Introduction to Transcendental

Numbers

Student Seminar Presentation

Romeo Schilling
October 28, 2021

Lecturer: Özlem Imamoglu

Department of Mathematics, ETH Zürich
 Abstract

This report includes a very brief introduction to transcendental numbers,

punctuated by some historical notes and some open problems. By no
means is it meant to be an exhaustive survey, and in that regard it
cannot reflect all the depth of the subject.
Of course, only a small part of the material summarized in this report
will be discussed in the 45-minute talk: I will try to choose what can
best be explained in the allotted time, avoiding too many technical
details and barely mentioning the most general or accurate version of
any presented result.

“Run to brilliance. Sprint to excellence. Soar to transcen-

dence.” — Matshona Dhliwayo

i
 Contents

Contents iii

1 Introduction 1
1.1 Recap of Some Algebraic Concepts . . . . . . . . . . . . . . . . 1
1.1.1 Algebraic Numbers . . . . . . . . . . . . . . . . . . . . . 1
1.1.2 Algebraic Integers . . . . . . . . . . . . . . . . . . . . . 8
1.1.3 Transcendental Numbers . . . . . . . . . . . . . . . . . 12
1.2 Some Historical Notes . . . . . . . . . . . . . . . . . . . . . . . 14

2 Transcendence Criteria 17
2.1 Irrationality Criterion . . . . . . . . . . . . . . . . . . . . . . . . 18
2.1.1 A Proof of the Irrationality of e . . . . . . . . . . . . . . 24
2.2 Liouville’s Approximation Theorem . . . . . . . . . . . . . . . 25
2.2.1 Liouville Numbers . . . . . . . . . . . . . . . . . . . . . 31
2.2.2 Generalizations of Liouville’s Approximation Theorem 36

3 Transcendence of e 39
3.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.2 The Original Proof of Hermite . . . . . . . . . . . . . . . . . . 45

iii
 Chapter 1

Introduction

Although Leonhard Euler wrote that transcendental numbers ”transcend

the power of algebraic methods”, any result about transcendental numbers
is linked to those about algebraic numbers and quite often uses tools of
algebra to be proven. Since a transcendental number is defined by what
it is not, i.e. an algebraic number, there is this indeniable and interesting
duality between those two concepts. For example, if one succeeds to prove
a necessary, respectively a sufficient condition for a number to be algebraic,
one will have directly gained a sufficient, respectively a necessary condition
for a number to be transcendental.

1.1 Recap of Some Algebraic Concepts

All the following concepts have been discussed in our algebra courses. But
we will repeat a few of them here.

1.1.1 Algebraic Numbers

Definition 1.1 (Algebraic numbers) A complex number α ∈ C is called an alge-

braic number if α is an algebraic element over the field Q.
In other words, a complex number α ∈ C is an algebraic number if there exists (at
least) one non-zero polynomial with rational coefficients f = f ( X ) ∈ Q[ X ] which
has α as a root, i.e.
∃ n ∈ N≥0 and ∃ a0 , a1 , a2 , . . . , an ∈ Q with ( a0 , a1 , . . . , an ) 6= (0, 0, . . . , 0),
such that f (α) = an αn + . . . + a1 α + a0 = 0

Or, equivalently, a complex number α ∈ C is an algebraic number if there exists (at

least) one non-constant1 polynomial with rational coefficients f = f ( X ) ∈ Q[ X ]
1 If α = 0, take f = f ( X ) = X which is a non-constant polynomial; otherwise, if α 6 = 0,

then we certainly cannot have n = 0 or else f (α) = a0 = 0 =⇒ f ≡ 0.

1
1. Introduction

which has α as a root, i.e.

∃ n ∈ N≥1 and ∃ a0 , a1 , a2 , . . . , an ∈ Q with an 6= 0,
such that f (α) = an αn + . . . + a1 α + a0 = 0
The subset of C consisting of all algebraic numbers is usually denoted by

A := {α ∈ C | α is an algebraic element over Q}

and is in fact a subfield of C.

Remark 1.2 We can prove that A is an algebraic closure of Q, which is why certain
authors use the notation Q instead of A.
Proposition 1.3 (Minimal polynomial) For each α ∈ A, there exists a unique
polynomial, denoted irrα,Q = irrα,Q ( X ), which satisfies the following properties:
(i) irrα,Q ∈ Q[ X ], i.e. its coefficients are rational
(ii) irrα,Q is monic, i.e. its leading coefficient is equal to 1
(iii) irrα,Q (α) = 0 , i.e. it has α as a root
(iv) irrα,Q has the lowest possible degree (which is necessarily greater or equal to 1)
among all polynomials satisfying properties (i) to (iii).
We call irrα,Q the minimal polynomial of α over Q.
To characterize the minimal polynomial irrα,Q of an algebraic number α over Q, it
only needs to satisfy properties (i) to (iii) as well as the following property:
(iv)’ irrα,Q is irreducible in Q[ X ].
Remark 1.4 We can prove that for an algebraic number α ∈ A and for any
polynomial f ∈ Q[ X ] with rational coefficients, we have the equivalence:

α is a root of f ⇐⇒ irrα,Q divides f in Q[ X ]

Definition 1.5 (Degree of an algebraic number) We say that α ∈ A is of degree n

(necessarily n ≥ 1) if
deg(irrα,Q ) = n,
i.e. if the degree of the minimal polynomial of α over Q is finite and equal to n.
We write: deg(α) = n.
Remark 1.6 One can prove that for α ∈ A:

deg(irrα,Q ) = n ⇐⇒ [Q(α) : Q] = n,

i.e. the degree of the simple field extension Q(α)/Q is finite and equal to n.
In this case, the extension Q(α)/Q has for Q-basis the set
n o
1, α, α2 , . . . , αn−1

2
 1.1. Recap of Some Algebraic Concepts

i.e. any x ∈ Q(α) can be written as a unique linear combination of the elements
1, α, α2 , . . . , αn−1 over Q, that is:
n −1
x= ∑ λi αi for some unique λi ∈ Q (i = 0, 1, . . . , n − 1)
i =0

Proposition 1.7 Let α ∈ A be complex algebraic number. Then we have:

deg(α) = 1 ⇐⇒ α ∈ Q

In particular, the degree of a real algebraic number is strictly more than 1 if and only
if α is an irrational number.

Proof One direction is obvious: if α ∈ Q, then f = f ( X ) := X − α satifies

properties (i ) to (iv) of Proposition 1.3, hence irrα,Q = f , which implies that
α has degree 1.
For the other direction, suppose that α has degree 1, which means that
irrα,Q = aX + b for some a, b ∈ Z with a 6= 0. But 0 = irrα,Q (α) = aα + b
implies that α = −ab ∈ Q. 
Definition 1.8 (Algebraic conjugates of an algebraic number) For a complex alge-
braic number α ∈ A, we call the (complex) roots of the minimal polynomial of α
over Q the algebraic conjugates of α.

Remark 1.9 Since char (Q) = 0, any polynomial in Q[ X ] is necessarily separable

(i.e. each of its irreducible factors in Q[ X ] has no repeated roots). Hence, for any
α ∈ A of degree n, its minimal polynomial irrα,Q has exactly n distinct (complex)
roots, which means that α has exactly n distinct algebraic conjugates.
√
Example 1.10 The golden ratio ϕ := 1+2 5 is an algebraic number of degree 2,
because its minimal polynomial can be proven to be

irr ϕ,Q = irr ϕ,Q ( X ) = X 2 − X − 1

√
1− 5
The number 2 is the other algebraic conjugate of ϕ.

Proposition 1.11 We have the equivalence:

α ∈ A ⇐⇒ Re(α) ∈ A and Im(α) ∈ A,

i.e. a number is algebraic if and only if its real and imaginary parts are algebraic
numbers.

Proof One direction is obvious, because if Re(α) ∈ A and Im(α) ∈ A, and

since obviously i ∈ A (with minimal polynomial irri,Q = X 2 + 1), then
it follows that α = Re(α) + i Im(α) remains in the field A, as a sum and
product of elements of A.

3
1. Introduction

For the other direction, let α be an algebraic number, i.e. it is a root of a

non-constant polynomial f with rational coefficients. But we know that any
polynomial with real coefficients has the property that if it has a complex
root, then it also has as root the complex conjugate of this complex root.
Therefore, since f ∈ Q[ X ] ⊆ R[ X ] and since f (α) = 0, then α is also a root of
f , implying that α is also an algebraic number. Therefore, since A is a field
containing both α, α and i, it certainly contains the elements α+2 α = Re(α)
and α− α
2i = Im( α ), as desired.
In fact, for any algebraic number α ∈ A, we have:

irrα,Q = irrα,Q

Indeed, since α ∈ C is a root of irrα,Q ∈ Q[ X ] ⊆ R[ X ], then α is also a root

of irrα,Q . So, by Remark 1.4, since irrα,Q (α) = 0, we have that irrα,Q divides
irrα,Q in Q[ X ]. Similarly, we deduce that irrα,Q divides irrα,Q in Q[ X ]. Hence
those two polynomials irrα,Q and irrα,Q are associated in Q[ X ], i.e. differ by
only a unit of Q (because the set of units of Q[ X ] is just the set of constant
and non-zero polynomials of Q[ X ], which is isomorphic to Q× ). Since irrα,Q
and irrα,Q are both monic, they are equal. 
Remark 1.12 From the proof of Proposition 1.11, we can highlight the fact that:
if α is an algebraic number, then α is also an algebraic number with the same minimal
polynomial over Q, namely irrα,Q = irrα,Q .
Similarly, if α and β are algebraic numbers such that β is an algebraic conjugate of
α, then they have the same minimal polynomial, namely irrα,Q = irr β,Q .
Proposition 1.13 If α is an algebraic number of degree n ≥ 1, then Re(α) is
algebraic number of degree m with 1 ≤ m ≤ n.
Proof It follows from Proposition 1.11, since α ∈ A, that the real number
Re(α) ∈ A is also an algebraic number with degree say m ≥ 1.
Since α is of degree n, let α1 := α, α2 , . . . , αn be the n distinct algebraic
conjugates of α (see Remark 1.9).
Consider the following polynomial of degree n:
n
g = g( X ) := ∏[2X − (αi + αi )]
i =1

Since αi + αi = 2 Re(αi ) ∈ R, the polynomial g has real coefficients. We

certainly have that g(Re(α)) = g(Re(α1 )) = g( 12 (α1 + α1 )) = 0. If we can
further verify that g is a polynomial in Q[ X ], then by Remark 1.4, the minimal
polynomial of Re(α) over Q satisfies:

irrRe(α),Q divides g in Q[ X ].

Hence, we deduce, as desired, that deg(irrRe(α),Q ) ≤ deg( g), that is m ≤ n,

because Re(α) and g are respectively of degree m and n.

4
 1.1. Recap of Some Algebraic Concepts

To prove that g ∈ Q[ X ], we need some Galois Theory. Let f be the minimal

polynomial of α over Q, and let E be the splitting field of this polynomial, i.e.

E = Q( α1 , . . . , α n )

Since all αi ’s are algebraic over Q, we even have:

E = Q[ α1 , . . . , α n ]

Recall that f is also the minimal polynomial of αi and, by Remark 1.12, of αi

for i = 1, . . . , n, and so the field E certainly contains αi and αi for i = 1, . . . , n.
We claim that the coefficients of g will remain unchanged under any permu-
tation of the αi ’s.
Indeed, for n = 1, this is trivial, because then α being of degre n = 1 is ra-
tional (by Proposition 1.7), hence we have in particular that α = α, implying
that:
g = g( X ) = (2X − (α + α)) = (2X − 2α),
which certainly remains unchanged under any permutation of the set {α}.
For n = 2, we have:

g = g( X ) = (2X − (α1 + α1 )) · (2X − (α2 + α2 ))

Since both α1 and α2 are roots of f , we must have:

α1 , α2 ∈ { α1 , α2 }

We cannot have α1 = α2 , otherwise this would imply that α1 = α2 , which is a

contradiction, because all algebraic conjugates of α are distinct.
Case 1: α1 = α1 and α2 = α2 , in which case:

g = (2X − 2α1 ) · (2X − 2α2 )

Case 2: α1 = α2 and α2 = α1 , in which case:

g = (2X − (α1 + α2 )) · (2X − (α2 + α1 ))

In both cases, any permutation of the αi ’s leaves g unchanged.

The proof of the general case n ≥ 3 is similar.
Now consider the Galois group of f , i.e. the group of all field automorphisms
of E leaving Q fixed:

Gal ( f ) := Gal ( E/Q) = {σ ∈ Aut( E) such that σ|Q = IdQ }

We know that any element σ ∈ Gal ( f ) permutes the roots of f , which are by
definition the algebraic conjugates of α, and in fact any σ ∈ Gal ( f ) is totally
determined by its values at the algebraic conjugates of α. This result comes
from the existence of a well-known injective group homomorphism Φ from

5
1. Introduction

the Galois group Gal ( f ) to the set of all permutations of the set of the roots
of f , namely R E ( f ) := { β ∈ E | f ( β) = 0}, given by:

Φ : Gal ( f ) ,−
→ SRE ( f )
σ 7→ σ
RE ( f )

Therefore, a permutation of the roots αi ’s corresponds to the natural action

of a field automorphim σ ∈ Gal ( f ) restricted to the set R E ( f ). To each such
σ, corresponds a unique ring automorphism σ∗ : E[ X ] → E[ X ] which sends
X to X and sends any element x of E to σ( x ). Hence:
n n
σ∗ ( g) = ∏[σ(2) · σ(X ) − (σ(αi ) + σ(αi ))] = ∏[2X − (σ(αi ) + σ(αi ))] = g
i =1 i =1

because, as we have seen, g remains unchanged under any permutation of

the αi ’s (which include all the αi ’s).
Since σ∗ ( g) = g, then, if we set

g = g( X ) = bn X n + bn−1 X n−1 + . . . + b1 X + b0 ∈ R[ X ],

we have that:

σ(bn ) X n + σ(bn−1 ) X n−1 + . . . + σ(b0 ) = σ∗ ( g) = g = bn X n + . . . + b1 X + b0

Hence, by comparison of the coefficients, we see that all coefficients of g

satisfy

σ ( bi ) = bi (i = 0, 1, . . . , n) for any element σ ∈ Gal ( f ),

therefore all bi ’s belong to the fixed field EGal (E/Q) of Gal ( f ) = Gal ( E/Q).
Since E is the splitting field of a polynomial in Q[ X ] and since char (Q) = 0,
we have that the field extension E/Q is Galois, which is equivalent to:

EGal (E/Q) = Q

Hence all coefficients of g belong to the field Q, which means that g ∈ Q[ X ],

as desired. 
Remark 1.14 Because C is the disjoint union of the following three sets

C = Q t (R \ Q) t (C \ R),

and since rational numbers are simply algebraic numbers of degree 1, the subfield A
of C is divided into three non-overlapping sets:

Q and (R \ Q) ∩ A and (C \ R) ∩ A

Furthermore, all of those three subsets of A are non-empty. For example,

6
 1.1. Recap of Some Algebraic Concepts

Figure 1.1: The complex algebraic numbers

• 2/3 belongs to Q with minimal polynomial irr2/3, Q = X − 2/3;

√
• 2 belongs to (R \ Q) ∩ A with minimal polynomial irr√2, Q = X 2 − 2;

• i belongs to (C \ R) ∩ A with minimal polynomial irri, Q = X 2 + 1.

We now give a result about the cardinality of the set of algebraic numbers.

Proposition 1.15 The field A of all algebraic numbers is countable.

Proof We only give the outline of the proof. We first recall that any polyno-
mial with coefficients in a given field has a finite number of roots which is at
most its degree. Now, since the set Q of all rational numbers is countable,
the set Q[ X ] of all polynomials with rational coefficients is also countable.
Therefore, since the set A of all algebraic numbers is the set of all roots of
a countable number of polynomials, each with a finite number of roots, we
have the desired result. 
There is another property of algebraic numbers using polynomials with
integral coefficients, instead of rational coefficients:

Proposition 1.16 We have the equivalence:

∃ a non-constant polynomial g ∈ Z[ X ]
α ∈ A ⇐⇒
such that g(α) = 0 and with positive leading coefficient

Watch out, g is not necessarily monic!

Proof One direction is obvious, because any polynomial with integral coeffi-
cients is certainly a polynomial with rational coefficients.
For the other direction, let α ∈ A be an algebraic number. That is, there
exists a polynomial f = f ( X ) = an X n + . . . + a1 X + a0 with n ∈ N≥1 and
an , . . . , a1 , a0 ∈ Q and an 6= 0 and such that f (α) = 0. We set ai := pi /qi

7
1. Introduction

with pi , qi ∈ Z and qi > 0, as well as gcd( pi , qi ) ∼ 1 for i = 0, . . . , n, and we

consider the new polynomial:
lcm(q0 , q1 , . . . , qn )
g = g( X ) := · f (X)
gcd( p0 , p1 , . . . , pn )
(This construction comes from the ”decomposition of f into its content and
its corresponding Z-primitive polynomial.”)
We certainly have that g ∈ Z[ X ] with degree at least 1 and with non-zero
leading coefficient, and that g(α) = 0. If the leading coefficient of g is strictly
negative, by considering the polynomial (− g), we get the desired result. 
We can specify Proposition 1.16 by adding irreducibility and by specifying
the degree of the polynomial in question.
Proposition 1.17 We have the equivalence:

∃ a non-constant polynomial g ∈ Z[ X ] of degree n

α ∈ A of degree n ⇐⇒ such that g(α) = 0 and with positive leading coefficient
and which is irreducible in Q[ X ]
Watch out, g is not necessarily monic!
Proof It suffices to redo the proof of Proposition 1.16 by taking f to be the
minimal polynomial irrα, Q of α over Q. Then g only differs from irrα,Q from
a non-zero constant, hence g has the same degree as irrα,Q and is irreducible
in Q[ X ] like irrα,Q . 
Remark 1.18 We note that the polynomial g in the proof of both Propositions 1.16
and 1.17 is a Z-primitive polynomial. Hence, by Gauss’s Lemma, the irreducibility
of g in Q[ X ] is equivalent to the irreducibility of g in Z[ X ] (because we also have
that Quot(Z) = Q)

1.1.2 Algebraic Integers

We have stressed twice that the polynomial g in the last two propositions
was not necessarily monic. By requiring that g be also monic, we get a new
definition:
Definition 1.19 (Algebraic integers) An algebraic number α ∈ A is called an
algebraic integer if it is the root of a monic non-constant polynomial g ∈ Z[ X ].
By definition, all algebraic integers are algebraic numbers. But the converse
is not true. There exist algebraic numbers which are not algebraic integers.
One easy example is the following:
√
Example 1.20 The irrational number 2/3 is not an algebraic integer, because
even though it is a root of a non-constant polynomial with integral coefficients,
namely 9X 2 − 2, we cannot find a monic such polynomial which does the trick!

8
 1.1. Recap of Some Algebraic Concepts

√
Proof Assume α := 2/3 is an algebraic integer. Then there exists a monic
polynomial g = g( X ) = X n + bn−1 X n−1 + . . . + b1 X + b0 ∈ Z[ X ] which has
α as root. So:
√ !n √ ! n −1
2 2
g(α) = + bn − 1 + . . . + b0 = 0
3 3

Hence, by multiplying both sides of this equation by 3n , we obtain by setting

bn = 1:
n √ √ √
∑ bi ( 2)i 3n−i = ( 2)n + bn−1 ( 2)n−1 · 3 + . . . + b0 · 3n = 0
i =0
√
For i = 0, . . . , n, we have that if i is odd, then ( 2)i is not an integer. So we
can separate our equation into two smaller equations, and we obtain:
√ √
∑ bi ( 2)i 3n−i = 0 and ∑ bi ( 2 ) i 3 n − i = 0
i odd i even
i =0,...,n i =0,...,n
√
By factorizing by 2, the first equation becomes:
√ √ i −1
∑ bi ( 2)i 3n−i = 0 ⇐⇒ 2 · ∑ bi 2 2 3n − i = 0
i odd i odd
i =0,...,n i =0,...,n

Since 3 divides 0 in Z, each sum above (the one for i even and the one for i
odd) must be divisible by 3. In particular, because each summand containing
bi , i 6= n, has a factor of 3, the number 3 must divide the summand containing
n −1 n
bn = 1. This tells us that 3 divides 2 2 in Z if n is odd, and 3 divides 2 2 in
Z if n√is even. In either case, this is false and hence we can conclude that
α := 2/3 is not an algebraic integer. 
We have to stress out
√ that algebraic integers are not always (ordinary) integers.
For example, i or 2 are algebraic integers
√ which are not ordinary integers.
Indeed, we have g1 (i ) = 0 and g2 ( 2) = 0 with g1 = g1 ( X ) = X 2 + 1
and g2 = g2 ( X ) = X 2 − 2 being two non-constant, monic polynomials with
integral coefficients.
In fact, we can specify the nature of real algebraic integers.

Proposition 1.21 A real algebraic integer is either an (ordinary) integer or an

irrational number. In other words, any non-integral root of a monic non-constant
polynomial with integral coefficients is necessarily irrational.

Proof Let α be a real algebraic integer, i.e. α ∈ R is a root of a non-constant

monic polynom g ∈ Z[ X ]. Suppose that α 6∈ R \ Q, i.e. , since α ∈ R, suppose
that α is rational (we have indeed restricted ourselves to real algebraic integers

9
1. Introduction

in this proposition). Let us show that α ∈ Z. Using a result from algebra,

since Quot(Z) = Q and since any monic polynom g ∈ Z[ X ] is necessarily
Z-primitive, then we know that if α = p/q ∈ Q (with p, q ∈ Z, q > 0 and
gcd( p, q) ∼ 1) is a rational root of g, then p divides the constant coefficient of
g and q divides the leading coefficient of g. Since g is monic, we get that q
divides 1 in Z, which implies that q = ±1. Since q > 0, we have q = 1, and
therefore α = p ∈ Z, as desired. 
Remark 1.22 As a fun fact, we can prove that the algebraic integers form a ring,
whereas the algebraic numbers form a field.
There is a small property which links algebraic numbers to algebraic integers:

Proposition 1.23 If α is an algebraic number, then there exists an integer c ∈ Z

such that c · α is an algebraic integer.

Proof Since α ∈ A, there exists a non-constant polynomial g ∈ Z[ X ] such

that
g ( α ) = a n α n + a n −1 α n −1 + . . . + a 0 = 0
with n ≥ 1 and an > 0 (we only need the fact that an 6= 0).
We define  
n −1 X
g̃ = g̃( X ) := an · g
an
More precisely, we have:
 n   n −1  
X X X
g̃ = ann−1 · an · + ann−1 · an−1 · + . . . + ann−1 · a1 · + ann−1 · a0
an an an

This polynomial has certainly integral coefficients and is monic. Furthermore,

we have:
 
n −1 an α
g̃( an α) = an g = ann−1 · g(α) = ann−1 · 0 = 0
an

Hence an α is an algebraic integer, which proves the proposition by taking

c := an . 
In fact, there is also a relation between algebraic numbers and algebraic
integers using the minimal polynomial:

Proposition 1.24 For any algebraic number α ∈ A, we have the equivalence:

α is an algebraic integer ⇐⇒ irrα,Q ∈ Z[ X ]

Proof One direction is obvious, because the minimal polynomial of any

algebraic number is always monic.
For the other direction, let α be an algebraic integer, i.e. there exists a monic

10
 1.1. Recap of Some Algebraic Concepts

non-constant polynomial g ∈ Z[ X ] such that g(α) = 0. Let also irrα, Q be the

minimal polynomial of the given algebraic number α over Q. By Remark 1.4,
since g ∈ Z[ X ] ⊆ Q[ X ] with g(α) = 0, we have that:

irrα, Q divides g and Q[ X ],

i.e. g( X ) = irrα, Q ( X ) · h( X ) for some polynomial h ∈ Q[ X ]. By ”decom-

posing” the two polynomials irrα, Q and h into their contents and their Z-
primitive parts, we get:
a c
irrα, Q ( X ) = · ϕ∗ ( X ) and h( X ) = · ψ∗ ( X )
b d
for some integers a, b, c, d ∈ Z with b, d > 0 and some Z-primitive polynomi-
als ϕ∗ , ψ∗ ∈ Z[ X ]. So we obtain from g = irrα, Q · h that:

bd · g = ac · ϕ∗ ψ∗

Because ϕ∗ and ψ∗ are Z-primitive, then by Gauss’ s Lemma, the product

ϕ∗ ψ∗ is also Z-primitive. But g is also Z-primitive (since g is monic), hence by
unicity of the decomposition of polynomials into their content and primitive
part up to association (i.e. up to units), we get:

bd ∼ ac in Z,

i.e. bd and ac are associated in Z. Now, because Z × = {±1}, we get

bd = ± ac, which implies that g = g( X ) = (±1) · ϕ∗ ψ∗ . So the leading
coefficient of both ϕ∗ and ψ∗ is ±1 (since g is monic).
Therefore, by comparing now the leading coefficients in irrα, Q = ba · ϕ∗ , we
get that:
a a
1 = · (±1) =⇒ = ±1 =⇒ irrα, Q = ± ϕ∗
b b
Since ϕ∗ ∈ Z[ X ], we deduce that irrα, Q ∈ Z[ X ], as desired. 
We should not talk of the minimal polynomial over Z, because Z is not a
field, but some authors do it in correlation with the last proposition. They
define the minimal polynomial of an algebraic number over Z as a monic
polynomial with integral coefficients having α as root and being irreducible
in Z[ X ], which is equivalent via Gauss’s Lemma to being irreducible in
Q[ X ]. But we prefer not to do that, because with this definition of minimal
polynomial over Z, its existence is not guaranteed anymore.
Up until now we have only considered roots of polynomials either with
rational coefficients or integral coefficients.

Proposition 1.25 If α ∈ C is a root of a polynomial h = h( X ) ∈ A[ X ] with

algebraic coefficients, then α must be an algebraic number.

11
1. Introduction

Proof We only outline the proof which uses some linear algebra. Let
E := Q( an , an−1 , . . . , a1 , a0 ) be the smallest field containing Q and all the
coefficients of h = h( X ) = an X n + an−1 X n−1 + . . . + a1 X + a0 . Since all co-
efficients of h are algebraic elements over Q, then the field extension E/Q
is a finite extension. Since the vector space formed by the powers of α is
finite-dimensional over E, it is also finite-dimensional over Q. That is, some
linear combination of the powers of α with rational coefficients vanishes, so
α is a root of a polynomial with rational coefficients. 

1.1.3 Transcendental Numbers

Having thoroughly recalled all we need about algebraic numbers, we can
now turn to transcendental numbers.
Definition 1.26 (Transcendental numbers) A complex number α ∈ C is called a
transcendental number if it is not an algebraic number, i.e. if there does not exist
any non-constant polynomial with rational coefficients which has α as a root.
The quality of a number being transcendental is called transcendence.
The fundamental theorem of algebra tells us that if f ∈ C[ X ] is a non-constant
polynomial with complex coefficients, then there exists a complex number
α ∈ C such that f (α) = 0. So in particular, for any non-constant polynomial
f ∈ Q[ X ] with rational coefficients, we have the existence of a complex
number α ∈ C such that f (α) = 0.
The subject of transcendental numbers focuses on the reverse question:
Given a complex number α ∈ C, does there exist a polynomial f ∈ Q[ X ] with
rational coefficients such that f (α) = 0? Simply by definition, the answer is
yes if and only if α is an algebraic number. In other words, the answer is no
if and only if α is a transcendental number.
We note that every real transcendental number must be irrational, since a
rational number is, by Proposition 1.7, an algebraic number of degree 1. The
converse is obviously not true: not all irrational numbers are real transcen-
√ √
1+ 5
dental numbers. For example, 2 and ϕ := 2 are both irrational numbers
which are not real transcendental, because they are both algebraic numbers
of degree 2.
Hence, making use of Proposition 1.7, we can say that the set of real numbers
R consists of three non-overlapping sets:
1. The set Q of all rational numbers, which is simply the set of algebraic
numbers of degree 1
2. The set (R \ Q) ∩ A of all algebraic irrational numbers, which is in fact
the set of real algebraic numbers of degree ≥ 2
3. The set of all real transcendental numbers, which is just the complement
of the set of real algebraic numbers with respect to R

12
 1.1. Recap of Some Algebraic Concepts

Figure 1.2: The real numbers

We have seen examples of elements in the first two sets, but a priori, we do
not know yet whether the set of real transcendental numbers is empty or not
(it is not!)
One can even wonder if the set of non-real transcendental numbers is empty
or not. To partially answer this question, we recall Proposition 1.11, which
says that a complex number is an algebraic number if and only if its real and
imaginary parts are both algebraic numbers. By taking the contraposite, we
obtain:

Proposition 1.27 Let α = a + ib ∈ C with a, b ∈ R be a given complex number.

Then we have the equivalence:

Re(α) = a is a transcendental number

α is a transcendental number ⇐⇒ OR
Im(α) = b is a transcendental number

As a consequence, if α is real transcendental number, then α + i and iα are

both complex non-real transcendental numbers, because in the first case,
Re(α + i ) = α is transcendental and in the second case, Im(iα) = α is tran-
scendental. Similarly, the number α + iα is also a non-real transcendental
number, because, here, even both real and imaginary parts are transcenden-
tal.
In other words, the existence of non-real transcendental numbers is a con-
sequence of the existence of real transcendental numbers. Because of this,
some authors only focus on the study of real transcendental numbers.
Another consequence of Proposition 1.25 in the light of transcendental num-
bers is the following:

Proposition 1.28 If α1 and α2 are both transcendental numbers, then either α1 + α2

or α1 · α2 is a transcendental number.

Proof Assume the contrary, i.e. α1 + α2 and α1 · α2 are both algebraic numbers.
Then the polynomial

h = h( X ) = ( X − α1 )( X − α2 ) = X 2 − (α1 + α2 ) X + α1 α2 ∈ A[ X ]

13
1. Introduction

would have algebraic coefficients. Hence, by Proposition 1.25, the roots

of h, which are α1 and α2 , would both be algebraic numbers, which is a
contradiction. 
That being said, the big question (which we have already mentioned) is:

Do real transcendental numbers exist?

If yes, can one give explicit examples of real transcendental numbers? And
even more difficult: given a specific (necessarily irrational) real number, like
e or π, how can you determine whether this number is transcendental or not?
We will answer to all these questions in this report, but we start by giving a
brief historical outlook on the subject of the transcendental numbers.

1.2 Some Historical Notes

The name ”transcendental” comes from the Latin transcendĕre, which means
’to climb over or beyond, to surmount’, and was first used in the context of a
function not being algebraic by Leibniz in his 1682 paper, in which he proved
that sin x is not an algebraic function of x.
Euler, in the 18th century, was probably the first person to define transcenden-
tal numbers in the modern sense, when he asserted in 1748 that the number
loga b was not algebraic for rational numbers a and b provided b is not of the
form b = ac for some rational c. Euler’s claim was not proved by the way
until the 20th century!
As already mentioned, a real transcendental number must be irrational. By
1744, Euler had already established the irrationality of e, and, by 1761, Lam-
bert had confirmed the irrationality of π. But both of these mathematicians
could not give a proof of the transcendence of those two numbers. Lambert
proposed a tentative sketch of a proof of π’s transcendence in his 1768 paper,
but it was far from being convincing.
The existence of transcendental numbers was already suspected by the end
of the 18th century, but it was the French mathematician Liouville in 1844
who exhibited the first class of transcendental numbers (now called Liouville
numbers in his honour) as a consequence of an approximation theorem that
he proved. The contribution of Liouville is significant because before 1844
we did not know for sure of the existence of transcendental numbers. This is
why we now consider 1844 to be the launch of the subject of transcendental
numbers.
We had to wait 29 more years for Hermite to finally give the first proof of
the transcendence of e in 1873, followed nine years later by Lindemann who
gave the first proof of the transcendence of π in 1882, by cleverly modifying

14
 1.2. Some Historical Notes

Hermite’s proof and by making use of the famous Euler equation linking
the old constant π to the modern constant e, namely eiπ + 1 = 0. The
proofs of the transcendence of e and π were considered among the greatest
achievements of 19th century mathematics. The transcendence of π especially
is a fact of great historical, as well as intrinsic, interest.
Indeed, one of the classical problems of Greek mathematics was to construct,
with compass and straightedge alone, a square whose area is that of a circle
√of
radius 1. This requires the construction of a line segment whose length is π,
which can be accomplished if a line segment of length π is constructible. The
Greeks were totally unable to decide whether such a line segment could be
constructed, and even the full resources of modern mathematics were unable
to settle this question until 1882, when π was proven to be transcendental
by Lindemann. Since the length of any line segment that can be constructed
√
with straightedge and compass can be written in terms of +, ×, −, ÷, and ,
and is therefore algebraic, this proves that a line segment of length π cannot
be constructed.
A totally different approach to the existence of transcendental numbers
was taken by Cantor who proved with his famous diagonal’s argument
(which dates back to 1874) that the real numbers are uncountable. Since
the real numbers are the union of the real algebraic numbers and the real
transcendental numbers, both of these sets cannot be countable. This implies
that the real transcendental numbers must be uncountable, since the real
algebraic numbers are countable (see Proposition 1.15). Cantor thus showed
in a non-constructive manner that almost all real numbers are transcendental,
without giving any explicit examples.
From the end of the 19th century up until the 1930’s, not much new impetus
has been given to the subject of transcendental numbers: we can only mention
a dozen of new and simpler proofs of the transcendence of e and π which rely
more and more on the property of the exponential function, and therefore
less succeptible of being generalized.
But in 1934 new impulse was given to the subject, when Hilbert’s 7th problem
(posed in 1900) was solved, namely:

Is ab always a transcendental number,

for any algebraic number a 6∈ {0, 1}
and any irrational algebraic number b?

The question was answered in the affirmative by Gelfond in 1934, following

the ideas of Gelfond and Siegel, and refined by Schneider in 1935, as well
as extended by Baker in 1966. A neat √application of Hilbert’s 7th problem
implies for example that the number 2 2 is transcendental.

15
1. Introduction

Among the important results of the remaining of the 20th century, we can
cite the classification of transcendental numbers by Mahler (1932) and later
by Koksma (1939); Schneider’s theorem on elliptic functions and Abelian
integrals (1941) and the Thue-Siegel-Roth Theorem on the approximation of
algebraic numbers by rationals (1955), which generalizes Liouville’s original
approximation theorem.
Open problems in the subject of transcendental numbers are very interesting
and surprisingly easy to state: they form a list of given numbers which have
yet to be proven to be either transcendental or algebraic.
For example, most sums, products, powers, etc. of √
the number π and the
e e 2 2
number e, e.g. eπ, e + π, π − e, π/e, π , e , π , π , eπ , are not known to
π

be√rational, algebraic, irrational or transcendental. A notable exception is

eπ n (for any positive integer n) which has been proven transcendental. By
the way, even though both eπ and e + π are of unknown status, we certainly
have by Proposition 1.28 that either eπ or e + π is transcendental. We also do
not know whether e and π are algebrically dependent or not.
Another open problem concerns the Euler’s constant γ defined as the limiting
difference between the harmonic series and the natural logarithm, namely:
! Z 
n ∞

1 1 1
γ = lim − log n + ∑ = − + dx.
n→∞
k =1
k 1 x bxc

Here, b x c represents the floor function. The number γ has not been proved
algebraic or transcendental. In fact, it is not even known whether γ is
irrational. Of course, some progress have been made, but a definite answer
has still not been given.
In the same vein, the Riemann zeta function at odd integers strictly greater
than 3, namely ζ (5), ζ (7), . . ., has not even been proven to be irrational.
The subject of transcendental numbers is a branch of number theory that
investigates transcendental numbers, in both qualitative and quantitative
ways. As yet, there seems to be no aspect to the study of transcendental
numbers which could be described as a general theory. However, there
are some methods which are quite powerful and useful in other domain
of mathematics. For example, many approximation theorems used in the
study of transcendental numbers come in handy in the study of diophantine
equations.

16
 Chapter 2

Transcendence Criteria

Usually the numbers for which we do not know of their transcendence arise
from analysis and are given by infinite processes, involving limits, infinite
series, infinite products or integrals. So it is very difficult in general to deter-
mine whether those numbers are even irrational, let alone transcendental, or
not.
Since being irrational is a necessary condition for a real number to be tran-
scendental, we start by looking at some irrationality criteria, hoping that one
of them can be modified into a transcendence criterion.
There exists a well-known criterion to determine whether a real and (for
simplification) non-negative number x is irrational or not, using its expansion
in an integer basis b ∈ N≥2 (e.g. in binary or decimal expansion), namely:

dj
x = bxc + ∑ j
= : [ s · d1 d2 d3 . . . ] b
j≥1 b

for some integers s and 0 ≤ dn ≤ b − 1 for n = 1, 2, . . ., where b x c denotes

the floor of x (which is equal to the integral part of x, since x ≥ 0).
Indeed, we have the equivalence: x is rational if and only if x has an eventu-
ally periodic (including finite) expansion in some (or equivalently in every)
integer basis. In other words, x is rational if and only if x is of the form

[ s · d 1 d 2 d 3 . . . d r d r +1 d r +2 . . . d r + p d r +1 d r +2 . . . d r + p d r +1 d r +2 . . . d r + p d r +1 . . . ] b

for some (or equivalently for every) integer b ≥ 2.

√ √
For example, the real numbers 2 and the golden ratio ϕ := 1+2 5 are
√
both irrational, because their decimal expansion are respectively 2 =
1.4142135623730950488 . . . and ϕ = 1.618033988749894 . . . and are infinite
aperiodic.

17
2. Transcendence Criteria

There is also another well-known criterion for irrationality using the contin-
ued fraction expansion. Any real and (for simplification) non-negative x can
be uniquely written as a simple continued fraction of the form

1
x = a0 + = : [ a0 ; a1 , a2 , a3 , . . . ]
1
a1 +
1
a2 +
1
a3 +
.
a4 + . .

for some integers a0 , a1 , a2 , . . .

We have the equivalence: x is irrational if and only if x has an infinite simple
continued fraction expansion.
√ √
For example, the real numbers 2 and the golden ratio ϕ := 1+2 5 are both
irrational,
√ because their simple continued fraction expansion are respectively
2 = [1; 2, 2, 2, . . .] and ϕ = [1; 1, 1, 1, . . .] and are not finite.
Unfortunately, we do not know the expansion of most constants arising from
analysis , either in an integer basis or in simple continued fractions. So these
two irrationality criteria are of no use. We thus need to find another criterion
for irrationality, which is, as already mentioned, a necessary condition for a
number to be transcendental.

2.1 Irrationality Criterion

The most efficient criterion for irrationality, or equivalently for rationality,
involves rational approximation.
Since Q is dense in R, then of course for any real number θ ∈ R and any
ε > 0, there exists a rational number p/q ∈ Q with p, q ∈ Z, and q > 0, as
well as gcd( p, q) ∼ 1, for which

p
|θ − | < ε (2.1)
q

p
The problem is that, as we try to make q closer and closer to θ, we may
have to use larger and larger integers p and q. So, the reasonable question
to ask here is how well can we approximate θ by rationals without too large
denominators.
p
Trivially, every real number θ can be approximated by a rational number q
1
with a given denominator q ≥ 1 with an error not exceeding 2q .

18
 2.1. Irrationality Criterion

Let
 indeed q ∈ Z>0 be any positive integer. Note that the closed interval
qθ − 2 , qθ + 12 has length 1 and, therefore, contains at least one integer.
1

Choosing p to be that integer, we immediately get the result:

1 p 1
|qθ − p| ≤ =⇒ |θ − | ≤
2 q 2q

This estimate
p 1
|θ − | ≤ (2.2)
q 2q
certainly implies (2.1). Indeed, for any ε > 0, just take q ∈ Z>0 large enough
1
so that 2q < ε and you get:

p 1
|θ − | ≤ <ε
q 2q

Considering rational approximations p/q of real numbers with an error of

order 1q is the key to find a criterion which will distinguish rational from
irrational numbers.
The idea is the following: if you have a rational number, then it has a very
bad approximation by any other rational numbers, in the following sense:
Let θ = a/b ∈ Q with a, b ∈ Z, b > 0 and gcd( a, b) ∼ 1 be any rational
p
number. Then for all other rational numbers q 6= θ (with p, q ∈ Z, q > 0 and
gcd( p, q) ∼ 1), we claim that the distance between θ and p/q is not too small.
Indeed, we have:
p | aq − pb|
|θ − | =
q qb
p
Now, since q 6= θ = ba , we have, using the fact that the product of means
equals the product of the extremes, that aq 6= bp, hence aq − bp 6= 0. So, since
this number is a non-zero integer, we have that | aq − bp| ≥ 1. So we obtain:
p 1 1 1 1 1
|θ − | ≥ = · > ·
q bq b q 2b q
1
By taking c := 2b > 0 (which is a positive constant only depending on
θ = a/b), we get that:
p 1 1 c
|θ − | > · = (2.3)
q 2b q q
On the other hand, we will show that irrational numbers are well approx-
imable by rationals.
Another interesting question is whether or not we can get a smaller error
of approximation than 1q . Surprisingly enough, it is possible, if not for all q,
then, at least for some of them.

19
2. Transcendence Criteria

This way of distinguishing rational from irrational numbers using rational ap-
proximations by specifying the order of the approximation error is formalised
in the next proposition:
Proposition 2.1 (Criterion for irrationality / rationality) Let θ ∈ R be a real
number. Then the following conditions are equivalent:
(i) θ 6∈ Q
(ii) For any ε > 0 there exists a rational number p/q ∈ Q (with p, q ∈ Z, q > 0
and gcd( p, q) ∼ 1) such that
p ε
0 < |θ − |<
q q
i.e. θ is said to be well approximable by rational numbers.
p
(The inequality > 0 guarantees that we consider rationals q 6= θ)
(iii) (Asymptotic Dirichlet’s Theorem or Dirichlet’s Theorem on Diophantine
Equations) There exist infinitely many rational numbers p/q ∈ Q (with
p, q ∈ Z, q > 0 and gcd( p, q) ∼ 1) such that
p 1
|θ − | < 2
q q
p
(We do not need to specify here that q 6= θ, since we have infinitely many of
them.)
(iv) (Uniform Dirichlet’s Theorem) For any real number Q > 1, there exists a
rational number p/q ∈ Q (with p, q ∈ Z, q > 0 and gcd( p, q) ∼ 1) such
that 1 ≤ q < Q and with the property that
p 1
0 < |θ − |<
q qQ
Proof (ii ) ⇒ (i ): This result has already been proven, because we have
explained the contraposite ¬(i ) =⇒ ¬(ii ) (see Equation (2.3)), namely:

there exists a positive constant c > 0 such that

for all other rational numbers p/q ∈ Q \ {θ }
θ ∈ Q =⇒
(with p, q ∈ Z, q > 0 and gcd( p, q) ∼ 1)
p
the inequality |θ − q | > qc holds

By the way the implication (ii ) ⇒ (i ) is actually the useful part of the
irrationality/rationality criterion, and paradoxically also the easiest one to
prove! The other easy implications are (iv) ⇒ (iii ) ⇒ (ii ) ⇒ (i ).
Indeed, let us prove that (iv) =⇒ (iii ):
Let Q > 1 be any real number. Then by the Uniform Dirichlet’s Theo-
rem, there exists a rational number p/q ∈ Q (with p, q ∈ Z, q > 0 and

20
 2.1. Irrationality Criterion

1 1
gcd( p, q) ∼ 1) such that 1 ≤ q < Q (which implies that 1 ≥ q > Q) and such
that
p 1 1
0 < |θ − | < < 2.
q qQ q
p p
Let Q0 be any real number exceeding |θ − q |−1 (which implies that Q10 < |θ − q |).
A second application of Uniform Dirichlet’s Theorem shows that there exists
a rational number p0 /q0 ∈ Q (with p0 , q0 ∈ Z, q0 > 0 and gcd( p0 , q0 ) ∼ 1) such
that 1 ≤ q0 < Q0 (which implies that 1 ≥ q10 > Q10 ) and such that
p
p0 1 |θ − q | p
0 < |θ − 0 | < 0 0 < 0
≤ |θ − |
q qQ q q
p0 p
Thus, necessarily, we get: q0 6= q .
Furthermore, we have that:
p0 1 1
|θ − 0
|< 0 0 < 0 2
q qQ (q )
By iterating this process, we obtain a sequence { qii }i∞=1 of distinct ratio-
p

nal numbers (with pi , qi ∈ Z, qi > 0 and gcd( pi , qi ) ∼ 1) such that for all
i = 1, 2, . . .
pi p i −1 p1

 | θ − q i | < | θ − q i −1 | < . . . < | θ − q 1 | ,

and
 0 < | θ − pi | < 1 2

qi (q ) i

Hence we have found infinitely many rational numbers p/q ∈ Q (with

p, q ∈ Z, q > 0 and gcd( p, q) ∼ 1) such that
p 1
|θ − | < 2 .
q q

Let us now prove that (iii ) =⇒ (ii ): Since there exist infinitely many rational
numbers pi /qi ∈ Q (with pi , qi ∈ Z, qi > 0 and gcd( pi , qi ) ∼ 1) such that
pi 1
|θ − |< ,
qi ( q i )2
then the set
qi ∈ Z>0 | i = 1, 2, . . .


is infinite.
Otherwise, if there were only finitely many such q1 , q2 , . . . , q M ∈ Z>0 ,
then, since there can only exist for each qi (i = 1, 2, . . . , M) finitely many
(i ) (i ) (i )
p1 , . . . pri ∈ Z with gcd( p j , qi ) ∼ 1 (j = 1, . . . , ri ) which satisfy
(i )
pj 1
|θ − |< ,
qi ( q i )2

21
2. Transcendence Criteria

(i )
p
there would be only finitely many rationals qji (i =, 1 . . . , M and j = 1, . . . , ri )
which satisfy the inequality in (iii ), which is a contradiction.
For that reason, since {qi }i∞=1 is an infinite sequence of positive integers, we
have (by considering a subsequence if necessary): qi → ∞, or equivalently
qi → 0, as i → ∞.
1

So, for any ε > 0, there exists an integer i0 ∈ N such that 1

q i0 < ε. Hence:

p i0 1 1 1 ε
0 < |θ − |< 2
= · <
q i0 ( q i0 ) q i0 q i0 q i0

which proves (ii ).

Let us now prove that (i ) =⇒ (iv): This part uses the pigeonhole or box
principle. Let θ ∈ R \ Q and Q > 1 be given.
Define N := d Qe: this means that N is the integer such that N − 1 < Q ≤ N.
Since Q > 1, we have N ≥ 2.
We recall, for any x ∈ R, the following equality: x = b x c + { x } with b x c ∈ Z
(the floor of x) and 0 ≤ { x } < 1 (the fractional part of x).
Consider the subset E of the unit interval [0, 1] which consists of the N + 1
elements
0, {θ }, {2θ }, {3θ }, . . . , {( N − 1)θ }, 1.
Since θ is irrational, these N + 1 elements are pairwise distinct. Split the
interval [0, 1] into N subintervals of length 1/N:
 
j j+1
Ij := , (0 ≤ j ≤ N − 1).
N N

By the pigeonhole principle, since we have N+1 elements and only N subin-
tervals, one of these N subintervals, say Ij0 , contains at least two elements
of E. Apart from 0 and 1, all elements {qθ } in E with 1 ≤ q ≤ N − 1 are
j j +1
irrational, hence belong to the union of the open intervals ( N , N ) with
0 ≤ j ≤ N − 1.
Case 1: If j0 = N − 1, then the interval
 
1
Ij0 = IN −1 = 1 − , 1
N

contains 1 as well as another element of E of the form {q∗ θ } with 1 ≤ q∗ ≤ N − 1.

Set p∗ := bq∗ θ c + 1. Then we have that 1 ≤ q∗ ≤ N − 1 < Q by choice of N
and we also have that:

p∗ − q∗ θ = bq∗ θ c + 1 − bq∗ θ c − {q∗ θ } = 1 − {q∗ θ }

But we have for the fractional part 0 ≤ {q∗ θ } < 1, implying: 1 − {q
∗ θ } > 0.
Moreover, 1 and {q θ } lie in 1 − N1 , 1 with {q∗ θ } ∈ 1 − N1 , 1 , hence,
∗


22
 2.1. Irrationality Criterion

since Q ≤ N:
1 1
0 < p∗ − q∗ θ = 1 − {q∗ θ } < ≤ .
N Q
So by dividing by q∗ > 0, we certainly have:
p∗ 1
0 < |θ − ∗
| < ∗ with 1 ≤ q∗ < Q,
q q Q
as desired.
Case 2: If j0 6= N − 1, we have that 0 ≤ j0 ≤ N − 2 and Ij0 contains two
elements {q1 θ } and {q2 θ } with 0 ≤ q1 < q2 ≤ N − 1. We have, since {q2 θ } is
j j +1
an irrational number lying in the open interval ( N0 , 0N ), that:
1
|{q2 θ } − {q1 θ }| < (2.4)
N
Now set
q : = q2 − q1 , p : = b q2 θ c − b q1 θ c .
Then we have 0 < q = q2 − q1 ≤ N − 1 < Q and
|qθ − p| = |q2 θ − q1 θ − bq2 θ c + bq1 θ c| = |{q2 θ } − {q1 θ }|
So we have from (2.4) and since Q ≤ N:
1 1
0 < |qθ − p| < ≤ ,
N Q
which implies that
p 1
0 < |θ − |< with 1 ≤ q < Q,
q qQ
as desired.
In both cases, the coprimality of the numerator and denominator of the
rational number p/q is obtained easily by dividing through by gcd( p, q). 
A refined version of the Asymptotic Dirichlet’s Theorem (which dates back
to 1842) was found by Adolf Hurwitz in 1891. One can prove it using either
continued fractions or Farey sequences. We do not give a proof here, only
the statement:
Proposition 2.2 (Hurwitz’s Theorem) Let θ be a real number. Then the following
conditions are equivalent:
(a) θ is irrational
(b) There exist infinitely many rational p/q ∈ Q (with p, q ∈ Z, q > 0 and
gcd( p, q) ∼ 1) such that
p 1
0 < |θ − |< √
q 5q2

23
2. Transcendence Criteria

Remark 2.3 Of course Condition (b) in Proposition 2.2 looks stronger than con-
1
dition (iii ) in the irrationality criterion (since √5q 2
≤ q12 ), so we certainly have
(b) =⇒ (iii ) =⇒ θ ∈ R \ Q. But (b) is in fact equivalent to (iii ), and that is what
Hurwitz proved, by proving that ( a) =⇒ (b). √
Moreover, the estimate in (b) with the constant 1/ 5 is optimal, in the sense that √
it is the best possible approximation we can have for the golden ratio ϕ := 1+2 5 .
Indeed, one can prove that for any real number c < √15 , there are only finitely many
p
rationals p/q (with p, q ∈ Z, q > 0 and gcd( p, q) ∼ 1) such that | ϕ − q | < qc2 .

2.1.1 A Proof of the Irrationality of e

The real number
∞
1
e := ∑ i!
i =0

is well-known to be irrational. One can find many proofs of this fact, even
very short ones. We will present a different proof of the irrationality of e
using the irrationality criterion.

Assuming e ∈ Q, then by the contraposite of condition (ii ) of the irrationality

criterion, there exists a positive constant c > 0 such that for every rational
p
q 6 = e (with p, q ∈ Z, q > 0 and gcd ( p, q ) ∼ 1), we have:

p c p
|e − | > ⇐⇒ q · |e − | > c
q q q

Now, by the Archimedean principle, let n be a positive integer such that

1
0< <c
n

and consider the number

n
1
∑ i!
i =0

p∗
This real number is in fact a rational number of the form q∗ (with p∗ , q∗ ∈ Z,
p∗
q∗ > 0 and gcd( p∗ , q∗ ) ∼ 1) satisfying q∗ 6= e. To see this, just write the n+1
summands with the same denominator n!, and then set q∗ := n! and p∗ the

24
 2.2. Liouville’s Approximation Theorem

corresponding numerator. We have then:

∞
p∗ 1
c < q · |e −
q ∗
| = n! · ∑ i!
i = n +1
 
1 1 1
= n! · + + +...
( n + 1) ! ( n + 2) ! ( n + 3) !
 
1 1 1 1
= n! · 1+ + + +...
( n + 1) ! (n + 2) (n + 2)(n + 3) (n + 2)(n + 3)(n + 4)
 
1 1 1 1
< n! · 1+ + + +...
( n + 1) ! (n + 1) (n + 1)(n + 1) (n + 1)(n + 1)(n + 1)
∞
1 1
= n! · ·∑
( n + 1 ) ! k =0 ( n + 1 ) k
1 1 1
= · 1
= ,
n + 1 1 − n+ 1
n

a contradiction to the fact that 1/n < c, implying the irrationality of e.

2.2 Liouville’s Approximation Theorem

The aim is to generalize the irrationality criterion into a criterion for tran-
scendental numbers or equivalently into a criterion for algebraic numbers.
We recall that a real number θ is irrational if and only if it is an algebraic
number of degree 1 (see Proposition 1.7). This is equivalent, by Proposition
1.17, to θ being the root of a polynomial g ∈ Z[ X ] of degree 1 with integral
coefficients and positive leading coefficient (since a polynomial of degree 1 is
certainly non-constant and irreducible in Q[ X ]).
We can actually bring out the relevant polynomials of degree 1 with integral
coefficients for any irrational number θ simply by rewriting the inequalities
of conditions (ii ) and (iv) of the irrationality criterion as follows:
p ε
0 < |θ − | < ⇔ 0 < |qθ − p| < ε
q q
and
p 1 1
0 < |θ − |< ⇔ 0 < |qθ − p| <
q qQ Q
So in fact, we have the following equivalence for the conditions (ii ) and (iv)
of the irrationality criterion:

for any ε > 0

∃ g = g( X ) := qX − p ∈ Z[ X ]
(ii ) ⇐⇒ with degree 1 and even with positive leading coefficient q > 0
which is Z -primitive (i.e. gcd( p, q) ∼ 1)
such that 0 < | g(θ )| < ε

25
2. Transcendence Criteria

and
for any real number Q > 1
∃ g = g( X ) := qX − p ∈ Z[ X ]
(iv) ⇐⇒ with degree 1 and even with leading coefficient 1 ≤ q < Q
which is Z -primitive (i.e. gcd( p, q) ∼ 1)
and such that 0 < | g(θ )| < Q1

By considering polynomials not just of degree 1, but of any degree, one can
modify the criterion for a number to be rational into a criterion for a number
to be algebraic (of any degree). This is exactly what Liouville did, as well as
later mathematicians who generalized Liouville’s result.
Liouville’s criterion essentially takes up the idea that not only rational num-
bers, but also algebraic numbers in general cannot be very well approximated
by rational numbers. So if a number can be very well approximated by
rational numbers, then it must be transcendental. The exact meaning of ”very
well approximated” is explained in the next proposition called ”Liouville’s
Approximation Theorem”.
Liouville’s criterion is unfortunately only a necessary condition for a number
to be algebraic, and hence by taking the contrapositive, Liouville’s criterion
can be rewritten as a sufficient condition for a number to be transcendental.
But his criterion was not strong enough to be necessary too, and indeed it fails
to detect the transcendence of e, for example. However, it did provide a large
class of real transcendental numbers. Liouville numbers have the privilege of
a tight approximation by rational numbers. This fact is against our intuitive
expectations, because it shows that in some respects real transcendental
numbers are nearer to rational numbers than algebraic irrational numbers.
Here is the classical version of Liouville’s criterion using an asymptotic
estimate:

Theorem 2.4 Let α ∈ C be an algebraic number of degree n ≥ 1. Then there exists

a positive constant c = c(α) > 0 depending only on α such that the inequality
p c c
|α − | > n = deg(α)
q q q

holds for all rational numbers p/q ∈ Q \ {α} (with p, q ∈ Z, q > 0, gcd( p, q) ∼ 1)
which are different from α.

Proof We can break up the proof into three different cases:

1. The case where α ∈ R is an algebraic number of degree n = 1.
That means (by Proposition 1.7) that α is rational. Hence, as we have
seen before, by the contrapositive of condition (ii ) of the irrationality

26
 2.2. Liouville’s Approximation Theorem

criterion, there exists ε 0 > 0 such that for any rational numbers p/q
p
(with p, q ∈ Z, q > 0 and gcd( p, q) ∼ 1), we have: |α − q | = 0 or
|α − qp | ≥ εq0 . So for any rational number p/q (with p, q ∈ Z, q > 0 and
gcd( p, q) ∼ 1) which are different from α, we must have:

p ε0
|α − | ≥
q q

By taking c := ε 0 /2 > 0 (which is a constant only depending on α), we

get:
p c c c
|α − | > = 1 = deg(α) ,
q q q q
as wanted.
2. The case where α ∈ R is a real algebraic number of degree n ≥ 2.
In this case, we have seen in Proposition 1.7 that α is necessarily an
irrational number, because an algebraic number is of degree 1 if and
only if it is a rational number. It follows, by Proposition 1.17, that there
exists a non-constant polynomial g ∈ Z[ X ] with integral coefficients
of degree n ≥ 2 which is irreducible in Q[ X ] and which has α as root
as well as a positive leading coefficient. Because deg( g) = n ≥ 2 and
because g is irreducible in Q[ X ], we know that this implies that g has
no roots in Q. Hence, we have that g( p/q) 6= 0 for all rational p/q
(with p, q ∈ Z, q > 0 and gcd( p, q) ∼ 1).
Let now p/q (with p, q ∈ Z, q > 0 and gcd( p, q) ∼ 1) be any rational
number. We want to find a positive constant c > 0 depending only on
α such that:
p c c
|α − | > n = deg(α)
q q q
We do not need to consider rational numbers p/q different from α
because α is irrational in our case.
There are two subcases to consider:
p
• |α − q | ≥ 1

In this subcase, since q > 0 is a positive integer, we have q ≥ 1,

hence q2 ≥ q ≥ 1, and so qn ≥ q ≥ 1 for all n ≥ 2. Therefore, we
have
1
0< n ≤1
q
We then deduce that:
p 1 1 1
|α − | ≥ 1 ≥ n > · n
q q 2 q

27
2. Transcendence Criteria

By simply taking the constant c := 1/2 > 0 which is certainly

independent of p/q, we get what we desired, that is:
p c
|α − | > n
q q

p
• |α − q | < 1

In this subcase, we write: g = g( X ) = an X n + . . . a1 X + a0 with

ai ∈ Z (and an > 0). So we have:
   n   n −1  
p p p p
g = an + a n −1 + . . . + a1 + a0
q q q q
an pn + an−1 pn−1 q + . . . + a1 pqn−1 + a0 qn
=
qn
p
Since g( q ) 6= 0 and since qn > 0, then the numerator of the last
big fraction is a non-zero integer, so we have:

| an pn + an−1 pn−1 q + . . . a1 pqn−1 + a0 qn | ≥ 1

So we obtain:
| an pn + an−1 pn−1 q + . . . + a1 pqn−1 + a0 qn |
 
p 1
|g |= ≥ n
q qn q

Now, using the mean-value theorem for g, because g is certainly

differentiable on R, and hence on the open interval with distinct
p p
endpoints α and q , we obtain for some ξ between α and q (by
taking the absolute values):
     
p p p p
|g | = |0 − g | = | g(α) − g | = | g0 (ξ )| · |α − |
q q q q
p 1
So considering the inequality | g( q )| ≥ qn , we get:

1 p
≤ | g0 (ξ )| · |α − | (2.5)
qn q

We want to majorate | g0 (ξ )|. By the triangular inequality, we have:

|ξ | ≤ |ξ − α| + |α|. But in our subcase, |α − qp | < 1 and ξ lies
p
between α and q , so we have |ξ − α| < 1, and hence, we get the
inequality |ξ | < 1 + |α|.
Now, for the derivative

g0 ( X ) = nan X n−1 + (n − 1) an−1 X n−2 + . . . + a1 ,

28
 2.2. Liouville’s Approximation Theorem

evaluated at ξ, we get:

g0 (ξ ) = nan ξ n−1 + (n − 1) an−1 ξ n−2 + . . . + a1

We set A := max (| a1 |, | a2 |, . . . , | an |) which is a positive constant

(since an 6= 0), only depending on g, hence on α. So, again by the
triangular inequality, we get:

| g0 (ξ )| ≤ n| an | · |ξ |n−1 + (n − 1)| an−1 | · |ξ |n−2 + . . . + | a1 |

≤ nA · (|ξ |n−1 + |ξ |n−2 + . . . + |ξ |0 )
(since | ai | ≤ A for i = 1, 2, . . . , n)
≤ nA · [(1 + |α|)n−1 + (1 + |α|)n−2 + . . . + (1 + |α|)0 ]

But since (1 + |α|)n−i ≤ (1 + |α|)n−1 for all i = 1, . . . , n, because

1 + |α| ≥ 1 and n ≥ 2, we obtain:

| g0 (ξ )| ≤ n · A · n · (1 + |α|)n−1 = n2 · A · (1 + |α|)n−1

The constant K := n2 · A · (1 + |α|)n−1 is a positive constant only

depending on α which satisfies

| g0 (ξ )| ≤ K (2.6)
p
By multiplying both sides of (2.6) by |α − q | > 0 (because α is
irrational) and by considering (2.5), we get the inequalities:
1 p p
n
≤ | g0 (ξ )| · |α − | ≤ K · |α − |
q q q
which implies that
p 1 1 1 1
|α − | ≥ · n > · n
q K q 2K q
1
By taking c := 2K > 0, we get the desired result.
3. The case where α ∈ C \ R is a non-real algebraic number.
Then α is necessarily of degree n ≥ 2 (otherwise it is rational by Propo-
sition 1.7 and hence real, which is a contradiction). So by Proposition
1.13, since α is an algebraic number of degree n, then Re(α) is also an
algebraic number with degree m with 1 ≤ m ≤ n.
Since Re(α) is a real algebraic number of degree m ≥ 1, we can apply
the proven cases 1 and 2 of Liouville’s Approximation Theorem to get
a positive constant c̃ = c̃(Re(α)) > 0 depending only on Re(α) such
that the inequality
p c̃
|Re(α) − | > m (2.7)
q q

29
2. Transcendence Criteria

holds for all rational numbers p/q ∈ Q \ {Re(α)} (with p, q ∈ Z, q > 0

and gcd( p, q) ∼ 1) which are different from Re(α).
Now consider any rational number p/q (with p, q ∈ Z, q > 0 and
gcd( p, q) ∼ 1) (which is necessarily different from α in our case).
Again we distinguish two subcases:
p
• If q 6= Re(α), then we surely have, using (2.7):

p p
|α − | = |(Re(α) − ) + iIm(α)|
q q
s
p 2

= Re(α) − + (Im(α))2
q
 
p
≥ | Re(α) − |
q
c̃
> m
q
c̃
≥ n
q

because 1 ≤ m ≤ n and q > 0 implies that 1 ≤ qm ≤ qn , which in

turn implies that qc̃m ≥ qc̃n , since c̃ > 0.
p
• If q = Re(α), then we have trivially:

p p
|α − | = |(Re(α) − ) + iIm(α)|
q q
= |iIm(α)|
= |Im(α)|
Im(α)
> ·1
2
Im(α)
2
≥
qn

since q > 0 and n ≥ 2 imply that qn ≥ 1, which in turn implies

that q1n ≤ 1.
Im(α)
So by setting c := min(c̃, 2 ) > 0, we surely have for any rational
number p/q (with p, q ∈ Z, q > 0 and gcd( p, q) ∼ 1) that

p c
|α − | > n ,
q q

as wanted. 

30
 2.2. Liouville’s Approximation Theorem

2.2.1 Liouville Numbers

Liouville’s Approximation Theorem allowed Liouville to construct the first
transcendental number in 1844, now called the Liouville constant, which is
the number:
∞
1
α := ∑ k!
k =1
10

i.e.

α = 0.1100010000000000000000010000000000000000000000000000000000 y
0000000000000000000000000000000000000000000000000000000000001 . . .

with a digit 1 in its decimal expansion at the positions 1! = 1 and 2! = 2 and

3! = 6 and 4! = 24 and 5! = 120, etc.

Remark 2.5 The series defining α is convergent, since it is dominated by the geomet-
ric series ∑∞ 1
k =1 10k , hence α is a real number. Moreover, α is obviously an irrational
number, because its decimal expansion is not eventually periodic, because it has
arbitrarily long strings of 0’s since the 1’s appear at ever growing positions after the
decimal point.

Proposition 2.6 (Liouville constant) The Liouville constant α := ∑∞

k =1
1
10k!
is a
transcendental number.

Proof Suppose α is an algebraic number. Then necessarily, since α 6∈ Q, we

have that α is of degree n ≥ 2. Hence by Liouville’s Approximation Theorem,
we can find a positive constant c = c(α) > 0 such that the inequality

p c
|α − | > n (2.8)
q q

holds for all rational numbers p/q ∈ Q (with p, q ∈ Z, q > 0, gcd( p, q) ∼ 1).
We do not need to require that p/q be different from α, since α is irrational.
Let i ∈ N be large enough so that

2
<c (2.9)
10(n+i)! · i

(This is possible, since c > 0 and the sequence { 10(n2+i)! · i }i∞=1 goes to 0 as i
goes to ∞).
Set j := n + i so that, with this new notation, we have

2
<c
10 j! · i

31
2. Transcendence Criteria

Now consider the number p/q with q := 10 j! and

j
1
p := 10 · j!
∑ 10k!
k =1
1
j! 1 1
= 10 · ( + + . . . + j! )
101! 102! 10
= 10 j!−1! + 10 j!−2! + . . . + 10 j!− j!
= 10 j!−1! + 10 j!−2! + . . . + 1

so that
j
p 1 1 1 1
= 1! + 2! + . . . + j! = ∑ k!
q 10 10 10 k =1
10
We certainly have that p, q ∈ Z, q > 0 and gcd( p, q) ∼ 1. So p/q is a relevant
rational number which satisfies inequality (2.8):
p c
|α − | > n
q q

We will show that we also have:

p c
|α − | ≤ n (2.10)
q q

which is a contradiction, hence α is not an algebraic number, hence it is

transcendental.
To prove inequality (2.10), we write:
∞
p 1
|α − | = ∑
q k = j +1
10k!
1 1 1 1
= + + + +...
10( j+1)! 10( j+2)! 10( j+3)! 10( j+4)!
1 1 1 1
= + + + +...
10( j+1)! 10( j+1)!( j+2) 10( j+1)!( j+2)( j+3) 10( j+1)!( j+2)( j+3)( j+4)
1 1 1 1
≤ + + + +...
10( j+1)! 10( j+1)! · 1 10( j+1)! · 2 10( j+1)! · 3
Hence we have:
p 1 1 1 1
| α − | ≤ ( j +1) ! · (1 + + 2 + 3 + . . .)
q 10 10 10 10

But the last series in parentheses is a geometric series equal to

1 1 1 1 10
1+ + 2 + 3 +... = 1
= <2
10 10 10 1 − 10 9

32
 2.2. Liouville’s Approximation Theorem

So we have, since j := n + i:

p 1 2 2 2 2 1
|α − | < ( j+1)! · 2 = j! · ( j+1) < j! · j = j! · (n+i) = j! · i · j! · n
q 10 10 10 10 10 10

2
Now because q := 10 j! and since 10 j! · i
< c, we obtain:

p 1
|α − | < c · n (2.11)
q q

which certainly implies the large inequality (2.10). 

The same idea can be used to prove that the number

∞
1
α := ∑ bk!
k =1

with b being any integer ≥ 2 is a transcendental number, or more generally,

that the number
∞
a
α := ∑ k!k
k =1
b

with b being any integer ≥ 2 and with { ak }∞ k=1 being any sequence of integers
such that ak ∈ {0, 1, 2, . . . , b − 1} for all k and ak 6= 0 for infinitely many k, is
also a transcendental number.
In fact, we can construct in a similar manner infinitely many transcendental
numbers. They form a class of transcendental numbers, which we now call
Liouville numbers in honour of Liouville.

Definition 2.7 (Liouville Numbers) A Liouville number is a real number α with

the property that, for every positive integer n ≥ 1, there exists a pair of integers
( p, q) with q > 1 (i.e. q ≥ 2) such that

p 1
0 < α− < n.
q q

Liouville numbers can be approximated ”quite closely” by rational numbers.

They are precisely the transcendental numbers that can be more closely
approximated by rational numbers than any algebraic irrational number.

Remark 2.8 The Liouville constant is an example of a Liouville number. Indeed, it

suffices to adapt the proof of Proposition 2.6, by taking i ∈ N large enough so that

2
<1 (2.12)
10(n+i)! · i

33
2. Transcendence Criteria

(This is possible, since the corresponding sequence goes to 0 as i goes to ∞).

Then by setting j := n + i and considering the rational number p/q with q := 10 j!
and
j
1
p := 10 j! · ∑ 10k!
k =1
1 1 1
= 10 j! · ( 1!
+ 2! + . . . + j! )
10 10 10
= 10 j!−1! + 10 j!−2! + . . . + 10 j!− j!
= 10 j!−1! + 10 j!−2! + . . . + 1
we can prove inequality (2.11) with c = 1, namely:
p 1
|α − | < n (2.13)
q q
as wanted.
Here is an expected proposition:
Proposition 2.9 Liouville numbers are transcendental.
Proof Let α be a Liouville number. First, we show that α must be irrational.
Assume the contrary, i.e. α = a/b for some integers a and b with b > 0. Let
n be a positive integer large enough so that 2n−1 > b (i.e. n > 1 + log2 (b)).
Then for any pair of integers ( p, q) with q > 1, we will prove that
p p 1
0 = α− or α − ≥ n, (2.14)
q q q
which is a contradiction to the definition of a Liouville number. Hence α is
irrational, as wanted.
Indeed, for any pair of integers ( p, q) with q > 1, we have;
p a p | aq − bp|
|α − | = | − | =
q b q bq
If | aq − bp| = 0, we would have
p | aq − bp|
|α − | = = 0,
q bq
meaning that such a pair ( p, q) satisfies the first equality in (2.14), irrespective
of any choice of n.
If, on the other hand, | aq − bp| > 0, then, since aq − bp is an integer, we can
assert the sharper inequality | aq − bp| ≥ 1. From this, it follows that
p | aq − bp| 1
|α − | = ≥
q bq bq

34
 2.2. Liouville’s Approximation Theorem

Now the last inequality implies

p 1 1 1
|α − | ≥ > n −1 ≥ n
q bq 2 q q

(by choice of n > 1 + log2 (b) and because q ≥ 2 =⇒ 2n−1 q ≤ qn ).

Therefore, in this case, such pair of integers ( p, q) satisfies the second inequal-
ity in (2.14), for some large enough positive n. Thus, we have established
(2.14), as wanted.
Now, assume α is an irrational algebraic number, i.e. assume α is a real
algebraic number of degree n ≥ 2. By Liouville’s Approximation Theorem,
there exists a positive constant c > 0 such that
p c
|α − | > n (2.15)
q q

holds for all integers p and q with q > 0 and gcd( p, q) ∼ 1. Let r be a
positive integer for which 2r ≥ 1/c. Since α is a Liouville number, then for
the positive integer n + r, there are integers p∗ and q∗ with q∗ > 1 such that

p∗ 1 1 c
|α − | < ∗ n +r ≤ r ∗ n ≤ ∗ n
q∗ (q ) 2 (q ) (q )

(by choice of r and because q∗ ≥ 2 =⇒ 2r ≤ (q∗ )r ).

This contradicts (2.15) and, hence, establishes that α is transcendental. 
We can ask ourselves whether the Liouville numbers form a big set of
transcendental numbers or not. The next proposition given without proof
answers this question:

Proposition 2.10 The set of Liouville numbers in [0, 1] has measure 0.

We have seen that all Liouville numbers are transcendental. But the converse
is not true. Indeed:

Corollary 2.11 There are transcendental numbers which are not Liouville numbers.

Proof Since the set A of algebraic numbers is countable, the set of real
algebraic numbers in [0, 1] has certainly measure 0. Hence, the set of real
transcendental numbers in [0, 1] has measure infinite (equal to the power of
the continuum), like the measure of [0, 1]. Hence the set of real transcendental
numbers in [0, 1] cannot be the set of Liouville numbers in [0, 1], because
their measure differ (enormously!) 
Finding a transcendental number which is not a Liouville number is not easy.
The numbers e and π are two such examples. One can prove for example that
e is not a Liouville number by considering its expansion in simple continued
fractions, but we will not give the proof here.

35
2. Transcendence Criteria

2.2.2 Generalizations of Liouville’s Approximation Theorem

Let α be an irrational algebraic real number of degree n ≥ 2.
Liouville’s Approximation Theorem states that there exists a positive constant
c = c(α) > 0 such that
p c
|α − | > n (2.16)
q q
holds for all rational numbers p/q ∈ Q (with p, q ∈ Z, q > 0 and gcd( p, q) ∼ 1).
On the other hand, since α is irrational, the Asymptotic Dirichlet’s Theorem
states that there exist infinitely many rational numbers p/q ∈ Q (with
p, q ∈ Z, q > 0 and gcd( p, q) ∼ 1) such that
p 1
|α − | < 2 (2.17)
q q
For n = 2, we see that both estimates (2.16) and (2.17) are sharp. This is
no longer true for n ≥ 3. We discuss here improvements of Liouville’s
Approximation Theorem; these improvements are deep and we state them
without proof.
Those statements, in which the exponent k will be specified below in Theo-
rems 2.12 to 2.15, have to be understood as follows:
For every irrational algebraic number α of degree n (necessarily ≥ 2) and
for any positive real number ε > 0, there are only finitely many rational
numbers p/q ∈ Q (with p, q ∈ Z, q > 0 and gcd( p, q) ∼ 1) which satisfy the
inequality:
p 1
| α − | < k+ε . (2.18)
q q
As a consequence, we can state:
For every irrational algebraic number α of degree n (necessarily ≥ 2) and for
any positive real number ε > 0, there exists a positive constant c = c(α, ε) > 0
depending only on α and ε such that the inequality
p c
| α − | > k+ε (2.19)
q q

holds for any rational number p/q ∈ Q (with p, q ∈ Z, q > 0 and gcd( p, q) ∼ 1).
We can convince ourselves that the case k = n in the inequality (2.19) gives a
version of Liouville’s Approximation Theorem.
Liouville’s Approximation Theorem has been strengthened over the years by
successively improving the exponent k:
n
Theorem 2.12 (Thue 1909) If n ≥ 3, the inequality (2.19) holds true for k = 2+1
√
Theorem 2.13 (Siegel 1921) If n ≥ 3, the inequality (2.19) holds true for k = 2 n

36
 2.2. Liouville’s Approximation Theorem

√
Theorem 2.14 (Dyson 1947) If n ≥ 3, the inequality (2.19) holds true for k = 2n

Theorem 2.15 (Roth 1955) If n ≥ 2, the inequality (2.19) holds true for k = 2
We note that all those theorems are not currently effective, i.e. there is no
bound known on the possible values of p, q given α.
Theorem 2.15 had been conjectured by Siegel in 1921. Even though Thue,
Siegel, and Dyson had successively improved Liouville’s original exponent
n, it was Roth who proved Siegel’s conjectured exponent in 1955, and won
a Fields Medal for this work. Still, Theorem 2.15 is often called the Thue-
Siegel-Roth Theorem. It is the best possible of its kind; the exponent 2 cannot
indeed be decreased, because Roth’s statement would fail on setting ε = 0
due to the Asymptotic Dirichlet’s Theorem.

There is a stronger conjecture of Serge Lang that

p 1
α− < 2
q q log(q)1+ε

can have only finitely many solutions in integers p and q.

37
 Chapter 3

Transcendence of e

We will explain the original proof of the transcendence of e, as discovered

by Hermite in 1873. Before starting with the proof, we need to state some
preliminary results.

3.1 Preliminaries
We start by introducing an auxiliary function which will happen to be very
useful in the proof of the transcendence of e.

Definition 3.1 (Auxiliary function) For any polynomial f ∈ R[ X ] (of whatever

degree ≥ 1) and for any complex number t ∈ C, let I (t; f ) denote the following
function Z
I (t; f ) := et−z f (z) dz,
γ

where γ is the line joining 0 and t in the complex plane, i.e.

γ : [0, 1] → C
θ 7→ γ(θ ) := θt

By definition of complex line integrals, we have:

Z 1 Z 1
I (t; f ) := et−γ(θ ) f (γ(θ ))γ0 (θ )dθ = et−θt f (θt)t dθ
0 0

Remark 3.2 By Cauchy’s theorem, since the function z 7→ et−z f (z) is holomorphic
on the domain C, the complex line integral I (t; f ) is the same over any piecewise
continuously differentiable path joining 0 and t in C and depends only on the
endpoints 0 and t. That is why we can simply write without ambiguity:
Z t
I (t; f ) = et−z f (z) dz
0

39
3. Transcendence of e

We will now mention some small properties of I (t; f ).

Lemma 3.3 With the same notations as before, we have:

I (t; f ) = et f (0) − f (t) + I (t; f 0 )

Proof We recall the formula about partial integration: Let v, w be two holo-
morphic functions on a domain G ⊆ C. Let γ : [ a, b] → G be continuous and
piecewise differentiable. Then the following holds:
Z Z
0 γ(b)
v(z)w (z)dz = [v(z)w(z)]γ(a) − v0 (z)w(z)dz
γ γ

By simply applying this formula to the line γ joining 0 and t and to the
following functions:

dv = f 0 (z)dz

v := f (z)
dw := et−z dz w = − et−z

we get:
Z
I (t; f ) = v(z)w0 (z)dz
γ
Z t
= [−et−z f (z)]0t − (−et−z ) f 0 (z)dz
0
= −et−t f (t) + et−0 f (0) + I (t; f 0 )
= et f (0) − f (t) + I (t; f 0 ),

as desired. 
By iterating the process in Lemma 3.3, we obtain for any polynom f ∈ R[ X ]
of degree m:

Lemma 3.4 With the same notations as before, and for any polynom f ∈ R[ X ] of
degree m, we have:
m m
I (t; f ) = et ∑ f ( j) (0) − ∑ f ( j) (t)
j =0 j =0

00
Proof We apply Lemma 3.3 to f , f 0 , f , . . . , f (m) to obtain:

I (t; f ) = et f (0) − f (t) + I (t; f 0 )

I (t; f 0 ) = et f 0 (0) − f 0 (t) + I (t; f 00 )
I (t; f 00 ) = et f 00 (0) − f 00 (t) + I (t; f 000 )
..
.
I (t; f (m) ) = et f (m) (0) − f (m) (t) + I (t; f (m+1) )

40
 3.1. Preliminaries

But deg( f ) = m, hence f (m+1) ≡ 0, so we have: I (t; f (m+1) ) = et−z · 0 dz = 0.

R
γ
Putting together all those equalities, we get:

I (t; f ) = et f (0) − f (t) + et f 0 (0) − f 0 (t) + et f 00 (0) − f 00 (t) + . . . + et f (m) (0) − f (m) (t)

which equals to:

m m
I (t; f ) = et ∑ f ( j) (0) − ∑ f ( j) (t),
j =0 j =0

as desired. 
Now we are interested in finding an upper bound for | I (t; f )| when f ∈ R[ X ]
is a polynomial with real coefficients of degree m:

Lemma 3.5 With the same notations as before, and for any polynom f ∈ R[ X ] of
degree m, we have:

| I (t; f )| ≤ |t| · e|t| · supξ ∈C,|ξ |≤|t| | f (ξ )|

Proof The proof is based on the triangular inequality for integrals. We have
by definition of the auxiliary function:
Z 1 Z 1
| I (t; f )| = | et−θt f (θt)t dθ | ≤ |et−θt | · | f (θt)| · |t| dθ
0 0

But we know that |ez | = eRe(z) and Re(z) ≤ |z| for all z ∈ C, so we obtain for
all θ ∈ [0, 1]:

|et−θt | = eRe(t−θt) ≤ e|t−θt| = e|t|·|1−θ | ≤ e|t|·1 = e|t|

Now, for the term | f (θt)|, we have for all θ ∈ [0, 1]:

| f (θt)| ≤ supθ ∈[0,1] | f (θt)| ≤ supξ ∈C,|ξ |≤|t| | f (ξ )|

since |θt| ≤ |t|. Now putting those inequalities together, we obtain:

Z 1
| I (t; f )| ≤ |t| · e|t| · supξ ∈C,|ξ |≤|t| | f (ξ )| · dθ,
0

which gives the desired upper bound. 

Now, we consider a particular polynom f ∈ R[ X ]:

f = f ( X ) = X p −1 ( X − 1 ) p . . . ( X − n ) p

where n ∈ N≥1 and p is any positive integer greater than 2.

We note that the degree of f is:

deg( f ) = p − 1 + np = (n + 1) p − 1

41
3. Transcendence of e

We also note that the roots of f are 0, 1, 2, . . . , n. The root 0 is a root of order
p − 1, and the root k for k = 1, 2, . . . , n is a root of order p. Therefore we
certainly have:

f ( j) (0) = 0 for 0 ≤ j < p − 1



f ( j) (k ) = 0 for 0 ≤ j < p (k = 1, 2, . . . , n)

We can give a formula for the successive derivatives of f at the root 0:

Lemma 3.6 For the particular polynom f = f ( X ) = X p−1 ( X − 1) p . . . ( X − n) p

(with p ∈ N≥2 ), we have:

 0 if 0 ≤ j ≤ p − 2
( j)
f (0) = ( p − 1)!(0 − 1) p . . . (0 − n) p = ( p − 1)!(−1)np (n!) p if j = p − 1
≡ 0 (mod p!) if j ≥ p


Proof The first case when 0 ≤ j ≤ p − 2 follows from the order p − 1 of the
root 0 of f , as already discussed.
We note that f is of the form: f ( X ) = X p−1 · ( g( X )) p with

g ( X ) = ( X − 1) · . . . · ( X − n ) ∈ Z[ X ]

Now, by the Leibniz’s formula, we get:

p −1 
p−1

f ( p −1)
(X) = ∑ r
( X p−1 )(r) · (( g( X )) p )( p−1−r)
r =0

But for the rth derivative of the monomial X p−1 , we have:

( X p−1 )(r) = ( p − 1)( p − 2) . . . ( p − r ) X p−1−r


for 0 ≤ r ≤ p − 1
( X p −1 ) (r ) ≡ 0 for r ≥ p

So, evaluated at X = 0, we get:

( X p −1 ) (r ) 6= 0 only when r = p − 1,
X =0

in which case we get:

( X p −1 ) ( p −1) = ( p − 1) !
X =0

42
 3.1. Preliminaries

This implies in the second case when j = p − 1 that:

p −1 
p−1

f p −1
(X) = ∑ ( X p −1 ) (r ) · (( g( X )) p )( p−1−r)
X =0 r =0 r X =0 X =0
p −2 
p−1

= ∑ ( X p −1 ) (r ) · (( g( X )) p )( p−1−r)
r =0 r X =0 X =0

+ ( X p −1 ) ( p −1) · (( g( X )) p )( p−1−( p−1))

X =0 X =0
= 0 + ( p − 1)! · ( g(0)) p
= ( p − 1) ! (0 − 1) p (0 − 2) p . . . (0 − n ) p
= ( p − 1)!(−1)np (1 · 2 · . . . · n) p
= ( p − 1)!(−1)np (n!) p ,

as desired.
Now, for the third case when j ≥ p, we get again by the Leibniz’s formula:

j  
j
f ( j) ( X ) = ∑ r (X p−1 )(r) · (( g(X )) p )( j−r)
r =0

Since we have that ( X p−1 )(r) ≡ 0 for r ≥ p, the sum in the formula for the jth
derivative of f with j ≥ p is reduced to the following sum:

p −1  
j
f ( j)
(X) = ∑ r
( X p−1 )(r) · (( g( X )) p )( j−r)
r =0

Now, evaluated at X = 0, we get:

p −1  
j
f ( j)
(X) = ∑ r
( X p −1 ) (r ) · (( g( X )) p )( j−r)
X =0 r =0 X =0 X =0

But since the term ( X p−1 )(r) is not zero only for r = p − 1, in which case
X =0
it is equal to ( p − 1)!, we get:
 
j
f ( j)
(X) = ( X p −1 ) ( p −1) · (( g( X )) p )( j−( p−1))
X =0 p−1 X =0 X =0
 
j
= ( p − 1)! · (( g( X )) p )( j−( p−1))
p−1 X =0

Now for j ≥ p, we certainly have that the order j − ( p − 1) of the derivative

of ( g( X )) p is strictly positive.

43
3. Transcendence of e

Let us calculate the ith derivative (( g( X )) p )(i) for any i > 0.

We have from the chain rule:

( g( X )) p )(i) = ((( g( X )) p )0 )(i−1)

= [ p · ( g( X )) p−1 · g0 ( X )](i−1)
= p · [( g( X )) p−1 · g0 ( X )](i−1)

So for i = j − ( p − 1) > 0 (in the case when j ≥ p), we get for the evaluation
at X = 0:

(( g( X )) p )( j−( p−1)) = p · [( g( X )) p−1 · g0 ( X )]( j−( p−1)−1)

X =0 X =0
= p · [( g(0)) p−1 · g0 (0)]( j− p)
= p · l,

where l := [( g(0)) p−1 · g0 (0)]( j− p) .

But the polynomial ( g( X )) p−1 · g0 ( X ) has integral coefficients, as a product
of two such polynomials, because g( X ) = ( X − 1) . . . ( X − n) ∈ Z[ X ], hence
not only g0 ( X ) ∈ Z[ X ], but also ( g( X )) p−1 ∈ Z[ X ].
So, the number l is in fact an integer, as it is the constant term of a polynomial
with integral coefficients.
Therefore, for the case j ≥ p, we have that:
 
( j) ( j) j
f (0) = f (X) = ( p − 1) ! · p · l
X =0 p−1

is an integer divisible by ( p − 1)! · p = p! (because ( p−j 1) and l are both

integers), which implies that, in the case when j ≥ p:

f ( j) (0) ≡ 0 (mod p!),

as desired. 
We can also give a formula for the successive derivatives of f at any other
roots k = 1, 2, . . . , n:

Lemma 3.7 For the particular polynom f = f ( X ) = X p−1 ( X − 1) p . . . ( X − n) p

(with p ∈ N≥2 ), we have for k = 1, . . . , n:

( j) 0 if 0 ≤ j ≤ p − 1
f (k) =
≡ 0 (mod p!) if j ≥ p

Proof The first case when 0 ≤ j ≤ p − 1 is obvious, as mentioned before,

because k (with k = 1, . . . , n) is a zero of f of order p.

44
 3.2. The Original Proof of Hermite

The second case, when j ≥ p, is the same as before. Indeed, as in the proof
of Lemma 3.6, we arrive at the conclusion:
   
j p −1 0 ( j− p) j
( j)
f (k) = ( p − 1)! · p · [( g(k)) · g (k)] = ( p − 1)! · p · l˜
p−1 p−1

where l˜ := [( g(k )) p−1 · g0 (k)]( j− p) ∈ Z. So f ( j) (k ) is certainly an integer

divisible by ( p − 1)! · p = p! (because ( p−j 1) and l˜ are both integers), hence
the desired result. 

3.2 The Original Proof of Hermite

Now we can turn to the main proof of the transcendence of e which is a proof
by contradiction.
Assume e is an algebraic number. Since e is irrational (see Section 2.1.1), we
know that if e is an algebraic number, it should be an algebraic number of
degree n ≥ 2. By Proposition 1.17, this means that there exists a polynomial

g = g ( X ) = a n X n + . . . + a1 X + a0 ∈ Z[ X ]

with integral coefficients of degree n ≥ 2, having e as a root, and being

irreducible in Q[ X ] (as well as with positive leading coefficient).
In other words, there exist n ∈ N≥2 and a0 , a1 , . . . , an ∈ Z with an 6= 0 (even
an > 0) such that:

g ( e ) = a n · e n + . . . + a1 · e + a0 = 0

We can further assume that a0 6= 0, because otherwise, g would not have any
constant term and hence g would have 0 as a root, which is a contradiction,
because any irreducible polynomial in Q[ X ] of degree at least 2 has no
rational root.
We now consider the previous polynomial f ∈ Z[ X ], namely:

f = f ( X ) = X p −1 ( X − 1 ) p . . . ( X − n ) p ,

but only for a large positive prime number p, such that p > n and also such
that p > | a0 |.
Let m denote the degree of f , i.e. m = (n + 1) p − 1. We certainly have, since
p ≥ 2, that m ≥ p.
We consider the following combination involving the auxiliary function I (t; f )
defined in (3.1) for t = 0, 1, 2, . . . , n:
n
J := a0 · I (0; f ) + a1 · I (1; f ) + . . . + an · I (n; f ) = ∑ ak · I (k; f )
k =0

45
3. Transcendence of e

where the ai ’s are the integral coefficients of g.

We claim that
n m
J=− ∑ ak ∑ f ( j) ( k ) (3.1)
k =0 j =0

Indeed, we have via Lemma 3.4:

J = a0 I (0; f ) + a1 I (1; f ) + . . . + an I (n; f )

! !
m m m m
= a0 e 0
∑f ( j)
(0) − ∑ f ( j)
(0) + a1 e 1
∑f ( j)
(0) − ∑ f ( j)
(1)
j =0 j =0 j =0 j =0
!
m m
+ . . . + an e n ∑ f ( j ) (0) − ∑ f ( j ) ( n )
j =0 j =0
m m m
= ∑ f ( j ) (0) · ( a0 e0 + a1 e1 + . . . + a n e n ) − a0 ∑ f ( j ) (0) − . . . − a n ∑ f ( j ) ( n )
j =0 j =0 j =0

Since a0 e0 + a1 e1 + . . . + an en = g(e) = 0, we get:

n m
J=− ∑ a k ∑ f ( j ) ( k ),
k =0 j =0

as claimed.
CLAIM 1: J ∈ Z
This is clearly true, using Equation (3.1), since all the ai ’s are integer and
since f ∈ Z[ X ] implies that f ( j) ∈ Z[ X ] (for j = 0, 1, . . . , m), which in turn
implies that the evaluation of f ( j) at the integers k = 0, 1, . . . , n remains an
integer.
CLAIM 2: J 6= 0
We have by taking out the term for k = 0 in Equation (3.1):
m n m
J = − a0 ∑ f ( j ) (0) − ∑ ak ∑ f ( j) ( k )
j =0 k =1 j =0

If J = 0, we would have:
m n m
− a0 ∑ f ( j ) (0) = ∑ ak ∑ f ( j) ( k ) (3.2)
j =0 k =1 j =0

We look at the left-handside of (3.2) and use the formula for the successive
derivatives of f at 0 from Lemma 3.6 (and noting that m ≥ p) to obtain:
" #
m m
− a0 ∑ f ( j) (0) = − a0 · ( p − 1)!(−1)np (n!) p + ∑ f ( j) (0)
j =0 j= p

46
 3.2. The Original Proof of Hermite

We claim that this integer composed of two terms is not divisible by p!.
Indeed, the second term
m
− a0 · ∑ f ( j ) (0)
j= p

is an integer equal to 0 modulo p!, hence it is divisible by p!, since we have

f ( j) (0) ≡ 0 (mod p!) for j ≥ p, via Lemma 3.6.
So the divisibility of − a0 ∑m ( j)
j=0 f (0) by p! depends on the divisibility by p!
of the first term
− a0 · ( p − 1)!(−1)np (n!) p
Since we have chosen p to be a large prime such that p > n, then p doesn’t
appear in the prime factorization of n!, and hence of (n!) p , (because p > n)
and of course p doesn’t appear in the prime factorization of ( p − 1)! either
(because p is prime), and so the product ( p − 1)!(−1)np (n!) p is not an integer
divisible by ( p − 1)! · p = p!. Furthermore, since we have chosen p prime
such that p > | a0 |, again p doesn’t appear in the prime factorization of a0 ,
and so the integer − a0 · ( p − 1)!(−1)np (n!) p is not divisible by ( p − 1)!p = p!,
i.e. the left handside of (3.2) is an integer not divisible by p!, i.e.
m
− a0 ∑ f ( j) (0) 6≡ 0 (mod p!) (3.3)
j =0

On the other hand, the right-handside of (3.2) is equal to:

p −1
! !
n m n m n m
∑ ak ∑ f ( j) ( k ) = ∑ ak ∑ f ( j) ( k ) + ∑ f ( j) ( k ) = ∑ ak 0+ ∑ f ( j) ( k )
k =1 j =0 k =1 j =0 j= p k =1 j= p

using the successive derivatives of f at k = 1, . . . , n from Lemma 3.7 (and

noting that m ≥ p). Moreover, by the same lemma, we know that: f ( j) (k ) is
an integer divisible by p! for all k = 1, . . . , n. So the right-handside of (3.2) is
an integer divisible by p!, i.e.
n m
∑ ak ∑ f ( j) (k ) ≡ 0 (mod p!) (3.4)
k =1 j =0

Equations (3.3) and (3.4) lead to a contradiction, since both handsides of (3.2)
should be divisible by the same integer, which implies that J 6= 0, as desired.
CLAIM 3: J is divisible by ( p − 1)!
As before, we take out the term for k = 0 in (3.1):
m n m
J = − a0 ∑ f ( j ) (0) − ∑ ak ∑ f ( j) ( k )
j =0 k =1 j =0

47
3. Transcendence of e

The first summation is equal as before to:

m
− a0 ∑ f ( j) (0) = − a0 · ( p − 1)!(−1)np (n!) p
j =0

which is certainly divisible by ( p − 1)!

The second summation
n m
− ∑ ak ∑ f ( j) ( k )
k =1 j =0

has been proven to be divisible by p! (see Equation (3.4)), and hence by

( p − 1)!. Hence J is the sum of two summations, each divisible by ( p − 1)!,
implying the desired claim.
Because of Claims 1,2,3, the integer J is a non-zero integer divisible by
( p − 1)!, so necessarily, we have:

| J | ≥ ( p − 1) ! (3.5)

We will arrive at a contradiction by finding an upper bound for | J |.

By using Lemma 3.5, we have for k = 0, 1, . . . , n:

| I (k; f )| ≤ |k|e|k| supξ ∈C,|ξ |≤|k| | f (ξ )|

So we have that:
n n
| J| ≤ ∑ |ak || I (k; f )| ≤ ∑ |ak ||k|e|k| supξ ∈C,|ξ |≤|k| | f (ξ )| (3.6)
k =0 k =0

Let A := max (| a0 |, . . . , | an |) > 0 (because an 6= 0 and even a0 6= 0).

So, for k = 0, 1, . . . , n, we certainly have:

| ak | ≤ A and e|k| = ek ≤ en and |k| = k ≤ n (3.7)

Moreover, we have:

supξ ∈C,|ξ |≤|k| | f (ξ )| ≤ supξ ∈C,|ξ |≤n | f (ξ )|

By definition of f , we also have:

| f (ξ )| = |ξ | p−1 · |ξ − 1| p . . . |ξ − n| p

So, for i = 0, . . . , n and for |ξ | ≤ n, we have:

|ξ − i | ≤ |ξ | + i ≤ n + i ≤ 2n,

which implies for |ξ | ≤ n that

| f (ξ )| ≤ (2n) p−1+np = (2n)m ,

48
 3.2. The Original Proof of Hermite

which in turn implies

supξ ∈C,|ξ |≤n | f (ξ )| ≤ (2n) p−1+np = (2n)m

However, we have the trivial estimate:

m = p − 1 + np ≤ p + np ≤ np + np = 2np

So we have:
supξ ∈C,|ξ |≤n | f (ξ )| ≤ (2n)2np = ((2n)2n ) p (3.8)
Hence, we have, using (3.6), (3.7) and (3.8):
n
| J | ≤ A · n · en · ((2n)2n ) p · ∑ 1 = A · en · n · ((2n)2n ) p · (n + 1)
k =0

By taking c1 := A · en · n · (2n)2n · (n + 1) > 0 and c2 := (2n)2n > 0, which

are two positive constants independent of p, we have:
p −1
| J | ≤ c1 · c2 (3.9)

But we know that the exponential of c2

c2 c2 ci
e c2 = 1 + + 2 +...+ 2 +...
1! 2! i!
is a convergent series. Hence, for its general term, we have:

c2i
→ 0 as i → ∞
i!
In particular, since c1 > 0, if p is large enough, then ( p − 1) is large enough
to have:
p −1
c2 1
<
( p − 1) ! c1
So we get
p −1
c2
c1 · <1 (3.10)
( p − 1) !
Hence, for p prime large enough such that p > n and p > | a0 | and such that
p −1
c
c1 · ( p2−1)! < 1 (this is possible since there are infinitely many primes), we get,
using (3.5) and (3.9):
p −1
p −1 p −1 c
( p − 1) ! ≤ | J | ≤ c1 · c2 =⇒ ( p − 1)! ≤ c1 · c2 =⇒ c1 · 2 ≥1
( p − 1) !
which is a contradiction to (3.10). Hence, e is not an algebraic number, hence
e is a transcendental number.

49