Motivation
The prime-sided polygon theorem uses a primitive root of unity and a quadratic Galois tower. Composite numbers of sides require one additional ingredient: the degree of a primitive th root over . Cyclotomic polynomials provide that degree. We will prove their irreducibility, identify the Galois group, and derive the full numerical criterion for a regular polygon.
Definition 8.2.1
(Primitive roots of unity and cyclotomic polynomials)
For , a primitive th root of unity is a complex number of multiplicative order exactly . Define
There are factors, where counts residues coprime to . The formula initially defines a complex polynomial; proving its integer coefficients and irreducibility is part of the work below. We use the fundamental theorem of arithmetic, Gauss's lemma and the finite Galois correspondence as background. The companion construction unit proves the connection between quadratic towers and straightedge-and-compass constructions.
Why the coefficients are integers
Every th root of unity has an order dividing , so the roots of split into disjoint groups of primitive th roots. Hence
in . This identity also proves integrality by induction. The base is . If all for proper divisors are monic integer polynomials, their product is monic in . Divide by that product using monic polynomial division in . The complex identity says the remainder is zero, so the quotient has integer coefficients. Its degree is by the number of primitive roots.
Worked example 8.2.2
(Cyclotomic polynomials for six and nine)
The proper divisors of six give
Using , and gives . Similarly, , so . The degree of the latter is .
A Möbius product formula
The factorization by exact orders can be inverted. Define , let when a prime square divides , and otherwise let for a product of distinct primes. As in the finite-field counting unit, is one for and zero for .
Proposition 8.2.3
(Cyclotomic Möbius product)
In the rational function field ,
Negative exponents in the displayed product mean reciprocals; the result is the polynomial .
Proof
(Proof)
Substitute into the right side and collect the exponent of each . For that exponent is
after setting . It is one for and zero for every proper divisor . Thus all factors cancel except . This calculation takes place in , where every nonzero polynomial is invertible; it does not assert that each individual reciprocal is a polynomial.
Worked example 8.2.4
(Computing the twelfth cyclotomic polynomial)
For , the nonzero Möbius terms give
The final equality follows by factoring and , then canceling . Its degree is four, consistent with .
Irreducibility by reducing modulo a prime
Theorem 8.2.5
(Cyclotomic irreducibility)
For every , the polynomial is irreducible over .
Proof
(Proof)
The case is immediate. Let , let be a primitive th root, and let be its monic minimal polynomial over . It has integer coefficients by Gauss's lemma because it is a monic factor of . We first prove that if is any primitive th root satisfying and is a prime not dividing , then .
Otherwise let be the monic minimal polynomial of . It is a different irreducible factor of . Since , the polynomial divides over and hence over by monic division. Reduce coefficients modulo . In the Frobenius identity gives
Thus the nonconstant monic polynomial divides . Choose an irreducible factor of ; unique factorization over makes divide as well. Because and are distinct monic factors of over , their product divides , so divides after reduction.
This is impossible: in characteristic , the derivative of is , which is nonzero because and is relatively prime to . The polynomial is squarefree. Hence as claimed.
For any integer coprime to , every prime factor of is also coprime to . Repeatedly applying the claim shows that . Thus has all primitive th roots as roots, so . On the other hand divides the monic polynomial of degree . Therefore , proving irreducibility.
Proof X-ray
Three moves in the irreducibility proof
The proof never computes . It shows that the minimal polynomial of already has roots, and it does so in three moves.
First, a reduction to one prime at a time. Every exponent coprime to is a product of primes , so it is enough to show that the root set of inside the primitive roots is closed under for each such prime.
Second, integrality. Gauss's lemma puts and in , and monic division keeps the divisibility inside . Without this, reduction modulo would not be defined.
Third, Frobenius. In every coefficient satisfies , and the th power map is additive, so . This converts the statement that is a root of into the statement that and share an irreducible factor. The argument closes because divides in , so a shared factor becomes a repeated factor of modulo , which the derivative test forbids exactly when .
The hypothesis is used only at the squarefree step. The proof also uses distinct minimal polynomials to ensure that their product divides before reduction. Reducing two distinct polynomials modulo a prime can make them share a factor; that possible collision is exactly what creates the contradiction here.
Counterexample mode
The prime step fails for a prime dividing the order
Claim: if is the minimal polynomial of a primitive th root , then for every prime . Take , , and . Then and , so the claim is false.
The proof breaks where expected. The minimal polynomial of is , and modulo two , so and do share the factor . There is no contradiction, because modulo two: the derivative is zero there, and is not squarefree. The correct statement restricts to primes , and these suffice because every exponent coprime to is a product of such primes.
Worked example 8.2.6
(A cyclotomic degree)
A primitive ninth root has minimal polynomial , so . The polynomial has degree nine and also contains roots of orders one and three; its degree is not the degree of a primitive ninth root. A primitive sixth root instead has degree .
The cyclotomic Galois group
Theorem 8.2.7
(Automorphisms of a cyclotomic field)
For a primitive th root , restriction to gives a group isomorphism
Proof
(Proof)
The field contains every th root , so it is the splitting field of and is Galois over . An automorphism preserves multiplicative order, so it sends to with . It is determined by this image. The assignment to is an injective homomorphism into ; composition multiplies exponents.
Cyclotomic irreducibility gives , and the finite-Galois automorphism-count theorem gives the same number of automorphisms. The target also has elements, so the injection is an isomorphism. The target is abelian even when it is not cyclic; the proof of the polygon criterion below does not require cyclicity.
Worked example 8.2.8
(The group for eight sides)
The units modulo eight are . Each nonidentity unit squares to one modulo eight, so the Galois group of is the Klein four group, not a cyclic group of order four. Its order is nevertheless a power of two, which is enough to build a chain of quadratic fixed fields. This example prevents the false generalization that every cyclotomic Galois group is cyclic.
Subfields and the real cyclotomic field
If , then is a primitive th root. Choosing roots compatibly gives , and the tower law yields
The ratio is an integer because it is an extension degree. Proper divisibility does not always make the field inclusion strict: .
Proposition 8.2.9
(The maximal real cyclotomic subfield)
For , the elements of fixed by complex conjugation form , and this field has degree over .
Proof
(Proof)
Put . It is real and lies in the cyclotomic field. The element satisfies over . It is nonreal for , while , so its degree over is exactly two. Complex conjugation is a nonidentity automorphism of the cyclotomic field fixing . Its fixed field has index two by Artin's fixed-field theorem. The inclusion , with both outer indices two, forces equality. The tower law gives degree over .
Worked example 8.2.10
(A real subfield of twelfth roots)
For , the sum . Thus the real subfield is and has degree . The full cyclotomic field has degree four. This distinction is useful when a geometric construction needs only a real coordinate rather than a full complex root.
The full polygon criterion
Definition 8.2.11
(Fermat prime)
A Fermat prime is a prime of the form for an integer ; examples include and .
We use only primality of the particular factors under discussion, not a claim that every number of this form is prime.
Theorem 8.2.12
(Gauss–Wantzel criterion)
For , a regular -gon is constructible by straightedge and compass from a unit segment if and only if
where and the are distinct Fermat primes. The product may be empty.
Proof
(Proof)
Let be a primitive th root. The polygon is constructible exactly when is constructible as a complex coordinate. If it is constructible, its real and imaginary parts lie in a finite real quadratic tower. Adjoining gives a finite complex tower whose degree over is a power of two and which contains . The tower law therefore makes a power of two.
Conversely suppose . The cyclotomic Galois group is abelian of that order. A finite abelian group of order a power of two has a chain of subgroups with consecutive indices two: take a maximal proper subgroup, whose quotient is a nontrivial simple abelian group and hence cyclic of prime order, necessarily two, and repeat inside the subgroup. The finite Galois correspondence reverses this chain to a tower of quadratic field extensions from to . The companion quadratic-tower theorem constructs the real and imaginary parts of . Thus the polygon is constructible exactly when is a power of two.
Write with distinct odd primes . The standard totient formula, obtained by counting multiples of each prime and applying inclusion–exclusion, is
where and for . This is a power of two exactly when every and each is a power of two. If an odd prime has with an odd factor , write . Then is a proper divisor of , a contradiction. Hence is itself a power of two and is a Fermat prime. Conversely each stated Fermat prime has a power of two, so the displayed product has this property. This proves the criterion.
Worked example 8.2.13
(Fifteen and nine sides)
A regular -gon is constructible because is a product of distinct Fermat primes. A regular -gon is not: repeats an odd prime, and equivalently is not a power of two. A regular -gon is constructible because . The criterion concerns exact constructions from a unit segment; approximations are a separate question.
Common mistake
(Testing instead of )
For a prime the degree of a primitive th root is , so the prime-sided test asks whether is a power of two. For composite this test is wrong. The number has , yet and the regular -gon is not constructible. The number has , yet and the regular -gon is constructible. Always compute the totient, or use the factorization in the criterion.
Common mistake
(Degree and the group must be used together)
A power-of-two degree alone does not generally place an arbitrary algebraic number in a quadratic tower. Here the relevant extension is Galois with an abelian group of power-of-two order, which supplies a chain of index-two subgroups. That special group structure is the sufficiency argument; the numerical degree is the necessary test.
Counterexample mode
A power-of-two degree is not enough
Claim: an algebraic number of power-of-two degree over is constructible. Let be a root of . Modulo two, has no root and differs from , so it is irreducible; by Gauss's lemma so is , and . But each sum over a pairing of the roots satisfies the resolvent cubic , which has no rational root, so three divides the degree of the splitting field. A quadratic tower containing would have a Galois closure that is again a quadratic tower, of power-of-two degree, containing that splitting field. The repair: the Galois closure of must have power-of-two degree. A cyclotomic field is already Galois, so there the degree test suffices.
Counting the units in a concrete case
The totient formula can be checked directly by inclusion–exclusion. Among the integers , the proportion divisible by a prime is . Excluding multiples of every distinct prime divisor gives ; expanding the product records the alternating intersections of the divisibility conditions. For , this is . The polygon is constructible because has distinct Fermat-prime odd factors.
For , the same count gives . The factor three in is the obstruction: the primitive th root has degree , which cannot divide a power-of-two tower degree. This is a necessary-degree argument; no assumption about the cyclicity of the Galois group is needed for the negative conclusion.
Practice and solutions
Quick check
(Why does have integer coefficients even though it is first defined as a product of complex linear factors?)
Use the factorization of by exact orders.
Solution · Answer
Induct on in . The product of the proper-divisor factors is monic with integer coefficients, so monic polynomial division in gives an integer quotient, and that quotient is .
Quick check
(What condition on a prime makes squarefree modulo in the irreducibility proof?)
Look at the derivative.
Solution · Answer
The condition is . Then the derivative is nonzero modulo , its only irreducible factor is , and does not divide ; so has no repeated factor.
Quick check
(Determine the degree of a primitive ninth root of unity over .)
Solution · Answer
Its minimal polynomial is , irreducible by the theorem above, so the degree is .
Quick check
(Is the cyclotomic Galois group always cyclic?)
Solution · Answer
No. For the unit group modulo eight is , and every element squares to one, so it is the Klein four group. The Galois group of is therefore not cyclic.
Quick check
(Decide whether regular polygons with and sides are constructible.)
Solution · Answer
has the allowed form, and , so the regular dodecagon is constructible. repeats an odd prime, and is not a power of two, so the regular -gon is not.
Quick check
(Why does a prime of the form require to be a power of two?)
Look for an odd factor of .
Solution · Answer
If with odd , then divides and lies strictly between and , contradicting primality. Thus has no odd factor greater than one and is a power of two.