CYCLOTOMIC POLYNOMIALS

1. Introduction
Let ζ = e2πi/n . Then ζ, ζ 2 , . . . , ζ n are the n nth roots of unity. They form the vertices
of a regular n-gon in the complex plane. If (a, n) > 1 then ζ a is a root of unity of order
n/(a, n) < n, but if (a, n) = 1 then ζ is not a root of lower order, and in this case we call
ζ a a primitive nth root of unity. We define the nth cyclotomic polynomial Φn (x) to be the
monic polynomial of degree φ(n) whose roots are the primitive nth roots of unity:

Y
n
(1) Φn (x) = (x − ζ a ).
a=1
(a,n)=1

Our first observation concerning cyclotomic polynomials is that

Y
(2) zn − 1 = Φd (x).
d|n

To see this, it suffices to classify roots of unity ζ a according to the value of (a, n). Thus
we see that
Y
n
n
z −1= (x − ζ a )
a=1
Y Y
n
= (x − ζ a ).
d|n a=1
(a,n)=n/d

Write a = bn/d where (b, d) = 1 and 1 ≤ b ≤ d. Then the above is

Y Y
d
= (x − ζ bn/d )
d|n b=1
(b,d)=1
Y
= Φd (x).
d|n

Our next task is to show that Φn (x) has integral coefficients. To establish this, we
induct on n. We note that Φ1 (x) = x − 1 has integral coefficients. Suppose that Φd (x) has
integral coefficients for all d < n. Put
Y
Gn (x) = Φd (x).
d|n
d<n

Then by the inductive hypothesis, Gd (x) has integral coefficients. Suppose that F (x)
and G(x) are polynomials with integral coefficients. Then by the division algorithm
1
2

there is a quotient polynomial Q(x) and a remainder polynomial R(x) such that F (x) =
G(x)Q(x) + R(x) and deg R < deg G. In general, the coefficients of Q(x) and R(x) are
rational numbers, but if G(x) is monic then the coefficients of Q(x) and R(x) are inte-
gers. We apply this with F (x) = xn − 1, G(x) = Gn (x). Then R(x) is identically 0 and
Q(x) = Φn (x).
The identity (2) can be inverted by the Möbius inversion formula to yield a formula for
Φn (x) in terms of the polynomials xd − 1:
Y
(3) Φn (x) = (xd − 1)µ(n/d) .
d|n

Since
1
= 1 + xd + x2d + x3d + · · · ,
1 − xd
it is evident that the right hand side of (3) is a power series with integral coefficients. This
provides a second means of seeing that Φn (x) has integral coefficients.
The first few cyclotomic polynomials are as follows:

Φ1 (x) = x − 1,
Φ2 (x) = x + 1,
Φ3 (x) = x2 + x + 1,
Φ4 (x) = x2 + 1,
Φ5 (x) = x4 + x3 + x2 + x + 1,
Φ6 (x) = x2 − x + 1,
Φ7 (x) = x6 + x5 + x4 + x3 + x2 + x + 1,
Φ8 (x) = x4 + 1,
Φ9 (x) = x6 + x3 + 1,
Φ10 (x) = x4 − x3 + x2 − x + 1,
Φ11 (x) = x10 + x9 + x8 + x7 + x6 + x5 + x4 + x3 + x2 + x + 1,
Φ12 (x) = x4 − x2 + 1,
Φ13 (x) = x12 + x11 + x10 + x9 + x8 + x7 + x6 + x5 + x4 + x3 + x2 + x + 1,
Φ14 (x) = x6 − x5 + x4 − x3 + x2 − x + 1,
Φ15 (x) = x8 − x7 + x5 − x4 + x3 − x + 1,
Φ16 (x) = x8 + 1.

It is noteworthy that the coefficients of the above polynomials take only the values ±1 and
0. At one time this was conjectured to apply to all cyclotomic polynomials, but Φ105 (x)
provides a counter-example, and we now know that there are cyclotomic polynomials with
very large coefficients. This occurs particularly when n is highly composite.
 3

2. Primitive roots modulo p

Euler used cyclotomic polynomials to prove the existence of primitive roots mod p. Gauss
dismissed Euler’s argument as incomplete, because Euler did not properly develop the
material in the preceding section. To reconstruct how Euler might have reasoned, we take
n = p − 1 in (2) to see that
Y
xp−1 − 1 = Φd (x).
d|(p−1)

Since the left hand side has p − 1 roots modulo p, each Φd (x) on the right hand side must
have φ(d) roots (mod p), and moreover if d and e are distinct divisors of p − 1, then the
roots of Φd (x) (mod p) must be disjoint from the roots (mod p) of Φe (x).
Lemma 1. If a has order h modulo p, then Φh (a) ≡ 0 (mod p).
Proof. From (2) with n = h and x = a, we see that the left hand side is a multiple of p,
and hence p|Φd (a) for some d|h. From (2) again we see that Φd (x) divides (xd − 1) in Z[x],
and hence Φd (a)|(ad − 1). Hence ad ≡ 1 (mod p). By the minimality of h, it follows that
d = h, so the proof is complete.
Lemma 2. If Φd (a) ≡ 0 (mod p) and d|(p − 1), then a has order d.
Proof. Let h denote the order of a modulo p. Since Φd (a)|(ad −1), we have ad ≡ 1 (mod p),
and hence h|d. From Lemma 1 we also know that p|Φh (a). That is, Φd (x) and Φh (x) have
a common root a (mod p). This implies that d = h.

On combining these lemmas we see that if h|(p − 1), then the roots (mod p) of Φh (x)
are precisely the residue classes of order h. Thus there are exactly φ(h) residue classes of
order h. In particular, there are exactly p − 1 residue classes of order p − 1, which is to
say primitive roots.
Lemma 2 is false if the hypothesis d|(p − 1) is omitted. To see this, note that Φp (1) ≡ 0
(mod p), but 1 does not have order p modulo p; it has order 1. Nevertheless, Lemma 2
can be extended, as follows.
Lemma 3. If Φn (a) ≡ 0 (mod p), then a has order n (mod p), or p|n.
Proof. Let h denote the order of a modulo p. Since an ≡ 1 (mod p), it follows that h|n.
If h = n, then we are done. Suppose that h < n. Then Φh (x) and Φn (x) are distinct
factors of xn − 1. Moreover, Φh (x) has a factor x − a modulo p, and so does Φn (x). Hence
Fn (x) = xn − 1 has a factor (x − a)2 modulo p. Hence Fn0 (x) has a factor x − a modulo p.
But

(4) nFn (x) − xFn0 (x) = −n

is a polynomial identity. Substitute x = a. Since p|Fn (a) and p|Fn0 (a), it follows that p|n.

One consequence of Lemma 3 is that if n - (p − 1) and p - n, then the congruence

Φn (x) ≡ 0 (mod p) has no solution.
4

3. Primes ≡ 1 (mod m)
We prove that there are infinitely many prime numbers p ≡ 1 (mod m). This is a special
case of a theorem of Dirichlet, which asserts that if (a, m) = 1 then there exist infinitely
many prime numbers p ≡ a (mod m). Since the assertion is trivial for m = 1, we may
suppose that m ≥ 2. Suppose that p1 , p2 , . . . , pr are primes that are ≡ 1 (mod m). We
show that there is at least one more such prime. Set x = mp1 p2 · · · pr , and let p be a prime
factor of Φm (x). Note that x ≥ 2, so that each factor on the right hand side of (1) has
absolute value > 1. Hence |Φm (x)| > 1, which ensures that Φm (x) does indeed have at
least one prime factor. By Lemma 3 it follows that p|m or that x has order m modulo p.
But x ≡ 0 (mod m), which implies that Φm (x) ≡ Φm (0) ≡ 1 (mod m). Hence p - m, and
consequently x has order m modulo p. But then m must divide p − 1, which is to say that
p ≡ 1 (mod m).
To see how this works numerically, we note that
Φ8 (101) = 2 · 89 · 584609, Φ10 (101) = 11 · 9367291,
Φ8 (102) = 5857 · 18481, Φ10 (102) = 61 · 251 · 7001,
Φ8 (103) = 2 · 56275441, Φ10 (103) = 1171 · 95191,
Φ8 (104) = 17 · 1657 · 4153, Φ10 (104) = 5 · 211 · 109831.
On the left we encounter 2 and primes ≡ 1 (mod 8), and on the right we have 5 and primes
≡ 1 (mod 10).

4. Irreducibility of cyclotomic polynomials

In this section we show that the cyclotomic polynomial Φn (x) is irreducible over the field
Q of rational numbers.
Let p denote a given prime number. For any polynomial F (x) with integral coefficients
let F (x) be the polynomial whose coefficients are the residue classes (mod p) determined
by the coefficients of F (x). Thus the assertion F = G means that there is a polynomial
H(x) with integral coefficients such that F (x) = G(x) + pH(x).
Lemma 4. (Schönemann, 1846) Let A(x) be a monic polynomial with integral coefficients,
say
Y
n
n n−1
A(x) = x + an−1 x + · · · + a0 = (x − αi ).
i=1
Let p be a prime number, and put
Y
n
C(x) = (x − αpi ).
i=1

Then C = A.
Proof. Let σk (α) denote the k th symmetric function of the αi . When σk (α)p is expanded
by the multinomial theorem, all coefficients except the extreme ones are divisible by p.
That is,
σk (α1 , α2 , . . . , αn )p − σk (αp1 , αp2 , . . . , αpn )
p
 5

is a symmetric polynomial in the αi with integral coefficients, and hence by the symmetric
function theorem this quantity is a rational integer.
Lemma 5. Put F (x) = xn − 1. Then F is squarefree if and only if p - n.
Proof. By the identity (4) we see that if p - n, then (F , F 0 ) = 1, and hence that F is
p
squarefree. On the other hand, if p|n, say n = mp, then F = xm − 1 , and hence F is not
squarefree.
Let Φn (x) denote the nth cyclotomic polynomial. Since Φn |F , it follows from the above
that if p - n, then Φn is squarefree.
Theorem. (Kronecker, 1854) The polynomial Φn (x) is irreducible over the field Q of
rational numbers.
Proof. Suppose that A and B are monic polynomials with rational coefficients such that
Φn = AB, and suppose also that deg A > 0. By Gauss’s lemma we know that A and B
have integral coefficients. Let Z denote the roots of A. Let C be the monic polynomial
whose roots are the numbers ζ p for ζ ∈ Z. Here p is an arbitrary prime not dividing n.
Our first step is to show that A = C. Since the map ζ 7→ ζ p merely permutes the roots of
Φn , we know that C | Φn . Let G = (B, C). Then G | B and G | C. But A = C by Lemma 4,
2
and hence G | A B = Φn . But Φn is squarefree, by Lemma 5. Hence G = 1, so G = 1,
and consequently C | A. But C and A have the same degree, so in fact A = C.
Now let ζ be a root of A, and ζ 0 a root of Φn . Then there exists a positive integer a,
(a, n) = 1, such that ζ 0 = ζ a . We factor a, a = p1 p2 · · · pk . Since ζ is a root of A, it follows
from the argument above that ζ p1 is also a root of A. Then by a second application of
the above argument, we see that ζ p1 p2 is also a root of A. Continuing in this manner, we
deduce that ζ 0 is a root of A. Since this is valid for every root ζ 0 of Φn , we conclude that
A = Φn . Hence Φn is irreducible.
Gauss proved that Φp is irreducible. The first proof of Kronecker’s theorem using
Schönemann’s theorem was given by Arndt in 1857. An alternative argument using
Schönemann’s theorem is found in the text of Nagell. Among the more elementary proofs,
the one of Landau (1929) is remarkable for its brevity: only 8 lines. A detailed account of
the various proofs has been given by K. Manteuffel, Wiss. Zeit. Tech. Hochschule Magde-
burg 1 (1957), 69–75.