diendantoanhoc.

net [VMF]

Dirichlet0s Theorem

Daniel Harrer (ZetaX)

24. Februar 2011

Symbols and used theorems

Z: the integers.
N: the set {1, 2, 3, ...} of positive integers.
N0 : the set {0, 1, 2, ...} of non-negative integers.
P: the primes in N.
Z/pZ or Fp : the (field of) residues mod p, p prime.
A sums and products of 0 numbers are always set to 0 respectively 1.

Theorem 1. (Unique factorisation)

For all n ∈ N there are (up to reordering) uniquely determined primes q1 , q2 , ..., qk such that
n = q1 q2 ...qk .

Theorem 2. (Binomial theorem)

n  
k n−k n
X
n
(a + b) = a b for all a, b ∈ C.
k=0
k

Preface
A lot of primes
A very fundamental result is

Theorem 3. There are infinitely many primes (in N).

There are a lot of proofs for this theorem, but probably the oldest and most famous one is:

1
 diendantoanhoc.net [VMF]

Proof. (Euklid) Assume that there are only finitely many primes, call them p1 , p2 , ..., pn . Consider
their product P = p1 · p2 · ... · pn . Since P + 1 > 1, there is (using theorem 1) at least one prime q
dividing P + 1. But this q is different from all primes pi because q and P are coprime, so q was
not in the initial set, a contradiction.
The proof above was known thousands of years ago. But in the 18th century Euler showed that
there is a much stronger result:
X1
Theorem 4. The sum diverges, in other words: grows to ∞.
p∈P
p

Proof. Every k ∈ N can be written as k = t · s2 with t not divisible by a square > 1. This gives
the inequality
n   n
X 1 Y 1 X 1
≤ 1+ .
k=1
k p∈P
p s=1 s2
p≤n

Since s12 ≤ s(s−1)

1 1
− 1s for s ≥ 2 we get ns=1 s12 ≤ 1 + ns=2 s−1
1 1 1
P P
= s−1 − s
=2− n
≤ 2. Together
x
with the easy to verify property 1 + x ≤ e for all x ∈ R this yields
n Y  n Y 
X 1 1 X 1 1 Y 1 P 1
≤ 1+ 2
≤ 2 1 + ≤ 2 · e p = 2 · e p

k=1
k p∈P
p s=1
s p∈P
p p∈P
p≤n p≤n p≤n

where the last sum also runs over all primes p ≤ n. To show that p∈P p1 diverges, it now suffices
P

to show that ∞ 1
P
k=1 k diverges. But the latter one is a well known property, shown by the
following since for all n ≥ 2m we have:
m −1 s+1
−1 s+1
n
X 1
2X m−1 2
1 X X 1 X 2 X−1 1
m−1 m−1
X1 m
≥ = ≥ s+1
= = ,
k=1
k k=1
k s=0 k=2s
k s=0 k=2s
2 s=0
2 2

giving the divergence because m can be chosen arbitrary large.

Dirichlet’s Theorem
It is a very natural question to ask if a given sequence contains infinitely many primes or not.
One of the easiest cases seems to be an arithmetic sequence a, a + m, a + 2m, a + 3m, .... In other
words, it is asked for a lot of primes p ≡ a mod m.
If d > 1 is a common divisor of a and m, then all terms of this sequence are divisible by d, thus
there can be only a finite set of primes in the sequence; in fact, the only prime can be d itself.
But for gcd(a, m) = 1, Dirichlet was able to prove that there are a lot of primes by giving a much
stronger result:

2
 diendantoanhoc.net [VMF]

Theorem 5. Let gcd(a, m) = 1, then there are infinitely many primes ≡ a mod m. More
X 1
exactly, the sum diverges and the primes are ”equally distributed” into the different
p∈P
p
p≡a mod m
residues a coprime to m.

All known proofs for this(these) theorem(s) require a lot of real or complex analysis, especially
concerning the so-called L-series L(s, χ) = ∞ χ(n)
P
n=1 ns with χ : N → C some function (for those
who know the term: χ is a ”character” from (Z/nZ)∗ to C here).
It’s an interesting question whether there exist more elementary proofs for special cases, possibly
based on the ideas of Euklid’s proof of theorem 3 (note that Dirichlet’s idea is more related to
Euler’s one).
Our goal is now to prove it for the cases a = 1 and a = −1 and arbitrary m.

Special cases
Let’s try some very special m, namely m = 4 and m = 3.

Theorem 6. There are ∞ many primes p with:

a) p ≡ −1 mod 4.

b) p ≡ −1 mod 3.

c) p 6≡ 1 mod m for any fixed m ∈ N in general.

Proof. It’s clear that c) implies a) and b) since there are only the residue classes 1 and −1
mod 3, 4, so we just need to prove this one.
Assume like before that there are only finitely many primes p1 , p2 , ..., pn 6≡ 1 mod m. Then let P
be their product and consider the number mP − 1: All its prime divisors q1 , q2 , ..., qk are ≡ 1
mod m because they are coprime to P , thus different from the initial primes pi . But
1 = 1 · 1 · ... · 1 ≡ q1 · q2 · ... · qk = mP − 1 ≡ −1 mod m, being impossible for m ≥ 3. So we got
our contradiction again.
Another idea has to be used to attack ≡ 1 mod m, which we will do first for m = 4:

Theorem 7. There are ∞ many primes ≡ 1 mod 4.

Before we can tackle this one, we need the following

Lemma 1. Let p be a prime dividing x2 + 1 for some x ∈ Z. Then p = 2 or p ≡ 1 mod 4.

3
 diendantoanhoc.net [VMF]

Proof. Assuming that p ≡ −1 mod 4, so that p−1

2
is odd, we want to bring p|x2 + 1 to something
2
absurd. We know that x ≡ −1 mod p, so we get using Fermat’s little theorem:

p−1 p−1
1 ≡ xp−1 ≡ x2 2 ≡ (−1) 2 ≡ −1 mod p,
our so much desired contradiction.
Now back to theorem 7:
Proof. Yes, it gets boring, but for the sake of proving the theorem assume that there are only
finitly many primes p1 , p2 , ..., pn ≡ 1 mod 4 and let P be their product. Then consider any prime
divisor q of (2P )2 + 1: q is clearly odd, coprime to P , and ≡ 1 mod 4 by the Lemma, done.

Requirements
Before we can give a general proof like that of theorem 7, we need some more stuff.
Definition 1. Let p be prime and p - a ∈ Z. Then the smallest k ∈ N with ak ≡ 1 mod p is
called the order of a mod p and is denoted by ordp (a) (note that the order always exists since
ap−1 ≡ 1 mod m by Fermat’s little theorem).
This definition can still be made if p is any integer coprime to a, but we will need only the case p
prime.
A very powerful principle with striking simplicity is the next
Lemma 2. (Order lemma mod p)
Take a, p as in the above definition and let k be given such that ak ≡ 1 mod p. Then ordp (a)|k.
This holds also for any p, but will, as said before, not be required.
Proof. Take division with remainder to write k = q · ordp (a) + r, 0 ≤ r < ordp (a). By the
definitions we get ar ≡ ak · a−q·ordp (a) ≡ 1 · 1−q ≡ 1 mod p. We can’t have r 6= 0 since then
0 < r < ordp (a) and ar ≡ 1 mod p, contradicting that ordp (a) is the minimal positive integer
with aordp (a) ≡ 1 mod p. So r = 0 and k = q · ordp (a), proving the lemma.
A very useful type of polynomial, closely related to orders, is now given by
2·πi
Definition 2. Set ζn = e n . Then the n-th cyclotomic polynomial Φn (x) is defined by
n
Y
Φn (x) = (x − ζnk ).
k=1
gcd(k,n)=1

Theorem 8. The cyclotomic polynomials Φn (x) fulfill the fundamental property

Y
xn − 1 = Φd (x).
d|n

4
 diendantoanhoc.net [VMF]

Proof. Since both sides are monic polynomials (their leading coefficient is 1), it suffices to show
2·π·k
that they have the same complex roots. The roots of xn − 1 are e n for k = 0, 1, 2, ..., n − 1,
2·π·k
thus let ζ = e n be a root of that polynomial.
2·π·k 2·π·dk0 2·π·k0
Let d = gcd(k, n) and n = d · n0 , k = d · k 0 . By that we have ζ = e n = e dn0 = e n0 and
gcd(k 0 , n0 ) = 1, thus ζ being a root of Φn0 (x) by definition. This shows that any root of the left
hand side (LHS from now) is one of the right hand side (RHS from now).
Now let ζ be a root of the RHS, let’s say Φn0 (ζ) = 0 for n0 |n. By this n = d · n0 for some integer d
0
d
and because of that ζ n = ζ n = 1d = 1, so it is a root of the LHS.
Theorem 9. The coefficients of the Φn (x) are integers.
Proof. We will use induction on n to show that Φn (x) is a polynomial with integers as coefficients.
Clearly Φ1 (x) = x − 1, so it is as nice as we want it to be. Thus Q assume the theorem to be proven
for all m < n, so especially for the divisors d 6= n of n. By that d|n,d6=n Φd (x) is a monic
polynomial with integer coefficients.
xn − 1
By theorem 8 we have Φn (x) = Q , and by making standard division of polynomials,
d|n,d6=n Φd (x)
we see that Φn (x) has indeed integers as coefficients.

Primes ≡ 1 mod n
Now the time has come to prove the infinity of primes ≡ 1 mod m for any m.
We will use cyclotomic polynomials for this, so let’s start collecting their properties mod p:
Lemma 3. Let n > 0 and a be integers, p prime, and g(x) a polynomial with integer coefficients.
If xn − 1 ≡ (x − a)2 · g(x) mod p, then p|n (polynomials mod p are archieved and handled by
reducing all coefficients mod p).
Proof. Set y = x − a, then we have (y + a)n − 1 ≡ y 2 · g(y + a) mod p. Expand the LHS by the
binomial theorem to be y n + ... + n · an−1 y + (an − 1). From the factor y 2 on the RHS we get that
the constant and linear coefficient are 0 mod p, thus an ≡ 1 mod p and n · an−1 ≡ 0 mod p. So
we get n · 1 ≡ n · an ≡ (n · an−1 ) · a ≡ 0 · a ≡ 0 mod p.
This means nothing else than p|n.
Continuing with a way to construct primes we want:
Theorem 10. (Main theorem for cyclotomic polynomials mod p)
If p is a prime divisor of Φn (a) with n ∈ N and a ∈ Z, then p|n or p ≡ 1 mod n.
Proof. Let o = ordp (a) (it exists because p|Φm (a) ⇒ p|am − 1 ⇒ p - am ⇒ p - a). Assume also
that o 6= n. By the order lemma we have o|n. Using that
Y
xn − 1 = Φn (x) · (xo − 1) · Φd (x)
d|n, d-o
d6=n

5
 diendantoanhoc.net [VMF]

(theorem 8) together with Φn (x) ≡ (x − a) · g(x) mod p and xo − 1 ≡ (x − a) · h(x) mod p (the
last two because x ≡ a mod p is a root mod p of the LHSes), we are led to
n 2
Q
x − 1 ≡ (x − a) · j(x) mod p (with j(x) = g(x) · h(x) · d|n, d-o Φd (x)). But this gives p|n by
d6=n
lemma 3, proving this theorem.

Theorem 11. There are infinitly many primes ≡ 1 mod n.

Proof. For the fourth time now, we assume that there are only finitly many primes
p1 , p2 , ..., pn ≡ 1 mod n. Then we take their product P = p1 · p2 · ... · pn and choose q to be any
prime divisor of Φn (k · n · P ), where k is an integer just chosen big enough such that
Φn (k · n · P ) 6= ±1 so that at least one prime divisor exists.
We have q|(knP )n − 1 ⇒ q - (knP )n ⇒ q - n and q 6= pi (for all i). Because of q - n but
q|Φn (k · n · P ) when applying theorem 10, only the second case can happen there, giving that
q ≡ 1 mod n. Since q is different from all the pi , this gives a contradiction.

Fields, Orders and Polynomials

The approach of constructing a polynomial having, up to finitly many exceptions, only divisors
≡ −1 mod n fails for n ≥ 3. Indeed it was proved that for given polynomial f (x) the set of
residues a mod n for which there are ∞ many primes p ≡ a mod n with p|f (k) for some k build
a group; especially, there will always be a lot of them ≡ 1 mod n.
Our idea is now to construct a polynomial only (again up to some exceptions) archieving prime
divisors ≡ ±1 mod n.
This chapter is probably the most theoretic one: most stuff will not be needed again, and it is
possible to show the important theorems only for the needed cases. But these proofs are neither
shorter nor more intuitive, so we will handle the general case.

Lemma 4. Let f (x) = ak xk + ... + a0 + ... + a−k x−k with ak , ak−1 , ..., a−k+1 , a−k ∈ Z be

”symmetric”, meaning that ak = a−k , ..., ai = a−i , ..., a0 = a0 or equivalently f (x) = f x1 . Then
there exists a polynomial g(x) (with integer coefficients) fulfilling g x + x1 = f (x).

Proof. This falls by induction:

It’s clearly true for k = 0: we just take g(x) = f (x) = a0 .
Now let it be proved for alle m < k. We note that g(y) = ak y k fulfills
   k k   k−1  
1 1 X k k−i −i k −k
X k k−2i
g x+ = ak x + = ak x · x = ak x + a−k x + ak x .
x x i=0
i i=1
i

Simple checking gives that the sum/difference (and the product, which we will not need) of
symmetric terms is again symmetric. This leads to the symmetry (which also can be checked

6
 diendantoanhoc.net [VMF]

directly) of
  k−1 k−1   k−1
1 X
i
X k k−2i X
f (x) − g x + = ai x − ak x = bi x i .
x i=1
i
i=−(k−1) i=−(k−1)

1 1



By induction hypothesis we have that f (x) − g x + x = g x + x with a polynomial g(y). Thus
we can take g(y) = g(y) + g(y) as our polynomial. When we look back, we never left the integers
by our operations since we never divided, just added, subtracted and multiplicated.
The polynomial of our choice is more or less the following one:
Corollary 1. For n ∈ N, n ≥ 3 there exists a polynomial πn (x) with integers as coefficients such
ϕ(n)
that x 2 · πn x + x1 = Φn (x). Here ϕ(n) just denotes the degree of Φn (x).

Proof. We remember that the roots of Φn (x) are ζnk with gcd(k, n) = 1, so we can pair up ζnk and
ζn−k : If ζnk = ζn−k , then ζn2k = 1, thus n|2k; since n and k are coprime, we get n|2, contradicting
n ≥ 3. So we get that ϕ(n) is indeed even. And we also get that
Y Y
Φn (x) = (x − ζnk )(x − ζn−k ) = (x2 − (ζnk + ζn−k )x + 1),
1≤k≤ n2
1≤k≤ n2
gcd(k,n)=1 gcd(k,n)=1

yielding
Φn (x) Y
r(x) := φ(n)
= (x − (ζnk + ζn−k ) + x−1 ),
x 2
1≤k≤ n2
gcd(k,n)=1
1


giving r(x) = r x
, so it is symmetric. Using the lemma, the result follows (we again never left
the integers).
As promised, we will prove the next theorems in a more general way. Thus we need fields and
some related stuff.
Definition 3. A field is a set K together with ”addition” + and ”multiplication” · such that:
- there are 0K and 1K with 0K + a = a = 1K · a for all a ∈ K.

- the known laws of associativity, commutativity and distrubitivity hold.

- for all a ∈ K there is an (−a) ∈ K with a + (−a) = 0.

- for all a ∈ K\{0K } there is an a−1 ∈ K with a · a−1 = 1K .

To shorten things, for n ∈ N0 one often writes an for |a · a {z
· ... · a} and also often simplifies a · b to
n times
ab as one was always used to.

7
 diendantoanhoc.net [VMF]

Thus fields are just things we can calculate in as we always did.

Examples are:

- the rationals Q, the reals R or the complex numbers C.

- the residues mod p for p prime; this field, from now denoted by Fp , is more or less the
only field we will need.

Some properties we will leave to the reader:

Properties 1. For all a ∈ K:

- (−a) = (−1K ) · a.
- 0K · a = 0K .
- (−1K )2 = 1K .
- (−a)2 = a2 .
- ab = 0 =⇒ a = 0 or b = 0.

Definition 4. Let n ∈ Z. Then for any field K, n can be seen as some element nK of K by
nK := 1K + 1K + ... + 1K if n ≥ 0 and by nK := −(1K + 1K + ... + 1K ) otherwise. An easy check
| {z } | {z }
n times (−n) times
gives (−n)K = −(nK ), (m + n)K = mk + nK and (m · n)K = mk · nK for all m, n ∈ Z (one says
that nK is a ring homomorphism Z → K).

From now on K will always be a field and we will often just write n instead of nK for all integers
when it is clear that we work in K.
Since integers can be seen as elements of K, especially the binomial coefficients can be seen so.

Theorem P12. (Binomial
theorem)

n n k n−k n
(a + b) = k=0 a b k
for all a, b ∈ K.

Proof.

This
is proved

exactly in the same way it is done for complex numbers inductively using
n+1 n n
k
= k
+ k−1
.

Corollary 2. If p is a prime with pK = 0K , then (a + b)p = ap + bp for all a, b ∈ K.

 
p! p
Proof. For k = 1, 2, ..., p − 1 we have p| = since the numerator is divisible by p,
k!(p − k)! k
whereas the denominator is not. As a result kp K = 0K for those k. Now the binomial theorem


p  
k p−k p
X
p
finishes the proof by (a + b) = a b = ap + b p .
k=0
k K

Lemma 5. If x2 = a has a solution b ∈ K, then all solutions are given by x = ±b.

8
 diendantoanhoc.net [VMF]

Proof. Let b be a solution, then b2 = a, thus (−b)2 = b2 = a, so −b is also a solution.

Now let c be any solution, thus c2 − a = 0 ⇔ c2 − b2 = 0 ⇔ (c − b)(c + b) = 0. If c 6= b, then
c − b 6= 0, thus (c − b)−1 exists, leading to
b + c = (c − b)−1 (c − b)(c + b) = (c − b)−1 · 0 = 0 ⇒ c = −b, proving the lemma.
We will now treat the lemmata we used and proved before in a more general way.
Definition 5. Let a ∈ K, we call the smallest k ∈ N with ak = 1 the order of a in K and write
ordK (a) for it. For those a for which there is no such k, we just write ordK (a) = ∞.
Lemma 6. (order lemma)
If a ∈ K and n ∈ N such that an = 1K , then ordK (a)|n.
Proof. The same as we did before when proving it mod p:
Let o = ordk (a) and take division with remainder to get n = o · q + r with 0 ≤ r < o. We get
ar = an−o·q = an · (ao )−q = 1K · 1−q
K = 1K , contradicting the minimality of o again if r > 0. Thus
r = 0 and o|n = qo.
Definition 6. A polynomial over K, or sometimes also called a polynomial with coefficients in
K, is a term of the type ak xk + ... + a1 x + a0 with ai ∈ K for all i. Polynomials can in general be
added and multiplied exactly in the same way as we are used to in the complex numbers.
Lemma 7. Let n ∈ N, a ∈ K, and g(x) a polynomial with coefficients in K. If
xn − 1 = (x − a)2 g(x) as polynomials (so if xn − 1 has a double root), then nK = 0K .
Proof. We will just mimic the proof of lemma 3:
Set y = x − a, then (y + a)n − 1 = y 2 · g(y + a). By expanding the LHS we get
y n + ... + nK · an−1 y + (an − 1), and the RHS gives ... + 0K · y + 0K , thus nK · an−1 = 0K and
an = 1, giving nK = nK · 1K = nK · an = (nK · an−1 ) · a = 0K · a = 0K .
Since they have integer coefficients, we can treat the cyclotomic polynomials as polynomials over
any field K. The same holds for all the other polynomials that will come and be viewed in some
field.
Theorem 13. (main theorem on cyclotomic polynomials)
Let n ∈ N again. If there is an a ∈ K with Φn (a) = 0, then nK = 0K or ordK (a) = n.
Proof. Another one we can the proof copy for:
Assume that o := ordK (a) 6= n (the order exists since an = 1K ). Then o|n by the order lemma, so
we get that Y
xn − 1 = Φn (x) · (xo − 1) · Φd (x)
d|n, d-o
d6=n

by theorem 8. Now by definition Φn (a) = 0K and ao − 1 = 0K , thus Φn (x) = (x − a) · g(x) and

xo − 1 = (x − a) · h(x) with n 2
Q g(x), h(x) polynomials over K. But by this x − 1 = (x − a) · j(x)
with j(x) = g(x) · h(x) · d|n, d-o Φd (x).
d6=n
Now using lemma 7 we get nK = 0K .

9
 diendantoanhoc.net [VMF]

If we have a field K, we can get another one containing K, e.g. by the following process:
2
Theorem √ 14. If K is a field and s √ ∈ K is such that
√ x = s has√ no solution x ∈ K, then the set
L = K[ s] of numbers of type a + b s√(with a + b √s = c + d s iff a = c,√ b = d) with the
canonical
√ and intuitive
√ addition√(a + b√ s) + (c
√ + d√s) = (a + c) + (b + d) s and √ multiplication
(a + b s) · (c + d s)[= ac + ad s + b sc + b s · d s] = (ac + sbd) + (ad + bc) s is again a field
√ 2
(with s = s). Additionally, K is a subset, 0K = 0L , 1K = 1L such√that the new addition and
multiplication are the old ones, too. One says K is a subfield of K[ s] then. This especially
implies that nK = 0 iff nL = 0, thus the property for an integer to be zero doesn’t change.

Proof. The easy checking of the properties of a field will be left to the reader:

- associativity, commutativity, distributivity are easy to check directly

√ √
- 0L = 0 + 0 s, 1L = 1 + 0 s
√ √ √ √
- −(a + b s) = (−a) + (−b) s, (a + b s)−1 = (a(a − bs2 )−1 ) + (−b(a − bs2 )−1 ) s (here it
has first to be shown that a2 − b2 s = 0 ⇔ a = 0K = b).
√
K is a subfield by the numbers of type a + 0 s (again just checking or simply intuition); the rest
follows from this.

Corollary 3. If 2K 6= 0K (in fact, this restriction is not necessary), a quadratic equation

y 2 − py + q = 0 has always exactly two solutions y1 , y2 (counting roots with multiplicity, thus
double roots are counted twice) in some field L ”containing” K. Then also y1 + y2 = a and
y1 y2 = b (Vieta’s theorem).

Proof. Exactly the same way one solves quadratic equations normally works (we need 2K 6= 0K
to be able to divide by 2K ):
2 2
a2 a2
 
2 a a
y − ay + b = 0 ⇔ y − +b− =0⇔ y− = − b.
2K 4K 2K 4K
q
2 2
Now if x2 = a4 − b has a solution a4 − b ∈ K, we could proceed (here we use Lemma 5) with
q q
a2 2
y − 2K = ± 4K − b ⇔ y = 2K ± 4aK − b and get our two solutions.
a a
hq i
a2
But also otherwise, we just consider L = K 4K
− b and proceed then.
The other properties follow directly from expanding (y − y1 )(y − y2 ) = y 2 − (y1 + y2 )y + (y1 y2 ),
where y1 , y2 are the solutions found before.

Primes ≡ −1 mod n
We are now able to construct the primes we want:

10
 diendantoanhoc.net [VMF]

Theorem 15. If πn (x) = 0 has a root a in Fp (p prime), then p|2n or p ≡ ±1 mod n.

Proof. We exclude p = 2 (we already have p|2n then).

We are looking for an b 6= 0 with b + 1b = a which happens iff b2 − ab + 1 = 0. Let y1 and y2 be
the solutions of y 2 − ay + 1 = 0 (either in Fp or in the field constructed in corollary 3); clearly
y = 0 is not a solution, so we can always set b = y1 to get b + 1b = a.
Now y1p is also a solution of the equation because using corollary 2 twice we get

p
0 = 0p = (y12 − ay1 + 1)p = y12 + (−ay1 + 1)p = (y1p )2 − ap y1p + 1p = (y1p )2 − ay1p + 1

(here we used that ap = a, which is just Fermat’s little theorem ap ≡ a mod p since a is a
standard residue mod p). But y 2 − ay + 1 = 0 has just two solutions (corollary 3 again), thus
y1p = y1 or y1p = y2 . Before we start considering those two cases, we see (by definition of πn (x))
that  
ϕ(n)
1 ϕ(n)
Φn (y1 ) = y1 πn y1 +
2
= y1 2 · 0 = 0.
y1
Now using the main theorem on cyclotomic polynomials yields p|n or ord(b) = n. Assume that
ord(b) = n from now on since otherwise (p|n) we are already done.

Case 1: y1p = y1 , thus y1p−1 = 1, giving n = ord(y1 )|p − 1 by the order lemma; but the latter means
nothing else than p ≡ 1 mod n.

Case 2: y1p = y2 . By Vieta’s theorem (see corollary 3 once time) we have y1 y2 = 1 ⇒ y2 = y11 . Using
this gives y1p = y11 , implying y1p+1 = 1. Again by the order lemma this gives n|p + 1, thus
p ≡ −1 mod n.

Now nothing more is to be shown.

Corollary 4. Let p be an odd (thus p 6= 2) prime divisor of πn (k) for some integer k. Then p|n
or p ≡ ±1 mod n. (We will never use this corollary, but the theorem itself.)

Our last problem is to ”seperate” the divisors ≡ 1 mod n from those ≡ −1 mod n. The
polynomials πn (x) are not the best to do this, so we will construct a similar one.

Lemma 8. There is a rational number t with πn (t) < 0.

ϕ(n)
Proof. By definition π(ζn + ζn−1 ) = ζn 2 · Φn (ζn ) = 0, and ζn + ζn−1 = ζn + ζn = 2Re(ζn ) ∈ R.
There are also no double roots of πn (x) in C because that would give double roots of Φn (x).
Since πn (x) is a polynomial without double roots, it exactly changes its sign at it’s roots. Since
there is a real root (e.g. 2Re(ζn )), there is a change of sign and thus also a real number r with
π(r) < 0. If we choose a rational t close enough to r (here we use that polynomials give
continuous functions), we still have πn (t) < 0.

11
 diendantoanhoc.net [VMF]

 
Definition 7. We take fixed integers an , bn such that πn abnn < 0 (such an , bn exist be the
previous lemma) and take k to be the degree of πn (x).
Then we define τn (x) := bkn πn x + abnn .
τ (cn x)
At last, we set cn = τn (0) < 0 and then Ψn (x) := .
−cn
Lemma 9. The polynomials Ψn (x) and τn (x) have integers as coefficients (especially cn is an
integer) and a positive leading coefficient.

Proof. At first we prove that τn (x) has integers as coefficients:

Let πn (x) = rk xk + ... + r1 x + r0 (all ri are integers then) such that we have
 k  
τn (x) = bkn rk x + abnn + ... + bkn r1 x + abnn + bkn r0 , so it suffices to show that the polynomials
 i
k an
bn ri x + bn are integral for i = 0, 1, ..., k. But this follows directly from the binomial theorem:

 i i   X i  
an X
i−m m −m i i
bkn ri x+ k
= bn r k x an b n = k−m i−m m
bn rk x an ,
bn m=0
m m=0
m

where only integers occure (bk−mn is always an integer if m ≤ k).

k
Let τn (x) = sk x + ... + s1 x + s0 with integers si now (s0 = τn (0) = cn ). Thus

τn (cn x) sk ckn xk s 1 cn x s0
Ψn (x) = = + ... + + ,
−cn −cn −cn −cn
i
has only integers as coefficients. Indeed −c s0
n
= −1 ∈ Z and s−ci cn
n
= −si cni−1 ∈ Z for i ≥ 1. The sign
of the leading coefficient of Φn (x), πn (x), τn (x) and Ψn (x) never changes (since −cn > 0), thus it
suffices to show that the one of Φn (x) is positive. But this is clear from the definition when we
look back how they are defined.
We just have to show that we didn’t lose too much of the properties of πn (x).

Theorem 16. Let p be a prime divisor of Ψn (k) not dividing 2bn cn n (k some given integer).
Then p ≡ ±1 mod n.

Proof. Lets work in the field Fp again. There we have 0 = Ψn (k) = τn−c (cn k)
n
, thus τn (cn k) = 0, thus
 
k
bkn πn k + abnn = 0. But bn 6= 0 in this field, so after multiplying with (b−1
n ) this gives
 
an
πn k + bn = 0. By theorem 15 we are finished.

Now the goal is near. But we will need all the developed techniques.

Theorem 17. There are infinitely many primes p ≡ −1 mod n.

12
 diendantoanhoc.net [VMF]

Proof. We can assume n to be greater than 2 since otherwise it’s trivial.

As expected, we assume that there are only finitely many such primes p1 , ..., pk ≡ −1 mod n and
call their product P . Lets take an integer k such that Ψn (k · 2bn cn nP ) > 1 (exists since the
leading coefficient of Φn (x) is positive)and factor Ψn (k · 2bn cn nP ) = q1 q2 ...qm into not neccessary
τ (0)
different primes qi . We have Ψn (k · 2bn cn nP ) ≡ Ψn (0) = −τ (0)
= −1 mod 2bn cn nP and especially
Ψn (k · 2bn cn nP ) ≡ −1 mod n. Thus Ψn (k · 2bn cn nP ) is coprime to 2bn cn n, thus by theorem 16
the qi are ≡ ±1 mod n. The same way we get that the qi are different from the pj , thus by
assumption qi ≡ 1 mod n for all i, giving that
−1 ≡ Ψn (k · 2bn cn nP ) = q1 q2 ...qm ≡ 1 · 1 · ... · 1 = 1 mod n, contradicting n > 2.

13