Dirichlet’s Theorem

Calvin Lin Zhiwei

August 18, 2007

Abstract
This paper provides a proof of Dirichlet’s theorem, which states
that when (m, a) = 1, there are infinitely many primes p such that
p ≡ a (mod m). The proof explained here mirrors Serre’s proof in [1].
However, one distinction is that the Riemann zeta function is used as
motivation to obtain results for the Dirichlet L-functions, which
yields the above result. In fact, this paper can be read from the
viewpoint of asking how do simple results of the zeta function
depend on basic properties.

For the rest of the paper, m is a fixed integer as given in the above theorem.
Let P deonote the set of primes, Pm,a denote the set of primes that are
congruent to a modulo m and Gm = (Z/mZ)× . We will use the Euler-phi
function φ(m) which counts the numbers of positive integers less than m
that are coprime to m. In particular, |Gm | = φ(m). A sequence of complex
numbers {an } is called strictly multiplicative if an · am = anm , ∀n, m ∈ N.
The concept of analytic continuation will be useful in getting a deeper
understanding of the proofs. However, prior knowledge is not necessary,
and comments relating to analytic continuation can be ignored. Basic
group theory is assumed. In particular, the result that abelian groups can
be written as the product of cyclic groups is used without proof.

1 Riemann-zeta Function
In this section, we recall and derive certain basic properties of the Riemann
zeta function. For each s ∈ C, Re(s) > 1, the Riemann zeta function is
defined as

1
 ∞
X 1
ζ(s) =
n=1
ns

Proposition 1. For Re(s) > 1, the zeta function converges.

Proof. Using the integral test,

Z ∞ ∞ Z n+1 ∞ ∞ ∞ Z l Z ∞
dt X dt X 1 X 1 X dt dt
= ≤ = 1+ ≤ 1+ = 1+
1 ts n=1 n ts
n=1
n s
n=2
n s
n=2 n−1
ts
1 ts

1
R∞ dt
The result follows as we know that s−1
= 1 ts
within the domain.

Proposition 2.
Y 1
ζ(s) =
p∈P
1 − p−s

Proof. Let N be any natural number, AN be the set of natural numbers

whose prime factors are not larger than N . Then we have
X 1 ∞
Y X 1 Y 1
= =
n∈A
ns p∈P,p≤N j=0 pjs p∈P,p≤N 1 − p−s
N

The result follows by letting N tend to infinity.

Comment. Notice that the numerators are {1, 1, 1, . . . } and form a strictly
multiplicative sequence, which allows the function to be written in a product
form.
1
Proposition 3. ζ(s) = s−1
+ ρ(s), where ρ(s) is holomorphic for Re(s) > 0.

Proof. Recall that

Z ∞ ∞ Z n+1
1 dt X dt
= =
s−1 1 ts n=1 n ts
Hence,

2
 ∞ ∞ Z n+1
X 1 X dt
ζ(s) = s
=
n=1
n n=1 n
ns
∞ Z n+1 ∞ Z n+1
X dt 1 X dt
= s
+ ( − s
)
n=1 l n s − 1 n=1 l t
∞ Z n+1
1 X 1 1
= + s
− s dt (∗)
s − 1 n=1 n n t

R l+1
Let ρn (s) = n=1 n1s − t1s , and ρ(s) = ∞
P
n=1 ρn (s). It is clear that each ρn (s)
is well defined and holomorphic for Re(s) > 0. Hence, it remains to show
that the series of functions converges uniformly on compact sets of
Re(s) > 0.
Observe that |ρn (s)| ≤ sup |n−s − t−s |, and n−s − t−s = 0 when t = n.
n≤t≤n+1
By considering the derivative of n−s − t−s which is st−(s+1) , and letting
x = Re(s),

Z n+1 Z n+1
s 1
|ρn (s)| ≤ sup dt ≤ |s| sup dt
n≤t≤n+1 n ts+1 n≤t≤n+1 n ts+1
n+1
|s|
Z
dt 1
= |s| sup ≤ |s| ≤
n≤t≤n+1 n tx+1 nx+1 n2

P
Thus, ρn (s) converges uniformly for |s| ≤ k, Re(s) ≥ , ∀ > 0.
Corollary. ζ(s) has a simple pole at s = 1.
Comment. The equation (∗) gives the analytic continuation of ζ(s) in
Re(s) > 0, s 6= 1.
Theorem 1. For Re(s) > 1, as s → 1, p∈P p1s ∼ log s−1
1
P
.
1
Proof. From proposition 3, since ρ(s) = ζ(s) = s−1 is holomorphic hence
1
bounded in a neighborhood of 1, thus log s−1 ∼ log ζ(s). Replacing ζ(s) by
its product form in Proposition 2,

3
 Y 1 X 1 X 1 X 1
log ζ(s) = log = log = = + η(s)
p∈P
1 − p−s p∈P
1 − p−s p∈P,k≥1 k · pks p∈P
ps

X 1
where η(s) = ks
. To show that η(s) is bounded, we majorize it
p∈P,k≥2
k · p
by the series, with x = Re(s)

X 1 X 1 X 1
|η(s)| ≤ ks
≤ ks
≤
p∈P,k≥2
kp p∈P,k≥2
p p∈P,k≥2
pkx
∞
X 1 X 1 X 1
≤ ≤ ≤ ≤1
p∈P
px (px − 1) p∈P
p(p − 1) n=2 n(n − 1)

Hence,
1 X 1
log ∼ log ζ(s) ∼
s−1 p∈P
ps
.

2 Characters of finite abelian groups

In this section, we obtain a basis for complex valued functions on
Gm = (Z/mZ)× and study their properties. These arise naturally as an
application of characters of Gm . Let G be an abelian group with the
operation written multiplicatively.
Definition. A character of a group G is a homomorphism χ : G → C× .
Example. The function that maps every element of G identically to 1 is
clearly a homomorphism. It is call the trivial character and denoted by χ1 .
It is easy to check that the characters of G form a group Hom(G, C× ) which
is denoted by Ĝ, and called the dual of G.
Proposition 4. Ĝ is a finite abelian group of the same order as G.
Moreover, they are isomorphic as groups.

4
Proof. We will first prove this for cyclic groups |G| = n . Let x be a
generator for the cyclic group G, then each χ ∈ Hom(G, C× ) is uniquely
determined by χ(x). Moreover, 1 = χ(1) = χ(xn ) = χ(x)n shows that there
are at most n possible characters of G. Conversely, given any nth root of
unity ω, the map χω : G → C× given by χω (xa ) = ω a is a homomorphism.

We will now prove the general statement for abelian groups |G|. By the
classification of abelian groups, G ≡ H1 × . . . × Hk , where each Hi is a
cyclic group of prime power hi . Let Hi be generated by xi , then each
χ ∈ Hom(G, C× ) is uniquely determined by (χ(x Q 1 ), χ(x2 ), . . . , χ(xk )). Since
χ(hi ) can take on at most Q |Hi | values, |Ĝ| ≤ |Hi | = |G|.
Conversely, let π : G → Hi be the projection map sending g to
(x1 g1 , x2 g2 , . . . , xk gk ). Then, given any (n1 , n2 , . . . , nk ) ∈ Nk , 1 ≤ ai ≤ hi ,
Q 2πi ni g
consider χ(n1 , n2 , . . . , nk ) which takes g to e hi i clearly defines a
character. Thus, |Ĝ| = |G|, with the isomorphism given by the generators
(hence not canonically)

Proposition 5. Let G be a finite abelian group. Then G is canonically

ˆ
isomorphic to Ĝ.

Proof. Given x ∈ G, the map φx : χ → χ(x) is a character of Ĝ. By the

ˆ
previous proposition, since |G| = |Ĝ|, it remains to show that the map
x → φx is injective.
With Qnotation as in the previous proposition, consider the projection map
G → Hi sending g to (xa11 , . . . , xakk ). For g 6= 1, ∃aj 6= 0. Without loss of
generality, j = 1. Then, χ(1, 0, . . . , 0)(g) 6= 1, which proves injectivity.

Proposition 6. Let |G| = n, χ ∈ Ĝ. Then,


X n χ=1
χ(x) =
0 χ 6= 1
x∈G

Proof. The first formula follows from adding n 10 s. For the second, choose
y ∈ G such that χ(y) 6= 1. Then
X X X
χ(y) χ(x) = χ(xy) = χ(x)
x∈G x∈G x∈G

5
 X
which implies that χ(x) = 0.
x∈G

Corollary. Let |G| = n, χ ∈ Ĝ. Then


X n x=1
χ(x) =
6 1
0 x=
χ∈Ĝ

Proof. This follows from the previous proposition applied to the dual
group, and using the isomorphism given in proposition 5.

Comment. Now consider functions from G to C. Since χ ∈ Ĝ form a

linearly independent set of functions from G to C and |G| = |Ĝ|, we get a
basis set of multiplicative functions which satisfy χ(xy) = χ(x)χ(y).
For Gm , we can extend the domain of these characters from Gm to N, by
defining
(
0 (n, m) 6= 1
χ(n) =
χ(a) n = a (mod m)
Furthermore, it is clear that {χ(n)} is strictly multiplicative. These χ will
be used to define the Dirichlet L-functions in Section 3.4.

3 Dirichlet series
3.1 Lemmas
The following lemmas are useful for dealing with convergence of
summations over complex numbers.

Lemma 3.1. Let {an }, {bn } be two sequences of complex numbers and
P0
define Aj,k = kn=j an , and Sj,j 0 = jn=j an bn . Then,
P
P 0 −1
Sj,j 0 = Aj,j 0 · bj 0 + jn=j Aj,n (bn − bn+1 ).

Proof. The statement follows by replacing an with Aj,n − Aj,n−1 and

regrouping terms.

6
Lemma 3.2. α, β ∈ R such that 0 < α < β. Let z = x + iy, x, y ∈ R and
x > 0, then
z
|e−αz − e−βz | ≤ | |(e−αx − e−βx )
x
Proof. Z β
−αz −βz
e −e =z e−tz dt
α

hence by taking absolute values, with x = Re(z),

Z β Z β
−αz −βz −tz z
|e − e | ≤ |z| e dt = |z| e−tx dt = | |(e−αx − e−βx )
α α x

The following is a standard lemma from complex analysis.

Lemma 3.3. Let U be an open subset of C and fn a sequence of

holomorphic functions on U which converge unifromly on every compact set
to a function f . Then, f is holomorphic in U and the derivatives fn0
converge uniformly on all compact subsets to the derivative f 0 of f .

Proof. Let D be a closed disc contained in U and ∂D be its boundary with

usual orientation. By Cauchy’s formula, ∀z in the interior of D,
Z
1 fn (z)
fn (z0 ) = dz
2πi ∂D z − z0
Taking limits, uniform convergence allows us to exchange the integral at
lmits, hence we get

Z
1 fn (z)
f (z0 ) = lim fn (z0 ) = lim dz
2πi ∂D z − z0
Z Z
1 fN (z) 1 f (z)
= lim dz = dz
2πi ∂D z − z0 2πi ∂D z − z0

which shows that f is holomorphic in the interior of D. The statement for

derivatives is proved similarly using

7
 Z
0 1 f (z)
f (z0 ) = dz
2πi ∂D (z − z0 )2

3.2 Dirichlet series

Let (λn ) be a increasing sequence of positive real numbers tending to
infinity. A Dirichlet series with exponents (λn ) is a series of the form
∞
X
f (z) = an e−λn z , an ∈ C, z ∈ C
n=1

Example. The zeta function is a Dirichlet series with an = 1 and

λn = log n. The Dirichlet series with these exponents will be studied in
Section 3.3
Proposition 7. If the Dirichlet series f (z) converges for z = z0 , then it
converges uniformly in every domain of the form Re(z − z0 ) ≥ 0,
Arg(z − z0 ) ≤ α, where α < π2 .
Proof.
P Without loss of generality, z0 = 0 and the hypothesis states that
an is convergent. By Lemma 3.3, we have to show that there is uniform
|z|
convergence in every domain of the form Re(z) ≥ 0, Re(z) ≤ k. Given  > 0,
0
there exists N such that ∀j, j ≥ N, |Aj,j 0 | ≤ . Applying Lemma 3.1 with
bn = e−λn z , we obtain
j 0 −1
−λ0j z
X
Sj,j 0 = Aj,j 0 e + Aj,n (e−λn z − e−λn+1 z )
n=j

Applying Lemma 3.2, with x = Re(z),

j 0
z X −λn x
|Sj,j 0 | ≤ (1 + | | (e − e−λn+1 x )) ≤ (1 + k)
x n=j
Thus, we have uniform convergence.
Corollary. It follows that if f converges for z = z0 , then it converges for
Re(z) > Re(z0 ) by Lemma 3.3. Hence, the function is holomorphic in an
open half plane, and its analytic continuation is given by the same formula.

8
Proposition 8. If the Dirichlet series f (z) has positive real coefficients
{aj } and converges for R(z) > ρ, ρ ∈ R, and the function f can be extended
analytically to a function holomorphic in a neighborhood of the point z = ρ,
then there exists  > 0 such that f converges for R(z) > ρ − .

Proof. Without loss of generality, assume ρ = 0, hence f is holomorphic in

a disc |z − 1| ≤ 1 +  and has a convergent Taylor series. The derivatives
can be calculated using lemma 3.3, obtaining
X X
f (n) (z) = a` (−λ` )n e−λ` ·z f (n) (1) = a` (−1)n (λ` )n e−λ` ·1
` `

The Taylor series expansion is

∞
X 1
f (z) = (z − 1)n f (n) (1), when |z − 1| ≤ 1 + 
n=0
n!
Hence,
∞
X 1 X
f (−) = [ (1 + )n (−1)n a` (−1)n (λ` )n e−λ` ·1 ]
n=0
n! `

is a double series with positive coefficient. By rearranging terms,

∞
X
−λ`
X 1
f (−) = [a` e (1 + )n λn` ]
` n=0
n!
X
= a` e−λ` eλ` (1+)
`
X
= a` eλ` 
`

Hence, the Dirichlet series converges for z = − and thus for Re(z) > 
from Proposition 7.

Corollary. The set of convergence of a Dirichlet series is an open half

plane bounded by a singularity.

9
3.3 Ordinary Dirichlet Series
The ordinary Dirichlet series is obtained by setting λn = log n, which gives
us
∞
X an
f (s) = , an ∈ C, s ∈ C
n=1
ns

Corollary. Since ∞ 1
P
n=1 nα converges for α > 1, if {an } is bounded, then
there is absolute convergence for Re(s) > 1.

Proposition 9. If the absolute value of the partial sums Aj,k = kn=j an

P
are bounded by K, then there is convergence for Re(s) > 0.

Proof. This is similar to the proof given for ρ(s) in proposition 3. By

applying Lemma 3.1,
j 0 −1
!
X 1 1 1
|Sj,j 0 | ≤ K | s− s
+ | 0s |
n=j
n (n + 1) j

By Proposition 7, we may assume that s is real, hence the inequality

simplifies to |Sj,j 0 | ≤ jKs , showing convergence.

3.4 Dirichlet L-function

Recall that the strictly multiplicative function χ : N → C was defined at
the end of section 2.
(
0 (n, m) 6= 1
χ(n) =
χ(a) n = a (mod m)
The corresponding Dirichlet L-function is defined by
∞
X χ(n)
L(s, χ) =
n=1
ns

Example. In the specific case where m = 1, ζ(s) = L(s, χ1 ).

We now establish results for the L-functions that are analogous to

Propositions 1, 2 and 3 given for the zeta function.

10
Corollary.
Q From proposition 2, L(s, χ1 ) = F (s)ζ(s) where
−s
F (s) = p|m (1 − r ). Thus L(s, χ1 ) extends analytically for Re(s) > 0 and
has a simple pole at s = 1.

Proposition 10. For χ 6= χ1 , L(s, χ) converges for Re(s) > 0 and

converges absolutely for Re(s) > 1. Moreover,
Y 1
L(s, χ) = χ(p)
for Re(s) > 1
p∈P 1− ps

Proof. Since the absolute value of the coefficients |χ(n)| are bounded by 1,
it follows that the series is absolutely convergent for Re(s) > 1.
Let N be any natural number, let AN be the set of natural numbers whose
prime factors are not larger than N . Then we have
∞
X χ(n) Y X χ(p)` Y 1
s
= `s
= χ(p)
n∈AN
n p∈P,p≤N `=0
p p∈P,p≤N 1 − ps

As N tends to infinity, we establish the product formula of L(s, χ) as above.

From Proposition 6, j+m−2

P
i=j χ(i) = 0, hence |Aj,j 0 | is bounded by φ(m). By
Proposition 9, L(1, χ) converges in Re(s) > 0. Thus, we have convergence
in Re(s) > 0 from Proposition 8.

3.5 Product of Dirichlet L-functions for fixed m

For any prime p - m, let p be the image of p in (Z/mZ)× , f (p) be the order
of p, and g(p) = φ(m)/f (p).
Q
Proposition 11. Let ζm (s) = χ L(s, χ). Then ζ(s) is a ordinary
Dirichlet series, with positive integral coefficients, converging in the half
plane Re(s) > 1, and
Y 1
ζm (s) =  g(p)
1
p-m 1− pf (p)s

Proof. For a fixed p, let ωp denote the f (p)-th roots of unity. From the
construction of characters in Proposition 4, it follows that the multi-set
{χ(p)} is g(p) copies of ωp .

11
Since ω∈ωp (1 − ωT ) = (1 − T f (p) ), we have
Q
f (p) g(p)
Q
χ (1 − χ(p)T ) = (1 − T ) . Convergence in Re(s) > 1 is guaranteed as
we are multiplying finitely many convergent functions. Replace each
L-function by its product form and set T = p1s ,

YY 1
ζm (s) = χ(p)
χ p-m 1− ps
YY 1
= χ(p)
p-m χ
1− ps
Y 1
=  g(p)
1
p-m 1− pf (p)s

Expansion of this product form shows that ζm (s) is a ordinary Dirichlet

series with has positive coefficients.

Theorem 2. ∀χ 6= 1, L(1, χ) 6= 0.

Proof. Proof by contradiction. Suppose that for some χ 6= 1, L(1, χ) = 0,

then ζm would be holomorphic at s = 1, and hence in Re(s) > 0 by
Proposition 10. The product form given in Proposition 11 would thus be
the analytic continuation of ζm (s). By Proposition 8, since the Dirichlet
series has positive coefficients, the series would converge for all Re(s) > 0.
However,
1 1
g(p) is dominated by ,
1 − p−φ(m)s

1
1− pf (p)s
1
Q
it follows that the coefficients of ζm are greater than p-m 1−p−φ(m)s This
1
sequence diverges for s = φ(m) as
Y 1 Y
−1
X1
= (1 − p )
1 − p−1 n
n
p-m p|m

12
Corollary. ζm (s) has a simple pole at s = 1

Comment. Observe that ζ1 (s) = ζ(s), so we have a proof of the corollary

to Proposition 3, which is independent of Proposition 3.

4 Proof of Dirichlet’s Theorem

The result stated in Theorem 1 allows us to define the analytic density of a
set of primes A ⊂ P. We say that A has an analytic density k if the ratio
X 1 X 1
/
p∈A
ps p∈P ps

tends to k as s → 1+ on the reals. It is clear that if A is finite, then it has

an analytic density of 0. We can now formulate a stronger version of
Dirichlet’s Theorem.
1
Theorem 3. The set Pm,a has an analytic density of φ(m)
.

Comment. In the spirit of this paper, this result is analogous to Theorem

1. To imitate the proof, we need to understand
P log ζm (s). The ‘principal
1 n
determination’ of the logarithm is log 1−α = α as given by the Taylor
expansion. For L(s, χ), define
Y 1 X 1 X χ(p)l
log L(s, χ) = log χ(p)
= log χ(p)
=
1− 1− lpls
p∈P ps p∈P ps p∈P,l∈N

This series is clearly convergent in Re(s) > 1.

Equivalently, we take the ‘branch’ of log L(s, χ) in Re(s) > 1 which becomes
0 as s tends to infinity on the real axis.

For χ ∈ Gˆm , define

X χ(p) X χ(p)
fχ (s) = =
ps p∈P
ps
p-m,p∈P

This function is convergent for s > 1.

Proof. We require the following 2 lemmas.

13
 1
Lemma 4.1. For χ = 1, fχ (s) ∼ log s−1 .
1
P
Proof. This follows from Theorem 1 as fχ (s) differs from ps
by the
finitely many terms p|m p1s .
P

Lemma 4.2. For χ 6= 1, fχ (s) is bounded as s → 1.

Proof. To prove this, we consider the logarithm of L(s, χ).
X X χ(pl )
log L(s, χ) = fχ (s) + Fχ (s) where Fχ (s) =
p∈P l≥2
lpls
P |1|
|Fχ (s)| ≤ l≥2 |lpls |
≤ 1, and by Theorem 2, log L(s, χ) is bounded, hence
so is fχ (s).
1
P
Now, for the proof of Dirichlet’s theorem, let g(s) = p∈Pm,a ps . Then,

1 X 1 X X χ(a)−1 χ(p)
χ(a)−1 fχ (s) =
φ(m) φ(m) p∈P
ps
χ∈Ĝ χ∈Ĝ

1 X X χ(a−1 p)
=
φ(m) p∈P ps
χ∈Ĝ
1 X φ(m)
=
φ(m) ps
p∈P,a−1 p≡1

= g(s)

where the third equality follows from Proposition 6.

Thus,

1 X 1 1 1
g(s) = χ(a)−1 fχ (s) ∼ 1(a)−1 f1 (s) ∼ · log
φ(m) φ(m) φ(m) s−1
χ∈Ĝ

References
[1] Jean-Pierre Serre, A course in Arithmetic, 1973.

14