Past exam of the number theory course of the University of Oxford / 2008 / A (Galois Theory)

For an irreducible polynomial $f\in K[X]$, take $L=K[X]/(f)$ and let $\alpha$ be the class of $X$. The quotient is a field, is generated by $\alpha$ over $K$, and satisfies $f(\alpha )=0$. Its degree is $\boxed{[L:K]=\mathrm{deg}f}$, with basis $1,\alpha ,\dots ,{\alpha }^{\mathrm{deg}f-1}$.

A splitting field is a field generated over $K$ by all roots of the polynomial and in which the polynomial is a product of linear factors. Prove existence and the factorial bound together by induction on the degree. Constant nonzero polynomials and linear polynomials require no extension. For a reducible degree-$n$ polynomial write $g=hk$ with positive degrees $r,n-r$. Split $h$ by induction, then split $k$ over that field. The resulting field is generated by all roots of $g$, and its degree over $K$ is at most $r!(n-r)!\le n!$. If $g$ is irreducible, adjoin one root in degree $n$, divide out its linear factor, and split the degree-$(n-1)$ quotient over that extension. The tower formula gives degree at most $n(n-1)!=n!$. This also works in inseparable characteristic and proves the required bound without assuming Galoisness.

For the sextic with constant term three, Eisenstein's criterion at three gives degree six. Choose a root $\alpha$, so ${\alpha }^3$ is a square root of $-3$. Thensatisfies ${\zeta }^2-\zeta +1=0$, and is a primitive sixth root of unity: ${\zeta }^3=-1$ and $\zeta \ne -1$. The six distinct roots are $\alpha {\zeta }^j$ for $0\le j<6$, all in $\mathbb{Q}(\alpha )$. Hence the simple extension is itself the splitting field and $\boxed{[L:\mathbb{Q}]=6}$.

For the constant term two, Eisenstein at two gives $[\mathbb{Q}(\gamma ):\mathbb{Q}]=6$ for a root $\gamma$. Its splitting field is $\mathbb{Q}(\gamma ,{\zeta }_6)$. The supplied noncontainment of a root of $X^2+3$ means that adjoining ${\zeta }_6=(1+\sqrt{-3})/2$ has degree two. Thus

Every automorphism fixes zero and one, and preserves sums, products, negatives and multiplicative inverses of nonzero elements. Consequently elements fixed by every member of $G$ are closed under these operations and form a field. Artin fixed-field theorem gives $\boxed{[L:L^G]=|G|}$ for a finite automorphism group.

Here $\mathbb{Q}(t)$ is the fraction field of $\mathbb{Q}[t]$, with $t$ transcendental; its elements are fractions of rational-coefficient polynomials, with nonzero denominator and the usual equality by cross-multiplication. Substitution of $1-t$ or $1/t$ defines an automorphism, each its own inverse. With $\sigma \tau$ meaning first apply $\tau$ to a field element and then $\sigma$, its image on $t$ is $1/(1-t)$. Iterating gives $(t-1)/t$ and then $t$, so its order is three. The relations reduce every word to at most six possibilities, and the six resulting images areThey are distinct rational functions, so the group has six elements. Its faithful permutation action on $\{0,1,\infty \}$ identifies it with $\boxed{S_3}$.

For the first fixed field put $u=t(1-t)$. It is fixed by $\sigma$, and $t$ satisfies $X^2-X+u=0$ over $\mathbb{Q}(u)$. The degree is exactly two because the nontrivial automorphism $\sigma$ fixes that field. The fixed-field degree is also two, soFor the inversion put $v=t+1/t$. It is fixed and $t$ satisfies $X^2-vX+1=0$. The same degree argument givesThe source requests polynomial generators in both cases. The second requirement is impossible. If a nonconstant polynomial $p(t)=a_d{t}^d+\cdots +a_0$ were fixed by inversion, $p(t)=p(1/t)$ would equate a Laurent polynomial with highest exponent $d>0$ to one with no positive exponent, a contradiction. Thus the only polynomial invariants of inversion are constants, which cannot generate its nonconstant fixed field. The fixed field of inversion in a rational function field therefore has the second displayed rational-function generator.

A finite extension is Galois when it is both normal and separable; over $\mathbb{Q}$ separability is automatic, so it is equivalently a splitting field of a polynomial with distinct roots. Its Galois group consists of the field automorphisms fixing $\mathbb{Q}$. If $G=\mathrm{Gal}(L/\mathbb{Q})$, the Fundamental theorem of Galois theory gives mutually inverse inclusion-reversing bijectionsThey satisfy $[L:L^H]=|H|$ and $[L^H:\mathbb{Q}]=[G:H]$. The intermediate extension $K/\mathbb{Q}$ is Galois exactly when $H=\mathrm{Gal}(L/K)$ is normal, and then $\boxed{\mathrm{Gal}(K/\mathbb{Q})\cong G/H}$ via restriction.

Every permutation of the four roots permutes the three unordered partitions into two pairs and hence permutes the three ${\beta }_j$. Thus each coefficient of $g(X)={\prod }_{j=1}^3(X-{\beta }_j)$ is a symmetric polynomial in the four roots. The Fundamental theorem of symmetric polynomials expresses it in the elementary symmetric functions, which are rational coefficients of $f$. Hence $g\in \mathbb{Q}[X]$. Each product ${\alpha }_i{\alpha }_j$, $i<j$, occurs exactly twice in ${\beta }_1+{\beta }_2+{\beta }_3$, so that sum is $2q$. Therefore the quadratic coefficient is $\boxed{-2q}$.

For the given cubic write $g(X)=(X-2{)}^3+12$. The translated cubic $Y^3+12$ is Eisenstein at three, hence irreducible. It has one real root and a nonreal conjugate pair. Its Galois group is transitive on three roots, so its order is divisible by three; complex conjugation is a transposition. A subgroup of $S_3$ with both a three-cycle and a transposition is all of $S_3$. HenceLet $G=\mathrm{Gal}(L/\mathbb{Q})\le S_4$. Irreducibility of the quartic makes the root action transitive, so four divides $|G|$. The cubic splitting field $K$ lies inside $L$ by the definitions of the ${\beta }_j$, and restriction onto its Galois group gives a quotient of order six. Thus six also divides $|G|$. Since $|G|$ divides 24, it is 12 or 24. If it were 12, the supplied index-two result would give $G=A_4$. The restriction kernel would then be a normal subgroup of $A_4$ of index six, contrary to the other supplied result. Thus

A straightedge and compass construction uses finitely many steps drawing lines through known points and circles with known centres and constructible radii, and adjoining their intersection points. The coordinates obtained by line–line intersections require only field operations; line–circle and circle–circle intersections require at most a quadratic extension after eliminating a linear equation. Thus all coordinates lie in a tower of quadratic extensions of the initial coordinate field. The degree of any finite intermediate coordinate field containing that initial field divides a power of two and is therefore $\boxed{2^r}$ for some $r\ge 0$.

For the source's notation this assumes $K\supseteq K_0$, as is customary when previously constructed points are retained. If only the coordinates of a final subset are called $K$, replace it by the compositum $K_{0}K$; otherwise the expression $[K:K_0]$ need not be defined. The power-of-two condition is necessary and is not being asserted as sufficient for an arbitrary algebraic number.

Starting from a unit cube, doubling its volume would construct a length $\sqrt[3]{2}$. Its polynomial $X^3-2$ is Eisenstein at two, so its degree over $\mathbb{Q}$ is three, which cannot divide a power of two. Thus doubling the cube is impossible by ruler and compass.

Set $c=\cos \theta$, $s=\sin \theta$. The real part of $(c+is{)}^5$ is $c^5-10c^3{s}^2+5c{s}^4$. Replacing $s^2$ by $1-c^2$ givesChoose the given angle to have cosine $5/6$. It is constructible, since its unit circle point has coordinates $(5/6,\sqrt{11}/6)$. If a fifth of that angle were constructible and had cosine $c$, the identity would implyThis polynomial is Eisenstein at five: all nonleading coefficients are divisible by five and the constant term is not divisible by 25, while 96 is not divisible by five. Hence $[\mathbb{Q}(c):\mathbb{Q}]=5$. But the initial angle field is $\mathbb{Q}(\sqrt{11})$, itself of degree two, so adjoining any finite collection of constructible coordinates to it would still give total degree over $\mathbb{Q}$ a power of two. It could not contain a degree-five element. Thus general angle quintisection is impossible with straightedge and compass.