You are on page 1of 11

smo

Swiss Mathematical Olympiad
osm

Roots of Unity, Cyclotomic Polynomials

and Applications

The task to be done here is to give an introduction to the topics in the title. This paper is nei-
ther complete nor very rigorous concerning the proofs. It's a kind of brainstorming. Although
roots of unity are easy to dene, they are quite involved objects. Nevertheless I believe you
should know how to work with them, because they lead to powerful tools in number theory and
the theory of polynomials.

A complex number α is called a n-th root of unity if αn = 1. Equivalently we can say that the
n-th roots of unity are precisely the complex zeros (=roots) of the polynomial xn − 1. This
polynomial does not have multiple roots, therefore there are exactly n such numbers. It is
easily seen that the product of two n-th roots of unity (rou for short) is again an n-th rou, and
a simple check shows that the set of these is in fact a commutative group of order n, which
we denote by µn . Explicitly by de Moivres formulas the n-th rou are given by α = e2kπi/n ,
0 ≤ k ≤ n − 1, which means that the all have absolute value 1 and the argument is a multiple
of 2π/n. Therefore the are the vertices of a regular n-gon inscribed into the unit circle.

We dene the order of a rou α to be the smallest positive integer n, such that αn = 1. We call
a n-th rou primitive if its order is exactly n. There exists a primitive n-th rou, for example
α = e2πi/n . Because 1 ∈ µn is the neutral element of this group, the order of α in our sense
is the same as the order of α as an element of the group µn . Especially, the existence of a
primitive n-th rou implies that the group µn is cyclic, meaning that each element of µn can
be written as a power αk of a primitive n-th rou α. Another consequence is that the order of
α ∈ µn does not depend on n.
Lemma 0.1. Let α be a root of unity.
(a) Let d be the order of α and let k be an integer. The order of αk equals d/(k, d).
(b) If α is a primitive n-th rou, then αk is a primitive n-th rou if and only if (k, n) = 1.
There are exactly ϕ(n) primitive n-th rou (Euler ϕ-function)
(c) We have α ∈ µn if and only if the order d of α is a positive divisor of n.

Proof. We prove rst that αm = 1 if and only if d | m. Write m = ld + r with 0 ≤ r < d. Then
αm = (αd )l · αr = αr . Now by minimality of d we have αm = 1 if and only if r = 0.
(a) We may assume k > 0. Let e be the order of αk . Then (αk )e = 1 and e is minimal with
this property. Therefore e is minimal such that ke is divisible by d. Obviously we must have
e = d/(d, k).
(b) Is a direct consequence of (a) and the existence of a primitive n-th rou.
(c) This is just a reformulation of the result at the beginning of the proof because α ∈ µn ⇔

1

Let p(x) be a polynomial with arbitrary (complex) coecients. We give now a reinterpretation of the above proof in terms rou. We can make this correspondence more concrete in the following way. Now by identifying each β ∈ µϕ(n) with its image in (Z/nZ)∗ . In case there exists a primitive root mod n. We have to nd all n. Let a be one of them. Next we show how rou may used to compute the coecients of a polynomial knowing its values on the complex unit circle. But how can we reconstruct the coecients of p explicitly? We 2 . then p is completely determined by its values on d + 1 dierent points. If n = 3k then A = 1 + 1 + 1 6= 0. Choose a primitive root g mod n and a primitive ϕ(n)-th root of unity α and consider the map h : µϕ(n) → (Z/nZ)∗ . As an example. If n = 3k + 1 then A = a2 + a + 1 = 0. γ ∈ µϕ(n) . Example 1. Therefore the problem statement is equivalent to the fact that both complex roots of x2 + x + 1 are also roots of x2n + xn + 1.αn = 1. I hope that all of this looks familiar to you. the number of primitive ϕ(n)-th rou in µϕ(n) is the same as the number of primitive roots mod n. In addition we have h(β · γ) = h(β) · h(γ) for all β. We have to check whether or not A = a2n + an + 1 = 0 and distinguish three cases. We have a2 + a + 1 = (a3 − 1)/(a − 1). Find all natural numbers n such that the polynomial x2n + xn + 1 is divisible by x2 + x + 1 . they are always 3n-th rou. αk 7→ g k . so the question reduces to decide when they are even n-th rou. Solution. therefore a3 = 1 (a is a primitive third rou!). then a is also a root of (x − 1)(x2 + x + 1) = x3 − 1. Analogously we have a2n + an + 1 = (a3n − 1)/(an − 1). then the number of such roots equals ϕ(ϕ(n)). Obviously. it respects the mul- tiplications of both groups. so they correspond by the map h. the multiplications coincide and there is no dierence anymore between these two groups. By denition of g and α this map is well dened and bijective. An easy check shows that x2 + x + 1 has no multiple roots. Finally if n = 3k + 2 then A = a4 + a2 + 1 = a + a2 + 1 = 0. We have seen a lemma like this in the structural theory of the groups (Z/nZ)∗ where we also dened orders and obtained similar relations. It is well known that if the degree of p equals d. the latter is a cyclic group of order ϕ(n) and is structurally identical to µϕ(n) . because both are characterized as generators of the respective groups. We rst give an ad hoc solution requiring no knowledge on rou. The groups µm and (Z/nZ)∗ are both nite and commutative. By lemma 1 (c) this is the case if and only if 3 | n. in other words the two primitive third rou. so the roots of this polynomial are the 3n-th rou which are not n-th rou. As a consequence we get that if there exists a primitive root mod n. This is no coincidence. All we know about the rst is also true for the second and vice versa. As result we get that the division is possible if and only if n is not divisible by 3. such that the two primitive third rou are of this form. therefore the roots of this polynomial are precisely the third roots of unity which are not rst rou. Therefore h is a so called homomorphism.

then αm 6= 1 and by the formula for a geometric series we get n−1 n−1 X k m X m k (αm )n − 1 (α ) = (α ) = =0 k=0 k=0 αm − 1 This two calculations prove the lemma for the monomials p(x) = xm .. but will come back to this later. β∈µn k=0 Proof. 0 3 6 n Solution. . p[2] = 1 and p[n] = 0 for n ≥ 3. . For example p(x) = x2 − 2x + 3. + p[n] = 3 where α is a primitive third rou. maybe via factorizations etc. Let n be a multiple of 3. + .2. m ≥ 0. a kind of innite polynomial) such that the numbers we are looking for appear as coecients of the monomials xn . Example 2.. . The rst equality is a direct consequence of the fact that µn is cyclic and α is a generator of this group. Consider the Polynomial p(x) = (x + 1)n . We calculate the right hand side in another way. Now the argument is two sided.) . k=0 k=0 If m is not a multiple of n. We give an example which uses a special feature of third roots of unity. Lemma 0. The sum S in the problem is precisely p[0] + p[3] + . One the one hand the above lemma allows us to express certain combinations of these coecients in terms of roots of unity (more precisely in terms of the values of this function at certain roots of unity). + p[n]. The general case now follows from this and repeated application of the facts (p + q)[n] = p[n] + q[n] for any polynomials p. Of course we have p(1) = (1 + 1)n = 2n . Compute the sum         n n n n + + + . then p[0] = 3. q and (a · p)[n] = a(p[n]) for any polynomial p and any complex number a. . Let p be a polynomial and let α be a primitive n-th root of unity. Now dene a polynomial or more generally a power series (Taylor series. We may 3 . On the other hand it is often possible to compute these rou-expressions in another way. then we have n−1 X n−1 X (αk )m = 1 = n.  S = p[0] + p[3] + . Next we use a special feature on third roots of unity. p[1] = −2. Then we have X n−1 X p(β) = p(αk ) = n (p[0] + p[n] + p[2n] + . If m is a multiple of n. . So what? This has very important applications for the computation of so called generating functions.denote by p[n] the coecient of the term xn in p. . By the above lemma we have 1 p(1) + p(α) + p(α2 ) . The idea behind this concept is the following: Imagine we want to count something. Generating functions are one of the most powerful tools in combinatorics and often lead to short but a bit stupid solutions of counting problems. We will not discuss this in detail.

Prove that n Y |P Ai | = |P O|n − Rn . We have n Y n Y n Y n |P Ai | = |Rx − Rαi | = R |x − αi | i=1 i=1 i=1 . The coordinate of P is a real number. Let x > 1 be such that the coordinate of P is Rx. 3 There is an important geometric applications of roots of unity. such that 1 ∈ C is one of the vertices. Remember that multiplication with a complex number eiϕ of absolute value 1 describes a rotation of the complex plane with angle ϕ (counter- clockwise). More precisely. . Therefore. n are the n-th rou. i = 1. α1 = 1. then the points a + bα with α ∈ µn are the vertices of a regular n-gon with center a and circumradius |b|. They form a regular n-gon inscribed in the unit circle. √ √ assume that α = e2πi/3 = −1/2 + i 3/2. general regular n-gons may be described with roots of unity. On the half-line OA1 choose P such that A1 is between O and P . An be a regular n-gon inscribed in the circle of center O and radius R. b ∈ C are complex numbers. We identify the plane with C. (Putnam 55) Let A1 A2 . . . An analogous calculation gives p(1 + α2 ) = 1 if n is even and = −1 if n is odd. Obviously any regular n-gon may be described in this way. a + b is one of the n vertices. . Therefore. if a. Then α + 1 = 1/2 + i 3/2 = e2πi/6 is a primitive 6-th rou! So we get p(1 + α) = (e2πi/6 )n and this equals 1 if n is even and −1 if n is odd (remember that n is divisible by 3). Finally 2n + 2(−1)n S= . a being unique and b being unique only up to multiplication with an n-th rou. i=1 Solution. This is why many problems involving regular n-gons may be solved using rou. Example 3. . . After a translation and a rotation we may assume that the vertices of the n-gon have coordinates Ai = Rαi where αi .

n .

.

Y .

= Rn .

(x − αi ).

= Rn |xn − 1| = Rn (xn − 1) .

.

.

.

(USA 76) Let P. Q. Prove that P (1) = 0. i=1 = (Rx) − Rn = |P O|n − Rn . S be polynomials such that for all x P (x5 ) + xQ(x5 ) + x2 R(x5 ) = (x4 + x3 + x2 + x + 1)S(x). For which n do we have x2 + x + 1 | (x − 1)n − xn − 1? 2. n Exercises 1. R. 4 .

. Let f : R2 → R be a function. . . + en an . An in the plane f (A1 ) + f (A2 ) + . 5 . + f (An ) = 0. Let a1 . 5. Prove that f is the zero function. such that mn | e1 a1 + e2 a2 + . . such that for every regular n-gon A1 A2 . en . Finitely many points are given on the unit circle so that the product of the distances from any point on the circle to the given points does not exceed 2. Show that there are integers e1 . an be integers.3. a2 . (Romania 96) Let n > 2 be a natural number. . none of which is a multiple of mn−1 . 4. . . . . . . e2 . (Shortlist 02) Let m and n be integers > 1. Prove that the points are the vertices of a regular polygon. not all zero. . . . with |ei | < m for 1 ≤ i ≤ n.

We shall not prove this. (a) Φn (x) has integer coecients. + xp + 1. But (b) requires knowledge in Galois theory which is well beyond our reach. Now let p be a prime number. This is why they are so important. because the roots of xn − 1 are precisely the n-th rou and each of them has an order dividing n by lemma 1 (c) we get the factorization Y xn − 1 = Φd (x). We dene the n-th cyclotomic polynomial Φn (x) as follows: Y Φn (x) = (x − α). although (a) is not too dicult. To illustrate the method we will do some easy cases.3. + 1)(xq−1 + . let p. d>0 Funny. d>0 By the way. . d>0 in fact the full factorization in irreducible factors over Z respectively Q. q be dierent primes. . . Moreover. But rst we need a method to compute Φn (x) explicitly. so xpq − 1 Φpq = . α where the product is taken over all primitive n-th rou. the theorem implies that Y xn − 1 = Φd (x) d | n. The important point now is the following deep result: Theorem 0. x−1 (already in this simple case the irreducibility is quite hard to prove). (x − 1)(x + 1)(x2 + x + 1) Some more formulas are collected in the following lemma: 6 . . which is rather simple in fact. Taken together. . Let n be a positive integer. (x − 1)(xp−1 + . (xp − 1) As a last example. d | n. We have xpq − 1 = Φpq (x)Φp (x)Φq (x)Φ1 (x). First of all. so it lies in Z[X]. . Next we have for example xp − 1 = Φp2 (x)Φp (x)Φ1 (x) = Φp2 (x)(xp − 1) and 2 (xp )p − 1 Φp2 (x) = = xp(p−1) + xp(p−2) + . + x + 1. .Now we shall turn to the denition and properties of the cyclotomic polynomials. This has important consequences as we shall see. isn't it. of course Φ1 (x) = x − 1. (a) Φn (x) is irreducible over Z and therefore also over Q. + 1) for example x6 − 1 Φ6 = = x2 − x + 1. . d | n. These give a complete factorization of the polynomials xn − 1 in Z[X] as we shall see. By denition and lemma 1 (b) this is a normalized polynomial of degree ϕ(n). comparing the degrees on both sides of this equation we get the following formula: X ϕ(d) = n. We have xp − 1 = Φ1 (x)Φp (x) = (x − 1)Φp (x) and therefore p x −1 Φp (x) = = xp−1 + xp−2 + . Using the above formula we can compute them recursively on the number of prime factors of n.

xn − 1 d | 2n. We get n n n−1 pn k Y Y x = Φ1 (x) · Φpk (x) = (x − 1) Φp (xp ). (d) Φpn (x) = Φn (xp )/Φn (x) for p prime and p6 | n.4. the last expression equals Y Φ2k+1 d (x). ) for a prime factorization n = pr11 · · · prss with dierent primes r −1 (b) Φn (x) = Φp1 ···ps (x p11 ···prss −1 pi and positive exponents ri > 0. (a) Φpr (x) = Φp (xp ) for p prime and r > 0. For example we get     2n x −1  Y Y xn + 1 = = Φd (x) /  Φd (x) . (c) Φ2n (x) = Φn (−x) for n ≥ 3 odd. d>0 Now write n = 2k · m with m odd. k=1 k=0 concretely for p = 2 and p = 3 we obtain n−1 n−1 n k n k k Y Y x2 − 1 = (x − 1) (x2 + 1). x3 − 1 = (x − 1) (x2·3 + x3 + 1). d>0 7 . Combining (b) and (d) one can compute explicitly all cyclotomic polynomials. d | m. We list the rst twelve: 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 + x 3 + 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 Most important are the factorizations of polynomials of the form xp − 1 for a prime p. d>0 d | n. r−1 Lemma 0. k=0 k=0 Using cyclotomic polynomials we may also factor other polynomials which are related to xn − 1.

(m + 1. d>0 Y = Φd (x). d | n(m+1). d6 |n. d>0 d | n. such that 2n + 1 is divisible by n2 . . We have   m+1 x −1  Y Y 1 + x + x2 + . (IMO 90) Find all integers n > 1. we write n = 2k 3l m with m odd and not divisible by 3:     3n x +1  Y Y x2n − xn + 1 = = Φ2k+1 d (x) /  Φ2k+1 d (x) xn + 1 d | 3l+1 m. We want to prove that k = 1. Write n = 3k m with 36 | m. n) of positive integers. . . + xm | 1 + xn + x2n + . the rst polynomial divides the second if and only if none of the divisors d > 1 of m + 1 is also a divisor of n (else there would be a factor of the former polynomial that would not occur in the factorization of the latter). 8 . This gives 2k ≤ k + 1 and k = 1. d>0 We will give an application: Example 4. On the other hand this must be divisible by n2 and therefore by 32k . . Now proceed as in the rst step to prove that in fact n = 3. Solution. . they do not share any common complex roots). We already proved at the last meeting that the smallest prime divisor of n equals 3. + xmn = = Φd (x) /  Φd (x) xn −1 d | n(m+1).Proceeding further. Solution. + xm = = Φd (x) /Φ1 (x) = Φd (x). d | m. n) = 1. d>0 Y = Φ2k+1 3l+1 d (x). (USA 77) Find all pairs (m. d>0 d | 3l m. We set y = 2m and get from the above factorization k 3k Y n 2 +1 = y + 1 = Φ2 (y) · Φ2·3l (y) l=1 k−1 k−1 2·3l 3l l l Y Y m = (y + 1) · (y − y + 1) = (2 + 1) · (22·3 m − 23 m + 1). l=0 l=0 Now its easy to check that each of the factors on the right hand side is divisible by 3 but not by 9 (use the fact that m is not divisible by 3). . . Example 5. x−1 d | m+1. So 3k+1 is the largest power of 3 dividing 2n + 1. But this in turn is true if and only if m + 1 and n are coprime. . d>1 And analogously     n(m+1) x −1  Y Y 1 + xn + x2n + . d>0 Because dierent cyclotomic polynomials are coprime (by denition. + xmn . such that 1 + x + x2 + . d>0 d | m+1.

Exercises 1. Prove that for every positive integer n. Prove the formulas in Lemma 4 by similar arguments as in the examples. . (Shortlist 02) Let p1 . the integer pn (2) has at n+1 n least n dierent prime factors. The following is quite a deep result which you may use (but hardly prove) here: If r and s are positive integers. Remark : First prove by elementary means that it has at least 2n dierent divisors. A fancy approach via cyclotomic polynomials is also possible. With this approach you will get a far better bound for the number of divisors! 9 . Prove that 2p1 p2 ···pn + 1 has at least 4n dierent positive divisors. Give the complete factorizations of the polynomials pn (x) = x2 + x2 + 1 and qn (x) = n+1 n x2 − x2 + 1 over Z. pn be distinct primes > 3. This is easy by induction. . Use induction. Then try to rene your argumentation to get 4n . . . 2. s is divisible by the other and the quotient is a prime power. then Φr (2) and Φs (2) are coprime unless one of r. 3.

Describe a regular n-gon with rou as explained in the text. en ) with ei ≥ 0 (compare IMO 86 Nr 3. But it is not easy at all to rule out this case. zn have absolute value 1. . Now use the hypothesis (and lemma 2 if you want) to obtain precise information about the coecients of this polynomial. Use a standard box-principle argument for sequences (e1 . 10 . Instead consider the polynomial n Y p(x) = (1 + xai + x2ai + . . . Some points may be counted more than once! 4. as in the solution to example 1. There is a unique case for which this argument breaks down (as in the latter two problems in the bracket). . maybe you should perform a rotation rst. A simple double counting argument is not enough. z ∈ C arbitrary. Oh. Identify R2 with C. . Use complex numbers. Obtain 4 dierent equations in this way and try to manipulate them to get P (1). Use third roots of unity. i=1 If multiplied out. . The given points z1 . This argument is somehow similar to this generating function thing in example 2. Then consider the sum aof all f (a) where a runs through all vertices of all n-gons constructed this way. so you can prescribe the value of z1 · zn (with the obvious restriction that this number has absolute value 1 of course). IMO 01 Nr 4 or SMO 03 Nr 8). If x is a fth root of unity. Derive a contradiction. . + x(m−1)ai ). . 2. then the right hand side vanishes. 5. consider an n-gon with center z and consider the n translations of it having the vertices of the rst n-gon as centers. . what are the exponents? What do you know additionally about them? Use this to obtain another expression for p where the exponents are determined modulo mn . 3.Hints for the Exercises on page 5 1. For an arbitrary point z in C the product of the distances to all zi is a polynomial function of z . Now to get an equation for f (z). .

This is a bit tedious to be done directly. which is in fact true much more generally: (a2 + ab + b2 )(a2 − ab + b2 ) = a4 + a2 b2 + b4 .Hints for the Exercises on page 9 1. it would be enough to prove that the factors are pairwise coprime. 3. (Shortlist 02) Use an appropriate factorization as in the last exercise to prove that in- creasing n by 1 produces a new prime divisor. 2. the two factors on the left hand side are coprime. This already gives the bound 2n . For the rst part: express this polynomials in terms of polynomials of the form xk − 1 as in the examples given in the text. Specialized to our specic situation. Instead prove and use the following additional observation: If n > m then the number mn has at least twice as many divisors than m. But this does not seem to be easy at all. This can be done directly using the denitions. To get the 4n one could try to prove that at least 2 new prime factors are produced when increasing n by 1. Instead use the building rule for the factorization. 11 . For the second part: Knowing the factorization. Use induction on the number of prime factors of n in all cases except for (c).