Lift finite-field cycles to a Galois group

Try

For f(X)=X³−2, reduction modulo 5 gives (X−3)(X²+3X+4). Multiply these factors. The quadratic discriminant is 9−16=3 modulo 5, a nonsquare because the squares are 0,1,4. Thus the factors are irreducible of degrees 1 and 2. Why should a calculation in characteristic 5 say anything about automorphisms of a field of characteristic zero?

The conclusion will be precise: the characteristic-zero Galois group contains a permutation fixing one root and exchanging the other two. It does not say that reduction identifies the whole field with a finite field, or that one factorization determines the entire group.

Reuse integral elements, finite-field Frobenius and Galois fixed fields. For the immediate prerequisite, work through factorization as Frobenius orbits. We need arithmetic closure of algebraic integers and the fact that a rational algebraic integer is an integer. We do not assume a theorem about the full integer ring of an arbitrary number field.

Predict

Let f∈ℤ[X] be monic of degree n≥1, with distinct roots α₁,…,αₙ in its splitting field L over ℚ. Put G=Gal(L/ℚ). Choose a prime p for which the reduction f̄ is squarefree. For monic f this is equivalent to p not dividing its nonzero discriminant. State these hypotheses before applying the result.

Predict which assertion is sufficient: for every i there is some σᵢ with the desired reduced root image, or there is one σ working for all i. Only the second gives a permutation matching the entire Frobenius action. Our proof keeps all root coordinates together using formal variables; choosing separate automorphisms would leave a gap.

Experiment

In the worked cubic, write a for the image of X in k=F₅[X]/(X²+3X+4). Then a²=2a+1. Compute successive powers before checking the table. The other quadratic root is 2−a; the linear root is 3.

Explain why the irreducible quadratic makes k a field and why none of these three roots coincide. The identity a³=2 is also an independent check on the factorization. Do not yet claim a lift merely because the table looks like a permutation.

Understand

Build a finite residue field from the roots

Set A=ℤ[α₁,…,αₙ]⊂L. Monicity makes every αᵢ integral. Reducing each power by f shows that products α₁^e₁⋯αₙ^eₙ with 0≤eᵢ<n span A as a ℤ-module. A finite generating list need not be a basis. In particular A/pA is a finite ring.

It is not the zero ring. Otherwise 1=pa for some a∈A, so a=1/p would be a rational algebraic integer, contradicting the rational-integrality lemma. Choose a maximal proper ideal of the finite nonzero ring A/pA: there is one because there are finitely many ideals and the zero ideal is proper. Pull it back to an ideal m of A. The quotient k=A/m is a finite field of characteristic p; write ρ:A→k for reduction.

Reducing f=∏(X−αᵢ) gives f̄=∏(X−ρ(αᵢ)) in k[X]. Since f̄ is squarefree, these n reduced roots are distinct. The map ρ is certainly not injective on A, but it is injective on this finite root set. Also k is generated by these roots over F_p, so it is a splitting field for f̄. Every element of G preserves A, because it permutes its generators.

Force a simultaneous lift with formal variables

Introduce independent formal variables T₁,…,Tₙ and Z. Define

P(Z,T)=∏ over σ∈G of (Z−Σᵢ Tᵢ σ(αᵢ)).

Every coefficient is an algebraic integer, being a finite sum of products of roots. Every element of G permutes the factors, so fixes every coefficient. The fixed-field theorem puts the coefficients in ℚ, and rational integrality then puts them in ℤ. Thus P∈ℤ[Z,T₁,…,Tₙ]. No averaging by |G| is used, so p dividing |G| causes no difficulty here.

The identity factor gives P(ΣᵢTᵢαᵢ,T)=0. Reduce this identity and apply Frobenius to its coefficients in k, keeping the formal T variables fixed. This coefficientwise operation is a ring automorphism of k[T₁,…,Tₙ]. It is not the operation that raises the entire polynomial to the pth power, which would also replace Tᵢ by Tᵢ^p. Since the coefficients of P̄ lie in F_p, the result is

P̄(ΣᵢTᵢρ(αᵢ)^p,T)=0.

Using the reduced product formula, this says

∏ over σ∈G of Σᵢ Tᵢ(ρ(αᵢ)^p−ρ(σ(αᵢ)))=0.

The polynomial ring over the field k is an integral domain. Therefore at least one whole factor is the zero polynomial. Comparing its coefficients gives a single σ∈G with

ρ(σ(αᵢ))=ρ(αᵢ)^p for every i.

This is a formal polynomial argument, not evaluation at all tuples from the finite field. A nonzero polynomial such as T^p−T can vanish at every element of F_p; that shortcut would be invalid.

Since integers and roots generate A, the equality extends to ρ∘σ=Frob∘ρ on A. Injectivity of Frobenius shows σ⁻¹(m)=m, hence σ(m)=m. So σ genuinely induces Frobenius on the residue field, not merely separate coincidences on roots. We have proved the needed lift without proving a general theorem about residue automorphisms of full integer rings.

Read the cycle lengths

For completeness, if h∈F_p[X] is monic irreducible and β is a root, let d be its Frobenius orbit length. The product B(X)=∏ from j=0 to d−1 of (X−β^(p^j)) has coefficients fixed by Frobenius, hence in F_p. Thus h divides B and deg h≤d. Conversely all d orbit members are distinct roots of h, so d≤deg h. The orbit length equals deg h. Squarefree factors partition all roots into these orbits.

The bijection αᵢ↦ρ(αᵢ) between the two root sets intertwines σ with Frobenius. Therefore factor degrees are exactly the cycle lengths of σ. In the worked cubic this supplies a transposition. At p=2, however, X³−2 reduces to X³ and all root images collide; the good-prime argument cannot read its multiplicity three as a three-cycle.

Apply

Change the polynomial to g(X)=X³−3 and the prime to 7. Check every cube in F₇, then predict a cycle type. Identify where monicity, irreducibility of the reduction, separability and the simultaneous lift enter your argument. Does the calculation alone identify the whole characteristic-zero Galois group?

Check the changed argument

The cubes modulo 7 are 0,1,6, so ḡ has no root. A cubic without a root is irreducible. Its derivative is 3X², which has no common root with ḡ because 0 is not a root of ḡ. The squarefree irreducible factor of degree three gives one Frobenius orbit and hence a lifted three-cycle. Both C₃ and S₃ contain such an element; this calculation alone does not distinguish them. A separate discriminant argument could finish the classification, but it is additional evidence.

Continue

Write the theorem as a chain of certificates: finite integral root algebra → nonzero finite residue field → distinct root images → one simultaneous automorphism → orbit degrees. Explain why no arrow may be replaced by a numerical observation. The choice of residue ideal and root labeling need not produce a unique named automorphism.

Return to discriminant and Galois arguments for parity information, and solvability by radicals only after the group has been justified. In particular S₄ is solvable; identifying it is not a nonsolvability proof. Continue to quartic Galois certificates to combine good-prime evidence with a complete subgroup argument. General ramification, a decomposition-group exact sequence and the assertion that every conjugacy class occurs at primes are not established here.