Math 2.0

Field theory and polynomials Proof

Proof of the Gauss–Wantzel theorem

Builds on Constructible numberThe constructible numbers form a field closed under square rootsSum of the roots of unityThe degree of a constructible number is a power of twoCyclotomic polynomialIrreducibility of the cyclotomic polynomials over the rationalsEuler's totient functionMinimal polynomial of an algebraic elementThe minimal polynomial divides every polynomial vanishing at the elementDegree of a simple extension is the degree of the minimal polynomialMultiplicativity of the totient functionTotient of a prime powerFactor theoremFermat primeEuler's formulade Moivre's formulaThe nth roots of unityNumber of nth roots of unityA polynomial over a field has at most as many roots as its degreeSplitting fieldSeparable polynomialCharacterizations of Galois extensionsGalois groupOrder of the Galois group of a Galois extension equals its degreeK(a) is isomorphic to K[x] modulo the minimal polynomialAbelian groupCauchy's theoremNormal subgroupFixed fieldFundamental theorem of Galois theoryTower law

Proof

Let n≥3n \ge 3, let ζ=e2πi/n\zeta = e^{2\pi i/n}, and let 𝒞\mathcal{C} be the set of the constructible numbers. By the field theorem, 𝒞\mathcal{C} is a subfield of ℂ\C closed under square roots. By Euler's formula and de Moivre's formula, the powers ζk=e2πik/n\zeta^k = e^{2\pi i k/n}, 0≤k<n0 \le k < n, are the vertices of a regular nn-gon with centre 00. If ζ∈𝒞\zeta \in \mathcal{C}, they all lie in the field 𝒞\mathcal{C}, so this nn-gon is constructible. Conversely, let a regular nn-gon with centre cc be constructed, and let vv and ww be adjacent vertices. Its vertices are c+(v−c)ζkc + (v - c)\zeta^k for 0≤k<n0 \le k < n. Their sum is ncnc, because ∑k=0n−1ζk=0\sum_{k=0}^{n-1} \zeta^k = 0. So c∈𝒞c \in \mathcal{C}. Also (w−c)/(v−c)(w - c)/(v - c) is ζ\zeta or ζ−1\zeta^{-1}, so ζ∈𝒞\zeta \in \mathcal{C}. So we must show: ζ∈𝒞\zeta \in \mathcal{C} if and only if nn has the stated form. Let φ\varphi be Euler's totient function.

The degree of ζ\zeta.

  1. The cyclotomic polynomial Φn\Phi_n is monic, has the root ζ\zeta (the factor with k=1k = 1), and has degree φ(n)\varphi(n). Indeed, for n≥2n \ge 2 neither k=nk = n nor 00 is coprime to nn, so both ranges give the same count. It is [[irreducible in ℚ[x]\Q[x]|0SK1VHZ5R2]].
  2. So the minimal polynomial of ζ\zeta, which divides Φn\Phi_n, equals Φn\Phi_n. By the degree of a simple extension, [ℚ(ζ):ℚ]=φ(n)[\Q(\zeta) : \Q] = \varphi(n).

Necessity.

  1. Suppose ζ∈𝒞\zeta \in \mathcal{C}. Then φ(n)\varphi(n) is a power of 22, say 2r2^r.
  2. Write n=2kp1a1⋯ptatn = 2^k p_1^{a_1} \cdots p_t^{a_t} with distinct odd primes pjp_j and aj≥1a_j \ge 1. By multiplicativity and the totient of a prime power, φ(n)=φ(2k)∏jpjaj−1(pj−1)\varphi(n) = \varphi(2^k) \prod_j p_j^{a_j - 1} (p_j - 1).
  3. Each factor pjaj−1(pj−1)p_j^{a_j - 1}(p_j - 1) divides 2r2^r, so it is a power of 22. Since pjp_j is odd, aj=1a_j = 1, and pj−1=2mjp_j - 1 = 2^{m_j} with mj≥1m_j \ge 1.
  4. Suppose that m=mjm = m_j has an odd divisor b>1b > 1, say m=cbm = cb. The polynomial yb+1y^b + 1 has the root −1-1, as bb is odd. By the factor theorem, y+1y + 1 divides yb+1y^b + 1 in ℤ[y]\Z[y].
  5. Putting y=2cy = 2^c shows that 2c+12^c + 1 divides 2m+1=pj2^m + 1 = p_j. Since c<mc < m, we have 1<2c+1<pj1 < 2^c + 1 < p_j, which contradicts the primality of pjp_j.
  6. So mj=2sjm_j = 2^{s_j} for some sj≥0s_j \ge 0, and pj=22sj+1p_j = 2^{2^{s_j}} + 1 is a Fermat prime. Hence n=2kp1⋯ptn = 2^k p_1 \cdots p_t with distinct Fermat primes pjp_j.

Sufficiency.

  1. Let n=2kp1⋯ptn = 2^k p_1 \cdots p_t with distinct Fermat primes pj=22sj+1p_j = 2^{2^{s_j}} + 1. By step 4, φ(n)=φ(2k)∏j22sj\varphi(n) = \varphi(2^k) \prod_j 2^{2^{s_j}}, where φ(2k)=2k−1\varphi(2^k) = 2^{k-1} for k≥1k \ge 1 and φ(1)=1\varphi(1) = 1. So φ(n)=2r\varphi(n) = 2^r for some r≥0r \ge 0.
  2. Let L=ℚ(ζ)⊆ℂL = \Q(\zeta) \subseteq \C and f=xn−1f = x^n - 1. By the description of the roots of unity, every root of ff is ζk\zeta^k for some kk with 0≤k<n0 \le k < n.
  3. There are exactly nn roots of ff, so ζ0,…,ζn−1\zeta^0, \dots, \zeta^{n-1} are distinct. Then f−∏k(x−ζk)f - \prod_k (x - \zeta^k) has degree less than nn and nn roots, so it is 00 by the root bound.
  4. So ff splits over LL, and L=ℚ(ζ0,…,ζn−1)L = \Q(\zeta^0, \dots, \zeta^{n-1}). Thus LL is a splitting field of ff over ℚ\Q.
  5. Also (−1)⋅f+(x/n)⋅f′=1(-1) \cdot f + (x/n) \cdot f' = 1, so ff is separable. By the characterisation of Galois extensions, L/ℚL/\Q is a finite Galois extension.
  6. Let G=Gal⁡(L/ℚ)G = \operatorname{Gal}(L/\Q), the Galois group. By the order of the Galois group and step 2, ∣G∣=φ(n)=2r\abs{G} = \varphi(n) = 2^r.
  7. For σ∈G\sigma \in G, σ(ζ)n=1\sigma(\zeta)^n = 1, so σ(ζ)=ζaσ\sigma(\zeta) = \zeta^{a_\sigma} for some integer aσa_\sigma by step 10. Since [[every element of LL is h(ζ)h(\zeta) with h∈ℚ[x]h \in \Q[x]|0ZN39QDTP9]], σ\sigma is determined by σ(ζ)\sigma(\zeta).
  8. For σ,τ∈G\sigma, \tau \in G, we get στ(ζ)=ζaσaτ=τσ(ζ)\sigma\tau(\zeta) = \zeta^{a_\sigma a_\tau} = \tau\sigma(\zeta). So στ=τσ\sigma\tau = \tau\sigma, and GG is abelian.
  9. Claim: for every r≥0r \ge 0, every subfield E⊆ℂE \subseteq \C such that E/ℚE/\Q is a finite Galois extension with abelian Galois group of order 2r2^r satisfies E⊆𝒞E \subseteq \mathcal{C}. We prove it by induction on rr.
  10. If r=0r = 0, then [E:ℚ]=1[E : \Q] = 1 by the order of the Galois group. So E=ℚE = \Q, and ℚ⊆𝒞\Q \subseteq \mathcal{C} because the subfield 𝒞\mathcal{C} contains 11.
  11. Let r≥1r \ge 1 and GE=Gal⁡(E/ℚ)G_E = \operatorname{Gal}(E/\Q). By Cauchy's theorem, some h∈GEh \in G_E has order 22. As GEG_E is abelian, H={1,h}H = \set{1, h} is a normal subgroup.
  12. Let E′=EHE' = E^H be the fixed field. By the fundamental theorem of Galois theory, [E:E′]=2[E : E'] = 2, E′/ℚE'/\Q is normal, and Gal⁡(E′/ℚ)≅GE/H\operatorname{Gal}(E'/\Q) \cong G_E/H, which is abelian of order 2r−12^{r-1}.
  13. Every element of E′E' lies in EE, so it is separable over ℚ\Q. So E′/ℚE'/\Q is a finite Galois extension, and E′⊆𝒞E' \subseteq \mathcal{C} by the induction hypothesis.
  14. Choose a∈Ea \in E with a∉E′a \notin E'. By the tower law, [E′(a):E′][E'(a) : E'] divides [E:E′]=2[E : E'] = 2 and is not 11. So [E′(a):E′]=2[E'(a) : E'] = 2, [E:E′(a)]=1[E : E'(a)] = 1, and E=E′(a)E = E'(a).
  15. Let x2+bx+cx^2 + bx + c be the minimal polynomial of aa over E′E', and let s=2a+bs = 2a + b. Then s2=4(a2+ab)+b2=b2−4c∈E′⊆𝒞s^2 = 4(a^2 + ab) + b^2 = b^2 - 4c \in E' \subseteq \mathcal{C}.
  16. So s∈𝒞s \in \mathcal{C}, as a square root of an element of 𝒞\mathcal{C}. Then a=(s−b)/2∈𝒞a = (s - b)/2 \in \mathcal{C}, since 𝒞\mathcal{C} is a field containing bb.
  17. By the degree of a simple extension, every element of E=E′(a)E = E'(a) is u+vau + va with u,v∈E′u, v \in E'. All of these lie in 𝒞\mathcal{C}. So E⊆𝒞E \subseteq \mathcal{C}, which proves the claim.
  18. By steps 13, 14 and 16, the claim applies to E=LE = L. So ζ∈𝒞\zeta \in \mathcal{C}, and the regular nn-gon is constructible.

Idea

The regular nn-gon is constructible exactly when ζ=e2πi/n\zeta = e^{2\pi i/n} is, and ζ\zeta has degree φ(n)\varphi(n) over ℚ\Q. If ζ\zeta is constructible, φ(n)\varphi(n) is a power of 22, which forces the stated form of nn. Conversely, if φ(n)\varphi(n) is a power of 22, the abelian Galois group of ℚ(ζ)\Q(\zeta) can be cut down by subgroups of order 22. This gives a chain of quadratic extensions, and each quadratic step only needs a square root.

Where this leads

1 page builds on this one.

Further reading

  1. Web https://en.wikipedia.org/wiki/Constructible_polygon, Conditions for constructibility
About this page