The Farey Sequence

Jonathan Ainsworth, Michael Dawson,

John Pianta, James Warwick

Year 4 Project
School of Mathematics
University of Edinburgh
March 15, 2012
Abstract

The Farey sequence (of counting fractions) has been of interest to modern math-
ematicians since the 18th century. This project is an exploration of the Farey
sequence and its applications. We will state and prove the properties of the Farey
sequence and look at their application to clock-making and to numerical approx-
imations. We will further see how the sequence is related to number theory (in
particular, to the Riemann hypothesis) and examine a related topic, namely the
Ford circles.

This project report is submitted in partial fulfilment of the requirements for the
degree of BSc Mathematics except Pianta who is MA Mathematics

2
Contents

Abstract 2

1 Introduction to Farey sequences 5

1.1 The Ladies’ Diary . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Flitcon’s solution . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Farey sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Properties of the Farey Sequence . . . . . . . . . . . . . . . . . . 8
1.5 Farey History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.6 Length of the Farey Sequence . . . . . . . . . . . . . . . . . . . . 11

2 Clocks and Farey 12

2.1 Creation of the Stern-Brocot tree . . . . . . . . . . . . . . . . . . 12
2.2 Navigating the Tree . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3 Gear Ratios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

3 Approximation and Fibonacci 23

3.1 Approximating irrationals using Farey Sequences . . . . . . . . . 23
3.2 Farey Sequences of Fibonacci Numbers . . . . . . . . . . . . . . . 25

4 Ford Circles 27
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.2 Motivation and definition . . . . . . . . . . . . . . . . . . . . . . . 27
4.3 Properties of Ford circles . . . . . . . . . . . . . . . . . . . . . . . 29
4.4 From algebra to geometry and back to algebra . . . . . . . . . . . 32

3
5 The Farey Sequence and The Riemann Hypothesis 36
5.1 Equivalent statement to the Riemann Hypothesis using the Farey
Sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
5.2 Forward Proof of Equivalent Statement . . . . . . . . . . . . . . . 40
5.3 Backward Proof of Equivalent Statement . . . . . . . . . . . . . . 41

6 Appendix 48

4
Chapter 1

Introduction to Farey sequences

1.1 The Ladies’ Diary

Figure 1.1: Ladies Diary Cover 1747

“The Ladies Diary: or, the Woman’s Almanack” was an annual publication
printed in London from 1704 to 1841. It contained calendars, riddles, mathemat-
ical problems and other “Entertaining Particulars Peculiarly adapted for the Use
and Diversion of the fair-sex” [1]
In the 1747 edition, the following mathematical question appeared.

5
 The question of how many non-unique fractions occur with denominator ≤ 99
is a simple one and is given by the sum 99
P
n=1 n = 4950 (this sum can be quickly

solved in your head using the method made famous by Gauss [2]). However, the
question specifically asks for the number of fractions with different values, i.e.
the unique fractions, a problem that took 4 years to answer.
The 1751 edition of the Diary published three solutions to this problem, two of
which are flawed [3]. A writer going by the name of Flitcon provided the correct
solution (see Appendix) that there are 3003 simplified fractions. In other words,
1947/4950 of the possible fractions cancel.
The topic of the next section will be an analysis of this solution and hence an
answer to the original question.

1.2 Flitcon’s solution

Some of the following proofs have been omitted in this report. For the proofs see,
for example, Guthery [4].

Definition 1. Euler’s totient function is the function φ such that φ(n) is the
number of integers < n which are coprime to n.

Definition 2. A function, f , is multiplicative if f (a).f (b) = f (a.b).

Proposition 1. Euler’s totient function is multiplicative, provided that the inte-

gers a and b are coprime.

Euler’s totient function φ : n 7→ “number of integers less than n which are

coprime to n” can be computed as follows
Qk 1
Theorem 3. φ(n) = n. i=1 (1 − pi
) where Π denotes the product and pi are the
k distinct prime factors of n (this used the well-known “Fundamental Theorem
of Arithmetic”).

6
 Now we can prove that Flitcon was correct, using the methods outlined in his
solution.

Proposition 2 (Ladies’ diary 1747). There are 3003 unique rationals of the form
m
n
with m ≤ n < 100 on the interval (0, 1).

Proof. Sketch proof (Flitcon’s method):

• Construct a function, say φ, which maps n to ”number of integers less than

n which are coprime to n”. We have defined Euler’s totient function to do
this. Flitcon used an equivalent function*.

• Construct a table with three columns. The first column contains the integers
from 2 to 99. The second column contains the prime decomposition of n.
The third column contains φ(n). When constructing this table it is helpful
to notice (as Flitcon did) that the function φ is multiplicative.

• Sum the third column for all entries 2 to 99. This gives the total number
of simplified fractions between 2 and 99.

Proof by Maple: See the Appendix for our Maple code which verifies Flitcon’s
original solution. The code can also be used to compute any Farey sequence and
its length.

*Flitcon’s method appears to use Euler’s totient function (published 17 years

previously, in 1734) however, Flitcon doesn’t cite it and may have derived a non-
general form of the function for himself.

Example 4. φ(36) = 36.(1 − 21 )(1 − 13 ) = 12

Thus there are 12 unique fractions with denominator 36.

We would like to generalise the problem of how many simplified fractions there
are between 0 and 1 given any restriction of the denominator. Such a sequence
of numbers is called a Farey sequence.

7
1.3 Farey sequences
p
Definition 5. A Farey sequence Fn is the set of rational numbers q
with p and
q coprime, with 0 < p < q < n , ordered by size.

Example 6. F1 = { 01 , 11 }
F2 = { 10 , 12 , 11 }
F3 = { 10 , 13 , 12 , 32 , 11 }
F4 = { 10 , 14 , 13 , 21 , 23 , 34 , 11 }

Remark 7. The question in the Ladies’ diary is equivalent to asking what is the
length of F99 . Example 4 says that F36 has 12 more elements than F35 .

1.4 Properties of the Farey Sequence

a c a c
If we have two fractions b
and d
with the properties that b
< d
and |bc − ad| = 1
Then the fractions are known as Farey neighbours, they appear next to each other
in some Farey sequence. The mediant or Freshman’s sum of these two fractions
is given by
aMc a+c
= .
b d b+d
a c a+c
Theorem 8 (Mediant Property). If b
< d
then their mediant b+d
lies between
a a+c
them, b
< b+d
< dc .

Proof.

a+c a bc − ad c a+c bc − ad
− = > 0 and − = >0
b+d b b(b + d) d b+d d(b + d)

Proof comes from Tom M. Apostol [5].

A consequence of this is that if two fractions in a Farey sequence are Farey
neighbours they will remain so until their mediant separates them in a later Farey
0 1
Sequence. For example 1
< 2
are Farey neighbours in F2 , and their mediant is
1
3
which separates them in F3 . This is an important property that will feature
throughout the report.

8
Remark 9. The mediant always generates simplified fractions. So for example
2
4
will not be generated by taking mediants.

a c a c
Theorem 10 (Neighbours Property). Given 0 ≤ b
< d
≤ 1, b
and d
are Farey
neighbours in Fn if and only if bc − ad = 1.

Proof. If pq , a
b
and c
d
are in some Farey sequence, with a
b
< p
q
< c
d
and bp − aq =
qc − pd = 1, then

bp + pd = qc + aq

p(b + d) = q(a + c)
p a+c
=
q b+d

a p p c
Hence b
and q
are neighbours and q
and d
are neighbours.
We will prove the converse by induction.
First F1 = { 01 , 11 } and |bc − ad| = |1 − 0| = 1 so the result is true for n = 1.
Assuming the result for Fn we next prove that the result follows for Fn+1 . We have
Fn = {. . . ab , dc . . .} and |bc−ad| = 1. If b+d ≤ n+1 then Fn+1 = {. . . ab , (a+b) , c . . .}
(c+d) d

and |b(a + c) − a(b + d)| = |bc − ad| = 1. The only other possibility is that
b + d > n + 1 in which case Fn+1 = {. . . ab , dc . . .} with |bc − ad| = 1, so the result
is true for n + 1.
Hence, by the axiom of induction the result is true for all positive integer
values of n.

Proof adapted from NRich [6].

1.5 Farey History

So how did this list of simple or “vulgar” fractions become known as the Farey
sequence? As with many things in life, history may have been too kind to those
who were simply remembered, and too harsh to those forgotten.
Whilst Mr Flitcon’s solution got the precise number of elements of F99 (not
including 0 and 1), it did not give us an explicit formula to find those elements,

9
or indeed even list them.
The French Revolution acted as an unlikely catalyst for the first ever publish-
ing of F99 , but when France’s new regime legislated that the whole country was
to switch to the metric system instead of imperial measurements in 1791 it fell to
Charles Haros to create a mathematical table to convert between fractions and
decimals.
Published in “Journal de l’Ecole Polytechnique”, Haros’ table contained every
irreducible fraction with denominators from 2 to 99 and their decimal approxima-
0
tion. Aside from including 1
and 11 , Charles Haros had formulated F99 . To do this
he used the mediant property to find the fractions with higher denominators and
even provided a sketch proof that it worked. He also noted that if two numbers
a c
b
and d
are neighbours in the table then |bc − ad| = 1.

From here the history of the Farey sequence travels to Britain, and to a man
called Henry Goodwyn. Henry Goodwyn ran and owned a brewery and made
mathematical tables in his spare time. In his retirement he set out (much like
Charles Haros) to create a table of fractions and decimal equivalents. However,
Goodwyn’s tables were to contain every irreducible fraction with denominators
between 1 and 1024. His paper “The First Centenary of a Series of concise and
useful Tables of all the complete decimal Quotients which can arise from dividing a
Unit or any whole Number less than each Divisor by all Integers from 1 to 1024”
was presented to the Royal Society on 25th April 1816, having had a private
printing circulated a year previously.
Less than a month after Henry Goodwyn’s presentation to the Royal Society
a geologist by the name of John Farey wrote into The Philosophical Magazine
and Journal with a note entitled “On a curious Property of the vulgar Fractions”.
In the note, John Farey pointed out a “curious property”, but offered no proof
(see a later remark). He finished the letter as in Figure 1.2.
John Farey’s note was then republished in the French magazine “Bulletin de
la Société Philomatique”. It was from here that Cauchy saw it and provided a
proof that the mediant property holds (crediting John Farey) in August 1816 [7].

10
 Figure 1.2: Excerpt from Farey’s Letter

1.6 Length of the Farey Sequence

Definition 11. The length of a given Farey sequence is given by the recursion
formula
|Fn | = |Fn−1 | + φ(n)

Where φ(n) is Euler’s Totient Function, see Theorem 3.

Theorem 12. The length of the Farey sequence behaves asymptotically with

3n2
|Fn | ∼
π2

The following three graphs represent what the length of the Farey sequence
behaves asymptotically with, what the length of the Farey sequence actually is
along with the error between the two. A code in M aple was written to calculate
the exact way that the function was behaving.

Figure 1.3: Approxi- Figure 1.4: Exact Figure 1.5: Error in Ap-
mated Length Length proximation

11
Chapter 2

Clocks and Farey

Mathematics has always been important in forming time-based instruments such

as calendars and plays an important part in the making of clocks. Clockmakers
not only borrowed some of the ideas from mathematics but actually developed
some of their own that then became part of mathematics. This resulted in the
creation of the Stern-Brocot tree, which is surprisingly useful in the construction
of clocks.

2.1 Creation of the Stern-Brocot tree

The Stern-Brocot tree was first described by Moritz Stern in 1858, who explained
it’s relation to other areas in number theory. Clock maker Achielle Brocot, dis-
covered the tree independently in 1861[8] and used it practically to approximate
gear ratios, but never realised it to be of any other mathematical significance.
We saw earlier how the Farey sequence is constructed using Farey neighbours
and mediants, when this process is extended to the whole real line we get the
Stern-Brocot tree. A similar process of mediant insertion, starting with a different
pair of interval endpoints [ 10 , 10 ], may also be seen to describe the construction of
the vertices at each level of the Stern-Brocot tree.[9]
To construct the Stern-Brocot tree, we need to recall the mediant of two
rational numbers to be

12
 a a+c c
< <
b b+d d
0
We start to construct the tree in steps beginning with the two fractions 1
1 1
and 0
. It is useful to think of 0
here as representing infinity. We then con-
tinue the construction forming the next level by inserting the mediant of any two
consecutive rational numbers that are already in the tree as in Figure 2.1.

Figure 2.1

And we continue to add the mediants in Figure 2.2.

Figure 2.2

And eventually we end up with a tree like in Figure 2.3.

13
 Figure 2.3

Notice that this has constructed a binary search tree with the Farey sequence
0 1 1
between 1
and 1
and a periodic extension of the Farey sequence between 1
and
1
0
. When a rational number r appears, it is the Farey neighbour of two rationals.

Definition 13. An Ancestor of a number in the Stern-Brocot tree is any number

that appears above it in the same branch.
A Child of a number in the Stern-Brocot tree is a number that appears directly
below it. Each number has a Right Child (R) and a Left Child (L).
Moreover, the Left Child of any number is the mediant of that number and its
first Ancestor to the left (and first Ancestor to the right for the Right Child).

1
Example 14. 3
has Left Child 41 , Right Child 2
5
and Ancestors 12 , 11 , 10 and 10 .

The Left Child of any number is the mediant of that number and its first
Ancestor to the left (and first Ancestor to the right for the Right Child).

3 3 1
Example 15. The Right Child of 4
is the mediant of 4
and 1
(the first ancestor
to its right) = 45 .

Every rational number appears in this tree exactly once as the new rationals
added are always between the consecutive numbers that have already appeared.
Thus the Stern-Brocot sequences differ from the Farey sequences in two ways.
It includes all positive rationals, not just those within the interval [0,1], and at

14
the nth step all mediants are included, not only the ones with denominator less
than or equal to n. The Farey sequence of order n may be found by an in order
traversal of the left subtree of the Stern-Brocot tree, backtracking whenever a
number with denominator greater than n is reached.

2.2 Navigating the Tree

Now we will see how the tree can be navigated and show how it contains every
number. We will start with an example.[10]
Where can we find 47 ?
The construction always begins with the fractions

0 1 1
, , .
1 1 0
We will begin at the fraction 11 . Since 0
1
< 4
7
< 11 , we will move left to the
0
Farey neighbour of 1
and 11 , which is 1
2
as shown in Figure 2.4. This is a move to
the Left Child of 11 , denoted L.

Figure 2.4

1 4 1
Now we have 2
< 7
< 1
so we will move to 23 , as in Figure 2.5. We have
1 1 2
moved to the Right Child (R) of 2
. The path from 1
to 3
can be described as
LR.
1 4
In the same way we now have 2
< 7
< 32 . So we move to the left to 35 , and
the path is LRL. With one more step to the left, we arrive at 47 , as in Figure 2.6,
which we may represent as LRLL.
How does this show that every positive rational appears in the tree? If we are
p
looking for the rational r. At every step, r is always between two rationals q
and

15
 Figure 2.5

Figure 2.6

p0 p p0
q0
. If r is equal to the mediant of q
and q0
, then we have found r. If not we move
from the mediant to the left if r is less than the mediant and right otherwise. If
we never find r then this process will continue indefinitely, and the denominators
p p0
of the rationals q
and q0
that bracket r grow will arbitrarily large.
Let’s illustrate this geometrically. Since q and q 0 grow arbitrarily large, the
p p0
points (q, p) and (q 0 , p0 ) are eventually to the right of r. And since q
<r< q0
, it
follows that r lies in the parallelogram as shown in Figure 2.7.

Figure 2.7

Remember that the area of the shaded parallelogram above is qp0 − pq 0 = 1.

p
Consider now the area of the parallelogram defined by q
and r, shaded darkly in

16
Figure 2.8.

Figure 2.8

The area of the dark parallelogram must be a positive integer, obeying the
mediant property, that is less than 1. Since this is impossible, r must have
appeared somewhere earlier in the tree.
We now have a way to label strings of positive rational numbers with strings
of right and left movements L’s and R’s.
If we consider infinite strings of L’s and R’s we see that they correspond to
positive irrational numbers. If we have a positive irrational number α we can find
it’s corresponding string with the following algorithm:[11]

1
1. Begin at r = 1
and let S be the empty string.

2. If α < r, replace r by its Left Child and replace S by S=L. Otherwise,

replace r by its Right Child and S by S=R.

3. Repeat step 2.

In this way we can find some well known strings, for example, S = RLRL-
1
RLRLRLR.... Here following the tree from 1
we move to 21 , 32 , 53 , 58 ,.... Since we
are alternating from right to left moves we are always finding the mediant of the
previous two terms in the tree and this should be recognised as ratios of consec-
utive Fibonacci numbers. So we can see that this converges to the golden ratio
and it’s string is Φ = RLRLRLRLRLR....
The key use of the Stern-Brocot tree in clock making is that any sequence
found in this way is optimal in the following sense. Given any rational approxi-

17
mation to a rational number x, the sequence usually contains an element that is
a more accurate approximation with a smaller numerator and denominator.
We will show why this property holds by considering the process in which
we create a sequence that converges to x. At every step there are consecutive
p p0
rationals q
and q0
that bracket x. Since r is a rational number different from x,
eventually r will not be bracketed by these two rational numbers.
p p0 p00
Now consider the last step at which r is bracketed by q
and q0
. If q 00
is the
next mediant obtained in the process, then it must lie between r and x.
p00
Now r will be found in the Stern-Brocot tree under q 00
, which means that the
numerator of r is no smaller than p00 and the denominator is no smaller than q 00 .
p00
Thus, q 00
is closer to x than r is and the numerator and denominator are no larger
than r’s.
We now also explain why the rationals that appear in the tree are expressed
in their simplest form. i.e If we have the ratio pq , then p and q have no common
sp
factors. Suppose instead that the rational sq
with s > 1. Suppose that the
a sp
fraction b
is the rational that follows sq
when it appears in the tree. Then we
have |sqa−spb| = s|qa−pb| = 1 (as they are Farey neighbours) which is impossible
if s > 1 which implies that the rationals in the tree are in their simplest form.

2.3 Gear Ratios

So why would a clock maker be interested in this tree?
Clocks typically have a source of energy, such as a spring, a suspended weight,
or battery that turns a shaft at a fixed rate. If the clock has a minute hand and
an hour hand we need some mechanism that will speed or slow down the motion
of the shaft as it is transferred to the hands.
Gears are used to slow the motion of the shaft. In Figure 2.9, the small green
gear, with 20 teeth, drives the blue gear, with 60 teeth. In the clock making
language the smaller gear is called a pinion and the larger gear is a wheel.
Every time the green pinion advances by one tooth, so does the blue wheel.
20 1
Therefore, one revolution of the green pinion produces 60
or 3
of a revolution of
the blue wheel or the green wheel turns three times to turn the blue wheel once.

18
 Figure 2.9

If we wish to have a shaft that rotates once a second that drives a minute
1
hand, we could use a pinion and wheel whose ratio of teeth is 60
. But there is
another option. Figure 2.10 illustrates a gear train; the green pinion turns the
blue wheel slowing the speed by a factor 16 . However, the blue wheel turns the
red pinion at the same rate, and this pinion, with 10 teeth, turns the gray wheel,
1
with 100 teeth thus slowing the speed by another factor of 10
. Therefore, the
1 1 1
overall ratio of the speed of the green pinion to the grey wheel is 6
× 10
= 60
.

Figure 2.10

19
 In theory any number of stages can be added to a gear train. For example we
could add a pinion on the grey wheel that can drive another wheel.
To illustrate how this is used in clock making, imagine that we have a shaft
that rotates once per minute and we wish to drive an hour hand, which rotates
once every 12 hours.[9] The ratio of these speeds is 1/720. The most obvious
solution is a single pinion with one tooth to drive a wheel with 720 teeth or a
pinion with 10 teeth driving a wheel with 7200 teeth. But the larger gears would
be impractical, so instead we use a gear train:

1 1 1 1 1
= = × ×
720 10 × 8 × 9 10 8 9
The key to this is that we can easily factor 720.
Consider the situation where we require a shaft that rotates once every 23
minutes to turn a wheel that rotates once every 191 minutes (an example Brocot
used in his paper).[13] Since these are both prime, it is not possible to use this
required ratio as a gear train.
First we shall approximate 23
191
with pq . So we have a pinion with p teeth on
the shaft and it’s turning a wheel with q teeth. The pinion makes one revolution
every 23 minutes. This means that the wheel turns once every 23 pq minutes,
making the error

q 23q − 191p E( pq , 191

23
)
23 − 191 = = minutes.
p p p
Where E is the mediant property of the Farey sequence mentioned earlier,
E( ab , dc ) = |bc − ad|, which is being used in this case as a way to measure the error
of an approximation.
We may easily compute E( pq , 191
23
) as we descend the tree. Since 23
191
is between
1
8
and 19 , Brocot began by making a table like this:
p q E p q E
1 9 16 1 8 -7
.. .. ..
2 17 9 . . .
.. .. .. .. .. ..
. . . . . .
23 191 0

20
 23 2
This shows that if we were to approximate 191
by 17
, the error will result in
2 23 23·17−191·2 9
the wheel taking E( 17 , 191 )/p = 2
= 2
= 4.5 too many minutes to rotate.
Brocot then continued in this way until the completed table looks like this:
p q E p q E
1 9 16 1 8 -7
2 17 9 4 33 -5
3 25 2 7 58 -3
13 108 1 10 83 -1
23 191 0
23
Brocot’s algorithm reveals that the closest approximations to 19
are ratios of
10 13
83
(which runs a tenth of a minute fast) and 108
(a thirteenth of a minute slow).
It is possible to do better than this using a gear train and Brocot originally did
this taking mediants between the approximation and the exact fraction. However
finding the fraction that had a nicely factorisable numerator and denominator was
trial and error. He realised that all the work for this could be done beforehand
and wrote a table showing all fractions with numerator and denominator less
than 100, ordered in magnitude.
Here is an example of a basic problem from Camus’ “A treatise on the teeth
of wheels”[14]. “To find the number of the teeth...of the wheels and pinions of a
machine, which being moved by a pinion, placed on the hour wheel, shall cause
a wheel to make a revolution in a mean year, supposed to consist of 365 days, 5
hours, 49 minutes.”
720
Multiplying out the days and years we find that we need a ratio of 525949
.
The numerator factors well but we have a problem with the prime denominator.
So we need to find an approximation to this fraction, but with that both the
numerator and denominator that has small factors. Camus original solution to
this was a series of repeated trials which Brocot thought was defective and so
solved it similarly as above which is more efficient.
In this case the error in approximation is E = q(720) − p(525949).

21
 p q E p q E
0 1 720 1 0 -525,949
.. .. .. .. .. ..
. . . . . .
33 24106 3 163 119069 -7
262 191387 2 196 143175 -4
491 358668 1 229 167281 -1
720 525949 0
196
We choose the fraction to be 143,175
as it can be factored nicely and both
numbers in the fraction would have appeared the Brocot table mentioned earlier.
2 2 7 7
We can then make this into a gear train of 3
× 25
× 23
× 83
and the error in the
4
gear train is 196
. This results in a gear train that is just over a second too fast.
Not a bad approximation at all.
Methods like this are hardly used these days as there is so much computing
power. A gearing problem can be solved using brute force that would have been
horrific in the times of Camus or Brocot. If you need to approximate a ratio have
a computer try all pairs of gears with no more than 100 teeth. There are 10,000
combinations which a computer can churn out in seconds. For a two-stage gear
train, running through the 100 million possibilities takes a computer minutes. As
is said by Hayes in “On the teeth of wheels,” “The whirling gears of progress have
put the gear-makers out of work.”[12]

22
Chapter 3

Approximation and Fibonacci

3.1 Approximating irrationals using Farey Se-

quences
Given an irrational on the interval [0, 1], we can find a rational approximation
using the Farey sequences, as mentioned in the previous section. We can choose
to limit the size of the denominator in our approximation, by restricting the size
of the Farey sequence, Fn , i.e. we restrict the value of n. For example, we are
1
asked to find an approximation of π
= 0.318309886183791 where the denominator
32 8
is no larger than 100. The first and easiest approximation is probably 100
= 25
.
But is there a more suitable approximation?
It is possible to use the Farey sequence as a tool for approximation by using
the mediant.
As we saw with the Stern-Brocot section, given an irrational number between
0 and 1 it is possible to find a rational approximation using iterations of the
mediant in the Farey sequence. See for example, Cook [15].
1
Example 16. Look for the closest rational approximation of π
where the de-
nominator is less than 100. One can iterate using Theorem 8 to find the closest
rational approximation. That is to say that if given an interval in which the ratio-
nal number is known to be present, the mediant of that interval will provide two
smaller intervals. The rational number will then lie in one of the intervals upon
which the mediant is carried out a further time. This provides a way of getting

23
closer to the rational.
0 1

The first interval is the first Farey sequence, in other words ,
1 1
. Its cor-
responding mediant is 21 . The next step is to decide whether π1 lies in 01 , 12 or

1 1
, . Since π1 < 12 the new interval becomes 01 , 12 and the current approxi-

2 1

mation is 12 . With this interval a new mediant can be found and subsequently a
new approximation. The new interval is 01 , 21 and its corresponding mediant is


1
3
. These steps are repeated until the denominator of the new approximation lies
just under 100 and a further iteration forces the denominator to be more than
1 29
100. After a few steps the rational approximation of π
is 91
. It may be noticed
7 29
that 22
is still a better approximation than 91
and therefore shows that despite
the fact that a new approximation is found after each step, the best approxima-
tion is not necessarily the newest fraction found after each iteration. The next
7 57
approximation which is more accurate than 22
is in fact 179
.
The one big drawback of this process is that in order to find the most suitable
rational approximation the exact value of the irrational to be approximated needs
to be known in order to choose the correct intervals. This process searches for
fractions around the irrational but does not necessarily find a better one after
each iteration.
As we have seen in the previous section, the Stern-Brocot tree has been made
redundant using developments in computing. We wrote a M atlab code which
takes as its input an irrational number to be approximated and the limit of the
denominator at which to stop. The code is shown in the Appendix. Figure 6.3.
The following is the output of the Matlab code:

24
 1
Figure 3.1: Output from Matlab for approximating π

This code outputs the limit of the iteration process, its path down the Stern-
Brocot tree, the best fractional approximate, its path down the Stern-Brocot tree
and how long the iteration took. It took M atlab only 0.000183 seconds to find
an approximate for π1 . This code is a modern way of using the Stern-Brocot tree
to approximate values. The code only considers the left-hand branch of the tree
because it only looks at Farey fractions and hence fractions between 0 and 1.

3.2 Farey Sequences of Fibonacci Numbers

The Fibonacci sequence is given by ϕ = 1, 1, 2, 3, 5, 8, 13, 21, 34, 55, . . . . The next
term in the sequence is the sum of the previous two, ϕm = ϕm−1 + ϕm−2 .

Definition 17. Define the sequence of Fibonacci fractions as:

, , , , . . . , ϕϕm+2
1 1 2 3
2 3 5 8
m
, ϕϕm+1
m+3
,...

1 1
We know from F3 that 2
and 3
are Farey neighbours.

25
Theorem 18. It can be shown that any two neighbouring fractions in the sequence
of Fibonacci fractions are neighbours in the Farey sequence.

ϕn ϕn+1
We need to show that for all n, ϕn+2
and ϕn+3
are Farey neighbours. That is
|ϕn+1 ϕn+2 − ϕn ϕn+3 | = 1.
P roof by induction.
Take n = 1:
|1 ∗ 2 − 1 ∗ 3| = 1
Suppose it is true for n = k
Take n = k + 1:
|ϕk+2 ϕk+3 − ϕk+1 ϕk+4 |

Since ϕn = ϕn−1 + ϕn−2

|ϕk+2 ϕk+3 − ϕk+1 ϕk+4 | = |(ϕk+1 + ϕk )ϕk+3 − ϕk+1 (ϕk+2 + ϕk+3 )|

= |ϕk+1 ϕk+3 + ϕk ϕk+3 − ϕk+1 ϕk+2 − ϕk+1 ϕk+3 |

= |ϕk ϕk+3 − ϕk+1 ϕk+2 |

Therefore by induction the statement is true for all n. End of proof . [17]

26
Chapter 4

Ford Circles

4.1 Introduction
The aim of this part of the project is to explore how circles - arguably the most
studied objects in mathematics - are intimately linked to the rationals in a non-
obvious way. The usual way to represent rationals geometrically is as points on
the real numberline. We will look at a different geometric representation of the
rationals, as described by Lester Ford [18]. We will discuss some of the properties
of this geometry as well as using a group theoretic approach to describe it. We
will see that these circles are in bijection with the Farey sequences.

4.2 Motivation and definition

Throughout this section we will continue to consider sets of rational numbers on
p
the interval [0, 1] in their simplified form. For each simplified rational number q
p
we can define a circle in the upper-half plane which is tangent to the point q
on
the real line.
We can see that these circles intersect.
1
If we define the radius of the circles as 2q 2
we get some interesting results -
all the circles which intersect do so trivially, that is, any two circles are either
disjoint or tangent.
1
Figure 4.3 is an image of the seven largest such circles with radius 2q 2
on the

27
 √
Figure 4.1: r = 1/q 2 Figure 4.2: r = 1/ 3q 2

interval [0, 1] .

Figure 4.3: The seven largest Ford circles, taken from our Maple worksheet

We call sets of circles of this form, as in Figure 4.3, sets of Ford circles,
named after the American mathematician Lester Ford, who worked at Edinburgh
University in 1914 [20].

Remark 19. The images from Züllig’s paper [19] suggest a motivation for choos-
1 1
ing the diameter to be q2
. We can see that as the radius approaches 2q 2
the cir-
cles approach tangency. This is not the motivation that Ford had for working
with these sets of circles - a brief description of this will come at the end of the
chapter.

We will use the obvious notation to represent a Ford circle:

Definition 20. Let the circle C( pq ) be defined as the circle lying in the upper-half
p 1
plane tangent to the point q
on the real line with radius 2q 2
. Each such circle is

28
a Ford circle.

Example 21. The green circle in Figure 4.3 is called C( 12 ).

Definition 22. Cn is the set of Ford circles which contains all the circles with
1
diameter ≥ n2
.

Example 23. Figure 4.3 represents the set of Ford circles C4 .

4.3 Properties of Ford circles

We can see intuitively how Ford circles are related to the Farey sequence. For
example the circles are symmetric about 21 , as are the elements in Fn . We will
proceed to show that there exists a bijection between the Ford circles and the
Farey sequence. This result lies close to the heart of mathematics - linking number
theory and geometry.

Proposition 3. The set of Ford circles Cn is in one-to-one correspondence with

the Farey sequence Fn .

Example 24. If we look back to Figure 4.3 we can map the Ford circles one-to-
one with the following Farey sequence (from left to right according to colour):
F4 = { 10 , 14 , 13 , 12 , 23 , 34 , 11 }

Just as with the Farey sequence, we can generate the set of Ford circles it-
eratively (“in layers”). The set of circles Cn is precisely the set Cn−1 in union
1
with the set of circles of radius 2n2
. In other words, the “new elements” in each
Farey sequence (the ones in Fn which are not in Fn−1 ) are given by the set of
Ford circles which have the smallest radius in the diagram for Cn .

Example 25. Again, looking back to Figure 4.3 we can see the set of circles C1
in red, the set C2 − C1 in green, the set C3 − C2 in blue and the set C4 − C3 in
black.

Here are the two main properties of the Farey sequences which we saw in
Chapter 1 - the mediant property (Theorem 8) and the neighbours property
(Theorem 10) expressed in geometric language of Ford circles:

29
 Figure 4.4: Geometric proof of neighbours

Theorem 26 (Neighbours property for Ford circles). The intersection of two Ford
circles is a singleton iff |bc − ad| = 1. That is, we have an analogous definition
to Farey neighbours and thus we can describe two circles as Ford neighbours iff
|bc − ad| = 1.

We have seen the algebraic proof in the context of Farey sequences. Now let’s
look at a geometric proof.

Proof. Consider two tangent Ford circles C( ab ) and C( dc ) with a

b
< dc . As in Figure
4.4. Recall that we know the radii of the circles in terms of a,b,c, and d. Consider
the right-angled triangle DEF , where D is the center of C( ab ) and F is the center
of C( dc ) and E marks the intersection of the horizontal line through F and the
vertical line through D. Since we know the radius of the circles we can calculate
1
the coordinates of D, E and F . We have that length E = 2b2
− 2d12 , D = 1
2b2
+ 2d12 ,
c
and F = d
− ab . From the Pythagorean Theorem we can generate the identity:
F 2 + E 2 = D2 which can be written algebraically as ( dc − ab )2 + ( 2b12 − 2d12 )2 =
( 2b12 + 1 2
2d2
) If we expand the brackets and cancel like terms we get c2 b2 + a2 d2 −
2acbd − 1 = 0 and we can factorise this to (ad − bc)2 = 1 Hence it follows that
|bc − ad| = 1, as required.
The argument is reversible. Assume that |bc − ad| = 1 holds. Using the usual

30
notation for lines, DE and EF are defined according to the points described
above and the line DF (the line connecting the centers of the two circles) will
be equal to the sum of their radii. Which directly implies that the circles are
tangent to each other.

The above proof is based on the proof which can be found on Cut the Knot
[21].

Example 27 (Computing the lengths). Consider the right-angled triangle whose

vertices are the center of C(1), the center of C( 12 ) and the point (1, 81 ). Recall
c
that C(1) is tangent to the real line at 1 which in this example is d
and C( 12 ) is
1
tangent at 2
= ab .
We can compute the lengths of the sides of the triangle to be 48 , 3
8
, 58 .

Remark 28 (Ex-neighbours in the Farey sequence). We can see from the theorem
above that the Ford circles contain more information than the Farey sequences.
In the general Farey sequence Fn there is no way of telling from the written set
0 1
of fractions whether 1
and 1
were once neighbours (they were, in F1 ). In the
diagram for the general set of Ford circles Cn we can see that C(0) and C(1)
are tangent and hence were once neighbours in a Farey sequence. We could say
that the information about “ex-neighbours” is preserved in the diagrams of Ford
circles.

Theorem 29 (Mediant property for Ford circles). If C( ab ) and C( dc ) are tangent

Ford circles (i.e. they are Ford neighbours) then C( a+c
b+d
) is the unique circle
tangent to the real line and both the circles. That is, C( a+c
b+d
) is the mediant Ford
circle.

Proof. For any fraction we can define a Ford circle. Hence the mediant Ford
circle exists. Since |bc − ad| = 1 implies that both (b + d)c − (a + c)d = 1 (i.e.
the equation obtained by plugging in the values of the two kissing circles C( a+b
c+d
)
and C( dc ) and (a + c)b − (b + d)a = 1 (i.e. the same equation again describing
the kissing circles C( a+b
c+d
) and C(a/b), it is clear that if there is a mediant Ford
circle then it touches the other two circles iff they are tangent to each other.

This proof is adapted from Cut the Knot [21].

31
 Figure 4.5: The right-angled triangles represent distinct pythagorean triples

Remark 30 (Pythagorean triples). If we look back to example 27, the sides of

the triangle correspond to the primitive Pythagorean triple 3, 4, 5.
There is a unique (up to the symmetries of the diagram) primitive pythagorean
triple for any pair of Ford neighbours. A description of how to find all primitive
Pythagorean triples can be found at Cut the Knot [22].

Remark 31 (Ladies’ diary and a geometric picture of Q). Note that if we look
back to the problem in the Ladies’ Diary then by the neighbours property of Ford
circles we have shown that each of the 3005 circles contained in C99 are either
tangent or disjoint. In general, there is a Ford circle for every rational. Fur-
thermore, recall the fact that the mediant generates all of the rationals and the
mediant never produces a fraction which is not in simplified form e.g. we never
produce 42 . If we start with C(1) and C(0) and a straight line then we can generate
the rational points on the real line by constructing Ford circles.

4.4 From algebra to geometry and back to alge-

bra
In this section we will sketch a group theoretic approach to the analysis of Ford
circles. We will see that this approach leads immediately to some new results. The
following approach is described by Carnahan [23]. This section assumes the reader
has some familiarity with Groups, Group actions and Möbius transformations.
Recall that the matrix group SL2 (Z) is the set of 2 × 2 matrices with integer

32
 Figure 4.6: This image depicts the Ford circle C( 10 ) in black

coefficients such that |bc − ad| = 1. The elements of the rows and columns of the
matrices are coprime.

Definition 32. The Ford circle of infinite radius C( 01 ) is equal to the line R + i.

We will proceed to consider the Ford circles as curves in the complex plane
rather than in the Cartesian plane as before.

a c


Proposition 4. Given a matrix b d ∈ SL2 (Z) the corresponding Möbius trans-
formation maps C( 01 ) to C( ab ).

This proposition can be verified directly: take a point x + i which lies on the
line R + i = C( 10 ). The point maps to z = ax+ai+c
bx+bi+d
, which lies on C( ab ) whose
center is ( ab , 2b12 ) and whose radius is 1
2b2
hence the equation of the circle is given
by |z − ( ab + i( 2b12 ))|2 = ( 2b12 )2 and indeed it can be checked that the points lie on
this circle, see Carnahan [23].

Proposition 5. The action induced on two Ford circles C( ab ) and C( dc ) preserves

|bc − ad| = 1

The above propositions shows that Ford circles map to Ford circles. With
these two results we can succinctly prove the two key properties of Ford circles
(the neighbour and mediant properties). We will look at the latter here, the
former is covered by Carnahan [23].

Theorem 33. The mediant property holds for Ford Circles.

33
 Figure 4.7: The three Ford circles used in the proof of Theorem 33

c a
Figure 4.8: The semicircle whose diameter is d
− b

Proof. We can consider the neighbours C( 10 ) and C(0) as well as their mediant
C(1). Since we are working with groups and groups actions it is enough to show
that ab dc maps these circles to C( ab ), C( dc ) and C( a+c


b+d
) respectively, which we
have already demonstrated in the proposition above. Hence we have the required
result.

What are the advantages of this group-theoretic approach? There are three
advantages:

1. We can find the points of intersection of any two intersecting Ford circles
immediately. From observing that C(1) ∩ C(0) = i we can conclude that

34
 C( ab ) ∩ C( dc ) = ai+c
bi+d

c a
2. We can see that the semicircle whose diameter is d
− b
goes through the
point of intersection of the two circles C( ab ) and C( dc ), see Figure 4.8. This
ultimately gives rise to an interesting graph (the Farey Diagram). See, for
example, Hatcher [24].

3. We can use Ford circles to express the irrationals by continued fraction

chain of rational Ford circles. We have a natural way to do this which Ford
discusses in his paper Fractions (1938) [18]. This is related to the earlier
section in this report on irrational approximations. Ian Short provides a
method for constructing these approximations and suggests a way to extend
this approach to approximating irrational complex numbers with rational
complex numbers [25].

When we construct irrational complex numbers from rational complex num-

bers in the way described above it requires a definition of Ford Spheres, which
are described in Ford’s paper, Fractions (1938) [18].
Ford first found the Ford spheres and then simplified these spheres to the
two-dimensional version - the Ford circles. Which indeed provides the historically
accurate motivation for the Ford circles, as promised earlier in the report. A more
detailed account of Ford’s discovery can be found in the introduction to his 1938
paper [18]. Disclaimer! Ford circles were found in ancient Japanese Sangaku
mathematics too [26] .

35
Chapter 5

The Farey Sequence and The

Riemann Hypothesis

The Riemann Hypothesis is one of the most famous mathematical problems in

recent history. It has remained unproven since Bernhard Riemann (1826-1866)
first conjectured it in his 1859 paper, “Ueber die Anzahl der Primazahlen unter
einer gegebenen Grösse” (On the Number of Primes Less Than a Given Magni-
tude) and is one of the Clay Mathematics Institute’s millennium problems, with
$1,000,000 being awarded to anyone who can prove it. [27]
As is obvious from the title, Riemann wanted to find the number of primes
less than a given number. To do this he turned to the zeta function (now called
the Riemann zeta function).

Definition 34. The Riemann Zeta Function is defined for all s ∈ C\{1} as
follows:
∞
X 1
ζ(s) =
n=1
ns

The Zeta function converges absolutely for all complex s with Re(s) > 1.

1 1 1
• ζ(1) = 1
+ 2
+ 3
+ · · · = ∞ (the harmonic series), thus s = 1 is a simple
pole

• ζ(s) = 0 when s = −2, −4, −6, . . . , these are known as the trivial zeros

The trivial zeros occur at s = −2, −4, −6, −8, . . . because of the relation between

36
the zeta function and Bernoulli numbers[28]:

−Bn
ζ(1 − n) = , n ∈ N, n ≥ 2
n

Where Bn = 0 for odd n ≥ 3.[29]

The Zeta function is related to the distribution of primes by equality to the
Euler Product (note p is prime):

Y 1 1 1 1
ζ(s) = −s
= −s
· −s
· ...
p
1−p 1−2 1−3 1 − 5−s

The Riemann Hypothesis All the non-trivial zeros to the Riemann zeta func-
1
tion lie on the critical line s = 2
+ it

It is important to note that this does not mean every point on this line is a
zero, but that every non-trivial zero lies on it for some t. 1,500,000,000 solutions
to ζ(s) = 0 have been found to lie on the critical line (and no where else), but a
formal proof for all cases has never been found.[27]
Whilst having never been proved, equivalent statements to the Riemann hy-
pothesis exist. One such statement is Mertens conjecture.

Notation (Little o): f (x) = o(g(x)) means f (x) has a smaller rate of growth
than g(x). In other words, for all constants C > 0:

|f (x)| ≤ C · |g(x)|

Which is equivalent to
f (x)
lim =0
x→∞ g(x)

Mertens Conjecture
Before stating Mertens conjecture, the Möbius Function and Mertens Function
must be defined.

37
Definition 35. Möbius Function

 0 if k is not square free
µ(k) =
 (−1)p if k is the product of p distinct primes

A number is square-free if it does not contain any squares in its prime decom-
position (all its prime factors are unique). For example 18 = 32 · 2, thus is not
square-free.

• µ(1) = 1

• µ(2) = (−1)1 = −1 as 2 is prime

• µ(4) = µ(22 ) = 0 as 4 is not square-free

• µ(6) = µ(3 · 2) = (−1)2 = 1 as 6 is the product of two distinct primes

Definition 36. Mertens Function

X
M (n) = µ(k) (5.1)
k≤n

• M (2) = µ(1) + µ(2) = 1 + (−1) = 0

• M (3) = µ(1) + µ(2) + µ(3) = 1 + (−1) + (−1) = −1

• M (6) = 1 + (−1) + (−1) + 0 + (−1) + 1 = −1

Mertens function moves slowly, only able to increase or decrease by 1 at the

most at each step, and there is no n such that |M (n)| > n.
In 1885, Dutch mathematician Thomas Joannes Stieltjes conjectured that
1
there was no n such that |M (n)| > n 2 , yet a hundred years later in 1985 this was
proven false by Andrew Odlyzko and Herman te Riele.[30]
However, the Riemann Hypothesis is equivalent to the weaker conjecture that:

1
M (n) = o(n 2 + ) (5.2)

∀ > 0. In other words, for all C:

38
 1
M (n) ≤ Cn( 2 +)
1
Meaning |M (n)| is bounded by n 2 + . It is using this that the equivalent state-
ment to the Riemann hypothesis using the Farey sequence is drawn.

5.1 Equivalent statement to the Riemann Hy-

pothesis using the Farey Sequence
As with Mertens conjecture, before we can draw the equivalent statement, it is
necessary to make a few definitions.

Definition 37. Let L(n) be the length of the Farey Sequence Fn and rv be the
v th Farey term. We define the difference, δv in the following way:

δv = rv − v/L(n) (5.3)

Example 38.

0 1 1 1 2 1 3 2 3 4 1
F5 = { , , , , , , , , , , }
1 5 4 3 5 2 5 3 4 5 1
L(5) = 11
0 1 1
δ1 = − =−
1 11 11
1 2 1
δ2 = − =
5 11 55
..
.
1 11
δ11 = − =0
1 11

In 1924, the Franel-Landau theorem was published[4], stating that:

L(n)
1
X
|δv | = o(n 2 + ) (5.4)
v=1

39
∀ > 0 and where n refers to Fn , which is equivalent to the Riemann hypothesis.
This is the link between the Farey sequence and the Riemann hypothesis. To
show its equivalence to the Riemann hypothesis, the paper contained the proof
that:
L(n)
X
|δv | = o(n1/2+ ) ⇔ M (n) = o(n1/2+ ) (5.5)
v=1

Thus, it is true if and only if Mertens conjecture is true, which is equivalent to

the Riemann hypothesis. We will follow the proof of this statement from H. M.
Edwards’ ”Riemann’s Zeta Function”[32], following it from left to right, then
right to left.
The key to the proof is to relate δv to M (x). If that is achieved then we can
start to prove 5.5.
To do this, we first look at a real-valued function defined on the interval [0, 1].
Let rv denote the elements of the Farey sequence, as above. The sum of the
function at the points in the Farey sequence can be related to Mertens function
by the following equality:

L(n) ∞ X
k
X X j n
f (rv ) = f ( )M ( ) (5.6)
v=1 k=1 j=1
k k

5.2 Forward Proof of Equivalent Statement

In this section, we want to show:

L(n)
X
|δv | = o(n1/2+ ) ⇒ M (n) = o(n1/2+ )
v=1

Taking formula 5.6 and applying it to the function f (u) = e2πiu gives

L(n) ∞ X
k
X X n
e2πiu = e2πij/k M ( ) (5.7)
v=1 k=1 j=1
k

Pk
Lemma 39. j=1 e2πij/k = 0 unless k = 1, in which case it equals 1.

40
 This simplifies the previous equation to:

L(n)
X
M (n) = e2πirv
v=1
L(n)
X
= e2πi[(v/L(n))+δv ]
v=1
L(n) L(n)
X X
2πiv/L(n) 2πiδv
= e (e − 1) + e2πiv/L(n)
v=1 v=1
L(n)
X
⇒ |M (n)| ≤ |(e2πiδv − 1)| + 0
v=1

Rearranging the right hand side gives:

L(n)
X
|M (n)| ≤ |eπiδv ||(eπiδv − e−πiδv )|
v=1
L(n)
X
= 2 |sin(πδv )|
v=1
L(n)
X
≤ 2π |δv |
v=1

Using 5.4 and letting 2π · K() = K 0 () we gain the inequality:

|M (n)| ≤ K 0 ()n1/2+

Therefore:
M (n) = o(n1/2+ )

∀ > 0

5.3 Backward Proof of Equivalent Statement

This time we want to show:

X
M (n) = o(n1/2+ ) ⇒ |δv | = o(n1/2+ )

41
Definition 40. Bernoulli Polynomials.
The nth Bernoulli polynomial Bn (u) satisfies the equation:
Z x+1
Bn (u)du = xn (5.8)
x

Bernoulli polynomials can be extended to be periodic with period 1, denoted here

as B̄n (u).

Example 41.

1
B1 (u) = u + (5.9)
2
1
B̄1 (u) = u − buc + (5.10)
2

Figure 5.1: Graph of B1 (u) Figure 5.2: Graph of B̄1 (u)

A property of Bernoulli polynomials that will be made use of later is the fol-
lowing:

1 k−1
Bn (ku) = k n−1 [Bn (u) + Bn (u + ) + ... + Bn (u + )] (5.11)
k k

It should be noted that this identity also holds for periodic Bernoulli polynomials.

Define the function G to be:

L(n)
X
G(u) = B̄1 (u + rv ) (5.12)
v=1

42
And I as: Z 1
I= [G(u)]2 du (5.13)
0

Using the form 5.6, it is possible to write G as:

∞ X
k
X j n
G(u) = B̄1 (u + )M ( )
k=1 j=1
k k

And then applying the identity 5.11:

∞
X n
G(u) = B̄1 (ku)M ( ) (5.14)
k=1
k

Which gives two expressions for G, and thus two ways to evaluate I.
First consider equation 5.12. This form shows that at every Farey fraction,
rv , the function jumps down by 1 and increases with a gradient of L(n) between
rv and rv+1 .
Between rv and rv+1 , G acts like the function G(u) = L(n)u − v − 1/2, as the
following graphs illustrate.

Figure 5.3: G(u) using elements of F3 Figure 5.4: G(u) using elements of F4

This gives us a new representation of I

L(n) Z rv
X 1
I = (L(n)u − v − )2 du
v=1 rv−1 2
L(n)
X 1 (L(n)u − v + 12 )3 rv
= |rv−1
v=1
L(n) 3

43
 Figure 5.5: G(u) using elements from F5

As the first term of every Farey sequence is 0, r0 = 0.

L(n)
X 1 (L(n)rv − v + 21 )3 1 (L(n)rv−1 − v + 21 )3
I= ( − )
v=1
L(n) 3 L(n) 3

v v
Substituting L(n)rv = L(n)[rv − L(n)
+ L(n)
] = L(n)δv + v:

L(n)
1 X 1 1
I = [(L(n)δv + )3 − (L(n)δv−1 − )3 ]
3L(n) v=1 2 2
L(n)
1 X 1 1
= [(L(n)δv + )3 − (L(n)δv − )3 ]
3L(n) v=1 2 2
L(n)
1 X 1 1
= [2 · 3(L(n)δv )2 · + ( )3 ]
3L(n) v=1 2 2

L(n)
X 1
= L(n) δv2 + (5.15)
v=1
12

Which provides us with an explicit formula for I.

Using G(u) from 5.14 and the fact that the sum is finite, we can evaluate I
as follows:
L(n) L(n) Z 1
XX n n
I= M ( )M ( ) B̄1 (au)B̄1 (bu)du (5.16)
a=1 b=1
a b 0

R1
Let Iab = 0
B̄1 (au)B̄1 (bu)du, which can be evaluated explicitly.
To find an explicit formula for Iab , it is essential to consider three cases.

1. b = 1

44
 2. b and a are coprime

3. a and b are not coprime

At first it may seem strange to consider these cases, but once the formula for
b = 1 is obtained, the other two soon follow, and give us the formula for any
values of a and b.

Case 1: b = 1

Z a
Ia1 = B̄1 (au)B̄1 (u)du (5.17)
0

Substituting v = au ⇒ du = a−1 dv
Z a
v
= B̄1 (v)B̄1 ( )a−1 dv
0 a

Substituting v = k + t ⇒ dv = dt

a−1 Z 1
−1
X k t
=a B̄1 (k + t)B̄1 ( + )dt
k=0 0 a a

Using the periodicity of B̄1 and 5.11.

Z 1
−1 t
= a B̄1 (t)B̄1 (a · )dt
a
Z0 1
1
= a−1 (t − )2 dt
0 2
−1
⇒ Ia1 = (12a)

Case 2: a and b are coprime.

Following the same steps as with Ia1 , the formula quickly becomes:
Z 1
−1 bt
Iab = a B̄1 (t)B̄1 (a · )dt
a
Z0 1
= a−1 B̄1 (t)B̄1 (bt)dt
0

45
Which is in the same form as 5.17, but with a = 1 instead. Therefore:

= a−1 I1b

= a−1 (12b)−1

= (12ab)−1

Case 3: a and b are not coprime. Let c = (a, b), the greatest common divisor or
a and b. Then a = αc and b = βc.
Z 1
Iab = B̄1 (αcu)B̄1 (βcu)du
0
Z c
−1
= c B̄1 (αt)B̄1 (βt)dt
0
= Iαβ

= (12αβ)−1

ab
Substituting αβ = c2
c2
=
12ab
Now we have an explicit evaluation of Iab for any values of a and b. If a and b
are coprime, we set c = 1, if they are not, we let c = gcd(a, b) as before.
This leads to the final equation for I:

∞ X
∞
X n n c2
I= M ( )M ( ) (5.18)
a=1 b=1
a b 12ab

Using M (n) = o(n1/2+ )∀ > 0, then ∃C such that M (n) < Cn(1/2)+ . Substitut-
ing this value for M (n) gives:

∞ X
∞
X n n c2
I < C( )(1/2)+ C( )(1/2)+
a=1 b=1
a b 12ab
∞ ∞
C2 X X c2
= n1+2
12 a=1 b=1 (αcβc)(3/2)+
2 ∞ X
∞ X
∞
1+2 C
X c2
= n
12 α=1 β=1 c=1
α3/2 β 3/2 c1+2

46
 C2
Letting K = 12
, taking our explicit formula for I from 5.15, and the fact that
the infinite sum above converges to 0:

L(n)
X
L(n) δv2 < Kn1+2
v=1

Taking the square root of each side gives:

L(n)
X
1/2
L(n) ( (δv2 ))1/2 < K 1/2 n1/2+
v=1

The final step is to show:

L(n) L(n)
X X
1/2
|δv | ≤ L(n) ( (δv2 ))1/2 (5.19)
v=1 v=1

This is done by applying the Cauchy-Schwarz inequality:

L(n) L(n)
X X
|δv | = | (±1)δv |
v=1 v=1
L(n)
X
1/2
≤ (1| + 1 +{z· · · + 1}) ( (δv )2 )1/2
L(n) times v=1

L(n)
X
1/2
= L(n) ( (δv )2 )1/2
v=1

Thus we obtain
L(n)
X
|δv | < K 1/2 n1/2+
v=1

for all  > 0 Therefore:

L(n)
X
|δv | = o(n1/2+ )
v=1

which concludes the proof of statement 5.5.[32]

47
Chapter 6

Appendix

Figure 6.1: Our Maple code

We based our code on JacquesC’s function [33] for a Farey sequence.

48
 49

Figure 6.2: Flitcon’s solution, 1751

Figure 6.3: Matlab code for approximation of irrationals using the Farey sequence.

50
Bibliography

[1] Ladies’ diary 1747 page 34, available on Google books: http:
//books.google.co.uk/books?id=HFpIAAAAYAAJ&lpg=RA2-PA46&ots=
eYrJcwCcbd&dq=ladies%20diary%201747&pg=RA1-PA34#v=onepage&q&f=
false Accessed on 12/3/12

[2] http://en.wikipedia.org/wiki/Carl_Friedrich_Gauss#Anecdotes Ac-

cessed 6/3/12

[3] Ladies’ diary 1751

[4] Scott B. Guthery, A Motif of Mathematics: History and Application of the

Mediant and the Farey Sequence (Docent Press, Boston, 2011)

[5] Tom M. Apostol Modular functions and Dirichlet series in number theory
Springer; 2nd edition (December 1, 1989)

[6] http://nrich.maths.org/6598 accessed on 7/03/12

[7] Augustin-Louis Cauchy, “Dmonstration d’un thorme curieux sur

les nombres” http://ebooks.cambridge.org/chapter.jsf?bid=
CBO9780511702518&cid=CBO9780511702518A033

[8] http://www.cut-the-knot.org/blue/Stern.shtml Accessed on

07/03/2012

[9] R. L. Graham, D. E. Knuth, and O. Patashnik, Concrete Mathematics, 2nd

ed., Addison-Wesley, Reading, MA, 1994.

[10] B. Bates, M. Bunder, and K. Tognetti, Locating terms in the Stern-Brocot

tree, European J. Combin. 31 (2010)

51
[11] http://www.ams.org/samplings/feature-column/fcarc-stern-brocot
Accessed on 07/03/2012

[12] Brian Hayes, On the Teeth of Wheels American Scientist, Vol. 88, No. 4
(July-August 2000).

[13] A. Brocot calcul des rouges par approximation, nouvelle methode (1861)

[14] Camus, Charles-thienne-Louis. 1842. A Treatise on the Teeth of Wheels,

Demonstrating the Best Forms Which Can Be Given to Them for the Pur-
poses of Machinery; Such as Mill-work and Clock-work, and the Art of Find-
ing Their Numbers. Translated from the French of M. Camus. A new edition
by John Isaac Hawkins, civil engineer. London

[15] Article by John D. Cook, http://www.johndcook.com/blog/2010/10/20/

best-rational-approximation/ Accessed 6/3/12

[16] The Mathematical Gazette, “Farey Rabbits” http://www.jstor.org/

stable/3621652?seq=1 Accessed 15/03/12

[17] NRich problem, http://nrich.maths.org/6599 Accessed 6/3/12

[18] “Fractions”, published in American Mathematical Monthly, volume 45, num-

ber 9, p586-601, November 1938. L. R. Ford

[19] Geometrische Deutung unendlicher Kettenbruche, J. Züllig, 1928

[20] http://www-history.mcs.st-andrews.ac.uk/Biographies/Ford.html
accessed 6/3/12

[21] http://www.cut-the-knot.org/proofs/fords.shtml accessed 29/2/12

[22] http://www.cut-the-knot.org/pythagoras/pythTriple.shtml Ac-

cessed 6/3/12

[23] Scott Carnahan, “Farey fractions, Ford circles, and SL2 .” Oc-
tober 2011. http://sbseminar.wordpress.com/2011/10/18/
farey-fractions-ford-circles-and-sl_2/#comment-17466 Accessed
on 29/2/12

52
[24] http://www.math.cornell.edu/~hatcher/TN/TNpage.html Topology of
Numbers, Allen Hatcher.

[25] Ian short, ”Ford circles, continued fractions, and best approximation of the
second kind”, Cornell Univeristy Library, December 2009.

[26] http://en.wikipedia.org/wiki/Sangaku#Select_examples Accessed

6/3/12

[27] http://www.claymath.org/millennium/Riemann_Hypothesis/ Accessed

14/3/12

[28] http://www.bernoulli.org/ Accessed 14/3/12

[29] Richard Courant. ”Differential and Integral Calculus (Volume 1)” (June 17,
1992)

[30] A. M. Odlyzko & H. J. J. te Riele, ”Disproof of Mertens Conjecture”,http://

www.dtc.umn.edu/~odlyzko/doc/arch/mertens.disproof.pdf Accessed
on 14/3/12

[31] http://www.cut-the-knot.org/arithmetic/algebra/SumOfVertices.
shtml Accessed 14/3/12

[32] H. M. Edwards, ”Riemann’s Zeta Function”, Academic Press, 1974

[33] Jacques C, http://www.mapleprimes.com/questions/

39552-Sequence-Procedure Accessed 6/03/12

53