?

proposition (Characterization of normal algebraic extensions):

For L/k algebraic: let \overline{k} be an algebraic closure containing L, then L/k is normal iff every k{\hbox{-}}embedding \sigma: L\to \overline{k} satisfies \operatorname{im}\sigma = L, so \sigma is a k{\hbox{-}}automorphism of L:

Link to Diagram

definition (Normal Closure):

If K/k is algebraic, then there is an extension N_k/K such that N_k/k is normal and N_k/K/k is a tower. N_k is referred to as the normal closure of K/k.

proposition (Characterization of finite normal extensions as splitting fields):

An extension L/k is finite and normal \iff L is the splitting field of some polynomial f\in k[x].

proof (?):

\implies:

  • Write L = k(a_1, \cdots, a_n) by finiteness.
  • Let m_i be the minimal polynomials of the a_i.
  • By normality, the m_i split in L[x].
  • Then L is the splitting field of f(x) \coloneqq\prod_i m_i(x).

\impliedby:

  • Suppose L/k = \operatorname{SF}(f), and pick any monic m\in L[x]^{{\mathrm{irr}}} with a root a\in L, so that m is the minimal polynomial of a.

  • Toward showing m splits in L: let M = \operatorname{SF}(m), we’ll show M=L.

  • To show that for any root b\in M we have b\in L, it suffices to show [L(b): L] = 1.

    The strategy: use that [L(a):L] = 1 since a\in L by assumption, and try to relate the two degrees.

  • We have L/k, and a number of towers to work with: \begin{align*} [L(a):k] &= [L(a): k(a)] [k(a): k] &= [L(a) : L] [L: k] \\ \\ [L(b):k] &= [L(b): k(b)] [k(b): k] &= [L(b) : L] [L: k] .\end{align*}

  • In the first set of equalities, note that k(a)_{/ {k}} \cong k(b)_{/ {k}} since a,b are conjugate roots over k. Moreover L(a)_{/ {k(a)}} \cong L(b) _{/ {k(b)}} since both are splitting fields for f.

  • Thus [L(a):k] = [L(b): k], which forces [L(a): L] = [L(b): L] after dividing by [L:k]. But [L(a): L] = 1.

proposition:

{\left\lvert {\mathop{\mathrm{Aut}}(L/k)} \right\rvert} \leq [L: k] with equality precisely when L/k is normal.

Separable Extensions

definition (Separable polynomials):

A polynomial f \in k[x] is separable iff f has no repeated roots.

example (of separable and inseparable polynomials):

    
  • x^2-1 is separable over {\mathbf{Q}}, but inseparable over { \mathbf{F} }_2 since it factors as (x-1)^2.
  • (x^2-2)^2 is inseparable over {\mathbf{Q}}
  • x^2-t is inseparable over { \mathbf{F} }_2(t).
  • f(x) \coloneqq x^n-1 is inseparable over { \mathbf{F} }_p when p\divides n.
    • Otherwise, f' = nx^{n-1} has only x=0 as roots, whereas 0 is not a root of f, so f is separable.
  • f(x) \coloneqq x^p-t is not separable over { \mathbf{F} }_p(t): it is irreducible by Eisenstein, but has only the single root t^{1\over p}.
  • f(x) \coloneqq x^{p^n}-x is separable over { \mathbf{F} }_p, since f'(x) = -1 has no roots at all.
definition (Separable Field Extension):

Let L/k be a field extension, \alpha \in L be algebraic over k, and f(x) \coloneqq\min(\alpha, k). The following are equivalent

  • L/k is a separable extension.
  • Every element {\alpha} \in L is separable over k, so \alpha has separable minimal polynomial m(x) in some splitting field of m.
  • Every finite subextension L'/k is separable.
fact:

If \alpha \in K/k is separable, then \alpha is separable in any larger field L/K/k since the minimal polynomial over the larger field will divide the minimal polynomial over the smaller field.

proposition (Separability test: \gcd with derivative):

f is separable iff \gcd(f, f')=1, so f, f' share no common roots. Moreover, the multiple roots of f are precisely the roots of \gcd(f, f').

proof (of separability test):

\not\implies: Suppose f has a repeated root r_i, so its multiplicity is at least 2. Then \begin{align*} f(x) = (x-r)^2 g(x) \implies f'(x) = 2(x-r)g(x) + (x-r)^2g'(x) ,\end{align*} so r is a root of f'.

\not\impliedby: Suppose r is a root of f, f'. Write f(x) = (x-r)p(x) and f'(x) = (x-r)p'(x) + p(x). Rearranging, f'(x) - (x-r)p'(x) = p(x), and since r is a root of the LHS it’s a root of the RHS. So p(x) = (x-r) q(x) and f(x) = (x-r)^2 q(x), making r a repeated root.

proposition (Separability test: identically zero derivative):

f\in k[x]^{{\mathrm{irr}}} is inseparable (so f has a repeated root) iff f'(x) \equiv 0.

proof (of separability test):

Assume f is monic, then f is inseparable iff f, f' have a common root a. So (x-a)\divides q\coloneqq\gcd(f, f'), and since f is irreducible, it must be the minimal polynomial of a. Since f'(a) = 0, this forces f'\divides f, and since \deg f' = \deg f - 1 < \deg f this forces f' \equiv 0.

proposition (Derivative completely detects separability):

    
  • For any field k, f\in k[x] is separable \iff f'\not\equiv 0 \in k[x].
  • For \operatorname{ch}k = 0, irreducible implies separable.
  • For \operatorname{ch}k = p, irreducibles f(x) are inseparable iff f(x) = g(x^p) for some g\in k[x].

Thus for an irreducible polynomial f, \begin{align*} f\text{ separable} \iff \gcd(f, f')=1 \iff f'\not\equiv 0 \iff_{\operatorname{ch}k = p} f(x) = g(x^p) .\end{align*}

proof (?):
  • First part:
    • \not A\implies \not B:
      • Let f be irreducible, and suppose f is separable. If d(x) \coloneqq\gcd(f, f') \neq 1, then f' can not divide f since f is irreducible, so f divides f'. But \deg f' < f and f\divides f' forces f'\equiv 0.
    • \not B \implies \not A:
      • If f'\equiv 0, then d(x) \coloneqq\gcd(f, f') = \gcd(f, 0) = f \neq 1 and f is not separable.
  • Second part:
    • If \operatorname{ch}k = 0 and f is irreducible, then \deg f \geq 2 and \deg f' \geq 1 so f' \neq 0 and f is thus separable.
  • Third part:
    • \impliedby: If f(x) = g(x^p) then f'(x) = g'(x^p)\cdot px^{p-1}\equiv 0.
    • \implies: Let f be irreducible and inseparable, so f' \equiv 0 \in k[x]. Then f(x) \coloneqq\sum_{k=0}^n a_k x^k implies f'(x) \coloneqq\sum_{k=1}^{n}ka_k x^{k-1}, which is zero iff ka_k \equiv 0 so p divides ka_k. So a_k\not\equiv 0 forces p\divides k, so f = a_0 + a_px^p + a_{2p}x^{2p} + \cdots.
corollary (Inseparable iff polynomial in characteristic powers):

If f\in k[x]^{{\mathrm{irr}}} and p\coloneqq\operatorname{ch}k > 0, then f inseparable \iff f(x) = q(x^{p^n}) for some unique n.

proof (of inseparable characterization):

\implies:

Use that f is inseparable iff f' \equiv 0. The claim is that f' \equiv 0 in characteristic p iff all exponents present in f are divisible by p. If f'\equiv 0, write \begin{align*} f(x) &= a_nx^n + a_{n-1} x^{n-1} + \cdots + a_1 x + a_0 \\ \implies f'(x) &= na_nx^{n-1} + (n-1)a_{n-1}x^{n-2} + \cdots + a_1 \\ &\equiv 0 ,\end{align*} which forces i a_i = 0 for all i. For any a_i\neq 0, this forces i\equiv 0 \operatorname{mod}p, so a_i can only be nonzero when p\divides i, so i=kp for some k. So reindex to write \begin{align*} f(x) = a_0 + a_1x^p + a_2x^{2p} + \cdots + a_n x^{np} = \qty{b_0 + b_1 x + b_2 x^{2} + \cdots + b_nx^{n}}^p \in \overline{k}[x] ,\end{align*} using (c+d)^p = c^p + d^p in characteristic p, and taking b_i \coloneqq a_i^{1\over p} \in \overline{k} So f' \equiv 0\implies f(x) = q(x^p) where q(t) \coloneqq\sum b_i t^i.

\impliedby: If f(x) = q(x^p) for some q, the previous calculation shows q has multiple roots, thus so does f, so f is inseparable.

fact (Irreducible implies separable in characteristic zero):

If \operatorname{ch}k = 0 and f\in k[x]^{{\mathrm{irr}}}, then f is automatically separable.

Why this is true: assuming f is irreducible, \gcd(f, f') = 1 or f. It can’t be f, since f\divides f' would force \deg f = \deg f' = 0 and make f a constant. So this \gcd is 1.

fact (Irreducible implies separable for perfect fields):

    
  • Use that irreducible polynomial f must have distinct roots, by the argument above. (In fact, it is the minimal polynomial of its roots.)

  • Toward a contradiction, suppose f is irreducible but inseparable.

  • Then f(x) = g(x^p) for some g(x) \coloneqq\sum a_k x^k.

  • Since Frobenius is bijective, write a_k = b_k^p for some b_k, then \begin{align*} f(x) = \sum a_k x^{pk} = \sum b_k^p x^{pk} =\qty{ \sum b_k x^k }^p ,\end{align*} making f reducible. \contradiction

fact (finite extensions of perfect fields are separable):

A finite extension of a perfect field is automatically separable, and one only needs to show normality to show it’s Galois.

proposition (Simplifications of separability for finite extensions):

If L/k is a finite extension, then, TFAE:

  • L/k is separable.
  • L = k( \alpha) for \alpha a separable element.
  • L = k( S ) for S some set of separable elements
  • [L: K] = [L:K]_s, i.e. the separable degree equals the actual degree. \begin{align*} [L: k] = \left\{{ L: k }\right\} \coloneqq{\sharp}\mathop{\mathrm{Aut}}_{{ \mathsf{Fields}_{k}}}(L) .\end{align*}
proposition (Separable splitting fields are Galois):

If L/k is separable, then \begin{align*} [L: k] = \left\{{ L: k }\right\} .\end{align*} If L/k is a splitting field, then \begin{align*} [L:K] = {\sharp}\mathop{\mathrm{Aut}}_{{ \mathsf{Fields}_{k}}}(L) \coloneqq{\sharp}{ \mathsf{Gal}}(L/k) .\end{align*}

proposition (Irreducible polynomials have separable splitting fields):

Irreducible polynomials have distinct roots after passing to a splitting field.

proposition (Algebraic extensions of perfect fields are separable):

If \operatorname{ch}k = 0 or k is finite, then every algebraic extension L/k is separable.

proposition (Irreducible implies separable for perfect fields):

If k is a perfect field, then every irreducible f\in k[x]^{{\mathrm{irr}}} is automatically separable.

proof (?):

If \operatorname{ch}k = 0 and f is irreducible, then since \deg f' < \deg f and f is irreducible we must have \gcd(f, f')=1 and f is separable.

If \operatorname{ch}k = p>0, then if f is irreducible and inseparable then f(x) = g(x^p) for some g. Write g(x) = \sum a_k x^k, and since k is perfect, write b_k \coloneqq a_k^{1\over p}, then \begin{align*} f(x) = \sum a_k x^{pk} = \sum b_k^p x^{pk} = \qty{\sum b_k x^k}^p ,\end{align*} so f is reducible. \contradiction.

definition (Separable degree):

The separable degree of an extension L/k is defined by fixing an embedding \sigma: k\hookrightarrow\overline{k} (the algebraic or separable closure) and letting [L:k]_s be the number of embeddings \sigma':L\to \overline{k}:

Link to Diagram

proposition (Separability is transitive.):

If L/K/k, then L/K is separable and K/k is separable \iff L/k is separable:

Link to Diagram

proof (?):

Use that L/k is separable \iff [L:k] = [L:k]_s.

\impliedby:

  • By definition, every \alpha \in L is separable over k.
  • K/k is separable:
    • Since K \subseteq L, any \alpha \in K is also separable over k.
  • L/K is separable:
    • If \alpha \in L, then \min_{\alpha, k}(x) is a separable polynomial over some splitting field.
    • Use that L/k implies \min_{\alpha, L}(x) divides \min_{\alpha, k}(x), so the former is separable, done.

\implies:

  • Now use that the separable degree is multiplicative in towers.

  • If all extensions in sight are finite, this direction is immediate: \begin{align*} [L:k]_s = [L:K]_s [K:k]_s = [L:K][K:k] = [L:K] .\end{align*}

  • For the infinite case, want to show every \alpha\in L is separable over k. It suffices to show \alpha is contained in some finite separable subextension. The strategy:

Link to Diagram

  • Let f(x) \coloneqq\min_{\alpha, K}(x) be the minimal polynomial of \alpha over the intermediate extension K, which by assumption is separable since L/K is separable.
    • So f\in K[x], and letting S be the finite set of coefficients of f, S \subseteq K.
    • Note that each coefficient s\in S is separable over k since K/k is separable by assumption.
  • Set F\coloneqq k(\alpha, S) \cap K. Note K/k is separable and F \subseteq K, so F/k is separable.
  • Moreover k(\alpha, S)/F is separable, since the minimal polynomial of \alpha over F is still f.
  • Now k(\alpha, S) / F /K is a tower of finite extensions where k(\alpha, S)/F and F/k are separable, so this reduces to the finite case.
proposition (Separability has the compositing property):

E/k and F/k are separable \iff EF/k is separable.

proof (?):

\impliedby: Separability always descends to subfields, and E \leq EF, F\leq EF.

\implies:

  • Write E = k(S) for some finite set S. Then EF = F(S).
  • Use that k(S)/k is separable iff s\in S is a separable element for all s.
    • Since E/k is separable, each s\in S is separable over k.
  • Since F/k is separable, each s\in S is separable over F.
  • So F(S)/F is separable.
  • Now use the tower F(S)/F/k to obtain F(S)/k separable, which is EF/k.

Galois Extensions

definition (Galois Extension and Galois Group):

Let L/k be a finite field extension. The following are equivalent:

  • L/k is a Galois extension.
  • L is normal, and separable.
  • The fixed field L^H of H\coloneqq\mathrm{Aut}(L/k) is exactly k.
  • L is the splitting field of a separable polynomial p\in K[x].
  • L is a finite separable splitting field of an irreducible polynomial.
  • There is a numerical equality: \begin{align*} {\sharp}\mathop{\mathrm{Aut}}_{{ \mathsf{Fields}_{k}}} (L) = [L: k] = \left\{{ L: k}\right\} ,\end{align*} where \left\{{E:F}\right\} is the number of isomorphisms to any field lifting \operatorname{id}_F:

figures/2021-08-09_22-29-40.png

In this case, we define the Galois group as \begin{align*} { \mathsf{Gal}}(L/k) \coloneqq\mathop{\mathrm{Aut}}_{{ \mathsf{Fields}_{k}}} (L/k) .\end{align*}

fact:

For L/k algebraic and \operatorname{ch}k = 0, L/k is Galois \iff L/k is normal.

proposition (Galois is upper transitive, characterization of when lower transitivity holds):

If L/k is Galois, then L/F is always Galois. Moreover, F/k is Galois if and only if { \mathsf{Gal}}(L/F) {~\trianglelefteq~}{ \mathsf{Gal}}(L/k)

Link to diagram

In this case, \begin{align*} { \mathsf{Gal}}(F/k) \cong \frac{{ \mathsf{Gal}}(L/k)}{{ \mathsf{Gal}}(L/F)} .\end{align*}

proposition (?):

Let L/K/k with L/k Galois. Then \begin{align*} K/k \text{ is Galois } \iff { \mathsf{Gal}}(L/K){~\trianglelefteq~}{ \mathsf{Gal}}(L/k) ,\end{align*} and moreover { \mathsf{Gal}}(K/k) = G.

proof (?):

    
  • Note separability is distinguished, so K/k is separable.
  • K/k is Galois \iff F/k is normal (since we already have separability).
  • \iff \sigma(K) = K for all \sigma\in G
  • \iff \sigma H \sigma^{-1}= H for all \sigma \in G.

So H is normal and G/H is a group. For the isomorphism, take \begin{align*} \rho: { \mathsf{Gal}}(L/k) &\to { \mathsf{Gal}}(K/k) \\ \rho &\mapsto { \left.{{\rho}} \right|_{{K}} } .\end{align*} This is well-defined since by normality \sigma(K) = K. Any f\in \ker \rho is the identity on K, so f\in { \mathsf{Gal}}(L/K) and \ker \phi = H. Since L/K is Galois, every f\in { \mathsf{Gal}}(K/k) lifts to { \mathsf{Gal}}(L/k), making \rho surjective.

example (?):

    
  • {\mathbf{Q}}(\zeta_3, 2^{1/3}) is normal but {\mathbf{Q}}(2^{1/3}) is not since the irreducible polynomial x^3 - 2 has only one root in it.
  • {\mathbf{Q}}(2^{1/3}) is not Galois since its automorphism group is too small (only of size 1 instead of 3?).
  • {\mathbf{Q}}(2^{1/4}) is not Galois since its automorphism group is too small (only of size 2 instead of 4). However, the intermediate extensions {\mathbf{Q}}(2^{1/4}) / {\mathbf{Q}}(2^{1/2}) and {\mathbf{Q}}(\sqrt 2) / {\mathbf{Q}} are Galois since they are quadratic. Slogan: “Being Galois is not transitive in towers.”
  • A quadratic extension that is not Galois: \operatorname{SF}(x^2 + y) \in { \mathbf{F} }_2(y)[x], which factors as (x - \sqrt{y})^2, making the extension not separable.

Fundamental Theorem of Galois Theory

theorem (Fundamental Theorem of Galois Theory):

Let L/k be a Galois extension, then there is a correspondence: \begin{align*} \left\{{\substack{\text{Subgroups } H \leq { \mathsf{Gal}}(L/k)}}\right\} &\rightleftharpoons \left\{{\substack{\text{Fields } F \text{ such}\\ \text{that } L/F/k}}\right\} \\ H &\rightarrow \left\{{\substack{E^H \coloneqq~\text{The fixed field of $H$}}}\right\} \\ \left\{{\substack{{ \mathsf{Gal}}(L/F) \coloneqq\left\{{ \sigma \in { \mathsf{Gal}}(L/k) {~\mathrel{\Big\vert}~}\sigma(F) = F}\right\}}}\right\} &\leftarrow F \end{align*}

  • This is contravariant with respect to subgroups/subfields.

  • [F: k] = [G: H], so degrees of extensions over the base field correspond to indices of subgroups.

  • [K : F] = {\left\lvert {H} \right\rvert}

  • L/F is Galois and Gal(K/F) = H

  • F/k is Galois \iff H is normal, and { \mathsf{Gal}}(F/k) = { \mathsf{Gal}}(L/k)/H.

  • The compositum F_1 F_2 corresponds to H_1 \cap H_2.

  • The subfield F_1 \cap F_2 corresponds to H_1 H_2.

remark:

A trick for remembering the degree/index correspondence:

Link to Diagram

theorem (Splitting + Perfect implies Galois):

    
  • If \operatorname{ch}k = 0 or k is finite, then k is perfect.

  • k = {\mathbf{C}},{\mathbf{R}}, {\mathbf{Q}}, { \mathbf{F} }_p are perfect, so any finite normal extension is Galois.

  • Every splitting field of a polynomial over a perfect field is Galois.

proposition (Composite Extensions):

If F/k is finite and Galois and L/k is arbitrary, then FL/L is Galois and \begin{align*} { \mathsf{Gal}}(FL/L) = { \mathsf{Gal}}(F / F\cap L) \subset { \mathsf{Gal}}(F/k) .\end{align*}

Quadratic Extensions

proposition (Classification of quadratic extensions):

If { \mathbf{F} } is a field with \operatorname{ch}({ \mathbf{F} })\neq 2 and E_{/{ \mathbf{F} }} is a degree 2 extension, then E is Galois and E = F(\sqrt{a}) for some squarefree a\in { \mathbf{F} }.

corollary (Quadratic extensions of rationals):

If E_{/{\mathbf{Q}}} is a quadratic extension, E = {\mathbf{Q}}(\sqrt q) for some q\in {\mathbf{Q}} squarefree. Explicitly, use the primitive element theorem to write E = {\mathbf{Q}}(\alpha), let f be the minimal polynomial, then take q=b^2-4ac. One can do slightly better by writing b^2-4ac = a/b so that \sqrt{b^2-4ac}=\sqrt{ab}/b and taking q=ab.

proposition (?):

For { \mathbf{F} }_p a finite field of prime order, all quadratic extensions E/{ \mathbf{F} }_p are isomorphic.

Distinguished Classes

See http://math.wsu.edu/students/jstreipel/notes/galoistheory.pdf

definition (Distinguished Classes):

A collection of field extensions \mathcal{S} is distinguished iff

  • (Transitive property) For any tower L/K/k, the extension L/k \in \mathcal{S} \iff L/K \in {\mathcal{S}} (upper transitivity) and K/k\in \mathcal{S} (lower transitivity):

Link to Diagram

  • (Lifting property) Lifts of distinguished extensions are distinguished: K/k\in \mathcal{S} and L/k any extension \implies LK/L \in \mathcal{S}:

Link to Diagram

  • (Compositing property) Whenever L/k, K/k\in {\mathcal{S}}, the amalgam LK/k \in {\mathcal{S}} as well:

Link to Diagram

One is supposed to think of LK/L as a “lift” of K/k.

example (of distinguished classes):

The following classes of extensions are distinguished:

  • Algebraic.
  • Finite.
  • Separable.
  • Purely inseparable.
  • Finitely generated.
  • Solvable.
warnings:

Normal extensions are not distinguished, since they fail the forward implication for (lower) transitivity. However, they do have the (forward implication) upper transitive, lifting, and compositing properties.

As a consequence, Galois extensions are also not distinguished.

fact (Normal/Algebraic/Galois extensions are upper transitive):

For L/F/k: L/k normal/algebraic/Galois \implies L/F normal/algebraic/Galois.

Algebraic Extensions

proposition (Transitivity of algebraic extensions, forward implication):

If L/K/k (not necessarily finite) with L/K and K/k both algebraic, then L/k is algebraic.

proof (?):

    
  • We want to show every \alpha \in L is algebraic over k, and it suffices to show \alpha is algebraic over some finite subextension k(S).
  • Pick \alpha\in L, then \alpha is algebraic over K by assumption, so it is a root of some f\in K[x].
  • Let S be the finitely many coefficients of f, then \alpha is algebraic over k(S).
  • Note that k(S)/k is finite and thus algebraic, and k(S,\alpha)/k(s) is finite and also algebraic, so we’re reduced to the finite case.
  • It suffices to show k(S, \alpha)/k(s)/k is finite, which follows from multiplicativity of degrees.
remark:

If L/K/k with \alpha algebraic over L, then \alpha is algebraic over K and \min_{\alpha, L} divides \min_{\alpha, K} (so minimal polynomials only get smaller in extensions).

Normal Extensions

corollary (Normality satisfies the lifting property):

E_1/k normal and E_2/k normal \implies E_1E_2/k normal and E_1 \cap E_2 / k normal.

Link to diagram

Issues with Normal Towers

example (Normal extensions are not transitive: failure of lower transitivity, forward implication):

One can similarly produce towers where the total extension is normal but the lower iterate is not normal: take \begin{align*} L/K/k \coloneqq{\mathbf{Q}}(2^{1\over 3}, \zeta_3) / {\mathbf{Q}}(2^{1\over 3}) / {\mathbf{Q}} .\end{align*} Now K/k isn’t normal, since { \mathsf{Gal}}(L/k) = S_3 but { \mathsf{Gal}}(L/K) = {\mathbf{Z}}/2 \not{~\trianglelefteq~}S_3.

Another example: let L/k be any algebraic extension that isn’t normal, and take N_k to be the normal closure to get N_k/L. Concretely, N_{\mathbf{Q}}/ {\mathbf{Q}}(2^{1\over 3})/{\mathbf{Q}} works.

example (Normal extensions are not transitive: failure of reverse implication):

One can produce towers of successively normal extensions whose total extension is not normal in a cheap way: take \begin{align*} L/K/k \coloneqq{\mathbf{Q}}(2^{1\over 4}) / {\mathbf{Q}}(2^{1\over 2}) / {\mathbf{Q}} .\end{align*} Each iterate is normal since it’s quadratic, but the overall extension misses complex roots and is thus not normal.

proposition (Normal extensions are upper transitive, forward implication):

For L/k finite,

Link to Diagram

proof (Finite case):

    
  • Use the fact that for finite extensions, L/k is normal and separable \iff L is the splitting field of a separable polynomial f\in k[x].
  • Now regard f as a polynomial in K[x]; then L is still the splitting field of f over K, done.

Alternatively,

  • Let \alpha\in L be a root of f\in K[x] with f irreducible, it suffices to show all roots of f are in L.
  • Let m\in K[x] be the minimal polynomial of \alpha over K, and let m'\in k[x] be the minimal polynomial of \alpha over k.
  • Since L/k is normal and \alpha\in L, m' splits in L.
  • Minimal polynomials are divisible in towers, so m divides m'. Since m' splits in L, so must m.
proof (General case):

    
  • Suppose L/K/k with L/k normal, we want to show L/K is normal.
  • Use the embedding characterization, it suffices to show that every embedding \sigma: L\hookrightarrow\overline{K} satisfies \operatorname{im}\sigma = L:

Link to Diagram

  • Now just use the fact that \overline{k} = \overline{K}, and since k\subseteq K, any K{\hbox{-}}morphism is also a k{\hbox{-}}morphism.
  • Since L/k is normal, \sigma(L) = L and L/K is thus normal.