Annales Mathematicae et Informaticae

36 (2009) pp. 117–121

http://ami.ektf.hu

A geometric proof to Cantor’s theorem and

an irrationality measure for some Cantor’s
series∗

Diego Marques

Departamento de Matemática, Universidade de Brasília

Submitted 18 November 2008; Accepted 22 May 2009

Abstract
Generalizing a geometric idea due to J. Sondow, we give a geometric proof
for the Cantor’s Theorem. Moreover, it is given an irrationality measure for
some Cantor series.
Keywords: Irrationality, irrationality measure, Cantor, Smarandache func-
tion.
MSC: Primary 11J72, Secondary 11J82

1. Introduction
In 2006, Jonathan Sondow gave a nice geometric proof that e is irrational.
Moreover, he said that a generalization of his construction may be used to prove
the Cantor’s theorem. But, he did not do that in his paper, see [2]. So we give a
geometric proof to Cantor’s theorem using a generalization to Sondow’s construc-
tion. After, it is given an irrationality measure for some Cantor series, for that we
generalize the Smarandache function. Also we give an irrationality measure for e
that is a bit better than the given one in [2].

2. Cantor’s Theorem
Definition 2.1. Let a0 , a1 , . . . , b1 , b2 , . . . be sequences of integers that satisfy the
inequalities bn > 2, and 0 6 an 6 bn − 1 if n > 1. Then the convergent series
a1 a2 a3
θ := a0 + + + + ... (2.1)
b1 b1 b2 b1 b2 b3
∗ The author is supported by Capes.

117
118 D. Marques

is called Cantor series.

Example 2.2. The number e is a Cantor series. For see that, take a0 = 2, an =
1, bn = n + 1 for n > 1.

We recall the following theorem due to Cantor [1].

Theorem 2.3 (Cantor). Let θ be a Cantor series. Suppose that each prime divides
infinitely many of the bn . Then θ is irrational if and only if both an > 0 and
an < bn − 1 hold infinitely often.
Proof. For proving the necessary condition, observe that if an = 0 for n > n0 ,
then the series is a finite sum, hence θ is rational. If an > 0 infinitely often, let
us to construct a nested sequence of closed intervals In with intersection θ. Let
I1 = [a0 + ab11 , a0 + a1b+1 1
]. Proceeding inductively, we have two possibilities, the
first one, if an = 0, so define In = In−1 . When an 6= 0, divide the interval In−1
into bn − an + 1 (> 2) subintervals, the first one with length b1a···b n
n
and the other
1
ones with equal length, namely, b1 ···bn , and let the first one be In . By construction,
1 1
|In | > b1 ···bn
, for all n ∈ N and when an 6= 0, the length of In is exactly b1 ···b n
. By
1
hypothesis on an , there exist infinitely many n ∈ N, such that |In | = b1 ···bn . Thus,
we have
h i h i
an +1 An +1
In = a0 + ab11 + . . . + b1a···b
n
n
, a 0 + a1
b1 + . . . + b1 ···bn = An
,
b1 ···bn b1 ···bn

where An ∈ Z for each n ∈ N. Also θ ∈ In for all n > 1. In fact, by hypothesis

it is easy see that θ > b1A···b
n
n
, for all n > 1. For the other inequality, note that
am 1
bm 6 1 − bm , for all m ∈ N, therefore

a1 an
b1 · · · bn (θ − (a0 + + ... + )) 6 1. (2.2)
b1 b1 · · · bn
Now if an = bn − 1 for n > n0 , then θ is the right-hand endpoint of In0 −1 , because
each In contains that endpoint and the lengths of the In tend to zero. Hence again
θ is rational. For showing the sufficient condition, note that if am < bm − 1, then
holds the strict inequality in (2.2), for each n < m. Since an > 0 holds infinitely
often,
∞
\
In = θ.
n=1

Suppose that θ = pq ∈ Q. Each prime number divides infinitely many bn , so there

exist n0 sufficiently large such that q|b1 · · · bn0 and an0 6= 0. Hence b1 · · · bn0 = kq
for some k ∈ N. Take N > n0 , such that, aN +1 < bN +1 − 1. Hence θ lies in
interior of IN . Also IN = In0 +k for some k > 0. Suppose IN = In0 . We can
kp An 0 kp A 0 +1
write θ = b1 ···bn
, thus b1 ···bn
< b1 ···bn
< b1n···bn
. But that is a contradiction.
0 0 0 0
n0 +1 kpbn0 +k ···b
If IN = In0 +k , for k > 1, then we write θ = b1 ···bn0 +k . But that is again a
contradiction. Therefore, it follows the irrationality of θ. 
A geometric proof to Cantor’s theorem . . . 119

3. Irrationality measure
The next step is to give an irrationality measure for some Cantor series. Now,
we construct an uncountable family of functions, where one of them is exactly a
well-known function for us.

Definition 3.1. Given σ = (b1 , b2 , . . . ) ∈ N∞ , satisfying

(∗) For all p prime number, the set {n ∈ N | p|bn } is infinite.

We define the function D(·, σ) : Z∗ → N, by

D(q, σ) := min{n ∈ N | q|b1 · · · bn }

Note that D(·, σ) is well defined, by condition (∗) and the well-ordering theorem.

In [2], J. Sondow showed that for all integers p and q with q > 1,
 
e − p  >
  1
, (3.1)
 q  (S(q) + 1)!

where S(q) is the smallest positive integer such that S(q)! is a multiple of q (the
so-called Smarandache function, see [3]). Note that if η = (1, 2, 3, . . . ), then
D(q, η) = S(q). Since e is a Cantor series and D(·, σ) is a generalization of Smaran-
dache function, it is natural to think in a generalization or an improvement to the
inequality in (3.1).

Lemma 3.2. Given n ∈ N, we have

     
m  θ − An  , θ − An + 1 
    
θ − > min (3.2)
 b1 · · · bn   b1 · · · bn   b1 · · · bn 

for all m ∈ Z.
m
Proof. Suppose that the result fail for some m ∈ Z. So, b1 ···bn lies in interior of
In . Contradiction. Hence (3.2) holds for all m ∈ Z. 

Proposition 3.3. Suppose that a Cantor series θ, like in (2.1) and satisfying (∗),
is an irrational number. For all integers p ∈ Z and q ∈ Z∗ , with D(q, σ) > 1, let
k be the smallest integer greater than D(q, σ) such that the interval Ik lies in the
interior of ID(q,σ) . Then
 

θ − p  min{ak , bk − ak − 1}
> (3.3)
 q b1 · · · bk

where σ = (b1 , b2 , . . . ).
120 D. Marques

pb1 ···bn
Proof. Let σ = (b1 , b2 , . . . ). Set n = D(q, σ) and m = q . Therefore m and
n are integers and

   

θ − p   m 
= θ−
 q  b1 · · · bn 
   
 An   An + 1 
> min θ −
 , θ− (3.4)
b1 · · · bn   b1 · · · bn 
min{ak , bk − ak − 1}
> . (3.5)
b1 · · · bk

The inequalities (3.4) and (3.5) follow respectively by Lemma 3.2 and the hypoth-
esis on k. 

The result below gives a slight improvement to (3.1).

Corollary 3.3. If p and q are integers, with q 6= 0, then

 
e − p  >
  1
, (3.6)
 q  (D(q, σ) + 2)!

where σ = (2, 3, 4, . . . ).

Proof. Since that minp∈Z |e − p| > 0.28 > 61 , then (3.6) holds in the case q = ±1.
In case q 6= ±1 the inequality also holds by Proposition 3.3 and Example 2.2.
Moreover, in this case we have S(q) − 1 ∈ {n ∈ N | q|(n + 1)!} and D(q, σ) + 1 ∈
{n ∈ N | q|n!}. Thus S(q) = D(q, σ) + 1. Hence
 
e − p  >
  1 1
= .
 q  (D(q, σ) + 2)! (S(q) + 1)!

Actually, the improvement happens only because (3.6) also holds for q = ±1.
1 1 1
Example 3.4. The number ξ := (1!) 5 + (2!)5 + (3!)5 +. . . = 1.031378 . . . is irrational,

moreover for p, q ∈ Z with q 6= 0, we have

 
ξ − p  >
  1
 q  (D(q, σ) + 2)!5

where σ = (25 , 35 , . . . ).

Acknowledgements. The author would like to thank Jonathan Sondow, Luiz

Antônio Monte and Ana Paula Chaves for their helpful comments.
A geometric proof to Cantor’s theorem . . . 121

References
[1] Cantor, G., Ueber die einfachen Zahlensysteme, Zeitschrift fur Mathematik und
Physik, 14 (1869) 121–128.
[2] Sondow, J., A geometric proof that e is irrational and a new measure of its irra-
tionality, Amer. Math. Monthly, 113 (2006) 637–641.
[3] Weisstein, E, W., Smarandache function, MathWorld– A Wolfram Web Resource,
published electronically at http://mathworld.wolfram.com/SmarandacheFunction.
html.

Diego Marques
Universidade de Brasília,
Brasília, DF,
Brazil
e-mail: diego@mat.unb.br