8.1 Field Extensions and Splitting Fields
Fields inside fields: the degree $[K:F]$ and the tower law, algebraic elements and minimal polynomials, $F(\alpha)\cong F[x]/\langle m\rangle$, algebraic closures, ruler-and-compass constructions, splitting fields and separability.
In §7.3 we learned to decide whether a polynomial is irreducible. This section answers the question that comes next: if $f\in F[x]$ is irreducible and has no root in $F$, where are its roots? The answer is to enlarge the field. Every such $f$ has a root in the field $F[x]/\langle f\rangle$ (Kronecker), every root $\alpha$ of $f$ generates a field $F(\alpha)$ that is a vector space over $F$ of dimension $\deg f$, and adjoining all the roots one after another produces the splitting field of $f$. Measuring the size of an extension by its dimension — the degree $[K:F]$ — turns questions about fields into linear algebra, and the single multiplicative rule $[L:F]=[L:K][K:F]$ (the tower law) settles problems that resisted geometers for two thousand years: no ruler-and-compass construction doubles the cube or trisects a $60^\circ$ angle, because $\sqrt[3]2$ and $\cos20^\circ$ have degree $3$, which is not a power of $2$.
CSIR NET asks about this material constantly and in a recognisable form: "the degree of $\mathbb{Q}(\sqrt2,\sqrt[3]2,i)$ over $\mathbb{Q}$ is…", "the degree of the splitting field of $x^4-2$ is…", "the minimal polynomial of $\sqrt2+\sqrt3$ over $\mathbb{Q}(\sqrt6)$ is…", "which of the following numbers is constructible?". Every one of them is answered with the same three tools: the tower law, the irreducibility tests of §7.3, and a careful look at which roots already lie in the field built so far. Chapter 8 then uses these tools twice: finite fields (§8.2) are splitting fields of $x^{p^n}-x$, and Galois theory (§8.3) studies the symmetries of splitting fields.
Field Extensions and the Tower Law
Throughout, "field" means commutative field, and a subfield of $K$ is a subring that is itself a field (it contains $1$, and inverses of its nonzero elements). The intersection of any family of subfields of $K$ is again a subfield, which allows us to speak of the smallest subfield containing a given set.
A field extension $K/F$ (read "$K$ over $F$") is a field $K$ together with a subfield $F\subseteq K$. Since $K$ is closed under addition and under multiplication by elements of $F$, it is a vector space over $F$. Its dimension
$$[K:F]=\dim_F K\in\{1,2,3,\dots\}\cup\{\infty\}$$is the degree of the extension; $K/F$ is finite if $[K:F]\lt\infty$ and infinite otherwise. A field $E$ with $F\subseteq E\subseteq K$ is an intermediate field, and $F\subseteq E\subseteq K$ is a tower. For $S\subseteq K$, $F(S)$ denotes the smallest subfield of $K$ containing $F$ and $S$ ("$F$ adjoin $S$"); $F(\alpha_1,\dots,\alpha_n)=F(\{\alpha_1,\dots,\alpha_n\})$, and $K/F$ is simple if $K=F(\alpha)$ for one $\alpha$, a primitive element.
Standard examples: $[\mathbb{C}:\mathbb{R}]=2$ with basis $\{1,i\}$; $[\mathbb{Q}(\sqrt2):\mathbb{Q}]=2$ with basis $\{1,\sqrt2\}$, since $\mathbb{Q}(\sqrt2)=\{a+b\sqrt2:a,b\in\mathbb{Q}\}$ ($\tfrac1{a+b\sqrt2}=\tfrac{a-b\sqrt2}{a^2-2b^2}$, and $a^2\ne2b^2$ for $(a,b)\ne(0,0)$ because $\sqrt2$ is irrational); $[\mathbb{R}:\mathbb{Q}]=\infty$, since a finite-dimensional $\mathbb{Q}$-vector space is countable while $\mathbb{R}$ is not (Unit 1, §1.1); $[F(t):F]=\infty$ for the field $F(t)$ of rational functions, because $1,t,t^2,\dots$ are linearly independent. Over a finite field, §7.3 gave the examples $\mathbb{F}_p[x]/\langle f\rangle$ of degree $\deg f$ over $\mathbb{F}_p$.
- Every ring homomorphism $\phi:F\to R$ from a field $F$ to a nonzero ring $R$ (with $\phi(1)=1$) is injective. Hence $F\cong\phi(F)$, and we may regard $R$ as containing a copy of $F$.
- Every field $K$ contains a smallest subfield, its prime subfield, which is isomorphic to $\mathbb{Q}$ if $\operatorname{char}K=0$ and to $\mathbb{F}_p=\mathbb{Z}_p$ if $\operatorname{char}K=p$. Thus every field is an extension of $\mathbb{Q}$ or of some $\mathbb{F}_p$.
Proof. (1) $\ker\phi$ is an ideal of $F$ not containing $1$ (because $\phi(1)=1\ne0$). The only ideals of a field are $0$ and $F$ (§7.1), so $\ker\phi=0$. (2) The map $\mathbb{Z}\to K$, $n\mapsto n\cdot1$, has kernel $c\,\mathbb{Z}$ where $c=\operatorname{char}K$ (§7.1). If $c=p$ is prime, the image is a subring isomorphic to $\mathbb{Z}/p\mathbb{Z}=\mathbb{F}_p$, which is a field; it is contained in every subfield (every subfield contains $1$, hence all $n\cdot1$). If $c=0$ the map is injective, and every subfield contains the fractions $(m\cdot1)(n\cdot1)^{-1}$, $n\ne0$; the map $\mathbb{Q}\to K$, $m/n\mapsto(m\cdot1)(n\cdot1)^{-1}$, is well defined (if $m/n=m'/n'$ then $mn'=m'n$ in $\mathbb{Z}$, hence in $K$) and is a ring homomorphism, so by (1) its image — the set of these fractions — is a subfield isomorphic to $\mathbb{Q}$. ∎
Part (1) is why "an extension of $F$" may be any field into which $F$ embeds: Kronecker's field $F[x]/\langle f\rangle$ below does not literally contain $F$, but it contains the isomorphic copy $\{a+\langle f\rangle:a\in F\}$, and we identify the two.
Proof. Spanning. Let $c\in L$. Write $c=\sum_j\lambda_jb_j$ with $\lambda_j\in K$ (finitely many nonzero), and each $\lambda_j=\sum_i\mu_{ij}a_i$ with $\mu_{ij}\in F$. Then $c=\sum_{i,j}\mu_{ij}\,a_ib_j$.
Independence. Suppose $\sum_{i,j}\mu_{ij}a_ib_j=0$ with $\mu_{ij}\in F$, finitely many nonzero. Regroup as $\sum_j\big(\sum_i\mu_{ij}a_i\big)b_j=0$. The coefficients $\sum_i\mu_{ij}a_i$ lie in $K$ and the $b_j$ are independent over $K$, so $\sum_i\mu_{ij}a_i=0$ for each $j$; the $a_i$ are independent over $F$, so every $\mu_{ij}=0$. The same argument shows that $a_ib_j=a_kb_l$ with $(i,j)\ne(k,l)$ is impossible (it is a nontrivial vanishing combination). Hence $\{a_ib_j\}$ is a basis of $L/F$ indexed by $I\times J$, and $[L:F]=|I|\,|J|$ when both are finite.
Infinite cases. If $[L:F]$ is finite, then $K$ is an $F$-subspace of $L$, so $[K:F]\le[L:F]$, and an $F$-basis of $L$ spans $L$ over the bigger field $K$, so $[L:K]\le[L:F]$. Contrapositively, if $[K:F]$ or $[L:K]$ is infinite then so is $[L:F]$. ∎
- For every intermediate field $E$, both $[E:F]$ and $[L:E]$ divide $[L:F]$.
- If $[L:F]$ is prime, the only intermediate fields are $F$ and $L$; hence $L=F(\alpha)$ for every $\alpha\in L\setminus F$.
- If $E_1,E_2$ are intermediate fields with $[E_1:F]=m$, $[E_2:F]=n$, then $\operatorname{lcm}(m,n)$ divides $[E_1E_2:F]$ and $[E_1E_2:F]\le mn$, where $E_1E_2$ (the compositum) is the smallest subfield of $L$ containing both. If $\gcd(m,n)=1$ then $[E_1E_2:F]=mn$ and $E_1\cap E_2=F$.
Proof. (1) is $[L:F]=[L:E][E:F]$. (2) By (1), $[E:F]\in\{1,p\}$; degree $1$ means $E=F$ (the vector space $E$ is spanned by $1$), and $[E:F]=p$ forces $[L:E]=1$, i.e. $E=L$. For $\alpha\notin F$, $E=F(\alpha)\ne F$. (3) $m$ and $n$ divide $[E_1E_2:F]$ by (1), hence so does their lcm. For the upper bound, if $\{a_1,\dots,a_m\}$ is an $F$-basis of $E_1$, the $E_2$-span $V$ of the $a_i$ is closed under multiplication (products $a_ia_k$ are $F$-combinations of the $a_l$), so $V$ is a ring containing $E_1$ and $E_2$ that is a finite-dimensional vector space over the field $E_2$; a finite-dimensional domain over a field is a field (multiplication by a nonzero $v$ is an injective linear map $V\to V$, hence onto, so $vw=1$ for some $w$). So $V=E_1E_2$ and $[E_1E_2:E_2]\le m$, giving $[E_1E_2:F]\le mn$. If $\gcd(m,n)=1$ the lcm is $mn$. Finally $[E_1\cap E_2:F]$ divides both $m$ and $n$, so it is $1$. ∎
Algebraic Elements, Minimal Polynomials and Simple Extensions
Let $K/F$ be an extension and $\alpha\in K$. Then $\alpha$ is algebraic over $F$ if $g(\alpha)=0$ for some nonzero $g\in F[x]$, and transcendental over $F$ otherwise. The extension $K/F$ is algebraic if every element of $K$ is algebraic over $F$. A complex number algebraic over $\mathbb{Q}$ is an algebraic number.
$F[\alpha]=\{g(\alpha):g\in F[x]\}$ is the smallest subring of $K$ containing $F$ and $\alpha$; it is the image of the evaluation homomorphism $\mathrm{ev}_\alpha:F[x]\to K$, $g\mapsto g(\alpha)$. Always $F[\alpha]\subseteq F(\alpha)$.
Proof. $I=\ker\mathrm{ev}_\alpha$ is an ideal of $F[x]$, nonzero because $\alpha$ is algebraic. $F[x]$ is a principal ideal domain (it is Euclidean, §7.3), so $I=\langle m\rangle$ for some $m\ne0$, unique up to a nonzero constant factor; making it monic fixes it. (a) Every nonzero element of $I$ is a multiple of $m$, hence has degree $\ge\deg m$. (b) $m$ is not constant (a nonzero constant does not vanish at $\alpha$). If $m=gh$ with $g,h\in F[x]$, then $g(\alpha)h(\alpha)=m(\alpha)=0$ in the field $K$, so $g(\alpha)=0$ or $h(\alpha)=0$; say $g\in I$, so $m\mid g$ and $\deg g\ge\deg m$, i.e. $h$ is a constant. So $m$ is irreducible. Conversely, if $p$ is monic irreducible with $p(\alpha)=0$, then $p\in I$, so $m\mid p$; irreducibility of $p$ and $\deg m\ge1$ force $p=m$. Finally $m_{\alpha,F}\in E[x]$ vanishes at $\alpha$, so it lies in the ideal $\langle m_{\alpha,E}\rangle$ of $E[x]$. ∎
Property (b) is how minimal polynomials are found in practice: produce some monic polynomial with root $\alpha$, then prove it irreducible with the tests of §7.3. For instance $m_{\sqrt[3]2,\mathbb{Q}}=x^3-2$ (Eisenstein), $m_{i,\mathbb{Q}}=x^2+1$, $m_{i,\mathbb{R}}=x^2+1$, but $m_{i,\mathbb{C}}=x-i$ — the minimal polynomial depends on the base field.
- If $\alpha$ is algebraic over $F$ with minimal polynomial $m$ of degree $n$, then $F[\alpha]=F(\alpha)$, $$F(\alpha)\cong F[x]/\langle m\rangle\quad(\text{via } g(\alpha)\leftrightarrow g+\langle m\rangle),$$ and $1,\alpha,\dots,\alpha^{n-1}$ is a basis of $F(\alpha)$ over $F$. In particular $[F(\alpha):F]=\deg m_\alpha$.
- If $\alpha$ is transcendental over $F$, then $F[\alpha]\cong F[x]$ is not a field, $F(\alpha)\cong F(x)$ (rational functions), and $[F(\alpha):F]=\infty$.
- (Kronecker's construction.) Conversely, for every irreducible $p\in F[x]$ the field $K=F[x]/\langle p\rangle$ is an extension of $F$ of degree $\deg p$ in which $\bar x=x+\langle p\rangle$ is a root of $p$. Hence every nonconstant polynomial over a field has a root in some finite extension.
Proof. (1) The evaluation map $\mathrm{ev}_\alpha:F[x]\to K$ has image $F[\alpha]$ and kernel $\langle m\rangle$ (Theorem 8.4), so $F[x]/\langle m\rangle\cong F[\alpha]$ by the first isomorphism theorem (§7.1). Since $m$ is irreducible, $F[x]/\langle m\rangle$ is a field (§7.3, "$F[x]/\langle f\rangle$ is a field iff $f$ is irreducible"). So $F[\alpha]$ is a subfield of $K$ containing $F$ and $\alpha$, hence contains $F(\alpha)$; with $F[\alpha]\subseteq F(\alpha)$ we get equality. Basis: for $g\in F[x]$ divide, $g=qm+r$ with $r=0$ or $\deg r\lt n$; then $g(\alpha)=r(\alpha)$ is an $F$-combination of $1,\alpha,\dots,\alpha^{n-1}$. If $\sum_{i\lt n}c_i\alpha^i=0$, the polynomial $\sum c_ix^i$ lies in $\langle m\rangle$ and has degree $\lt n$, so it is $0$.
(2) Now $\ker\mathrm{ev}_\alpha=0$, so $F[\alpha]\cong F[x]$, in which $x$ has no inverse; so $F[\alpha]\ne F(\alpha)$. The injective map $F[x]\to F(\alpha)$ extends to the fraction field, $g/h\mapsto g(\alpha)/h(\alpha)$ (defined because $h(\alpha)\ne0$ for $h\ne0$); its image is a subfield containing $F$ and $\alpha$, hence equals $F(\alpha)$. The powers $1,\alpha,\alpha^2,\dots$ are independent.
(3) $\langle p\rangle$ is maximal, so $K$ is a field; $a\mapsto a+\langle p\rangle$ embeds $F$ in $K$ (Theorem 8.1). Writing $p=\sum c_ix^i$, $p(\bar x)=\sum c_i\bar x^{\,i}=\overline{p}=0$ in $K$. The classes of $1,x,\dots,x^{n-1}$ form a basis by the division algorithm, as in (1). For a nonconstant $f$, apply this to an irreducible factor $p$ of $f$. ∎
- Let $\sigma:F\to F'$ be an isomorphism of fields, extended to $F[x]\to F'[x]$ by acting on coefficients ($f\mapsto f^\sigma$). Let $p\in F[x]$ be irreducible, $\alpha$ a root of $p$ in an extension of $F$, and $\beta$ a root of $p^\sigma$ in an extension of $F'$. Then there is exactly one isomorphism $\tau:F(\alpha)\to F'(\beta)$ with $\tau|_F=\sigma$ and $\tau(\alpha)=\beta$.
- In particular ($\sigma=\mathrm{id}$): if $\alpha,\beta$ have the same minimal polynomial over $F$ — they are conjugate over $F$ — then $F(\alpha)\cong F(\beta)$ by an isomorphism fixing $F$ and sending $\alpha\mapsto\beta$.
- Every homomorphism $\tau:F(\alpha)\to L$ that fixes $F$ sends $\alpha$ to a root of $m_\alpha$ in $L$, and is determined by $\tau(\alpha)$. So the number of such homomorphisms equals the number of distinct roots of $m_\alpha$ in $L$; it is at most $[F(\alpha):F]$.
Proof. (1) $p^\sigma$ is irreducible in $F'[x]$ (a factorisation of $p^\sigma$ would pull back along $\sigma^{-1}$ to one of $p$), so by Theorem 8.5 we have isomorphisms $F(\alpha)\cong F[x]/\langle p\rangle$ and $F'[x]/\langle p^\sigma\rangle\cong F'(\beta)$, and $g+\langle p\rangle\mapsto g^\sigma+\langle p^\sigma\rangle$ is an isomorphism between the two quotients (it is induced by the ring isomorphism $g\mapsto g^\sigma$, which maps $\langle p\rangle$ onto $\langle p^\sigma\rangle$). The composite $\tau$ sends $\sum c_i\alpha^i\mapsto\sum\sigma(c_i)\beta^i$, so $\tau|_F=\sigma$ and $\tau(\alpha)=\beta$. Any $\tau'$ with these properties agrees with $\tau$ on the basis $1,\alpha,\dots,\alpha^{n-1}$ and is additive and $\sigma$-semilinear, so $\tau'=\tau$. (2) is the case $F'=F$, $\sigma=\mathrm{id}$. (3) Applying $\tau$ to $m_\alpha(\alpha)=0$ gives $m_\alpha(\tau(\alpha))=0$ because $\tau$ fixes the coefficients; $\tau(\sum c_i\alpha^i)=\sum c_i\tau(\alpha)^i$ shows that $\tau$ is determined by $\tau(\alpha)$, and by (1) every root of $m_\alpha$ in $L$ occurs. ∎
So $\mathbb{Q}(\sqrt[3]2)$ and $\mathbb{Q}(\omega\sqrt[3]2)$, $\omega=e^{2\pi i/3}$, are isomorphic fields — both are $\mathbb{Q}[x]/\langle x^3-2\rangle$ — although they are different subfields of $\mathbb{C}$: the first lies in $\mathbb{R}$, the second does not. Part (3) says that $\mathbb{Q}(\sqrt[3]2)$ has exactly one embedding into $\mathbb{R}$ but three into $\mathbb{C}$; this counting of embeddings is the germ of Galois theory (§8.3).
- If $[K:F]=n$ is finite, then $K/F$ is algebraic and $\deg_F\alpha$ divides $n$ for every $\alpha\in K$.
- If $\alpha_1,\dots,\alpha_k\in K$ are algebraic over $F$, then $F(\alpha_1,\dots,\alpha_k)/F$ is finite, of degree at most $\prod_i\deg_F\alpha_i$. Thus: finite $\iff$ algebraic and finitely generated.
- The elements of $K$ that are algebraic over $F$ form a subfield of $K$ (the algebraic closure of $F$ in $K$): sums, products and quotients of algebraic elements are algebraic.
- (Transitivity.) If $F\subseteq K\subseteq L$, $K/F$ is algebraic and $\beta\in L$ is algebraic over $K$, then $\beta$ is algebraic over $F$. So an algebraic extension of an algebraic extension is algebraic.
Proof. (1) The $n+1$ vectors $1,\alpha,\dots,\alpha^n$ in the $n$-dimensional space $K$ are dependent, giving a nonzero $g\in F[x]$ with $g(\alpha)=0$. Then $F\subseteq F(\alpha)\subseteq K$ and $[F(\alpha):F]=\deg_F\alpha$ divides $n$ by Theorem 8.3(1). (2) Put $F_0=F$, $F_j=F(\alpha_1,\dots,\alpha_j)=F_{j-1}(\alpha_j)$. Then $[F_j:F_{j-1}]=\deg_{F_{j-1}}\alpha_j\le\deg_F\alpha_j$, because $m_{\alpha_j,F_{j-1}}$ divides $m_{\alpha_j,F}$ (Theorem 8.4); multiply using the tower law. Conversely a finite extension is algebraic by (1) and is generated by a basis. (3) If $\alpha,\beta$ are algebraic, $F(\alpha,\beta)$ is finite by (2), so $\alpha\pm\beta$, $\alpha\beta$ and $\alpha/\beta$ ($\beta\ne0$), which lie in it, are algebraic by (1). (4) Let $\beta$ be a root of $b_0+b_1x+\dots+b_mx^m\in K[x]$, nonzero. Then $\beta$ is algebraic over $E=F(b_0,\dots,b_m)$, which is finite over $F$ by (2); so $E(\beta)/E$ and $E/F$ are finite, $E(\beta)/F$ is finite by the tower law, and $\beta$ is algebraic over $F$ by (1). ∎
Part (3) is remarkable when one tries it by hand: it guarantees that $\sqrt2+\sqrt[3]3$ satisfies a rational polynomial (of degree $6$, see Exercise II.5) without producing it. The proof is pure linear algebra — the same linear algebra that Interactive 1 performs on the powers of $\alpha$.
- Find a polynomial. Isolate radicals and raise to powers ($\alpha=\sqrt2+\sqrt3\Rightarrow(\alpha^2-5)^2=24$), or use a known identity ($\cos3t=4\cos^3t-3\cos t$, $\zeta^n=1$).
- Prove it irreducible (§7.3): root test in degree $\le3$, Eisenstein with shifts, reduction mod $p$, Gauss's lemma — or avoid irreducibility altogether by computing the degree of a field you know: if $\mathbb{Q}(\alpha)=\mathbb{Q}(\sqrt2,\sqrt3)$ has degree $4$, a monic quartic with root $\alpha$ must be the minimal polynomial.
- Degrees of composite fields: build a tower adjoining one element at a time; at each step find the minimal polynomial of the new element over the field built so far (it divides the one over $\mathbb{Q}$, and may be smaller — $\omega\sqrt[3]2$ has degree $2$ over $\mathbb{Q}(\sqrt[3]2)$).
- Use divisibility: every degree in sight divides the total; coprime degrees multiply (Theorem 8.3); the total is at most the product of the individual degrees.
- Over a bigger base the minimal polynomial is the factor of $m_{\alpha,\mathbb{Q}}$ that contains $x-\alpha$; its degree is $[E(\alpha):E]$.
Algebraic Closures and Constructible Numbers
- If $\Omega\supseteq F$ is algebraically closed, then $A=\{\alpha\in\Omega:\alpha\text{ algebraic over }F\}$ is an algebraic closure of $F$. In particular the field $\overline{\mathbb{Q}}$ of algebraic numbers in $\mathbb{C}$ is an algebraic closure of $\mathbb{Q}$; it is countable, and $[\overline{\mathbb{Q}}:\mathbb{Q}]=\infty$. Also $\overline{\mathbb{R}}=\mathbb{C}$.
- (Steinitz, 1910.) Every field $F$ has an algebraic closure, and any two are isomorphic by an isomorphism fixing $F$.
- No finite field is algebraically closed.
Proof. (1) $A$ is a field (Theorem 8.7(3)) and algebraic over $F$. Let $g\in A[x]$ be nonconstant. It has a root $\beta\in\Omega$; $\beta$ is algebraic over $A$, hence over $F$ (Theorem 8.7(4)), so $\beta\in A$. Thus $A$ is algebraically closed. $\mathbb{C}$ is algebraically closed by the fundamental theorem of algebra (§2.3); $\overline{\mathbb{Q}}$ is countable because $\mathbb{Q}[x]$ is countable and each nonzero polynomial has finitely many roots; and $\sqrt[n]2\in\overline{\mathbb{Q}}$ has degree $n$ ($x^n-2$ is Eisenstein), so $[\overline{\mathbb{Q}}:\mathbb{Q}]\ge n$ for every $n$. Finally $[\mathbb{C}:\mathbb{R}]=2$, so $\mathbb{C}/\mathbb{R}$ is algebraic.
(2) Idea of the proof. For each nonconstant $f\in F[x]$ take a variable $x_f$, and in the polynomial ring $F[\{x_f\}]$ consider the ideal $I$ generated by all $f(x_f)$. A finite relation $1=\sum g_if_i(x_{f_i})$ is impossible (evaluate at roots of $f_1,\dots,f_k$ in a finite extension), so $I$ is proper and lies in a maximal ideal $M$ (Krull's theorem, §7.2, which rests on Zorn's lemma). The field $E_1=F[\{x_f\}]/M$ contains a root of every $f\in F[x]$. Repeating gives $F\subseteq E_1\subseteq E_2\subseteq\cdots$, whose union is algebraically closed (Artin's argument shows that $E_1$ already is), and the algebraic elements in it form an algebraic closure by (1). Uniqueness extends the identity of $F$ step by step with Theorem 8.6, again using Zorn's lemma to pass to the limit.
(3) If $F=\{a_1,\dots,a_q\}$, the polynomial $f=(x-a_1)\cdots(x-a_q)+1$ has degree $q\ge2$ and $f(a_i)=1\ne0$ for every $i$. ∎
Algebraic closures are used as a universe in which "all the roots" of all polynomials over $F$ live, so that splitting fields can be compared inside one field. For $F=\mathbb{Q}$ we simply work inside $\mathbb{C}$.
The oldest application of field degrees is to geometry. Starting from two points, which we call $0$ and $1$ in the plane $\mathbb{C}=\mathbb{R}^2$, a ruler-and-compass construction repeatedly draws (i) the line through two points already constructed, and (ii) the circle with centre a constructed point passing through another constructed point, and adds the intersection points of the lines and circles drawn.
- The constructible numbers form a subfield $\mathcal{C}$ of $\mathbb{C}$ that is closed under complex conjugation and under square roots; $a+bi\in\mathcal{C}$ ($a,b\in\mathbb{R}$) iff $a,b\in\mathcal{C}$.
- $\alpha\in\mathcal{C}$ if and only if there is a tower $\mathbb{Q}=K_0\subset K_1\subset\cdots\subset K_r\subseteq\mathbb{C}$ with $[K_j:K_{j-1}]=2$ for every $j$ and $\alpha\in K_r$.
- Consequently, if $\alpha$ is constructible then $[\mathbb{Q}(\alpha):\mathbb{Q}]$ is a power of $2$.
Proof (sketch with every key step). (1) The classical constructions give perpendiculars and parallels, hence the coordinates $a,b$ of a constructed point $a+bi$ and conversely the point from its coordinates. Sums and differences of real lengths are laid off on a line; products and quotients come from similar triangles (draw $1$ and $a$ on one ray, $b$ on another; the parallel through $a$ to the line joining $1$ and $b$ cuts the second ray at $ab$). For $a\gt0$, the semicircle on a diameter of length $1+a$ meets the perpendicular at the junction point at height $\sqrt a$ (the altitude of a right triangle is the geometric mean of the two pieces of the hypotenuse). A complex square root is obtained by bisecting the angle and taking the square root of the modulus.
(2) "$\Leftarrow$": each element of $K_j$ has the form $u+v\sqrt d$ with $u,v,d\in K_{j-1}$, so by induction and (1) every element of $K_r$ is constructible. "$\Rightarrow$": let $F\subseteq\mathbb{R}$ be a field containing all coordinates of the points constructed so far. A line through two such points has an equation $ax+by=c$ with $a,b,c\in F$; a circle has an equation $x^2+y^2+dx+ey+f=0$ with $d,e,f\in F$. Two lines meet at a point with coordinates in $F$ (Cramer's rule). A line and a circle: substituting the line into the circle gives a quadratic over $F$, so the new coordinates lie in $F(\sqrt D)$ for some $D\in F$, $D\ge0$ — an extension of degree $1$ or $2$. Two circles: subtracting their equations gives a line, reducing to the previous case. So the coordinates of all points constructed lie in the top of a tower of real fields, each of degree $1$ or $2$ over the previous; discard the steps of degree $1$ and finally adjoin $i$ (degree $2$) to reach $\alpha=a+bi$.
(3) $\mathbb{Q}(\alpha)\subseteq K_r$ and $[K_r:\mathbb{Q}]=2^r$, so $[\mathbb{Q}(\alpha):\mathbb{Q}]$ divides $2^r$ (Theorem 8.3). ∎
Proof. (a) $[\mathbb{Q}(\sqrt[3]2):\mathbb{Q}]=3$ is not a power of $2$. (b) $60^\circ$ is constructible (equilateral triangle). Trisecting it means constructing $c=\cos20^\circ$. From $\cos3t=4\cos^3t-3\cos t$ with $t=20^\circ$: $4c^3-3c=\tfrac12$, so $y=2c$ satisfies $y^3-3y-1=0$. This cubic has no rational root ($\pm1$ give $-3$ and $1$), so it is irreducible and $[\mathbb{Q}(c):\mathbb{Q}]=3$. (c) If $\sqrt\pi$ were constructible, so would be $\pi=(\sqrt\pi)^2$, which would then be algebraic; but $\pi$ is transcendental (Lindemann's theorem, which we quote). (d) The vertices of a regular $n$-gon inscribed in the unit circle with a vertex at $1$ are the $n$-th roots of unity; if it is constructible (starting from its centre and one vertex) then $\zeta_n=e^{2\pi i/n}\in\mathcal{C}$, and $[\mathbb{Q}(\zeta_n):\mathbb{Q}]=\deg\Phi_n=\varphi(n)$ because the cyclotomic polynomial $\Phi_n$ is irreducible over $\mathbb{Q}$ (§7.3). ∎
The converse of (d) — the regular $n$-gon is constructible iff $\varphi(n)$ is a power of $2$, i.e. iff $n=2^kp_1\cdots p_s$ with distinct Fermat primes $p_j=2^{2^{m}}+1$ (Gauss 1796, Wantzel 1837) — needs Galois theory and is proved in §8.3. The chapter introduction animates it.
Splitting Fields
Proof. Induction on $n$. For $n=1$, $K=F$. For $n\gt1$, let $p$ be an irreducible factor of $f$, and let $E=F(\alpha_1)$ be a field in which $p$ has a root $\alpha_1$ (Theorem 8.5(3)); $[E:F]=\deg p\le n$. Over $E$, $f=(x-\alpha_1)g$ with $\deg g=n-1$. By induction $g$ has a splitting field $K=E(\alpha_2,\dots,\alpha_n)$ over $E$ with $[K:E]\le(n-1)!$. Then $f$ splits over $K=F(\alpha_1,\dots,\alpha_n)$ and $[K:F]=[K:E][E:F]\le(n-1)!\,n$. Inside $\mathbb{C}$, the fundamental theorem of algebra supplies the roots, and $F(\alpha_1,\dots,\alpha_n)$ is the smallest subfield over which $f$ splits. ∎
Proof. Induction on $[K:F]$. If $[K:F]=1$, $f$ splits over $F$, so $f^\sigma$ splits over $F'$, $K'=F'$, and $\tau=\sigma$ is the only extension. If $[K:F]\gt1$, some irreducible factor $p$ of $f$ has degree $d\gt1$; pick a root $\alpha\in K$ of $p$. Then $p^\sigma$ is an irreducible factor of $f^\sigma$; it splits over $K'$, say with $e$ distinct roots $\beta_1,\dots,\beta_e$, where $e\le d$, and $e=d$ if $f^\sigma$ has no repeated roots. Any extension $\tau$ maps $\alpha$ to a root of $p^\sigma$ (apply $\tau$ to $p(\alpha)=0$). For each $j$, Theorem 8.6 gives exactly one $\sigma_j:F(\alpha)\to F'(\beta_j)$ extending $\sigma$ with $\alpha\mapsto\beta_j$. Now $K$ is a splitting field of $f$ over $F(\alpha)$ and $K'$ a splitting field of $f^{\sigma_j}=f^\sigma$ over $F'(\beta_j)$, and $[K:F(\alpha)]=[K:F]/d\lt[K:F]$. By induction each $\sigma_j$ extends to $K\to K'$, in at most $[K:F]/d$ ways (exactly that many in the separable case). Every extension of $\sigma$ arises from exactly one $j$, so there are at most $e\cdot[K:F]/d\le[K:F]$ of them, with equality when $e=d$ at every stage. ∎
The automorphisms of $K$ fixing $F$ form a group — the Galois group of §8.3 — and Theorem 8.12 already shows it has at most $[K:F]$ elements, exactly $[K:F]$ when the roots of $f$ are distinct.
Proof. The formula. A monic cubic $x^3+ax^2+bx+c$ becomes $x^3+px+q$ under $x\mapsto x-a/3$, which changes neither $F$, nor the field generated by the roots, nor their differences; so assume $f=x^3+px+q$. If $q=0$ then $x\mid f$; so $q\ne0$ and no root is $0$. Since $f=\prod(x-r_i)$, $f'(r_1)=(r_1-r_2)(r_1-r_3)$ and cyclically, so $\prod_if'(r_i)=\prod_{i\ne j}(r_i-r_j)=-D$. Using $r_i^3=-pr_i-q$, $$f'(r_i)=3r_i^2+p=\frac{3r_i^3+pr_i}{r_i}=\frac{-2pr_i-3q}{r_i}.$$ Here $\prod_ir_i=-q$, and for $p\ne0$, using $\prod_i(r_i-c)=-f(c)$ with $c=-\tfrac{3q}{2p}$, $$\prod_i(-2pr_i-3q)=-8p^3\prod_i(r_i-c)=8p^3f(c)=-27q^3-4p^3q$$ (for $p=0$ both ends equal $-27q^3$). Dividing by $-q$: $\prod f'(r_i)=4p^3+27q^2$, so $D=-4p^3-27q^2\in F$.
The degree. $[F(r_1):F]=3$. Over $F(r_1)$, $f=(x-r_1)g$ with $g=(x-r_2)(x-r_3)\in F(r_1)[x]$, and $K=F(r_1,r_2)$ since $r_3=-r_1-r_2$. So $[K:F]=3$ or $6$ according as $r_2\in F(r_1)$ or not. Put $\delta=(r_1-r_2)(r_1-r_3)(r_2-r_3)=g(r_1)\,(r_2-r_3)$, where $g(r_1)\ne0$ lies in $F(r_1)$ (an irreducible polynomial over a subfield of $\mathbb{C}$ has distinct roots, Theorem 8.15 below). Since $r_2+r_3=-r_1\in F(r_1)$, we get: $r_2\in F(r_1)\iff r_2-r_3\in F(r_1)\iff\delta\in F(r_1)$. Finally $\delta^2=D\in F$, so $[F(\delta):F]\in\{1,2\}$; if $\delta\in F(r_1)$ this degree also divides $3$, hence is $1$. Therefore $\delta\in F(r_1)\iff\delta\in F\iff D$ is a square in $F$. ∎
Proof. Let $\alpha=\sqrt[p]a\gt0$. The roots of $x^n-a$ are $\alpha\zeta_n^k$, $0\le k\lt n$ (for general $n$, $\alpha$ is any one root); they all lie in $\mathbb{Q}(\alpha,\zeta_n)$, and conversely $\zeta_n=(\alpha\zeta_n)/\alpha$ lies in the splitting field. Irreducibility: if $x^p-a=gh$ in $\mathbb{Q}[x]$ with $g$ monic of degree $k$, $0\lt k\lt p$, then $g$ is a product of $k$ factors $x-\alpha\zeta_p^{j}$, so $|g(0)|=\alpha^k\in\mathbb{Q}$. As $\gcd(k,p)=1$, write $1=uk+vp$ with $u,v\in\mathbb{Z}$: $\alpha=(\alpha^k)^u(\alpha^p)^v=(\alpha^k)^ua^v\in\mathbb{Q}$, so $a=\alpha^p$ is a $p$-th power in $\mathbb{Q}$, a contradiction. Degree: $[\mathbb{Q}(\alpha):\mathbb{Q}]=p$ and $[\mathbb{Q}(\zeta_p):\mathbb{Q}]=p-1$ ($\Phi_p$ is irreducible, §7.3) both divide $[K:\mathbb{Q}]$ and are coprime, so $p(p-1)\mid[K:\mathbb{Q}]$. But $K=\mathbb{Q}(\zeta_p)(\alpha)$ and $\alpha$ has degree $\le p$ over $\mathbb{Q}(\zeta_p)$, so $[K:\mathbb{Q}]\le p(p-1)$. ∎
Separability
Over $\mathbb{Q}$ an irreducible polynomial never has a repeated root. Over fields of characteristic $p$ it can, and that phenomenon — inseparability — must be excluded before Galois theory works smoothly.
- $f$ is separable $\iff\gcd(f,f')=1$ in $F[x]$. (The gcd may be computed in $F[x]$; it does not change in extensions.)
- If $f$ is irreducible, then $f$ is separable $\iff f'\ne0$.
- If $\operatorname{char}F=0$, every irreducible polynomial is separable: fields of characteristic $0$ are perfect.
- If $\operatorname{char}F=p$, an irreducible $f$ is inseparable $\iff f(x)=g(x^p)$ for some $g\in F[x]$.
Proof. (1) In a splitting field, $\alpha$ is a repeated root of $f$ iff $f(\alpha)=f'(\alpha)=0$ (§7.3, "repeated roots and the derivative"), iff $x-\alpha$ divides $\gcd(f,f')$ computed there. The Euclidean algorithm for $f,f'$ runs inside $F[x]$, so the monic gcd is the same over $F$ and over any extension; it is $1$ iff $f,f'$ have no common root. (2) If $f$ is irreducible, $\gcd(f,f')$ is $1$ or an associate of $f$; the latter happens iff $f\mid f'$, and since $\deg f'\lt\deg f$ that means $f'=0$. (3) In characteristic $0$, $f=\sum a_ix^i$ with $a_n\ne0$, $n\ge1$, has $f'$ with leading coefficient $na_n\ne0$. (4) $f'=\sum ia_ix^{i-1}=0$ iff $ia_i=0$ for all $i$ iff $a_i=0$ whenever $p\nmid i$, i.e. iff only powers $x^{pj}$ occur. ∎
- A field $F$ of characteristic $p$ is perfect iff every element of $F$ is a $p$-th power, $F=F^p$.
- Every finite field is perfect. Hence (with Theorem 8.15(3)) every algebraic extension of $\mathbb{Q}$ or of a finite field is separable.
- Let $F=\mathbb{F}_p(t)$ ($t$ transcendental). Then $x^p-t\in F[x]$ is irreducible and inseparable: in its splitting field it equals $(x-\tau)^p$, where $\tau^p=t$. So $\mathbb{F}_p(t)$ is not perfect.
Proof. (1) If some $a\in F$ is not a $p$-th power, take a root $\tau$ of $x^p-a$ in an extension; then $x^p-a=x^p-\tau^p=(x-\tau)^p$ ("freshman's dream": $\binom pk\equiv0\pmod p$ for $0\lt k\lt p$). Any monic factor in $F[x]$ is $(x-\tau)^k$ with $0\lt k\le p$, whose coefficient of $x^{k-1}$ is $-k\tau$; if $k\lt p$ this gives $\tau\in F$ and $a=\tau^p$, a contradiction. So $x^p-a$ is irreducible and inseparable. Conversely, if $F=F^p$ and $f=g(x^p)=\sum b_jx^{pj}$, write $b_j=c_j^p$; then $f=\big(\sum c_jx^j\big)^p$ is not irreducible. By Theorem 8.15(4), every irreducible polynomial is separable. (2) For finite $F$ of characteristic $p$ the Frobenius map $a\mapsto a^p$ is a ring homomorphism (freshman's dream again) and injective (Theorem 8.1), hence surjective on the finite set $F$: $F=F^p$. If $K/F$ is algebraic and $\alpha\in K$, $m_{\alpha,F}$ is irreducible over the perfect field $F$, hence separable. (3) $t$ is not a $p$-th power in $\mathbb{F}_p(t)$: if $t=(g/h)^p$ with polynomials $g,h$, then $th^p=g^p$, and the degree in $t$ of the left side is $\equiv1$, of the right side $\equiv0\pmod p$. Apply the argument of (1). ∎
The definition of a perfect field in (1) explains the name: in a perfect field of characteristic $p$, every element has a unique $p$-th root. Separability is what makes the following classical theorem work; for finite fields it also follows from the cyclicity of the multiplicative group (§8.2).
Proof. If $F$ is finite, so is $K$, and $K^\times$ is cyclic (§8.2); a generator $\gamma$ of $K^\times$ satisfies $K=F(\gamma)$. Let $F$ be infinite. Since $K$ is finitely generated (Theorem 8.7), induction on the number of generators reduces to $K=F(\alpha,\beta)$. Let $f=m_{\alpha,F}$, $g=m_{\beta,F}$, and in a field $L\supseteq K$ over which $fg$ splits let $f$ have roots $\alpha=\alpha_1,\dots,\alpha_r$ and $g$ the distinct roots $\beta=\beta_1,\dots,\beta_s$ ($g$ is separable). For $j\ne1$ the equation $\alpha_i+c\beta_j=\alpha+c\beta$ has at most one solution $c$; choose $c\in F$ avoiding these finitely many values, and put $\gamma=\alpha+c\beta$. The polynomial $h(x)=f(\gamma-cx)\in F(\gamma)[x]$ satisfies $h(\beta)=f(\alpha)=0$, and for $j\ne1$, $h(\beta_j)=0$ would mean $\gamma-c\beta_j=\alpha_i$ for some $i$, which we excluded. So the only common root of $g$ and $h$ in $L$ is $\beta$, and it is a simple root of $g$; hence $\gcd(g,h)=x-\beta$ in $L[x]$. But the gcd of two polynomials of $F(\gamma)[x]$ is computed in $F(\gamma)[x]$ by Euclid's algorithm, so $x-\beta\in F(\gamma)[x]$, i.e. $\beta\in F(\gamma)$, and then $\alpha=\gamma-c\beta\in F(\gamma)$. Thus $K=F(\gamma)$. ∎
- Degrees of composites are not products. $[\mathbb{Q}(\sqrt[3]2,\omega\sqrt[3]2):\mathbb{Q}]=6$, not $9$; $[\mathbb{Q}(\sqrt2,\sqrt8):\mathbb{Q}]=2$. The product is only an upper bound; coprimality of the degrees is what guarantees equality.
- $\mathbb{Q}(\alpha^2)$ can be smaller than $\mathbb{Q}(\alpha)$. $\alpha=\sqrt2+\sqrt3$: $\mathbb{Q}(\alpha^2)=\mathbb{Q}(\sqrt6)$ has degree $2$, while $\mathbb{Q}(\alpha)$ has degree $4$. For odd $[F(\alpha):F]$ this cannot happen.
- Isomorphic is not equal. $\mathbb{Q}(\sqrt[3]2)\cong\mathbb{Q}(\omega\sqrt[3]2)$ as fields, but they are different subfields of $\mathbb{C}$ (only the first is real), and neither contains the other.
- Same degree, not isomorphic. $\mathbb{Q}(\sqrt2)\not\cong\mathbb{Q}(\sqrt3)$: an isomorphism fixes $\mathbb{Q}$ and would send $\sqrt2$ to a square root of $2$ in $\mathbb{Q}(\sqrt3)$, which does not exist.
- $\mathbb{Q}(\text{one root})$ is rarely the splitting field. $\mathbb{Q}(\sqrt[3]2)\subset\mathbb{R}$ contains only one root of $x^3-2$; $\mathbb{Q}(\sqrt[4]2)$ contains two of the four roots of $x^4-2$. But for $x^3-3x+1$ and $x^4+1$ one root gives everything.
- Algebraic does not mean finite. $\overline{\mathbb{Q}}/\mathbb{Q}$ is algebraic of infinite degree; $\mathbb{Q}(\sqrt2,\sqrt3,\sqrt5,\dots)$ likewise.
- A field can be isomorphic to a proper subfield of itself. $\mathbb{Q}(t)\cong\mathbb{Q}(t^2)\subsetneq\mathbb{Q}(t)$, with $[\mathbb{Q}(t):\mathbb{Q}(t^2)]=2$. "Isomorphic and contained" implies equal only for finite extensions of the same base.
- Power of $2$ is necessary, not sufficient, for constructibility. The real roots of $x^4-4x+2$ (Eisenstein, degree $4$) are not constructible: the splitting field has degree $24$, which is not a power of $2$ (§8.3).
- Splitting-field degrees: small and large. It can be smaller than $\deg f$ ($x^6-1$: $2$) and as large as $n!$ ($x^4-4x+2$: $24$); it is divisible by $\deg f$ when $f$ is irreducible.
- Minimal polynomials depend on the base. $m_{\sqrt2,\mathbb{Q}}=x^2-2$ but $m_{\sqrt2,\mathbb{R}}=x-\sqrt2$; $m_{\sqrt[4]2,\mathbb{Q}(\sqrt2)}=x^2-\sqrt2$.
- Inseparability needs characteristic $p$ and an imperfect field. Over $\mathbb{Q}$, $\mathbb{R}$, $\mathbb{C}$ and every finite field all irreducible polynomials are separable; $x^p-t$ over $\mathbb{F}_p(t)$ is the standard exception. A reducible polynomial such as $(x+1)^2$ can be inseparable over any field.
- $f'=0$ does not make $f$ constant in characteristic $p$: $(x^p)'=px^{p-1}=0$.
- Algebraically closed fields are infinite; the algebraic closure of $\mathbb{Q}$ is the countable field $\overline{\mathbb{Q}}$, not $\mathbb{C}$.
Exercises
- 1 Find $[\mathbb{Q}(\sqrt2,\sqrt5):\mathbb{Q}]$ and a basis.
- 2 Show that $\mathbb{Q}(\sqrt2+\sqrt5)=\mathbb{Q}(\sqrt2,\sqrt5)$ and find the minimal polynomial of $\sqrt2+\sqrt5$ over $\mathbb{Q}$.
- 3 Find $[\mathbb{Q}(\sqrt[3]2,i):\mathbb{Q}]$.
- 4 Find $[\mathbb{Q}(\sqrt3,\sqrt[4]3):\mathbb{Q}]$.
- 5 Find $[\mathbb{Q}(\sqrt[6]2):\mathbb{Q}(\sqrt2)]$ and the minimal polynomial of $\sqrt[6]2$ over $\mathbb{Q}(\sqrt2)$.
- 6 Let $\operatorname{char}F\ne2$ and $[K:F]=2$. Prove that $K=F(\sqrt d)$ for some $d\in F$, i.e. $K=F(\beta)$ with $\beta^2=d\in F$.
- 7 Let $[K:F]=n$ and let $f\in F[x]$ be irreducible of degree $m$ with $\gcd(m,n)=1$. Prove that $f$ has no root in $K$ (if $m\gt1$) and is irreducible over $K$.
- 8 Prove in two ways that $[\mathbb{R}:\mathbb{Q}]=\infty$: by countability, and by exhibiting real numbers of arbitrarily large degree.
- 9 Find the minimal polynomial of $\sqrt2+\sqrt3+\sqrt6$ over $\mathbb{Q}$, given that it has degree $4$ — and prove that the degree is $4$.
- 1 Find the minimal polynomial of $1+i$ over $\mathbb{Q}$, over $\mathbb{R}$ and over $\mathbb{C}$.
- 2 Find the minimal polynomial of $\sqrt{2+\sqrt2}$ over $\mathbb{Q}$ and over $\mathbb{Q}(\sqrt2)$.
- 3 Let $\zeta=e^{2\pi i/5}$. Show that $\sqrt5\in\mathbb{Q}(\zeta)$ and find the minimal polynomial of $\zeta$ over $\mathbb{Q}(\sqrt5)$.
- 4 Let $\theta$ be a root of $x^3-x-1$. Prove that $[\mathbb{Q}(\theta):\mathbb{Q}]=3$ and write $\theta^5$, $\theta^{-1}$ and $(\theta+2)^{-1}$ in the basis $1,\theta,\theta^2$.
- 5 Prove that $[\mathbb{Q}(\sqrt2+\sqrt[3]3):\mathbb{Q}]=6$.
- 6 Let $\alpha$ be a root of $x^3-3x+1$. Show that $\alpha^2-2$ is also a root, and deduce that $\mathbb{Q}(\alpha)$ is the splitting field.
- 7 Factor $x^4+1$ into quadratics over $\mathbb{Q}(\sqrt2)$, over $\mathbb{Q}(i)$ and over $\mathbb{Q}(\sqrt{-2})$.
- 8 Assuming $\pi$ is transcendental, show that $[\mathbb{Q}(\pi):\mathbb{Q}(\pi^2)]=2$ and that $\mathbb{Q}(\pi^2)$ is not algebraic over $\mathbb{Q}$.
- 1 Find the degree of the splitting field of $x^4-3$ over $\mathbb{Q}$.
- 2 Find the degree of the splitting field of $x^6+x^3+1$ over $\mathbb{Q}$. (Hint: $x^9-1$.)
- 3 Show that $x^3-7x+7$ is irreducible over $\mathbb{Q}$ and find the degree of its splitting field.
- 4 Find the degree of the splitting field of $x^3+x+1$ over $\mathbb{Q}$, and of $x^4+4$ over $\mathbb{Q}$.
- 5 Find the degree of the splitting field of $(x^2-2)(x^2-3)(x^2-6)$, and of $x^p-1$ for a prime $p$.
- 6 Decide whether each of $\sqrt[3]{2+\sqrt5}$ and $\sqrt[3]{1+\sqrt2}$ (real cube roots) is constructible.
- 7 Show that the angle $90^\circ$ can be trisected with ruler and compass, but the angle $\arccos\tfrac13$ cannot. (Hint: $4c^3-3c=\tfrac13$.)
- 8 Decide which are separable: (a) $x^4+1$ over $\mathbb{F}_2$; (b) $x^3-2$ over $\mathbb{F}_5$; (c) $x^9-x$ over $\mathbb{F}_3$; (d) $x^p-x-a$ over $\mathbb{F}_p$, $a\in\mathbb{F}_p$.
- 9 How many field homomorphisms are there from $\mathbb{Q}(\sqrt[4]2)$ to $\mathbb{R}$? to $\mathbb{C}$? How many automorphisms does $\mathbb{Q}(\sqrt[3]2)$ have?
- 1 Let $K/F$ be algebraic and $R$ a ring with $F\subseteq R\subseteq K$. Prove that $R$ is a field.
- 2 Prove that an algebraically closed field is infinite, and that $\overline{\mathbb{Q}}\ne\mathbb{C}$.
- 3 For which $c\in\mathbb{Q}$ is $\mathbb{Q}(\sqrt2+c\sqrt3)=\mathbb{Q}(\sqrt2,\sqrt3)$?
- 4 Let $f\in F[x]$ be irreducible of degree $n$ with splitting field $K$. Prove that $n\mid[K:F]\le n!$, and that if $[K:F]=n$ then $K=F(\alpha)$ for every root $\alpha$ of $f$.
- 5 Let $\operatorname{char}F=p$ and let $\alpha$ be separable over $F$. Prove that $F(\alpha)=F(\alpha^p)$.
- 6 Let $f\in\mathbb{Q}[x]$ be irreducible of degree $n$ whose splitting field $K$ has degree $n!$. Prove that for every root $\alpha$, $f/(x-\alpha)$ is irreducible over $\mathbb{Q}(\alpha)$.
- 7 Prove that if $\alpha$ is algebraic over $F$ of odd degree, then so is every $\beta\in F(\alpha)$ — and that $F(\alpha)$ contains no root of $x^2-d$ for $d\in F$ not a square in $F$.
E1. CSIR Part B GATE MA The degree $[\mathbb{Q}(\sqrt2,\sqrt[3]2,i):\mathbb{Q}]$ is (A) $6$; (B) $8$; (C) $12$; (D) $24$.
E2. CSIR Part B The degree over $\mathbb{Q}$ of the splitting field of $x^6-3$ is (A) $6$; (B) $12$; (C) $18$; (D) $36$.
E3. CSIR Part B Which of the following real numbers is constructible by ruler and compass? (A) $\sqrt[3]3$; (B) $\cos(2\pi/7)$; (C) $\sqrt{3+\sqrt2}$; (D) the real root of $x^3-x-1$.
E4. CSIR Part C Let $\alpha=\sqrt3+\sqrt5$. Which of the following are true? (One or more options may be correct.) (A) $[\mathbb{Q}(\alpha):\mathbb{Q}]=4$. (B) $\mathbb{Q}(\alpha^2)=\mathbb{Q}(\alpha)$. (C) The minimal polynomial of $\alpha$ over $\mathbb{Q}(\sqrt3)$ is $x^2-2\sqrt3\,x-2$. (D) $x^4-16x^2+4$ is irreducible over $\mathbb{Q}(\sqrt{15})$.
E5. CSIR Part C GATE MA Which of the following statements are true? (A) The splitting field of $x^3-12x+8$ over $\mathbb{Q}$ has degree $3$. (B) $\mathbb{Q}(\sqrt[3]2)$ is a splitting field over $\mathbb{Q}$ of some polynomial in $\mathbb{Q}[x]$. (C) The splitting field of $x^5-1$ over $\mathbb{Q}$ has degree $5$. (D) If $f\in\mathbb{Q}[x]$ is irreducible of degree $n$, the degree of its splitting field is divisible by $n$.
E6. CSIR Part C Let $p$ be a prime and $F=\mathbb{F}_p(t)$, the field of rational functions in an indeterminate $t$. Which of the following are true? (A) $x^p-t$ is irreducible over $F$. (B) $x^p-t$ is separable over $F$. (C) Every irreducible polynomial in $\mathbb{F}_p[x]$ is separable. (D) $F$ is a perfect field.
Suggested plan — 3 lectures × 60 min.
- Extensions and degrees. Definition 8.1 with $\mathbb{C}/\mathbb{R}$, $\mathbb{Q}(\sqrt2)/\mathbb{Q}$, $\mathbb{R}/\mathbb{Q}$; Theorem 8.1 briefly; the tower law (Theorem 8.2) with a full proof on the board and Theorem 8.3; Example 1. Then Definition 8.2, Theorems 8.4–8.5 (stress that $F(\alpha)\cong F[x]/\langle m\rangle$ is the first isomorphism theorem applied to evaluation), Example 3 (inverses by Bézout).
- Algebraic elements. Theorem 8.6 with the three embeddings of $\mathbb{Q}(\sqrt[3]2)$ into $\mathbb{C}$; Example 2; project Interactive 1 — let the class predict the degree of $\sqrt2+\sqrt6$ and of $(1+\sqrt3)/2$ before running it. Theorem 8.7, the strategy box and Example 4. Close with algebraic closures (Theorem 8.8; the proof of (2) only as an idea) and constructible numbers (Theorems 8.9–8.10, Example 5).
- Splitting fields and separability. Theorems 8.11–8.14 with Interactive 2 ($x^3-2$, then $x^3-3x+1$ — ask why one root suffices — then $x^4-4x+2$). Example 6. Separability: Theorems 8.15–8.16, the example $x^p-t$, the primitive element theorem 8.17 and Example 7. Finish with E1, E4 and E5 done live.
Discussion questions. (1) $\mathbb{Q}(\sqrt[3]2)$ and $\mathbb{Q}(\omega\sqrt[3]2)$ are isomorphic but different. Which properties of a subfield of $\mathbb{C}$ are "intrinsic" (preserved by isomorphism), and which depend on how it sits in $\mathbb{C}$? (2) Why does the degree of the splitting field of an irreducible cubic depend only on whether one number is a square? What plays this role for quartics? (3) Is every number whose degree is a power of $2$ constructible? How would you look for a counterexample?
Misconceptions to watch for. Multiplying degrees without checking coprimality; assuming $\mathbb{Q}(\alpha)$ contains all conjugates of $\alpha$; confusing "isomorphic" with "equal" for subfields of $\mathbb{C}$; believing that the minimal polynomial does not depend on the base field; believing that "degree $2^k$" implies constructible; thinking that inseparability can occur over $\mathbb{Q}$ or over a finite field.
Suggested homework. I(1, 2, 5, 7, 9), II(2, 3, 4, 5), III(1, 3, 4, 6, 8), IV(1, 3, 5); Exam corner E2, E3, E6.
Research question. Given only the first few dozen decimal digits of a real number, can a computer discover its minimal polynomial — and how many digits are needed before the answer can be trusted?
Why it matters. Integer relation algorithms (PSLQ, by Ferguson and Bailey, and the LLL lattice-reduction algorithm of Lenstra, Lenstra and Lovász) search for integers $c_0,\dots,c_n$, not all zero, with $\sum c_k\alpha^k\approx0$. They are the engine of "experimental mathematics": the 1995 Bailey–Borwein–Plouffe formula for $\pi$ was found this way, and physicists use the same method to recognise high-precision values of Feynman integrals in particle physics as combinations of known constants. The method only suggests a polynomial; proving it is the minimal polynomial is the job of this section's theory.
What to do.
- In Python,
import mpmath, setmpmath.mp.dps = 30, and runmpmath.findpoly(2*mpmath.cos(2*mpmath.pi/7), 3, maxcoeff=100). Compare the answer with Example 5(c). - Repeat for $\sqrt2+\sqrt[3]3$ with degree bound $6$ (add
maxcoeff=10**4, maxsteps=10**5) at $15$, $20$ and $30$ digits. Record what you get each time. - For each polynomial returned, decide with the tools of §7.3 and §8.1 whether it can be the minimal polynomial (degree? irreducible? does $\alpha$ satisfy it to $50$ digits?).
- Find the real root of $x^4-4x+2$ near $0.518$ to $50$ digits with
findroot, recover its polynomial, and explain from Theorem 8.9 why this number, although of degree $4$, is still not constructible (§8.3 completes the argument).
Tools: Python with mpmath (bundled with sympy), or any system with a PSLQ/LLL routine (PARI/GP algdep, Mathematica RootApproximant).
Hand in: A table "digits used → polynomial found → genuine or spurious, and why", plus a paragraph on how many digits a degree-$n$ relation with coefficients of size $H$ needs (rule of thumb: comfortably more than $n\log_{10}H$).
Going further. Read about how integer relation methods showed that the third bifurcation point $3.5440903\ldots$ of the logistic map is an algebraic number of degree $12$, or use LLL to factor $x^4-10x^2+1$ over $\mathbb{Q}(\sqrt2)$ numerically.
[-1, -1, 2, 1], i.e. $x^3+x^2-2x-1$ up to sign. For $\sqrt2+\sqrt[3]3$, $15$ digits give the spurious cubic [185, -1307, 1334, 2542] and $20$ digits the spurious quartic [-668, 3597, -3191, -6708, 5835] — both impossible, since the degree is $6$ (Exercise II.5) — while $30$ digits give [1, 0, -6, -6, 12, -36, 1], the true minimal polynomial. The root of $x^4-4x+2$ is $0.51799935\ldots$ and findpoly returns [-1, 0, 0, 4, -2].🧪 Self-check quiz
Q1. $[\mathbb{Q}(\sqrt2,\sqrt3):\mathbb{Q}]$ equals
Q2. The minimal polynomial of $\sqrt[3]2$ over $\mathbb{Q}(\sqrt2)$ is
Q3. The splitting field of $x^3-2$ over $\mathbb{Q}$ has degree
Q4. Which number is not constructible with ruler and compass?
Q5. Over which field is there an irreducible polynomial with a repeated root?